diff --git a/comfy/conds.py b/comfy/conds.py index 1e3111baf..6cff25184 100644 --- a/comfy/conds.py +++ b/comfy/conds.py @@ -62,3 +62,18 @@ class CONDCrossAttn(CONDRegular): c = c.repeat(1, crossattn_max_len // c.shape[1], 1) #padding with repeat doesn't change result out.append(c) return torch.cat(out) + +class CONDConstant(CONDRegular): + def __init__(self, cond): + self.cond = cond + + def process_cond(self, batch_size, device, **kwargs): + return self._copy_with(self.cond) + + def can_concat(self, other): + if self.cond != other.cond: + return False + return True + + def concat(self, others): + return self.cond diff --git a/comfy/controlnet.py b/comfy/controlnet.py index 2a88dd019..098681582 100644 --- a/comfy/controlnet.py +++ b/comfy/controlnet.py @@ -132,6 +132,7 @@ class ControlNet(ControlBase): self.control_model = control_model self.control_model_wrapped = comfy.model_patcher.ModelPatcher(self.control_model, load_device=comfy.model_management.get_torch_device(), offload_device=comfy.model_management.unet_offload_device()) self.global_average_pooling = global_average_pooling + self.model_sampling_current = None def get_control(self, x_noisy, t, cond, batched_number): control_prev = None @@ -159,7 +160,10 @@ class ControlNet(ControlBase): y = cond.get('y', None) if y is not None: y = y.to(self.control_model.dtype) - control = self.control_model(x=x_noisy.to(self.control_model.dtype), hint=self.cond_hint, timesteps=t, context=context.to(self.control_model.dtype), y=y) + timestep = self.model_sampling_current.timestep(t) + x_noisy = self.model_sampling_current.calculate_input(t, x_noisy) + + control = self.control_model(x=x_noisy.to(self.control_model.dtype), hint=self.cond_hint, timesteps=timestep.float(), context=context.to(self.control_model.dtype), y=y) return self.control_merge(None, control, control_prev, output_dtype) def copy(self): @@ -172,6 +176,14 @@ class ControlNet(ControlBase): out.append(self.control_model_wrapped) return out + def pre_run(self, model, percent_to_timestep_function): + super().pre_run(model, percent_to_timestep_function) + self.model_sampling_current = model.model_sampling + + def cleanup(self): + self.model_sampling_current = None + super().cleanup() + class ControlLoraOps: class Linear(torch.nn.Module): def __init__(self, in_features: int, out_features: int, bias: bool = True, diff --git a/comfy/extra_samplers/uni_pc.py b/comfy/extra_samplers/uni_pc.py index 9d5f0c60b..1a7a83929 100644 --- a/comfy/extra_samplers/uni_pc.py +++ b/comfy/extra_samplers/uni_pc.py @@ -852,6 +852,12 @@ class SigmaConvert: log_std = 0.5 * torch.log(1. - torch.exp(2. * log_mean_coeff)) return log_mean_coeff - log_std +def predict_eps_sigma(model, input, sigma_in, **kwargs): + sigma = sigma_in.view(sigma_in.shape[:1] + (1,) * (input.ndim - 1)) + input = input * ((sigma ** 2 + 1.0) ** 0.5) + return (input - model(input, sigma_in, **kwargs)) / sigma + + def sample_unipc(model, noise, image, sigmas, sampling_function, max_denoise, extra_args=None, callback=None, disable=False, noise_mask=None, variant='bh1'): timesteps = sigmas.clone() if sigmas[-1] == 0: @@ -874,7 +880,7 @@ def sample_unipc(model, noise, image, sigmas, sampling_function, max_denoise, ex model_type = "noise" model_fn = model_wrapper( - model.predict_eps_sigma, + lambda input, sigma, **kwargs: predict_eps_sigma(model, input, sigma, **kwargs), ns, model_type=model_type, guidance_type="uncond", diff --git a/comfy/k_diffusion/external.py b/comfy/k_diffusion/external.py deleted file mode 100644 index 953d3db2c..000000000 --- a/comfy/k_diffusion/external.py +++ /dev/null @@ -1,194 +0,0 @@ -import math - -import torch -from torch import nn - -from . import sampling, utils - - -class VDenoiser(nn.Module): - """A v-diffusion-pytorch model wrapper for k-diffusion.""" - - def __init__(self, inner_model): - super().__init__() - self.inner_model = inner_model - self.sigma_data = 1. - - def get_scalings(self, sigma): - c_skip = self.sigma_data ** 2 / (sigma ** 2 + self.sigma_data ** 2) - c_out = -sigma * self.sigma_data / (sigma ** 2 + self.sigma_data ** 2) ** 0.5 - c_in = 1 / (sigma ** 2 + self.sigma_data ** 2) ** 0.5 - return c_skip, c_out, c_in - - def sigma_to_t(self, sigma): - return sigma.atan() / math.pi * 2 - - def t_to_sigma(self, t): - return (t * math.pi / 2).tan() - - def loss(self, input, noise, sigma, **kwargs): - c_skip, c_out, c_in = [utils.append_dims(x, input.ndim) for x in self.get_scalings(sigma)] - noised_input = input + noise * utils.append_dims(sigma, input.ndim) - model_output = self.inner_model(noised_input * c_in, self.sigma_to_t(sigma), **kwargs) - target = (input - c_skip * noised_input) / c_out - return (model_output - target).pow(2).flatten(1).mean(1) - - def forward(self, input, sigma, **kwargs): - c_skip, c_out, c_in = [utils.append_dims(x, input.ndim) for x in self.get_scalings(sigma)] - return self.inner_model(input * c_in, self.sigma_to_t(sigma), **kwargs) * c_out + input * c_skip - - -class DiscreteSchedule(nn.Module): - """A mapping between continuous noise levels (sigmas) and a list of discrete noise - levels.""" - - def __init__(self, sigmas, quantize): - super().__init__() - self.register_buffer('sigmas', sigmas) - self.register_buffer('log_sigmas', sigmas.log()) - self.quantize = quantize - - @property - def sigma_min(self): - return self.sigmas[0] - - @property - def sigma_max(self): - return self.sigmas[-1] - - def get_sigmas(self, n=None): - if n is None: - return sampling.append_zero(self.sigmas.flip(0)) - t_max = len(self.sigmas) - 1 - t = torch.linspace(t_max, 0, n, device=self.sigmas.device) - return sampling.append_zero(self.t_to_sigma(t)) - - def sigma_to_discrete_timestep(self, sigma): - log_sigma = sigma.log() - dists = log_sigma.to(self.log_sigmas.device) - self.log_sigmas[:, None] - return dists.abs().argmin(dim=0).view(sigma.shape) - - def sigma_to_t(self, sigma, quantize=None): - quantize = self.quantize if quantize is None else quantize - if quantize: - return self.sigma_to_discrete_timestep(sigma) - log_sigma = sigma.log() - dists = log_sigma.to(self.log_sigmas.device) - self.log_sigmas[:, None] - low_idx = dists.ge(0).cumsum(dim=0).argmax(dim=0).clamp(max=self.log_sigmas.shape[0] - 2) - high_idx = low_idx + 1 - low, high = self.log_sigmas[low_idx], self.log_sigmas[high_idx] - w = (low - log_sigma) / (low - high) - w = w.clamp(0, 1) - t = (1 - w) * low_idx + w * high_idx - return t.view(sigma.shape) - - def t_to_sigma(self, t): - t = t.float() - low_idx = t.floor().long() - high_idx = t.ceil().long() - w = t-low_idx if t.device.type == 'mps' else t.frac() - log_sigma = (1 - w) * self.log_sigmas[low_idx] + w * self.log_sigmas[high_idx] - return log_sigma.exp() - - def predict_eps_discrete_timestep(self, input, t, **kwargs): - if t.dtype != torch.int64 and t.dtype != torch.int32: - t = t.round() - sigma = self.t_to_sigma(t) - input = input * ((utils.append_dims(sigma, input.ndim) ** 2 + 1.0) ** 0.5) - return (input - self(input, sigma, **kwargs)) / utils.append_dims(sigma, input.ndim) - - def predict_eps_sigma(self, input, sigma, **kwargs): - input = input * ((utils.append_dims(sigma, input.ndim) ** 2 + 1.0) ** 0.5) - return (input - self(input, sigma, **kwargs)) / utils.append_dims(sigma, input.ndim) - -class DiscreteEpsDDPMDenoiser(DiscreteSchedule): - """A wrapper for discrete schedule DDPM models that output eps (the predicted - noise).""" - - def __init__(self, model, alphas_cumprod, quantize): - super().__init__(((1 - alphas_cumprod) / alphas_cumprod) ** 0.5, quantize) - self.inner_model = model - self.sigma_data = 1. - - def get_scalings(self, sigma): - c_out = -sigma - c_in = 1 / (sigma ** 2 + self.sigma_data ** 2) ** 0.5 - return c_out, c_in - - def get_eps(self, *args, **kwargs): - return self.inner_model(*args, **kwargs) - - def loss(self, input, noise, sigma, **kwargs): - c_out, c_in = [utils.append_dims(x, input.ndim) for x in self.get_scalings(sigma)] - noised_input = input + noise * utils.append_dims(sigma, input.ndim) - eps = self.get_eps(noised_input * c_in, self.sigma_to_t(sigma), **kwargs) - return (eps - noise).pow(2).flatten(1).mean(1) - - def forward(self, input, sigma, **kwargs): - c_out, c_in = [utils.append_dims(x, input.ndim) for x in self.get_scalings(sigma)] - eps = self.get_eps(input * c_in, self.sigma_to_t(sigma), **kwargs) - return input + eps * c_out - - -class OpenAIDenoiser(DiscreteEpsDDPMDenoiser): - """A wrapper for OpenAI diffusion models.""" - - def __init__(self, model, diffusion, quantize=False, has_learned_sigmas=True, device='cpu'): - alphas_cumprod = torch.tensor(diffusion.alphas_cumprod, device=device, dtype=torch.float32) - super().__init__(model, alphas_cumprod, quantize=quantize) - self.has_learned_sigmas = has_learned_sigmas - - def get_eps(self, *args, **kwargs): - model_output = self.inner_model(*args, **kwargs) - if self.has_learned_sigmas: - return model_output.chunk(2, dim=1)[0] - return model_output - - -class CompVisDenoiser(DiscreteEpsDDPMDenoiser): - """A wrapper for CompVis diffusion models.""" - - def __init__(self, model, quantize=False, device='cpu'): - super().__init__(model, model.alphas_cumprod, quantize=quantize) - - def get_eps(self, *args, **kwargs): - return self.inner_model.apply_model(*args, **kwargs) - - -class DiscreteVDDPMDenoiser(DiscreteSchedule): - """A wrapper for discrete schedule DDPM models that output v.""" - - def __init__(self, model, alphas_cumprod, quantize): - super().__init__(((1 - alphas_cumprod) / alphas_cumprod) ** 0.5, quantize) - self.inner_model = model - self.sigma_data = 1. - - def get_scalings(self, sigma): - c_skip = self.sigma_data ** 2 / (sigma ** 2 + self.sigma_data ** 2) - c_out = -sigma * self.sigma_data / (sigma ** 2 + self.sigma_data ** 2) ** 0.5 - c_in = 1 / (sigma ** 2 + self.sigma_data ** 2) ** 0.5 - return c_skip, c_out, c_in - - def get_v(self, *args, **kwargs): - return self.inner_model(*args, **kwargs) - - def loss(self, input, noise, sigma, **kwargs): - c_skip, c_out, c_in = [utils.append_dims(x, input.ndim) for x in self.get_scalings(sigma)] - noised_input = input + noise * utils.append_dims(sigma, input.ndim) - model_output = self.get_v(noised_input * c_in, self.sigma_to_t(sigma), **kwargs) - target = (input - c_skip * noised_input) / c_out - return (model_output - target).pow(2).flatten(1).mean(1) - - def forward(self, input, sigma, **kwargs): - c_skip, c_out, c_in = [utils.append_dims(x, input.ndim) for x in self.get_scalings(sigma)] - return self.get_v(input * c_in, self.sigma_to_t(sigma), **kwargs) * c_out + input * c_skip - - -class CompVisVDenoiser(DiscreteVDDPMDenoiser): - """A wrapper for CompVis diffusion models that output v.""" - - def __init__(self, model, quantize=False, device='cpu'): - super().__init__(model, model.alphas_cumprod, quantize=quantize) - - def get_v(self, x, t, cond, **kwargs): - return self.inner_model.apply_model(x, t, cond) diff --git a/comfy/k_diffusion/sampling.py b/comfy/k_diffusion/sampling.py index 937c5a388..dd6f7bbe5 100644 --- a/comfy/k_diffusion/sampling.py +++ b/comfy/k_diffusion/sampling.py @@ -717,7 +717,6 @@ def DDPMSampler_step(x, sigma, sigma_prev, noise, noise_sampler): mu += ((1 - alpha) * (1. - alpha_cumprod_prev) / (1. - alpha_cumprod)).sqrt() * noise_sampler(sigma, sigma_prev) return mu - def generic_step_sampler(model, x, sigmas, extra_args=None, callback=None, disable=None, noise_sampler=None, step_function=None): extra_args = {} if extra_args is None else extra_args noise_sampler = default_noise_sampler(x) if noise_sampler is None else noise_sampler @@ -737,3 +736,17 @@ def generic_step_sampler(model, x, sigmas, extra_args=None, callback=None, disab def sample_ddpm(model, x, sigmas, extra_args=None, callback=None, disable=None, noise_sampler=None): return generic_step_sampler(model, x, sigmas, extra_args, callback, disable, noise_sampler, DDPMSampler_step) +@torch.no_grad() +def sample_lcm(model, x, sigmas, extra_args=None, callback=None, disable=None, noise_sampler=None): + extra_args = {} if extra_args is None else extra_args + noise_sampler = default_noise_sampler(x) if noise_sampler is None else noise_sampler + s_in = x.new_ones([x.shape[0]]) + for i in trange(len(sigmas) - 1, disable=disable): + denoised = model(x, sigmas[i] * s_in, **extra_args) + if callback is not None: + callback({'x': x, 'i': i, 'sigma': sigmas[i], 'sigma_hat': sigmas[i], 'denoised': denoised}) + + x = denoised + if sigmas[i + 1] > 0: + x += sigmas[i + 1] * noise_sampler(sigmas[i], sigmas[i + 1]) + return x diff --git a/comfy/ldm/models/diffusion/__init__.py b/comfy/ldm/models/diffusion/__init__.py deleted file mode 100644 index e69de29bb..000000000 diff --git a/comfy/ldm/models/diffusion/ddim.py b/comfy/ldm/models/diffusion/ddim.py deleted file mode 100644 index 433d48e30..000000000 --- a/comfy/ldm/models/diffusion/ddim.py +++ /dev/null @@ -1,418 +0,0 @@ -"""SAMPLING ONLY.""" - -import torch -import numpy as np -from tqdm import tqdm - -from comfy.ldm.modules.diffusionmodules.util import make_ddim_sampling_parameters, make_ddim_timesteps, noise_like, extract_into_tensor - - -class DDIMSampler(object): - def __init__(self, model, schedule="linear", device=torch.device("cuda"), **kwargs): - super().__init__() - self.model = model - self.ddpm_num_timesteps = model.num_timesteps - self.schedule = schedule - self.device = device - self.parameterization = kwargs.get("parameterization", "eps") - - def register_buffer(self, name, attr): - if type(attr) == torch.Tensor: - if attr.device != self.device: - attr = attr.float().to(self.device) - setattr(self, name, attr) - - def make_schedule(self, ddim_num_steps, ddim_discretize="uniform", ddim_eta=0., verbose=True): - ddim_timesteps = make_ddim_timesteps(ddim_discr_method=ddim_discretize, num_ddim_timesteps=ddim_num_steps, - num_ddpm_timesteps=self.ddpm_num_timesteps,verbose=verbose) - self.make_schedule_timesteps(ddim_timesteps, ddim_eta=ddim_eta, verbose=verbose) - - def make_schedule_timesteps(self, ddim_timesteps, ddim_eta=0., verbose=True): - self.ddim_timesteps = torch.tensor(ddim_timesteps) - alphas_cumprod = self.model.alphas_cumprod - assert alphas_cumprod.shape[0] == self.ddpm_num_timesteps, 'alphas have to be defined for each timestep' - to_torch = lambda x: x.clone().detach().to(torch.float32).to(self.device) - - self.register_buffer('alphas_cumprod', to_torch(alphas_cumprod)) - self.register_buffer('alphas_cumprod_prev', to_torch(self.model.alphas_cumprod_prev)) - - # calculations for diffusion q(x_t | x_{t-1}) and others - self.register_buffer('sqrt_alphas_cumprod', to_torch(np.sqrt(alphas_cumprod.cpu()))) - self.register_buffer('sqrt_one_minus_alphas_cumprod', to_torch(np.sqrt(1. - alphas_cumprod.cpu()))) - self.register_buffer('log_one_minus_alphas_cumprod', to_torch(np.log(1. - alphas_cumprod.cpu()))) - self.register_buffer('sqrt_recip_alphas_cumprod', to_torch(np.sqrt(1. / alphas_cumprod.cpu()))) - self.register_buffer('sqrt_recipm1_alphas_cumprod', to_torch(np.sqrt(1. / alphas_cumprod.cpu() - 1))) - - # ddim sampling parameters - ddim_sigmas, ddim_alphas, ddim_alphas_prev = make_ddim_sampling_parameters(alphacums=alphas_cumprod.cpu(), - ddim_timesteps=self.ddim_timesteps, - eta=ddim_eta,verbose=verbose) - self.register_buffer('ddim_sigmas', ddim_sigmas) - self.register_buffer('ddim_alphas', ddim_alphas) - self.register_buffer('ddim_alphas_prev', ddim_alphas_prev) - self.register_buffer('ddim_sqrt_one_minus_alphas', np.sqrt(1. - ddim_alphas)) - sigmas_for_original_sampling_steps = ddim_eta * torch.sqrt( - (1 - self.alphas_cumprod_prev) / (1 - self.alphas_cumprod) * ( - 1 - self.alphas_cumprod / self.alphas_cumprod_prev)) - self.register_buffer('ddim_sigmas_for_original_num_steps', sigmas_for_original_sampling_steps) - - @torch.no_grad() - def sample_custom(self, - ddim_timesteps, - conditioning=None, - callback=None, - img_callback=None, - quantize_x0=False, - eta=0., - mask=None, - x0=None, - temperature=1., - noise_dropout=0., - score_corrector=None, - corrector_kwargs=None, - verbose=True, - x_T=None, - log_every_t=100, - unconditional_guidance_scale=1., - unconditional_conditioning=None, # this has to come in the same format as the conditioning, # e.g. as encoded tokens, ... - dynamic_threshold=None, - ucg_schedule=None, - denoise_function=None, - extra_args=None, - to_zero=True, - end_step=None, - disable_pbar=False, - **kwargs - ): - self.make_schedule_timesteps(ddim_timesteps=ddim_timesteps, ddim_eta=eta, verbose=verbose) - samples, intermediates = self.ddim_sampling(conditioning, x_T.shape, - callback=callback, - img_callback=img_callback, - quantize_denoised=quantize_x0, - mask=mask, x0=x0, - ddim_use_original_steps=False, - noise_dropout=noise_dropout, - temperature=temperature, - score_corrector=score_corrector, - corrector_kwargs=corrector_kwargs, - x_T=x_T, - log_every_t=log_every_t, - unconditional_guidance_scale=unconditional_guidance_scale, - unconditional_conditioning=unconditional_conditioning, - dynamic_threshold=dynamic_threshold, - ucg_schedule=ucg_schedule, - denoise_function=denoise_function, - extra_args=extra_args, - to_zero=to_zero, - end_step=end_step, - disable_pbar=disable_pbar - ) - return samples, intermediates - - - @torch.no_grad() - def sample(self, - S, - batch_size, - shape, - conditioning=None, - callback=None, - normals_sequence=None, - img_callback=None, - quantize_x0=False, - eta=0., - mask=None, - x0=None, - temperature=1., - noise_dropout=0., - score_corrector=None, - corrector_kwargs=None, - verbose=True, - x_T=None, - log_every_t=100, - unconditional_guidance_scale=1., - unconditional_conditioning=None, # this has to come in the same format as the conditioning, # e.g. as encoded tokens, ... - dynamic_threshold=None, - ucg_schedule=None, - **kwargs - ): - if conditioning is not None: - if isinstance(conditioning, dict): - ctmp = conditioning[list(conditioning.keys())[0]] - while isinstance(ctmp, list): ctmp = ctmp[0] - cbs = ctmp.shape[0] - if cbs != batch_size: - print(f"Warning: Got {cbs} conditionings but batch-size is {batch_size}") - - elif isinstance(conditioning, list): - for ctmp in conditioning: - if ctmp.shape[0] != batch_size: - print(f"Warning: Got {cbs} conditionings but batch-size is {batch_size}") - - else: - if conditioning.shape[0] != batch_size: - print(f"Warning: Got {conditioning.shape[0]} conditionings but batch-size is {batch_size}") - - self.make_schedule(ddim_num_steps=S, ddim_eta=eta, verbose=verbose) - # sampling - C, H, W = shape - size = (batch_size, C, H, W) - print(f'Data shape for DDIM sampling is {size}, eta {eta}') - - samples, intermediates = self.ddim_sampling(conditioning, size, - callback=callback, - img_callback=img_callback, - quantize_denoised=quantize_x0, - mask=mask, x0=x0, - ddim_use_original_steps=False, - noise_dropout=noise_dropout, - temperature=temperature, - score_corrector=score_corrector, - corrector_kwargs=corrector_kwargs, - x_T=x_T, - log_every_t=log_every_t, - unconditional_guidance_scale=unconditional_guidance_scale, - unconditional_conditioning=unconditional_conditioning, - dynamic_threshold=dynamic_threshold, - ucg_schedule=ucg_schedule, - denoise_function=None, - extra_args=None - ) - return samples, intermediates - - def q_sample(self, x_start, t, noise=None): - if noise is None: - noise = torch.randn_like(x_start) - return (extract_into_tensor(self.sqrt_alphas_cumprod, t, x_start.shape) * x_start + - extract_into_tensor(self.sqrt_one_minus_alphas_cumprod, t, x_start.shape) * noise) - - @torch.no_grad() - def ddim_sampling(self, cond, shape, - x_T=None, ddim_use_original_steps=False, - callback=None, timesteps=None, quantize_denoised=False, - mask=None, x0=None, img_callback=None, log_every_t=100, - temperature=1., noise_dropout=0., score_corrector=None, corrector_kwargs=None, - unconditional_guidance_scale=1., unconditional_conditioning=None, dynamic_threshold=None, - ucg_schedule=None, denoise_function=None, extra_args=None, to_zero=True, end_step=None, disable_pbar=False): - device = self.model.alphas_cumprod.device - b = shape[0] - if x_T is None: - img = torch.randn(shape, device=device) - else: - img = x_T - - if timesteps is None: - timesteps = self.ddpm_num_timesteps if ddim_use_original_steps else self.ddim_timesteps - elif timesteps is not None and not ddim_use_original_steps: - subset_end = int(min(timesteps / self.ddim_timesteps.shape[0], 1) * self.ddim_timesteps.shape[0]) - 1 - timesteps = self.ddim_timesteps[:subset_end] - - intermediates = {'x_inter': [img], 'pred_x0': [img]} - time_range = reversed(range(0,timesteps)) if ddim_use_original_steps else timesteps.flip(0) - total_steps = timesteps if ddim_use_original_steps else timesteps.shape[0] - # print(f"Running DDIM Sampling with {total_steps} timesteps") - - iterator = tqdm(time_range[:end_step], desc='DDIM Sampler', total=end_step, disable=disable_pbar) - - for i, step in enumerate(iterator): - index = total_steps - i - 1 - ts = torch.full((b,), step, device=device, dtype=torch.long) - - if mask is not None: - assert x0 is not None - img_orig = self.q_sample(x0, ts) # TODO: deterministic forward pass? - img = img_orig * mask + (1. - mask) * img - - if ucg_schedule is not None: - assert len(ucg_schedule) == len(time_range) - unconditional_guidance_scale = ucg_schedule[i] - - outs = self.p_sample_ddim(img, cond, ts, index=index, use_original_steps=ddim_use_original_steps, - quantize_denoised=quantize_denoised, temperature=temperature, - noise_dropout=noise_dropout, score_corrector=score_corrector, - corrector_kwargs=corrector_kwargs, - unconditional_guidance_scale=unconditional_guidance_scale, - unconditional_conditioning=unconditional_conditioning, - dynamic_threshold=dynamic_threshold, denoise_function=denoise_function, extra_args=extra_args) - img, pred_x0 = outs - if callback: callback(i) - if img_callback: img_callback(pred_x0, i) - - if index % log_every_t == 0 or index == total_steps - 1: - intermediates['x_inter'].append(img) - intermediates['pred_x0'].append(pred_x0) - - if to_zero: - img = pred_x0 - else: - if ddim_use_original_steps: - sqrt_alphas_cumprod = self.sqrt_alphas_cumprod - else: - sqrt_alphas_cumprod = torch.sqrt(self.ddim_alphas) - img /= sqrt_alphas_cumprod[index - 1] - - return img, intermediates - - @torch.no_grad() - def p_sample_ddim(self, x, c, t, index, repeat_noise=False, use_original_steps=False, quantize_denoised=False, - temperature=1., noise_dropout=0., score_corrector=None, corrector_kwargs=None, - unconditional_guidance_scale=1., unconditional_conditioning=None, - dynamic_threshold=None, denoise_function=None, extra_args=None): - b, *_, device = *x.shape, x.device - - if denoise_function is not None: - model_output = denoise_function(x, t, **extra_args) - elif unconditional_conditioning is None or unconditional_guidance_scale == 1.: - model_output = self.model.apply_model(x, t, c) - else: - x_in = torch.cat([x] * 2) - t_in = torch.cat([t] * 2) - if isinstance(c, dict): - assert isinstance(unconditional_conditioning, dict) - c_in = dict() - for k in c: - if isinstance(c[k], list): - c_in[k] = [torch.cat([ - unconditional_conditioning[k][i], - c[k][i]]) for i in range(len(c[k]))] - else: - c_in[k] = torch.cat([ - unconditional_conditioning[k], - c[k]]) - elif isinstance(c, list): - c_in = list() - assert isinstance(unconditional_conditioning, list) - for i in range(len(c)): - c_in.append(torch.cat([unconditional_conditioning[i], c[i]])) - else: - c_in = torch.cat([unconditional_conditioning, c]) - model_uncond, model_t = self.model.apply_model(x_in, t_in, c_in).chunk(2) - model_output = model_uncond + unconditional_guidance_scale * (model_t - model_uncond) - - if self.parameterization == "v": - e_t = extract_into_tensor(self.sqrt_alphas_cumprod, t, x.shape) * model_output + extract_into_tensor(self.sqrt_one_minus_alphas_cumprod, t, x.shape) * x - else: - e_t = model_output - - if score_corrector is not None: - assert self.parameterization == "eps", 'not implemented' - e_t = score_corrector.modify_score(self.model, e_t, x, t, c, **corrector_kwargs) - - alphas = self.model.alphas_cumprod if use_original_steps else self.ddim_alphas - alphas_prev = self.model.alphas_cumprod_prev if use_original_steps else self.ddim_alphas_prev - sqrt_one_minus_alphas = self.model.sqrt_one_minus_alphas_cumprod if use_original_steps else self.ddim_sqrt_one_minus_alphas - sigmas = self.model.ddim_sigmas_for_original_num_steps if use_original_steps else self.ddim_sigmas - # select parameters corresponding to the currently considered timestep - a_t = torch.full((b, 1, 1, 1), alphas[index], device=device) - a_prev = torch.full((b, 1, 1, 1), alphas_prev[index], device=device) - sigma_t = torch.full((b, 1, 1, 1), sigmas[index], device=device) - sqrt_one_minus_at = torch.full((b, 1, 1, 1), sqrt_one_minus_alphas[index],device=device) - - # current prediction for x_0 - if self.parameterization != "v": - pred_x0 = (x - sqrt_one_minus_at * e_t) / a_t.sqrt() - else: - pred_x0 = extract_into_tensor(self.sqrt_alphas_cumprod, t, x.shape) * x - extract_into_tensor(self.sqrt_one_minus_alphas_cumprod, t, x.shape) * model_output - - if quantize_denoised: - pred_x0, _, *_ = self.model.first_stage_model.quantize(pred_x0) - - if dynamic_threshold is not None: - raise NotImplementedError() - - # direction pointing to x_t - dir_xt = (1. - a_prev - sigma_t**2).sqrt() * e_t - noise = sigma_t * noise_like(x.shape, device, repeat_noise) * temperature - if noise_dropout > 0.: - noise = torch.nn.functional.dropout(noise, p=noise_dropout) - x_prev = a_prev.sqrt() * pred_x0 + dir_xt + noise - return x_prev, pred_x0 - - @torch.no_grad() - def encode(self, x0, c, t_enc, use_original_steps=False, return_intermediates=None, - unconditional_guidance_scale=1.0, unconditional_conditioning=None, callback=None): - num_reference_steps = self.ddpm_num_timesteps if use_original_steps else self.ddim_timesteps.shape[0] - - assert t_enc <= num_reference_steps - num_steps = t_enc - - if use_original_steps: - alphas_next = self.alphas_cumprod[:num_steps] - alphas = self.alphas_cumprod_prev[:num_steps] - else: - alphas_next = self.ddim_alphas[:num_steps] - alphas = torch.tensor(self.ddim_alphas_prev[:num_steps]) - - x_next = x0 - intermediates = [] - inter_steps = [] - for i in tqdm(range(num_steps), desc='Encoding Image'): - t = torch.full((x0.shape[0],), i, device=self.model.device, dtype=torch.long) - if unconditional_guidance_scale == 1.: - noise_pred = self.model.apply_model(x_next, t, c) - else: - assert unconditional_conditioning is not None - e_t_uncond, noise_pred = torch.chunk( - self.model.apply_model(torch.cat((x_next, x_next)), torch.cat((t, t)), - torch.cat((unconditional_conditioning, c))), 2) - noise_pred = e_t_uncond + unconditional_guidance_scale * (noise_pred - e_t_uncond) - - xt_weighted = (alphas_next[i] / alphas[i]).sqrt() * x_next - weighted_noise_pred = alphas_next[i].sqrt() * ( - (1 / alphas_next[i] - 1).sqrt() - (1 / alphas[i] - 1).sqrt()) * noise_pred - x_next = xt_weighted + weighted_noise_pred - if return_intermediates and i % ( - num_steps // return_intermediates) == 0 and i < num_steps - 1: - intermediates.append(x_next) - inter_steps.append(i) - elif return_intermediates and i >= num_steps - 2: - intermediates.append(x_next) - inter_steps.append(i) - if callback: callback(i) - - out = {'x_encoded': x_next, 'intermediate_steps': inter_steps} - if return_intermediates: - out.update({'intermediates': intermediates}) - return x_next, out - - @torch.no_grad() - def stochastic_encode(self, x0, t, use_original_steps=False, noise=None, max_denoise=False): - # fast, but does not allow for exact reconstruction - # t serves as an index to gather the correct alphas - if use_original_steps: - sqrt_alphas_cumprod = self.sqrt_alphas_cumprod - sqrt_one_minus_alphas_cumprod = self.sqrt_one_minus_alphas_cumprod - else: - sqrt_alphas_cumprod = torch.sqrt(self.ddim_alphas) - sqrt_one_minus_alphas_cumprod = self.ddim_sqrt_one_minus_alphas - - if noise is None: - noise = torch.randn_like(x0) - if max_denoise: - noise_multiplier = 1.0 - else: - noise_multiplier = extract_into_tensor(sqrt_one_minus_alphas_cumprod, t, x0.shape) - - return (extract_into_tensor(sqrt_alphas_cumprod, t, x0.shape) * x0 + noise_multiplier * noise) - - @torch.no_grad() - def decode(self, x_latent, cond, t_start, unconditional_guidance_scale=1.0, unconditional_conditioning=None, - use_original_steps=False, callback=None): - - timesteps = np.arange(self.ddpm_num_timesteps) if use_original_steps else self.ddim_timesteps - timesteps = timesteps[:t_start] - - time_range = np.flip(timesteps) - total_steps = timesteps.shape[0] - print(f"Running DDIM Sampling with {total_steps} timesteps") - - iterator = tqdm(time_range, desc='Decoding image', total=total_steps) - x_dec = x_latent - for i, step in enumerate(iterator): - index = total_steps - i - 1 - ts = torch.full((x_latent.shape[0],), step, device=x_latent.device, dtype=torch.long) - x_dec, _ = self.p_sample_ddim(x_dec, cond, ts, index=index, use_original_steps=use_original_steps, - unconditional_guidance_scale=unconditional_guidance_scale, - unconditional_conditioning=unconditional_conditioning) - if callback: callback(i) - return x_dec \ No newline at end of file diff --git a/comfy/ldm/models/diffusion/dpm_solver/__init__.py b/comfy/ldm/models/diffusion/dpm_solver/__init__.py deleted file mode 100644 index 7427f38c0..000000000 --- a/comfy/ldm/models/diffusion/dpm_solver/__init__.py +++ /dev/null @@ -1 +0,0 @@ -from .sampler import DPMSolverSampler \ No newline at end of file diff --git a/comfy/ldm/models/diffusion/dpm_solver/dpm_solver.py b/comfy/ldm/models/diffusion/dpm_solver/dpm_solver.py deleted file mode 100644 index da8d41f9c..000000000 --- a/comfy/ldm/models/diffusion/dpm_solver/dpm_solver.py +++ /dev/null @@ -1,1163 +0,0 @@ -import torch -import torch.nn.functional as F -import math -from tqdm import tqdm - - -class NoiseScheduleVP: - def __init__( - self, - schedule='discrete', - betas=None, - alphas_cumprod=None, - continuous_beta_0=0.1, - continuous_beta_1=20., - ): - """Create a wrapper class for the forward SDE (VP type). - *** - Update: We support discrete-time diffusion models by implementing a picewise linear interpolation for log_alpha_t. - We recommend to use schedule='discrete' for the discrete-time diffusion models, especially for high-resolution images. - *** - The forward SDE ensures that the condition distribution q_{t|0}(x_t | x_0) = N ( alpha_t * x_0, sigma_t^2 * I ). - We further define lambda_t = log(alpha_t) - log(sigma_t), which is the half-logSNR (described in the DPM-Solver paper). - Therefore, we implement the functions for computing alpha_t, sigma_t and lambda_t. For t in [0, T], we have: - log_alpha_t = self.marginal_log_mean_coeff(t) - sigma_t = self.marginal_std(t) - lambda_t = self.marginal_lambda(t) - Moreover, as lambda(t) is an invertible function, we also support its inverse function: - t = self.inverse_lambda(lambda_t) - =============================================================== - We support both discrete-time DPMs (trained on n = 0, 1, ..., N-1) and continuous-time DPMs (trained on t in [t_0, T]). - 1. For discrete-time DPMs: - For discrete-time DPMs trained on n = 0, 1, ..., N-1, we convert the discrete steps to continuous time steps by: - t_i = (i + 1) / N - e.g. for N = 1000, we have t_0 = 1e-3 and T = t_{N-1} = 1. - We solve the corresponding diffusion ODE from time T = 1 to time t_0 = 1e-3. - Args: - betas: A `torch.Tensor`. The beta array for the discrete-time DPM. (See the original DDPM paper for details) - alphas_cumprod: A `torch.Tensor`. The cumprod alphas for the discrete-time DPM. (See the original DDPM paper for details) - Note that we always have alphas_cumprod = cumprod(betas). Therefore, we only need to set one of `betas` and `alphas_cumprod`. - **Important**: Please pay special attention for the args for `alphas_cumprod`: - The `alphas_cumprod` is the \hat{alpha_n} arrays in the notations of DDPM. Specifically, DDPMs assume that - q_{t_n | 0}(x_{t_n} | x_0) = N ( \sqrt{\hat{alpha_n}} * x_0, (1 - \hat{alpha_n}) * I ). - Therefore, the notation \hat{alpha_n} is different from the notation alpha_t in DPM-Solver. In fact, we have - alpha_{t_n} = \sqrt{\hat{alpha_n}}, - and - log(alpha_{t_n}) = 0.5 * log(\hat{alpha_n}). - 2. For continuous-time DPMs: - We support two types of VPSDEs: linear (DDPM) and cosine (improved-DDPM). The hyperparameters for the noise - schedule are the default settings in DDPM and improved-DDPM: - Args: - beta_min: A `float` number. The smallest beta for the linear schedule. - beta_max: A `float` number. The largest beta for the linear schedule. - cosine_s: A `float` number. The hyperparameter in the cosine schedule. - cosine_beta_max: A `float` number. The hyperparameter in the cosine schedule. - T: A `float` number. The ending time of the forward process. - =============================================================== - Args: - schedule: A `str`. The noise schedule of the forward SDE. 'discrete' for discrete-time DPMs, - 'linear' or 'cosine' for continuous-time DPMs. - Returns: - A wrapper object of the forward SDE (VP type). - - =============================================================== - Example: - # For discrete-time DPMs, given betas (the beta array for n = 0, 1, ..., N - 1): - >>> ns = NoiseScheduleVP('discrete', betas=betas) - # For discrete-time DPMs, given alphas_cumprod (the \hat{alpha_n} array for n = 0, 1, ..., N - 1): - >>> ns = NoiseScheduleVP('discrete', alphas_cumprod=alphas_cumprod) - # For continuous-time DPMs (VPSDE), linear schedule: - >>> ns = NoiseScheduleVP('linear', continuous_beta_0=0.1, continuous_beta_1=20.) - """ - - if schedule not in ['discrete', 'linear', 'cosine']: - raise ValueError( - "Unsupported noise schedule {}. The schedule needs to be 'discrete' or 'linear' or 'cosine'".format( - schedule)) - - self.schedule = schedule - if schedule == 'discrete': - if betas is not None: - log_alphas = 0.5 * torch.log(1 - betas).cumsum(dim=0) - else: - assert alphas_cumprod is not None - log_alphas = 0.5 * torch.log(alphas_cumprod) - self.total_N = len(log_alphas) - self.T = 1. - self.t_array = torch.linspace(0., 1., self.total_N + 1)[1:].reshape((1, -1)) - self.log_alpha_array = log_alphas.reshape((1, -1,)) - else: - self.total_N = 1000 - self.beta_0 = continuous_beta_0 - self.beta_1 = continuous_beta_1 - self.cosine_s = 0.008 - self.cosine_beta_max = 999. - self.cosine_t_max = math.atan(self.cosine_beta_max * (1. + self.cosine_s) / math.pi) * 2. * ( - 1. + self.cosine_s) / math.pi - self.cosine_s - self.cosine_log_alpha_0 = math.log(math.cos(self.cosine_s / (1. + self.cosine_s) * math.pi / 2.)) - self.schedule = schedule - if schedule == 'cosine': - # For the cosine schedule, T = 1 will have numerical issues. So we manually set the ending time T. - # Note that T = 0.9946 may be not the optimal setting. However, we find it works well. - self.T = 0.9946 - else: - self.T = 1. - - def marginal_log_mean_coeff(self, t): - """ - Compute log(alpha_t) of a given continuous-time label t in [0, T]. - """ - if self.schedule == 'discrete': - return interpolate_fn(t.reshape((-1, 1)), self.t_array.to(t.device), - self.log_alpha_array.to(t.device)).reshape((-1)) - elif self.schedule == 'linear': - return -0.25 * t ** 2 * (self.beta_1 - self.beta_0) - 0.5 * t * self.beta_0 - elif self.schedule == 'cosine': - log_alpha_fn = lambda s: torch.log(torch.cos((s + self.cosine_s) / (1. + self.cosine_s) * math.pi / 2.)) - log_alpha_t = log_alpha_fn(t) - self.cosine_log_alpha_0 - return log_alpha_t - - def marginal_alpha(self, t): - """ - Compute alpha_t of a given continuous-time label t in [0, T]. - """ - return torch.exp(self.marginal_log_mean_coeff(t)) - - def marginal_std(self, t): - """ - Compute sigma_t of a given continuous-time label t in [0, T]. - """ - return torch.sqrt(1. - torch.exp(2. * self.marginal_log_mean_coeff(t))) - - def marginal_lambda(self, t): - """ - Compute lambda_t = log(alpha_t) - log(sigma_t) of a given continuous-time label t in [0, T]. - """ - log_mean_coeff = self.marginal_log_mean_coeff(t) - log_std = 0.5 * torch.log(1. - torch.exp(2. * log_mean_coeff)) - return log_mean_coeff - log_std - - def inverse_lambda(self, lamb): - """ - Compute the continuous-time label t in [0, T] of a given half-logSNR lambda_t. - """ - if self.schedule == 'linear': - tmp = 2. * (self.beta_1 - self.beta_0) * torch.logaddexp(-2. * lamb, torch.zeros((1,)).to(lamb)) - Delta = self.beta_0 ** 2 + tmp - return tmp / (torch.sqrt(Delta) + self.beta_0) / (self.beta_1 - self.beta_0) - elif self.schedule == 'discrete': - log_alpha = -0.5 * torch.logaddexp(torch.zeros((1,)).to(lamb.device), -2. * lamb) - t = interpolate_fn(log_alpha.reshape((-1, 1)), torch.flip(self.log_alpha_array.to(lamb.device), [1]), - torch.flip(self.t_array.to(lamb.device), [1])) - return t.reshape((-1,)) - else: - log_alpha = -0.5 * torch.logaddexp(-2. * lamb, torch.zeros((1,)).to(lamb)) - t_fn = lambda log_alpha_t: torch.arccos(torch.exp(log_alpha_t + self.cosine_log_alpha_0)) * 2. * ( - 1. + self.cosine_s) / math.pi - self.cosine_s - t = t_fn(log_alpha) - return t - - -def model_wrapper( - model, - noise_schedule, - model_type="noise", - model_kwargs={}, - guidance_type="uncond", - condition=None, - unconditional_condition=None, - guidance_scale=1., - classifier_fn=None, - classifier_kwargs={}, -): - """Create a wrapper function for the noise prediction model. - DPM-Solver needs to solve the continuous-time diffusion ODEs. For DPMs trained on discrete-time labels, we need to - firstly wrap the model function to a noise prediction model that accepts the continuous time as the input. - We support four types of the diffusion model by setting `model_type`: - 1. "noise": noise prediction model. (Trained by predicting noise). - 2. "x_start": data prediction model. (Trained by predicting the data x_0 at time 0). - 3. "v": velocity prediction model. (Trained by predicting the velocity). - The "v" prediction is derivation detailed in Appendix D of [1], and is used in Imagen-Video [2]. - [1] Salimans, Tim, and Jonathan Ho. "Progressive distillation for fast sampling of diffusion models." - arXiv preprint arXiv:2202.00512 (2022). - [2] Ho, Jonathan, et al. "Imagen Video: High Definition Video Generation with Diffusion Models." - arXiv preprint arXiv:2210.02303 (2022). - - 4. "score": marginal score function. (Trained by denoising score matching). - Note that the score function and the noise prediction model follows a simple relationship: - ``` - noise(x_t, t) = -sigma_t * score(x_t, t) - ``` - We support three types of guided sampling by DPMs by setting `guidance_type`: - 1. "uncond": unconditional sampling by DPMs. - The input `model` has the following format: - `` - model(x, t_input, **model_kwargs) -> noise | x_start | v | score - `` - 2. "classifier": classifier guidance sampling [3] by DPMs and another classifier. - The input `model` has the following format: - `` - model(x, t_input, **model_kwargs) -> noise | x_start | v | score - `` - The input `classifier_fn` has the following format: - `` - classifier_fn(x, t_input, cond, **classifier_kwargs) -> logits(x, t_input, cond) - `` - [3] P. Dhariwal and A. Q. Nichol, "Diffusion models beat GANs on image synthesis," - in Advances in Neural Information Processing Systems, vol. 34, 2021, pp. 8780-8794. - 3. "classifier-free": classifier-free guidance sampling by conditional DPMs. - The input `model` has the following format: - `` - model(x, t_input, cond, **model_kwargs) -> noise | x_start | v | score - `` - And if cond == `unconditional_condition`, the model output is the unconditional DPM output. - [4] Ho, Jonathan, and Tim Salimans. "Classifier-free diffusion guidance." - arXiv preprint arXiv:2207.12598 (2022). - - The `t_input` is the time label of the model, which may be discrete-time labels (i.e. 0 to 999) - or continuous-time labels (i.e. epsilon to T). - We wrap the model function to accept only `x` and `t_continuous` as inputs, and outputs the predicted noise: - `` - def model_fn(x, t_continuous) -> noise: - t_input = get_model_input_time(t_continuous) - return noise_pred(model, x, t_input, **model_kwargs) - `` - where `t_continuous` is the continuous time labels (i.e. epsilon to T). And we use `model_fn` for DPM-Solver. - =============================================================== - Args: - model: A diffusion model with the corresponding format described above. - noise_schedule: A noise schedule object, such as NoiseScheduleVP. - model_type: A `str`. The parameterization type of the diffusion model. - "noise" or "x_start" or "v" or "score". - model_kwargs: A `dict`. A dict for the other inputs of the model function. - guidance_type: A `str`. The type of the guidance for sampling. - "uncond" or "classifier" or "classifier-free". - condition: A pytorch tensor. The condition for the guided sampling. - Only used for "classifier" or "classifier-free" guidance type. - unconditional_condition: A pytorch tensor. The condition for the unconditional sampling. - Only used for "classifier-free" guidance type. - guidance_scale: A `float`. The scale for the guided sampling. - classifier_fn: A classifier function. Only used for the classifier guidance. - classifier_kwargs: A `dict`. A dict for the other inputs of the classifier function. - Returns: - A noise prediction model that accepts the noised data and the continuous time as the inputs. - """ - - def get_model_input_time(t_continuous): - """ - Convert the continuous-time `t_continuous` (in [epsilon, T]) to the model input time. - For discrete-time DPMs, we convert `t_continuous` in [1 / N, 1] to `t_input` in [0, 1000 * (N - 1) / N]. - For continuous-time DPMs, we just use `t_continuous`. - """ - if noise_schedule.schedule == 'discrete': - return (t_continuous - 1. / noise_schedule.total_N) * 1000. - else: - return t_continuous - - def noise_pred_fn(x, t_continuous, cond=None): - if t_continuous.reshape((-1,)).shape[0] == 1: - t_continuous = t_continuous.expand((x.shape[0])) - t_input = get_model_input_time(t_continuous) - if cond is None: - output = model(x, t_input, **model_kwargs) - else: - output = model(x, t_input, cond, **model_kwargs) - if model_type == "noise": - return output - elif model_type == "x_start": - alpha_t, sigma_t = noise_schedule.marginal_alpha(t_continuous), noise_schedule.marginal_std(t_continuous) - dims = x.dim() - return (x - expand_dims(alpha_t, dims) * output) / expand_dims(sigma_t, dims) - elif model_type == "v": - alpha_t, sigma_t = noise_schedule.marginal_alpha(t_continuous), noise_schedule.marginal_std(t_continuous) - dims = x.dim() - return expand_dims(alpha_t, dims) * output + expand_dims(sigma_t, dims) * x - elif model_type == "score": - sigma_t = noise_schedule.marginal_std(t_continuous) - dims = x.dim() - return -expand_dims(sigma_t, dims) * output - - def cond_grad_fn(x, t_input): - """ - Compute the gradient of the classifier, i.e. nabla_{x} log p_t(cond | x_t). - """ - with torch.enable_grad(): - x_in = x.detach().requires_grad_(True) - log_prob = classifier_fn(x_in, t_input, condition, **classifier_kwargs) - return torch.autograd.grad(log_prob.sum(), x_in)[0] - - def model_fn(x, t_continuous): - """ - The noise predicition model function that is used for DPM-Solver. - """ - if t_continuous.reshape((-1,)).shape[0] == 1: - t_continuous = t_continuous.expand((x.shape[0])) - if guidance_type == "uncond": - return noise_pred_fn(x, t_continuous) - elif guidance_type == "classifier": - assert classifier_fn is not None - t_input = get_model_input_time(t_continuous) - cond_grad = cond_grad_fn(x, t_input) - sigma_t = noise_schedule.marginal_std(t_continuous) - noise = noise_pred_fn(x, t_continuous) - return noise - guidance_scale * expand_dims(sigma_t, dims=cond_grad.dim()) * cond_grad - elif guidance_type == "classifier-free": - if guidance_scale == 1. or unconditional_condition is None: - return noise_pred_fn(x, t_continuous, cond=condition) - else: - x_in = torch.cat([x] * 2) - t_in = torch.cat([t_continuous] * 2) - if isinstance(condition, dict): - assert isinstance(unconditional_condition, dict) - c_in = dict() - for k in condition: - if isinstance(condition[k], list): - c_in[k] = [torch.cat([unconditional_condition[k][i], condition[k][i]]) for i in range(len(condition[k]))] - else: - c_in[k] = torch.cat([unconditional_condition[k], condition[k]]) - else: - c_in = torch.cat([unconditional_condition, condition]) - noise_uncond, noise = noise_pred_fn(x_in, t_in, cond=c_in).chunk(2) - return noise_uncond + guidance_scale * (noise - noise_uncond) - - assert model_type in ["noise", "x_start", "v"] - assert guidance_type in ["uncond", "classifier", "classifier-free"] - return model_fn - - -class DPM_Solver: - def __init__(self, model_fn, noise_schedule, predict_x0=False, thresholding=False, max_val=1.): - """Construct a DPM-Solver. - We support both the noise prediction model ("predicting epsilon") and the data prediction model ("predicting x0"). - If `predict_x0` is False, we use the solver for the noise prediction model (DPM-Solver). - If `predict_x0` is True, we use the solver for the data prediction model (DPM-Solver++). - In such case, we further support the "dynamic thresholding" in [1] when `thresholding` is True. - The "dynamic thresholding" can greatly improve the sample quality for pixel-space DPMs with large guidance scales. - Args: - model_fn: A noise prediction model function which accepts the continuous-time input (t in [epsilon, T]): - `` - def model_fn(x, t_continuous): - return noise - `` - noise_schedule: A noise schedule object, such as NoiseScheduleVP. - predict_x0: A `bool`. If true, use the data prediction model; else, use the noise prediction model. - thresholding: A `bool`. Valid when `predict_x0` is True. Whether to use the "dynamic thresholding" in [1]. - max_val: A `float`. Valid when both `predict_x0` and `thresholding` are True. The max value for thresholding. - - [1] Chitwan Saharia, William Chan, Saurabh Saxena, Lala Li, Jay Whang, Emily Denton, Seyed Kamyar Seyed Ghasemipour, Burcu Karagol Ayan, S Sara Mahdavi, Rapha Gontijo Lopes, et al. Photorealistic text-to-image diffusion models with deep language understanding. arXiv preprint arXiv:2205.11487, 2022b. - """ - self.model = model_fn - self.noise_schedule = noise_schedule - self.predict_x0 = predict_x0 - self.thresholding = thresholding - self.max_val = max_val - - def noise_prediction_fn(self, x, t): - """ - Return the noise prediction model. - """ - return self.model(x, t) - - def data_prediction_fn(self, x, t): - """ - Return the data prediction model (with thresholding). - """ - noise = self.noise_prediction_fn(x, t) - dims = x.dim() - alpha_t, sigma_t = self.noise_schedule.marginal_alpha(t), self.noise_schedule.marginal_std(t) - x0 = (x - expand_dims(sigma_t, dims) * noise) / expand_dims(alpha_t, dims) - if self.thresholding: - p = 0.995 # A hyperparameter in the paper of "Imagen" [1]. - s = torch.quantile(torch.abs(x0).reshape((x0.shape[0], -1)), p, dim=1) - s = expand_dims(torch.maximum(s, self.max_val * torch.ones_like(s).to(s.device)), dims) - x0 = torch.clamp(x0, -s, s) / s - return x0 - - def model_fn(self, x, t): - """ - Convert the model to the noise prediction model or the data prediction model. - """ - if self.predict_x0: - return self.data_prediction_fn(x, t) - else: - return self.noise_prediction_fn(x, t) - - def get_time_steps(self, skip_type, t_T, t_0, N, device): - """Compute the intermediate time steps for sampling. - Args: - skip_type: A `str`. The type for the spacing of the time steps. We support three types: - - 'logSNR': uniform logSNR for the time steps. - - 'time_uniform': uniform time for the time steps. (**Recommended for high-resolutional data**.) - - 'time_quadratic': quadratic time for the time steps. (Used in DDIM for low-resolutional data.) - t_T: A `float`. The starting time of the sampling (default is T). - t_0: A `float`. The ending time of the sampling (default is epsilon). - N: A `int`. The total number of the spacing of the time steps. - device: A torch device. - Returns: - A pytorch tensor of the time steps, with the shape (N + 1,). - """ - if skip_type == 'logSNR': - lambda_T = self.noise_schedule.marginal_lambda(torch.tensor(t_T).to(device)) - lambda_0 = self.noise_schedule.marginal_lambda(torch.tensor(t_0).to(device)) - logSNR_steps = torch.linspace(lambda_T.cpu().item(), lambda_0.cpu().item(), N + 1).to(device) - return self.noise_schedule.inverse_lambda(logSNR_steps) - elif skip_type == 'time_uniform': - return torch.linspace(t_T, t_0, N + 1).to(device) - elif skip_type == 'time_quadratic': - t_order = 2 - t = torch.linspace(t_T ** (1. / t_order), t_0 ** (1. / t_order), N + 1).pow(t_order).to(device) - return t - else: - raise ValueError( - "Unsupported skip_type {}, need to be 'logSNR' or 'time_uniform' or 'time_quadratic'".format(skip_type)) - - def get_orders_and_timesteps_for_singlestep_solver(self, steps, order, skip_type, t_T, t_0, device): - """ - Get the order of each step for sampling by the singlestep DPM-Solver. - We combine both DPM-Solver-1,2,3 to use all the function evaluations, which is named as "DPM-Solver-fast". - Given a fixed number of function evaluations by `steps`, the sampling procedure by DPM-Solver-fast is: - - If order == 1: - We take `steps` of DPM-Solver-1 (i.e. DDIM). - - If order == 2: - - Denote K = (steps // 2). We take K or (K + 1) intermediate time steps for sampling. - - If steps % 2 == 0, we use K steps of DPM-Solver-2. - - If steps % 2 == 1, we use K steps of DPM-Solver-2 and 1 step of DPM-Solver-1. - - If order == 3: - - Denote K = (steps // 3 + 1). We take K intermediate time steps for sampling. - - If steps % 3 == 0, we use (K - 2) steps of DPM-Solver-3, and 1 step of DPM-Solver-2 and 1 step of DPM-Solver-1. - - If steps % 3 == 1, we use (K - 1) steps of DPM-Solver-3 and 1 step of DPM-Solver-1. - - If steps % 3 == 2, we use (K - 1) steps of DPM-Solver-3 and 1 step of DPM-Solver-2. - ============================================ - Args: - order: A `int`. The max order for the solver (2 or 3). - steps: A `int`. The total number of function evaluations (NFE). - skip_type: A `str`. The type for the spacing of the time steps. We support three types: - - 'logSNR': uniform logSNR for the time steps. - - 'time_uniform': uniform time for the time steps. (**Recommended for high-resolutional data**.) - - 'time_quadratic': quadratic time for the time steps. (Used in DDIM for low-resolutional data.) - t_T: A `float`. The starting time of the sampling (default is T). - t_0: A `float`. The ending time of the sampling (default is epsilon). - device: A torch device. - Returns: - orders: A list of the solver order of each step. - """ - if order == 3: - K = steps // 3 + 1 - if steps % 3 == 0: - orders = [3, ] * (K - 2) + [2, 1] - elif steps % 3 == 1: - orders = [3, ] * (K - 1) + [1] - else: - orders = [3, ] * (K - 1) + [2] - elif order == 2: - if steps % 2 == 0: - K = steps // 2 - orders = [2, ] * K - else: - K = steps // 2 + 1 - orders = [2, ] * (K - 1) + [1] - elif order == 1: - K = 1 - orders = [1, ] * steps - else: - raise ValueError("'order' must be '1' or '2' or '3'.") - if skip_type == 'logSNR': - # To reproduce the results in DPM-Solver paper - timesteps_outer = self.get_time_steps(skip_type, t_T, t_0, K, device) - else: - timesteps_outer = self.get_time_steps(skip_type, t_T, t_0, steps, device)[ - torch.cumsum(torch.tensor([0, ] + orders)).to(device)] - return timesteps_outer, orders - - def denoise_to_zero_fn(self, x, s): - """ - Denoise at the final step, which is equivalent to solve the ODE from lambda_s to infty by first-order discretization. - """ - return self.data_prediction_fn(x, s) - - def dpm_solver_first_update(self, x, s, t, model_s=None, return_intermediate=False): - """ - DPM-Solver-1 (equivalent to DDIM) from time `s` to time `t`. - Args: - x: A pytorch tensor. The initial value at time `s`. - s: A pytorch tensor. The starting time, with the shape (x.shape[0],). - t: A pytorch tensor. The ending time, with the shape (x.shape[0],). - model_s: A pytorch tensor. The model function evaluated at time `s`. - If `model_s` is None, we evaluate the model by `x` and `s`; otherwise we directly use it. - return_intermediate: A `bool`. If true, also return the model value at time `s`. - Returns: - x_t: A pytorch tensor. The approximated solution at time `t`. - """ - ns = self.noise_schedule - dims = x.dim() - lambda_s, lambda_t = ns.marginal_lambda(s), ns.marginal_lambda(t) - h = lambda_t - lambda_s - log_alpha_s, log_alpha_t = ns.marginal_log_mean_coeff(s), ns.marginal_log_mean_coeff(t) - sigma_s, sigma_t = ns.marginal_std(s), ns.marginal_std(t) - alpha_t = torch.exp(log_alpha_t) - - if self.predict_x0: - phi_1 = torch.expm1(-h) - if model_s is None: - model_s = self.model_fn(x, s) - x_t = ( - expand_dims(sigma_t / sigma_s, dims) * x - - expand_dims(alpha_t * phi_1, dims) * model_s - ) - if return_intermediate: - return x_t, {'model_s': model_s} - else: - return x_t - else: - phi_1 = torch.expm1(h) - if model_s is None: - model_s = self.model_fn(x, s) - x_t = ( - expand_dims(torch.exp(log_alpha_t - log_alpha_s), dims) * x - - expand_dims(sigma_t * phi_1, dims) * model_s - ) - if return_intermediate: - return x_t, {'model_s': model_s} - else: - return x_t - - def singlestep_dpm_solver_second_update(self, x, s, t, r1=0.5, model_s=None, return_intermediate=False, - solver_type='dpm_solver'): - """ - Singlestep solver DPM-Solver-2 from time `s` to time `t`. - Args: - x: A pytorch tensor. The initial value at time `s`. - s: A pytorch tensor. The starting time, with the shape (x.shape[0],). - t: A pytorch tensor. The ending time, with the shape (x.shape[0],). - r1: A `float`. The hyperparameter of the second-order solver. - model_s: A pytorch tensor. The model function evaluated at time `s`. - If `model_s` is None, we evaluate the model by `x` and `s`; otherwise we directly use it. - return_intermediate: A `bool`. If true, also return the model value at time `s` and `s1` (the intermediate time). - solver_type: either 'dpm_solver' or 'taylor'. The type for the high-order solvers. - The type slightly impacts the performance. We recommend to use 'dpm_solver' type. - Returns: - x_t: A pytorch tensor. The approximated solution at time `t`. - """ - if solver_type not in ['dpm_solver', 'taylor']: - raise ValueError("'solver_type' must be either 'dpm_solver' or 'taylor', got {}".format(solver_type)) - if r1 is None: - r1 = 0.5 - ns = self.noise_schedule - dims = x.dim() - lambda_s, lambda_t = ns.marginal_lambda(s), ns.marginal_lambda(t) - h = lambda_t - lambda_s - lambda_s1 = lambda_s + r1 * h - s1 = ns.inverse_lambda(lambda_s1) - log_alpha_s, log_alpha_s1, log_alpha_t = ns.marginal_log_mean_coeff(s), ns.marginal_log_mean_coeff( - s1), ns.marginal_log_mean_coeff(t) - sigma_s, sigma_s1, sigma_t = ns.marginal_std(s), ns.marginal_std(s1), ns.marginal_std(t) - alpha_s1, alpha_t = torch.exp(log_alpha_s1), torch.exp(log_alpha_t) - - if self.predict_x0: - phi_11 = torch.expm1(-r1 * h) - phi_1 = torch.expm1(-h) - - if model_s is None: - model_s = self.model_fn(x, s) - x_s1 = ( - expand_dims(sigma_s1 / sigma_s, dims) * x - - expand_dims(alpha_s1 * phi_11, dims) * model_s - ) - model_s1 = self.model_fn(x_s1, s1) - if solver_type == 'dpm_solver': - x_t = ( - expand_dims(sigma_t / sigma_s, dims) * x - - expand_dims(alpha_t * phi_1, dims) * model_s - - (0.5 / r1) * expand_dims(alpha_t * phi_1, dims) * (model_s1 - model_s) - ) - elif solver_type == 'taylor': - x_t = ( - expand_dims(sigma_t / sigma_s, dims) * x - - expand_dims(alpha_t * phi_1, dims) * model_s - + (1. / r1) * expand_dims(alpha_t * ((torch.exp(-h) - 1.) / h + 1.), dims) * ( - model_s1 - model_s) - ) - else: - phi_11 = torch.expm1(r1 * h) - phi_1 = torch.expm1(h) - - if model_s is None: - model_s = self.model_fn(x, s) - x_s1 = ( - expand_dims(torch.exp(log_alpha_s1 - log_alpha_s), dims) * x - - expand_dims(sigma_s1 * phi_11, dims) * model_s - ) - model_s1 = self.model_fn(x_s1, s1) - if solver_type == 'dpm_solver': - x_t = ( - expand_dims(torch.exp(log_alpha_t - log_alpha_s), dims) * x - - expand_dims(sigma_t * phi_1, dims) * model_s - - (0.5 / r1) * expand_dims(sigma_t * phi_1, dims) * (model_s1 - model_s) - ) - elif solver_type == 'taylor': - x_t = ( - expand_dims(torch.exp(log_alpha_t - log_alpha_s), dims) * x - - expand_dims(sigma_t * phi_1, dims) * model_s - - (1. / r1) * expand_dims(sigma_t * ((torch.exp(h) - 1.) / h - 1.), dims) * (model_s1 - model_s) - ) - if return_intermediate: - return x_t, {'model_s': model_s, 'model_s1': model_s1} - else: - return x_t - - def singlestep_dpm_solver_third_update(self, x, s, t, r1=1. / 3., r2=2. / 3., model_s=None, model_s1=None, - return_intermediate=False, solver_type='dpm_solver'): - """ - Singlestep solver DPM-Solver-3 from time `s` to time `t`. - Args: - x: A pytorch tensor. The initial value at time `s`. - s: A pytorch tensor. The starting time, with the shape (x.shape[0],). - t: A pytorch tensor. The ending time, with the shape (x.shape[0],). - r1: A `float`. The hyperparameter of the third-order solver. - r2: A `float`. The hyperparameter of the third-order solver. - model_s: A pytorch tensor. The model function evaluated at time `s`. - If `model_s` is None, we evaluate the model by `x` and `s`; otherwise we directly use it. - model_s1: A pytorch tensor. The model function evaluated at time `s1` (the intermediate time given by `r1`). - If `model_s1` is None, we evaluate the model at `s1`; otherwise we directly use it. - return_intermediate: A `bool`. If true, also return the model value at time `s`, `s1` and `s2` (the intermediate times). - solver_type: either 'dpm_solver' or 'taylor'. The type for the high-order solvers. - The type slightly impacts the performance. We recommend to use 'dpm_solver' type. - Returns: - x_t: A pytorch tensor. The approximated solution at time `t`. - """ - if solver_type not in ['dpm_solver', 'taylor']: - raise ValueError("'solver_type' must be either 'dpm_solver' or 'taylor', got {}".format(solver_type)) - if r1 is None: - r1 = 1. / 3. - if r2 is None: - r2 = 2. / 3. - ns = self.noise_schedule - dims = x.dim() - lambda_s, lambda_t = ns.marginal_lambda(s), ns.marginal_lambda(t) - h = lambda_t - lambda_s - lambda_s1 = lambda_s + r1 * h - lambda_s2 = lambda_s + r2 * h - s1 = ns.inverse_lambda(lambda_s1) - s2 = ns.inverse_lambda(lambda_s2) - log_alpha_s, log_alpha_s1, log_alpha_s2, log_alpha_t = ns.marginal_log_mean_coeff( - s), ns.marginal_log_mean_coeff(s1), ns.marginal_log_mean_coeff(s2), ns.marginal_log_mean_coeff(t) - sigma_s, sigma_s1, sigma_s2, sigma_t = ns.marginal_std(s), ns.marginal_std(s1), ns.marginal_std( - s2), ns.marginal_std(t) - alpha_s1, alpha_s2, alpha_t = torch.exp(log_alpha_s1), torch.exp(log_alpha_s2), torch.exp(log_alpha_t) - - if self.predict_x0: - phi_11 = torch.expm1(-r1 * h) - phi_12 = torch.expm1(-r2 * h) - phi_1 = torch.expm1(-h) - phi_22 = torch.expm1(-r2 * h) / (r2 * h) + 1. - phi_2 = phi_1 / h + 1. - phi_3 = phi_2 / h - 0.5 - - if model_s is None: - model_s = self.model_fn(x, s) - if model_s1 is None: - x_s1 = ( - expand_dims(sigma_s1 / sigma_s, dims) * x - - expand_dims(alpha_s1 * phi_11, dims) * model_s - ) - model_s1 = self.model_fn(x_s1, s1) - x_s2 = ( - expand_dims(sigma_s2 / sigma_s, dims) * x - - expand_dims(alpha_s2 * phi_12, dims) * model_s - + r2 / r1 * expand_dims(alpha_s2 * phi_22, dims) * (model_s1 - model_s) - ) - model_s2 = self.model_fn(x_s2, s2) - if solver_type == 'dpm_solver': - x_t = ( - expand_dims(sigma_t / sigma_s, dims) * x - - expand_dims(alpha_t * phi_1, dims) * model_s - + (1. / r2) * expand_dims(alpha_t * phi_2, dims) * (model_s2 - model_s) - ) - elif solver_type == 'taylor': - D1_0 = (1. / r1) * (model_s1 - model_s) - D1_1 = (1. / r2) * (model_s2 - model_s) - D1 = (r2 * D1_0 - r1 * D1_1) / (r2 - r1) - D2 = 2. * (D1_1 - D1_0) / (r2 - r1) - x_t = ( - expand_dims(sigma_t / sigma_s, dims) * x - - expand_dims(alpha_t * phi_1, dims) * model_s - + expand_dims(alpha_t * phi_2, dims) * D1 - - expand_dims(alpha_t * phi_3, dims) * D2 - ) - else: - phi_11 = torch.expm1(r1 * h) - phi_12 = torch.expm1(r2 * h) - phi_1 = torch.expm1(h) - phi_22 = torch.expm1(r2 * h) / (r2 * h) - 1. - phi_2 = phi_1 / h - 1. - phi_3 = phi_2 / h - 0.5 - - if model_s is None: - model_s = self.model_fn(x, s) - if model_s1 is None: - x_s1 = ( - expand_dims(torch.exp(log_alpha_s1 - log_alpha_s), dims) * x - - expand_dims(sigma_s1 * phi_11, dims) * model_s - ) - model_s1 = self.model_fn(x_s1, s1) - x_s2 = ( - expand_dims(torch.exp(log_alpha_s2 - log_alpha_s), dims) * x - - expand_dims(sigma_s2 * phi_12, dims) * model_s - - r2 / r1 * expand_dims(sigma_s2 * phi_22, dims) * (model_s1 - model_s) - ) - model_s2 = self.model_fn(x_s2, s2) - if solver_type == 'dpm_solver': - x_t = ( - expand_dims(torch.exp(log_alpha_t - log_alpha_s), dims) * x - - expand_dims(sigma_t * phi_1, dims) * model_s - - (1. / r2) * expand_dims(sigma_t * phi_2, dims) * (model_s2 - model_s) - ) - elif solver_type == 'taylor': - D1_0 = (1. / r1) * (model_s1 - model_s) - D1_1 = (1. / r2) * (model_s2 - model_s) - D1 = (r2 * D1_0 - r1 * D1_1) / (r2 - r1) - D2 = 2. * (D1_1 - D1_0) / (r2 - r1) - x_t = ( - expand_dims(torch.exp(log_alpha_t - log_alpha_s), dims) * x - - expand_dims(sigma_t * phi_1, dims) * model_s - - expand_dims(sigma_t * phi_2, dims) * D1 - - expand_dims(sigma_t * phi_3, dims) * D2 - ) - - if return_intermediate: - return x_t, {'model_s': model_s, 'model_s1': model_s1, 'model_s2': model_s2} - else: - return x_t - - def multistep_dpm_solver_second_update(self, x, model_prev_list, t_prev_list, t, solver_type="dpm_solver"): - """ - Multistep solver DPM-Solver-2 from time `t_prev_list[-1]` to time `t`. - Args: - x: A pytorch tensor. The initial value at time `s`. - model_prev_list: A list of pytorch tensor. The previous computed model values. - t_prev_list: A list of pytorch tensor. The previous times, each time has the shape (x.shape[0],) - t: A pytorch tensor. The ending time, with the shape (x.shape[0],). - solver_type: either 'dpm_solver' or 'taylor'. The type for the high-order solvers. - The type slightly impacts the performance. We recommend to use 'dpm_solver' type. - Returns: - x_t: A pytorch tensor. The approximated solution at time `t`. - """ - if solver_type not in ['dpm_solver', 'taylor']: - raise ValueError("'solver_type' must be either 'dpm_solver' or 'taylor', got {}".format(solver_type)) - ns = self.noise_schedule - dims = x.dim() - model_prev_1, model_prev_0 = model_prev_list - t_prev_1, t_prev_0 = t_prev_list - lambda_prev_1, lambda_prev_0, lambda_t = ns.marginal_lambda(t_prev_1), ns.marginal_lambda( - t_prev_0), ns.marginal_lambda(t) - log_alpha_prev_0, log_alpha_t = ns.marginal_log_mean_coeff(t_prev_0), ns.marginal_log_mean_coeff(t) - sigma_prev_0, sigma_t = ns.marginal_std(t_prev_0), ns.marginal_std(t) - alpha_t = torch.exp(log_alpha_t) - - h_0 = lambda_prev_0 - lambda_prev_1 - h = lambda_t - lambda_prev_0 - r0 = h_0 / h - D1_0 = expand_dims(1. / r0, dims) * (model_prev_0 - model_prev_1) - if self.predict_x0: - if solver_type == 'dpm_solver': - x_t = ( - expand_dims(sigma_t / sigma_prev_0, dims) * x - - expand_dims(alpha_t * (torch.exp(-h) - 1.), dims) * model_prev_0 - - 0.5 * expand_dims(alpha_t * (torch.exp(-h) - 1.), dims) * D1_0 - ) - elif solver_type == 'taylor': - x_t = ( - expand_dims(sigma_t / sigma_prev_0, dims) * x - - expand_dims(alpha_t * (torch.exp(-h) - 1.), dims) * model_prev_0 - + expand_dims(alpha_t * ((torch.exp(-h) - 1.) / h + 1.), dims) * D1_0 - ) - else: - if solver_type == 'dpm_solver': - x_t = ( - expand_dims(torch.exp(log_alpha_t - log_alpha_prev_0), dims) * x - - expand_dims(sigma_t * (torch.exp(h) - 1.), dims) * model_prev_0 - - 0.5 * expand_dims(sigma_t * (torch.exp(h) - 1.), dims) * D1_0 - ) - elif solver_type == 'taylor': - x_t = ( - expand_dims(torch.exp(log_alpha_t - log_alpha_prev_0), dims) * x - - expand_dims(sigma_t * (torch.exp(h) - 1.), dims) * model_prev_0 - - expand_dims(sigma_t * ((torch.exp(h) - 1.) / h - 1.), dims) * D1_0 - ) - return x_t - - def multistep_dpm_solver_third_update(self, x, model_prev_list, t_prev_list, t, solver_type='dpm_solver'): - """ - Multistep solver DPM-Solver-3 from time `t_prev_list[-1]` to time `t`. - Args: - x: A pytorch tensor. The initial value at time `s`. - model_prev_list: A list of pytorch tensor. The previous computed model values. - t_prev_list: A list of pytorch tensor. The previous times, each time has the shape (x.shape[0],) - t: A pytorch tensor. The ending time, with the shape (x.shape[0],). - solver_type: either 'dpm_solver' or 'taylor'. The type for the high-order solvers. - The type slightly impacts the performance. We recommend to use 'dpm_solver' type. - Returns: - x_t: A pytorch tensor. The approximated solution at time `t`. - """ - ns = self.noise_schedule - dims = x.dim() - model_prev_2, model_prev_1, model_prev_0 = model_prev_list - t_prev_2, t_prev_1, t_prev_0 = t_prev_list - lambda_prev_2, lambda_prev_1, lambda_prev_0, lambda_t = ns.marginal_lambda(t_prev_2), ns.marginal_lambda( - t_prev_1), ns.marginal_lambda(t_prev_0), ns.marginal_lambda(t) - log_alpha_prev_0, log_alpha_t = ns.marginal_log_mean_coeff(t_prev_0), ns.marginal_log_mean_coeff(t) - sigma_prev_0, sigma_t = ns.marginal_std(t_prev_0), ns.marginal_std(t) - alpha_t = torch.exp(log_alpha_t) - - h_1 = lambda_prev_1 - lambda_prev_2 - h_0 = lambda_prev_0 - lambda_prev_1 - h = lambda_t - lambda_prev_0 - r0, r1 = h_0 / h, h_1 / h - D1_0 = expand_dims(1. / r0, dims) * (model_prev_0 - model_prev_1) - D1_1 = expand_dims(1. / r1, dims) * (model_prev_1 - model_prev_2) - D1 = D1_0 + expand_dims(r0 / (r0 + r1), dims) * (D1_0 - D1_1) - D2 = expand_dims(1. / (r0 + r1), dims) * (D1_0 - D1_1) - if self.predict_x0: - x_t = ( - expand_dims(sigma_t / sigma_prev_0, dims) * x - - expand_dims(alpha_t * (torch.exp(-h) - 1.), dims) * model_prev_0 - + expand_dims(alpha_t * ((torch.exp(-h) - 1.) / h + 1.), dims) * D1 - - expand_dims(alpha_t * ((torch.exp(-h) - 1. + h) / h ** 2 - 0.5), dims) * D2 - ) - else: - x_t = ( - expand_dims(torch.exp(log_alpha_t - log_alpha_prev_0), dims) * x - - expand_dims(sigma_t * (torch.exp(h) - 1.), dims) * model_prev_0 - - expand_dims(sigma_t * ((torch.exp(h) - 1.) / h - 1.), dims) * D1 - - expand_dims(sigma_t * ((torch.exp(h) - 1. - h) / h ** 2 - 0.5), dims) * D2 - ) - return x_t - - def singlestep_dpm_solver_update(self, x, s, t, order, return_intermediate=False, solver_type='dpm_solver', r1=None, - r2=None): - """ - Singlestep DPM-Solver with the order `order` from time `s` to time `t`. - Args: - x: A pytorch tensor. The initial value at time `s`. - s: A pytorch tensor. The starting time, with the shape (x.shape[0],). - t: A pytorch tensor. The ending time, with the shape (x.shape[0],). - order: A `int`. The order of DPM-Solver. We only support order == 1 or 2 or 3. - return_intermediate: A `bool`. If true, also return the model value at time `s`, `s1` and `s2` (the intermediate times). - solver_type: either 'dpm_solver' or 'taylor'. The type for the high-order solvers. - The type slightly impacts the performance. We recommend to use 'dpm_solver' type. - r1: A `float`. The hyperparameter of the second-order or third-order solver. - r2: A `float`. The hyperparameter of the third-order solver. - Returns: - x_t: A pytorch tensor. The approximated solution at time `t`. - """ - if order == 1: - return self.dpm_solver_first_update(x, s, t, return_intermediate=return_intermediate) - elif order == 2: - return self.singlestep_dpm_solver_second_update(x, s, t, return_intermediate=return_intermediate, - solver_type=solver_type, r1=r1) - elif order == 3: - return self.singlestep_dpm_solver_third_update(x, s, t, return_intermediate=return_intermediate, - solver_type=solver_type, r1=r1, r2=r2) - else: - raise ValueError("Solver order must be 1 or 2 or 3, got {}".format(order)) - - def multistep_dpm_solver_update(self, x, model_prev_list, t_prev_list, t, order, solver_type='dpm_solver'): - """ - Multistep DPM-Solver with the order `order` from time `t_prev_list[-1]` to time `t`. - Args: - x: A pytorch tensor. The initial value at time `s`. - model_prev_list: A list of pytorch tensor. The previous computed model values. - t_prev_list: A list of pytorch tensor. The previous times, each time has the shape (x.shape[0],) - t: A pytorch tensor. The ending time, with the shape (x.shape[0],). - order: A `int`. The order of DPM-Solver. We only support order == 1 or 2 or 3. - solver_type: either 'dpm_solver' or 'taylor'. The type for the high-order solvers. - The type slightly impacts the performance. We recommend to use 'dpm_solver' type. - Returns: - x_t: A pytorch tensor. The approximated solution at time `t`. - """ - if order == 1: - return self.dpm_solver_first_update(x, t_prev_list[-1], t, model_s=model_prev_list[-1]) - elif order == 2: - return self.multistep_dpm_solver_second_update(x, model_prev_list, t_prev_list, t, solver_type=solver_type) - elif order == 3: - return self.multistep_dpm_solver_third_update(x, model_prev_list, t_prev_list, t, solver_type=solver_type) - else: - raise ValueError("Solver order must be 1 or 2 or 3, got {}".format(order)) - - def dpm_solver_adaptive(self, x, order, t_T, t_0, h_init=0.05, atol=0.0078, rtol=0.05, theta=0.9, t_err=1e-5, - solver_type='dpm_solver'): - """ - The adaptive step size solver based on singlestep DPM-Solver. - Args: - x: A pytorch tensor. The initial value at time `t_T`. - order: A `int`. The (higher) order of the solver. We only support order == 2 or 3. - t_T: A `float`. The starting time of the sampling (default is T). - t_0: A `float`. The ending time of the sampling (default is epsilon). - h_init: A `float`. The initial step size (for logSNR). - atol: A `float`. The absolute tolerance of the solver. For image data, the default setting is 0.0078, followed [1]. - rtol: A `float`. The relative tolerance of the solver. The default setting is 0.05. - theta: A `float`. The safety hyperparameter for adapting the step size. The default setting is 0.9, followed [1]. - t_err: A `float`. The tolerance for the time. We solve the diffusion ODE until the absolute error between the - current time and `t_0` is less than `t_err`. The default setting is 1e-5. - solver_type: either 'dpm_solver' or 'taylor'. The type for the high-order solvers. - The type slightly impacts the performance. We recommend to use 'dpm_solver' type. - Returns: - x_0: A pytorch tensor. The approximated solution at time `t_0`. - [1] A. Jolicoeur-Martineau, K. Li, R. Piché-Taillefer, T. Kachman, and I. Mitliagkas, "Gotta go fast when generating data with score-based models," arXiv preprint arXiv:2105.14080, 2021. - """ - ns = self.noise_schedule - s = t_T * torch.ones((x.shape[0],)).to(x) - lambda_s = ns.marginal_lambda(s) - lambda_0 = ns.marginal_lambda(t_0 * torch.ones_like(s).to(x)) - h = h_init * torch.ones_like(s).to(x) - x_prev = x - nfe = 0 - if order == 2: - r1 = 0.5 - lower_update = lambda x, s, t: self.dpm_solver_first_update(x, s, t, return_intermediate=True) - higher_update = lambda x, s, t, **kwargs: self.singlestep_dpm_solver_second_update(x, s, t, r1=r1, - solver_type=solver_type, - **kwargs) - elif order == 3: - r1, r2 = 1. / 3., 2. / 3. - lower_update = lambda x, s, t: self.singlestep_dpm_solver_second_update(x, s, t, r1=r1, - return_intermediate=True, - solver_type=solver_type) - higher_update = lambda x, s, t, **kwargs: self.singlestep_dpm_solver_third_update(x, s, t, r1=r1, r2=r2, - solver_type=solver_type, - **kwargs) - else: - raise ValueError("For adaptive step size solver, order must be 2 or 3, got {}".format(order)) - while torch.abs((s - t_0)).mean() > t_err: - t = ns.inverse_lambda(lambda_s + h) - x_lower, lower_noise_kwargs = lower_update(x, s, t) - x_higher = higher_update(x, s, t, **lower_noise_kwargs) - delta = torch.max(torch.ones_like(x).to(x) * atol, rtol * torch.max(torch.abs(x_lower), torch.abs(x_prev))) - norm_fn = lambda v: torch.sqrt(torch.square(v.reshape((v.shape[0], -1))).mean(dim=-1, keepdim=True)) - E = norm_fn((x_higher - x_lower) / delta).max() - if torch.all(E <= 1.): - x = x_higher - s = t - x_prev = x_lower - lambda_s = ns.marginal_lambda(s) - h = torch.min(theta * h * torch.float_power(E, -1. / order).float(), lambda_0 - lambda_s) - nfe += order - print('adaptive solver nfe', nfe) - return x - - def sample(self, x, steps=20, t_start=None, t_end=None, order=3, skip_type='time_uniform', - method='singlestep', lower_order_final=True, denoise_to_zero=False, solver_type='dpm_solver', - atol=0.0078, rtol=0.05, - ): - """ - Compute the sample at time `t_end` by DPM-Solver, given the initial `x` at time `t_start`. - ===================================================== - We support the following algorithms for both noise prediction model and data prediction model: - - 'singlestep': - Singlestep DPM-Solver (i.e. "DPM-Solver-fast" in the paper), which combines different orders of singlestep DPM-Solver. - We combine all the singlestep solvers with order <= `order` to use up all the function evaluations (steps). - The total number of function evaluations (NFE) == `steps`. - Given a fixed NFE == `steps`, the sampling procedure is: - - If `order` == 1: - - Denote K = steps. We use K steps of DPM-Solver-1 (i.e. DDIM). - - If `order` == 2: - - Denote K = (steps // 2) + (steps % 2). We take K intermediate time steps for sampling. - - If steps % 2 == 0, we use K steps of singlestep DPM-Solver-2. - - If steps % 2 == 1, we use (K - 1) steps of singlestep DPM-Solver-2 and 1 step of DPM-Solver-1. - - If `order` == 3: - - Denote K = (steps // 3 + 1). We take K intermediate time steps for sampling. - - If steps % 3 == 0, we use (K - 2) steps of singlestep DPM-Solver-3, and 1 step of singlestep DPM-Solver-2 and 1 step of DPM-Solver-1. - - If steps % 3 == 1, we use (K - 1) steps of singlestep DPM-Solver-3 and 1 step of DPM-Solver-1. - - If steps % 3 == 2, we use (K - 1) steps of singlestep DPM-Solver-3 and 1 step of singlestep DPM-Solver-2. - - 'multistep': - Multistep DPM-Solver with the order of `order`. The total number of function evaluations (NFE) == `steps`. - We initialize the first `order` values by lower order multistep solvers. - Given a fixed NFE == `steps`, the sampling procedure is: - Denote K = steps. - - If `order` == 1: - - We use K steps of DPM-Solver-1 (i.e. DDIM). - - If `order` == 2: - - We firstly use 1 step of DPM-Solver-1, then use (K - 1) step of multistep DPM-Solver-2. - - If `order` == 3: - - We firstly use 1 step of DPM-Solver-1, then 1 step of multistep DPM-Solver-2, then (K - 2) step of multistep DPM-Solver-3. - - 'singlestep_fixed': - Fixed order singlestep DPM-Solver (i.e. DPM-Solver-1 or singlestep DPM-Solver-2 or singlestep DPM-Solver-3). - We use singlestep DPM-Solver-`order` for `order`=1 or 2 or 3, with total [`steps` // `order`] * `order` NFE. - - 'adaptive': - Adaptive step size DPM-Solver (i.e. "DPM-Solver-12" and "DPM-Solver-23" in the paper). - We ignore `steps` and use adaptive step size DPM-Solver with a higher order of `order`. - You can adjust the absolute tolerance `atol` and the relative tolerance `rtol` to balance the computatation costs - (NFE) and the sample quality. - - If `order` == 2, we use DPM-Solver-12 which combines DPM-Solver-1 and singlestep DPM-Solver-2. - - If `order` == 3, we use DPM-Solver-23 which combines singlestep DPM-Solver-2 and singlestep DPM-Solver-3. - ===================================================== - Some advices for choosing the algorithm: - - For **unconditional sampling** or **guided sampling with small guidance scale** by DPMs: - Use singlestep DPM-Solver ("DPM-Solver-fast" in the paper) with `order = 3`. - e.g. - >>> dpm_solver = DPM_Solver(model_fn, noise_schedule, predict_x0=False) - >>> x_sample = dpm_solver.sample(x, steps=steps, t_start=t_start, t_end=t_end, order=3, - skip_type='time_uniform', method='singlestep') - - For **guided sampling with large guidance scale** by DPMs: - Use multistep DPM-Solver with `predict_x0 = True` and `order = 2`. - e.g. - >>> dpm_solver = DPM_Solver(model_fn, noise_schedule, predict_x0=True) - >>> x_sample = dpm_solver.sample(x, steps=steps, t_start=t_start, t_end=t_end, order=2, - skip_type='time_uniform', method='multistep') - We support three types of `skip_type`: - - 'logSNR': uniform logSNR for the time steps. **Recommended for low-resolutional images** - - 'time_uniform': uniform time for the time steps. **Recommended for high-resolutional images**. - - 'time_quadratic': quadratic time for the time steps. - ===================================================== - Args: - x: A pytorch tensor. The initial value at time `t_start` - e.g. if `t_start` == T, then `x` is a sample from the standard normal distribution. - steps: A `int`. The total number of function evaluations (NFE). - t_start: A `float`. The starting time of the sampling. - If `T` is None, we use self.noise_schedule.T (default is 1.0). - t_end: A `float`. The ending time of the sampling. - If `t_end` is None, we use 1. / self.noise_schedule.total_N. - e.g. if total_N == 1000, we have `t_end` == 1e-3. - For discrete-time DPMs: - - We recommend `t_end` == 1. / self.noise_schedule.total_N. - For continuous-time DPMs: - - We recommend `t_end` == 1e-3 when `steps` <= 15; and `t_end` == 1e-4 when `steps` > 15. - order: A `int`. The order of DPM-Solver. - skip_type: A `str`. The type for the spacing of the time steps. 'time_uniform' or 'logSNR' or 'time_quadratic'. - method: A `str`. The method for sampling. 'singlestep' or 'multistep' or 'singlestep_fixed' or 'adaptive'. - denoise_to_zero: A `bool`. Whether to denoise to time 0 at the final step. - Default is `False`. If `denoise_to_zero` is `True`, the total NFE is (`steps` + 1). - This trick is firstly proposed by DDPM (https://arxiv.org/abs/2006.11239) and - score_sde (https://arxiv.org/abs/2011.13456). Such trick can improve the FID - for diffusion models sampling by diffusion SDEs for low-resolutional images - (such as CIFAR-10). However, we observed that such trick does not matter for - high-resolutional images. As it needs an additional NFE, we do not recommend - it for high-resolutional images. - lower_order_final: A `bool`. Whether to use lower order solvers at the final steps. - Only valid for `method=multistep` and `steps < 15`. We empirically find that - this trick is a key to stabilizing the sampling by DPM-Solver with very few steps - (especially for steps <= 10). So we recommend to set it to be `True`. - solver_type: A `str`. The taylor expansion type for the solver. `dpm_solver` or `taylor`. We recommend `dpm_solver`. - atol: A `float`. The absolute tolerance of the adaptive step size solver. Valid when `method` == 'adaptive'. - rtol: A `float`. The relative tolerance of the adaptive step size solver. Valid when `method` == 'adaptive'. - Returns: - x_end: A pytorch tensor. The approximated solution at time `t_end`. - """ - t_0 = 1. / self.noise_schedule.total_N if t_end is None else t_end - t_T = self.noise_schedule.T if t_start is None else t_start - device = x.device - if method == 'adaptive': - with torch.no_grad(): - x = self.dpm_solver_adaptive(x, order=order, t_T=t_T, t_0=t_0, atol=atol, rtol=rtol, - solver_type=solver_type) - elif method == 'multistep': - assert steps >= order - timesteps = self.get_time_steps(skip_type=skip_type, t_T=t_T, t_0=t_0, N=steps, device=device) - assert timesteps.shape[0] - 1 == steps - with torch.no_grad(): - vec_t = timesteps[0].expand((x.shape[0])) - model_prev_list = [self.model_fn(x, vec_t)] - t_prev_list = [vec_t] - # Init the first `order` values by lower order multistep DPM-Solver. - for init_order in tqdm(range(1, order), desc="DPM init order"): - vec_t = timesteps[init_order].expand(x.shape[0]) - x = self.multistep_dpm_solver_update(x, model_prev_list, t_prev_list, vec_t, init_order, - solver_type=solver_type) - model_prev_list.append(self.model_fn(x, vec_t)) - t_prev_list.append(vec_t) - # Compute the remaining values by `order`-th order multistep DPM-Solver. - for step in tqdm(range(order, steps + 1), desc="DPM multistep"): - vec_t = timesteps[step].expand(x.shape[0]) - if lower_order_final and steps < 15: - step_order = min(order, steps + 1 - step) - else: - step_order = order - x = self.multistep_dpm_solver_update(x, model_prev_list, t_prev_list, vec_t, step_order, - solver_type=solver_type) - for i in range(order - 1): - t_prev_list[i] = t_prev_list[i + 1] - model_prev_list[i] = model_prev_list[i + 1] - t_prev_list[-1] = vec_t - # We do not need to evaluate the final model value. - if step < steps: - model_prev_list[-1] = self.model_fn(x, vec_t) - elif method in ['singlestep', 'singlestep_fixed']: - if method == 'singlestep': - timesteps_outer, orders = self.get_orders_and_timesteps_for_singlestep_solver(steps=steps, order=order, - skip_type=skip_type, - t_T=t_T, t_0=t_0, - device=device) - elif method == 'singlestep_fixed': - K = steps // order - orders = [order, ] * K - timesteps_outer = self.get_time_steps(skip_type=skip_type, t_T=t_T, t_0=t_0, N=K, device=device) - for i, order in enumerate(orders): - t_T_inner, t_0_inner = timesteps_outer[i], timesteps_outer[i + 1] - timesteps_inner = self.get_time_steps(skip_type=skip_type, t_T=t_T_inner.item(), t_0=t_0_inner.item(), - N=order, device=device) - lambda_inner = self.noise_schedule.marginal_lambda(timesteps_inner) - vec_s, vec_t = t_T_inner.tile(x.shape[0]), t_0_inner.tile(x.shape[0]) - h = lambda_inner[-1] - lambda_inner[0] - r1 = None if order <= 1 else (lambda_inner[1] - lambda_inner[0]) / h - r2 = None if order <= 2 else (lambda_inner[2] - lambda_inner[0]) / h - x = self.singlestep_dpm_solver_update(x, vec_s, vec_t, order, solver_type=solver_type, r1=r1, r2=r2) - if denoise_to_zero: - x = self.denoise_to_zero_fn(x, torch.ones((x.shape[0],)).to(device) * t_0) - return x - - -############################################################# -# other utility functions -############################################################# - -def interpolate_fn(x, xp, yp): - """ - A piecewise linear function y = f(x), using xp and yp as keypoints. - We implement f(x) in a differentiable way (i.e. applicable for autograd). - The function f(x) is well-defined for all x-axis. (For x beyond the bounds of xp, we use the outmost points of xp to define the linear function.) - Args: - x: PyTorch tensor with shape [N, C], where N is the batch size, C is the number of channels (we use C = 1 for DPM-Solver). - xp: PyTorch tensor with shape [C, K], where K is the number of keypoints. - yp: PyTorch tensor with shape [C, K]. - Returns: - The function values f(x), with shape [N, C]. - """ - N, K = x.shape[0], xp.shape[1] - all_x = torch.cat([x.unsqueeze(2), xp.unsqueeze(0).repeat((N, 1, 1))], dim=2) - sorted_all_x, x_indices = torch.sort(all_x, dim=2) - x_idx = torch.argmin(x_indices, dim=2) - cand_start_idx = x_idx - 1 - start_idx = torch.where( - torch.eq(x_idx, 0), - torch.tensor(1, device=x.device), - torch.where( - torch.eq(x_idx, K), torch.tensor(K - 2, device=x.device), cand_start_idx, - ), - ) - end_idx = torch.where(torch.eq(start_idx, cand_start_idx), start_idx + 2, start_idx + 1) - start_x = torch.gather(sorted_all_x, dim=2, index=start_idx.unsqueeze(2)).squeeze(2) - end_x = torch.gather(sorted_all_x, dim=2, index=end_idx.unsqueeze(2)).squeeze(2) - start_idx2 = torch.where( - torch.eq(x_idx, 0), - torch.tensor(0, device=x.device), - torch.where( - torch.eq(x_idx, K), torch.tensor(K - 2, device=x.device), cand_start_idx, - ), - ) - y_positions_expanded = yp.unsqueeze(0).expand(N, -1, -1) - start_y = torch.gather(y_positions_expanded, dim=2, index=start_idx2.unsqueeze(2)).squeeze(2) - end_y = torch.gather(y_positions_expanded, dim=2, index=(start_idx2 + 1).unsqueeze(2)).squeeze(2) - cand = start_y + (x - start_x) * (end_y - start_y) / (end_x - start_x) - return cand - - -def expand_dims(v, dims): - """ - Expand the tensor `v` to the dim `dims`. - Args: - `v`: a PyTorch tensor with shape [N]. - `dim`: a `int`. - Returns: - a PyTorch tensor with shape [N, 1, 1, ..., 1] and the total dimension is `dims`. - """ - return v[(...,) + (None,) * (dims - 1)] \ No newline at end of file diff --git a/comfy/ldm/models/diffusion/dpm_solver/sampler.py b/comfy/ldm/models/diffusion/dpm_solver/sampler.py deleted file mode 100644 index e4d0d0a38..000000000 --- a/comfy/ldm/models/diffusion/dpm_solver/sampler.py +++ /dev/null @@ -1,96 +0,0 @@ -"""SAMPLING ONLY.""" -import torch - -from .dpm_solver import NoiseScheduleVP, model_wrapper, DPM_Solver - -MODEL_TYPES = { - "eps": "noise", - "v": "v" -} - - -class DPMSolverSampler(object): - def __init__(self, model, device=torch.device("cuda"), **kwargs): - super().__init__() - self.model = model - self.device = device - to_torch = lambda x: x.clone().detach().to(torch.float32).to(model.device) - self.register_buffer('alphas_cumprod', to_torch(model.alphas_cumprod)) - - def register_buffer(self, name, attr): - if type(attr) == torch.Tensor: - if attr.device != self.device: - attr = attr.to(self.device) - setattr(self, name, attr) - - @torch.no_grad() - def sample(self, - S, - batch_size, - shape, - conditioning=None, - callback=None, - normals_sequence=None, - img_callback=None, - quantize_x0=False, - eta=0., - mask=None, - x0=None, - temperature=1., - noise_dropout=0., - score_corrector=None, - corrector_kwargs=None, - verbose=True, - x_T=None, - log_every_t=100, - unconditional_guidance_scale=1., - unconditional_conditioning=None, - # this has to come in the same format as the conditioning, # e.g. as encoded tokens, ... - **kwargs - ): - if conditioning is not None: - if isinstance(conditioning, dict): - ctmp = conditioning[list(conditioning.keys())[0]] - while isinstance(ctmp, list): ctmp = ctmp[0] - if isinstance(ctmp, torch.Tensor): - cbs = ctmp.shape[0] - if cbs != batch_size: - print(f"Warning: Got {cbs} conditionings but batch-size is {batch_size}") - elif isinstance(conditioning, list): - for ctmp in conditioning: - if ctmp.shape[0] != batch_size: - print(f"Warning: Got {ctmp.shape[0]} conditionings but batch-size is {batch_size}") - else: - if isinstance(conditioning, torch.Tensor): - if conditioning.shape[0] != batch_size: - print(f"Warning: Got {conditioning.shape[0]} conditionings but batch-size is {batch_size}") - - # sampling - C, H, W = shape - size = (batch_size, C, H, W) - - print(f'Data shape for DPM-Solver sampling is {size}, sampling steps {S}') - - device = self.model.betas.device - if x_T is None: - img = torch.randn(size, device=device) - else: - img = x_T - - ns = NoiseScheduleVP('discrete', alphas_cumprod=self.alphas_cumprod) - - model_fn = model_wrapper( - lambda x, t, c: self.model.apply_model(x, t, c), - ns, - model_type=MODEL_TYPES[self.model.parameterization], - guidance_type="classifier-free", - condition=conditioning, - unconditional_condition=unconditional_conditioning, - guidance_scale=unconditional_guidance_scale, - ) - - dpm_solver = DPM_Solver(model_fn, ns, predict_x0=True, thresholding=False) - x = dpm_solver.sample(img, steps=S, skip_type="time_uniform", method="multistep", order=2, - lower_order_final=True) - - return x.to(device), None diff --git a/comfy/ldm/models/diffusion/plms.py b/comfy/ldm/models/diffusion/plms.py deleted file mode 100644 index 9d31b3994..000000000 --- a/comfy/ldm/models/diffusion/plms.py +++ /dev/null @@ -1,245 +0,0 @@ -"""SAMPLING ONLY.""" - -import torch -import numpy as np -from tqdm import tqdm -from functools import partial - -from ldm.modules.diffusionmodules.util import make_ddim_sampling_parameters, make_ddim_timesteps, noise_like -from ldm.models.diffusion.sampling_util import norm_thresholding - - -class PLMSSampler(object): - def __init__(self, model, schedule="linear", device=torch.device("cuda"), **kwargs): - super().__init__() - self.model = model - self.ddpm_num_timesteps = model.num_timesteps - self.schedule = schedule - self.device = device - - def register_buffer(self, name, attr): - if type(attr) == torch.Tensor: - if attr.device != self.device: - attr = attr.to(self.device) - setattr(self, name, attr) - - def make_schedule(self, ddim_num_steps, ddim_discretize="uniform", ddim_eta=0., verbose=True): - if ddim_eta != 0: - raise ValueError('ddim_eta must be 0 for PLMS') - self.ddim_timesteps = make_ddim_timesteps(ddim_discr_method=ddim_discretize, num_ddim_timesteps=ddim_num_steps, - num_ddpm_timesteps=self.ddpm_num_timesteps,verbose=verbose) - alphas_cumprod = self.model.alphas_cumprod - assert alphas_cumprod.shape[0] == self.ddpm_num_timesteps, 'alphas have to be defined for each timestep' - to_torch = lambda x: x.clone().detach().to(torch.float32).to(self.model.device) - - self.register_buffer('betas', to_torch(self.model.betas)) - self.register_buffer('alphas_cumprod', to_torch(alphas_cumprod)) - self.register_buffer('alphas_cumprod_prev', to_torch(self.model.alphas_cumprod_prev)) - - # calculations for diffusion q(x_t | x_{t-1}) and others - self.register_buffer('sqrt_alphas_cumprod', to_torch(np.sqrt(alphas_cumprod.cpu()))) - self.register_buffer('sqrt_one_minus_alphas_cumprod', to_torch(np.sqrt(1. - alphas_cumprod.cpu()))) - self.register_buffer('log_one_minus_alphas_cumprod', to_torch(np.log(1. - alphas_cumprod.cpu()))) - self.register_buffer('sqrt_recip_alphas_cumprod', to_torch(np.sqrt(1. / alphas_cumprod.cpu()))) - self.register_buffer('sqrt_recipm1_alphas_cumprod', to_torch(np.sqrt(1. / alphas_cumprod.cpu() - 1))) - - # ddim sampling parameters - ddim_sigmas, ddim_alphas, ddim_alphas_prev = make_ddim_sampling_parameters(alphacums=alphas_cumprod.cpu(), - ddim_timesteps=self.ddim_timesteps, - eta=ddim_eta,verbose=verbose) - self.register_buffer('ddim_sigmas', ddim_sigmas) - self.register_buffer('ddim_alphas', ddim_alphas) - self.register_buffer('ddim_alphas_prev', ddim_alphas_prev) - self.register_buffer('ddim_sqrt_one_minus_alphas', np.sqrt(1. - ddim_alphas)) - sigmas_for_original_sampling_steps = ddim_eta * torch.sqrt( - (1 - self.alphas_cumprod_prev) / (1 - self.alphas_cumprod) * ( - 1 - self.alphas_cumprod / self.alphas_cumprod_prev)) - self.register_buffer('ddim_sigmas_for_original_num_steps', sigmas_for_original_sampling_steps) - - @torch.no_grad() - def sample(self, - S, - batch_size, - shape, - conditioning=None, - callback=None, - normals_sequence=None, - img_callback=None, - quantize_x0=False, - eta=0., - mask=None, - x0=None, - temperature=1., - noise_dropout=0., - score_corrector=None, - corrector_kwargs=None, - verbose=True, - x_T=None, - log_every_t=100, - unconditional_guidance_scale=1., - unconditional_conditioning=None, - # this has to come in the same format as the conditioning, # e.g. as encoded tokens, ... - dynamic_threshold=None, - **kwargs - ): - if conditioning is not None: - if isinstance(conditioning, dict): - cbs = conditioning[list(conditioning.keys())[0]].shape[0] - if cbs != batch_size: - print(f"Warning: Got {cbs} conditionings but batch-size is {batch_size}") - else: - if conditioning.shape[0] != batch_size: - print(f"Warning: Got {conditioning.shape[0]} conditionings but batch-size is {batch_size}") - - self.make_schedule(ddim_num_steps=S, ddim_eta=eta, verbose=verbose) - # sampling - C, H, W = shape - size = (batch_size, C, H, W) - print(f'Data shape for PLMS sampling is {size}') - - samples, intermediates = self.plms_sampling(conditioning, size, - callback=callback, - img_callback=img_callback, - quantize_denoised=quantize_x0, - mask=mask, x0=x0, - ddim_use_original_steps=False, - noise_dropout=noise_dropout, - temperature=temperature, - score_corrector=score_corrector, - corrector_kwargs=corrector_kwargs, - x_T=x_T, - log_every_t=log_every_t, - unconditional_guidance_scale=unconditional_guidance_scale, - unconditional_conditioning=unconditional_conditioning, - dynamic_threshold=dynamic_threshold, - ) - return samples, intermediates - - @torch.no_grad() - def plms_sampling(self, cond, shape, - x_T=None, ddim_use_original_steps=False, - callback=None, timesteps=None, quantize_denoised=False, - mask=None, x0=None, img_callback=None, log_every_t=100, - temperature=1., noise_dropout=0., score_corrector=None, corrector_kwargs=None, - unconditional_guidance_scale=1., unconditional_conditioning=None, - dynamic_threshold=None): - device = self.model.betas.device - b = shape[0] - if x_T is None: - img = torch.randn(shape, device=device) - else: - img = x_T - - if timesteps is None: - timesteps = self.ddpm_num_timesteps if ddim_use_original_steps else self.ddim_timesteps - elif timesteps is not None and not ddim_use_original_steps: - subset_end = int(min(timesteps / self.ddim_timesteps.shape[0], 1) * self.ddim_timesteps.shape[0]) - 1 - timesteps = self.ddim_timesteps[:subset_end] - - intermediates = {'x_inter': [img], 'pred_x0': [img]} - time_range = list(reversed(range(0,timesteps))) if ddim_use_original_steps else np.flip(timesteps) - total_steps = timesteps if ddim_use_original_steps else timesteps.shape[0] - print(f"Running PLMS Sampling with {total_steps} timesteps") - - iterator = tqdm(time_range, desc='PLMS Sampler', total=total_steps) - old_eps = [] - - for i, step in enumerate(iterator): - index = total_steps - i - 1 - ts = torch.full((b,), step, device=device, dtype=torch.long) - ts_next = torch.full((b,), time_range[min(i + 1, len(time_range) - 1)], device=device, dtype=torch.long) - - if mask is not None: - assert x0 is not None - img_orig = self.model.q_sample(x0, ts) # TODO: deterministic forward pass? - img = img_orig * mask + (1. - mask) * img - - outs = self.p_sample_plms(img, cond, ts, index=index, use_original_steps=ddim_use_original_steps, - quantize_denoised=quantize_denoised, temperature=temperature, - noise_dropout=noise_dropout, score_corrector=score_corrector, - corrector_kwargs=corrector_kwargs, - unconditional_guidance_scale=unconditional_guidance_scale, - unconditional_conditioning=unconditional_conditioning, - old_eps=old_eps, t_next=ts_next, - dynamic_threshold=dynamic_threshold) - img, pred_x0, e_t = outs - old_eps.append(e_t) - if len(old_eps) >= 4: - old_eps.pop(0) - if callback: callback(i) - if img_callback: img_callback(pred_x0, i) - - if index % log_every_t == 0 or index == total_steps - 1: - intermediates['x_inter'].append(img) - intermediates['pred_x0'].append(pred_x0) - - return img, intermediates - - @torch.no_grad() - def p_sample_plms(self, x, c, t, index, repeat_noise=False, use_original_steps=False, quantize_denoised=False, - temperature=1., noise_dropout=0., score_corrector=None, corrector_kwargs=None, - unconditional_guidance_scale=1., unconditional_conditioning=None, old_eps=None, t_next=None, - dynamic_threshold=None): - b, *_, device = *x.shape, x.device - - def get_model_output(x, t): - if unconditional_conditioning is None or unconditional_guidance_scale == 1.: - e_t = self.model.apply_model(x, t, c) - else: - x_in = torch.cat([x] * 2) - t_in = torch.cat([t] * 2) - c_in = torch.cat([unconditional_conditioning, c]) - e_t_uncond, e_t = self.model.apply_model(x_in, t_in, c_in).chunk(2) - e_t = e_t_uncond + unconditional_guidance_scale * (e_t - e_t_uncond) - - if score_corrector is not None: - assert self.model.parameterization == "eps" - e_t = score_corrector.modify_score(self.model, e_t, x, t, c, **corrector_kwargs) - - return e_t - - alphas = self.model.alphas_cumprod if use_original_steps else self.ddim_alphas - alphas_prev = self.model.alphas_cumprod_prev if use_original_steps else self.ddim_alphas_prev - sqrt_one_minus_alphas = self.model.sqrt_one_minus_alphas_cumprod if use_original_steps else self.ddim_sqrt_one_minus_alphas - sigmas = self.model.ddim_sigmas_for_original_num_steps if use_original_steps else self.ddim_sigmas - - def get_x_prev_and_pred_x0(e_t, index): - # select parameters corresponding to the currently considered timestep - a_t = torch.full((b, 1, 1, 1), alphas[index], device=device) - a_prev = torch.full((b, 1, 1, 1), alphas_prev[index], device=device) - sigma_t = torch.full((b, 1, 1, 1), sigmas[index], device=device) - sqrt_one_minus_at = torch.full((b, 1, 1, 1), sqrt_one_minus_alphas[index],device=device) - - # current prediction for x_0 - pred_x0 = (x - sqrt_one_minus_at * e_t) / a_t.sqrt() - if quantize_denoised: - pred_x0, _, *_ = self.model.first_stage_model.quantize(pred_x0) - if dynamic_threshold is not None: - pred_x0 = norm_thresholding(pred_x0, dynamic_threshold) - # direction pointing to x_t - dir_xt = (1. - a_prev - sigma_t**2).sqrt() * e_t - noise = sigma_t * noise_like(x.shape, device, repeat_noise) * temperature - if noise_dropout > 0.: - noise = torch.nn.functional.dropout(noise, p=noise_dropout) - x_prev = a_prev.sqrt() * pred_x0 + dir_xt + noise - return x_prev, pred_x0 - - e_t = get_model_output(x, t) - if len(old_eps) == 0: - # Pseudo Improved Euler (2nd order) - x_prev, pred_x0 = get_x_prev_and_pred_x0(e_t, index) - e_t_next = get_model_output(x_prev, t_next) - e_t_prime = (e_t + e_t_next) / 2 - elif len(old_eps) == 1: - # 2nd order Pseudo Linear Multistep (Adams-Bashforth) - e_t_prime = (3 * e_t - old_eps[-1]) / 2 - elif len(old_eps) == 2: - # 3nd order Pseudo Linear Multistep (Adams-Bashforth) - e_t_prime = (23 * e_t - 16 * old_eps[-1] + 5 * old_eps[-2]) / 12 - elif len(old_eps) >= 3: - # 4nd order Pseudo Linear Multistep (Adams-Bashforth) - e_t_prime = (55 * e_t - 59 * old_eps[-1] + 37 * old_eps[-2] - 9 * old_eps[-3]) / 24 - - x_prev, pred_x0 = get_x_prev_and_pred_x0(e_t_prime, index) - - return x_prev, pred_x0, e_t diff --git a/comfy/ldm/models/diffusion/sampling_util.py b/comfy/ldm/models/diffusion/sampling_util.py deleted file mode 100644 index 7eff02be6..000000000 --- a/comfy/ldm/models/diffusion/sampling_util.py +++ /dev/null @@ -1,22 +0,0 @@ -import torch -import numpy as np - - -def append_dims(x, target_dims): - """Appends dimensions to the end of a tensor until it has target_dims dimensions. - From https://github.com/crowsonkb/k-diffusion/blob/master/k_diffusion/utils.py""" - dims_to_append = target_dims - x.ndim - if dims_to_append < 0: - raise ValueError(f'input has {x.ndim} dims but target_dims is {target_dims}, which is less') - return x[(...,) + (None,) * dims_to_append] - - -def norm_thresholding(x0, value): - s = append_dims(x0.pow(2).flatten(1).mean(1).sqrt().clamp(min=value), x0.ndim) - return x0 * (value / s) - - -def spatial_norm_thresholding(x0, value): - # b c h w - s = x0.pow(2).mean(1, keepdim=True).sqrt().clamp(min=value) - return x0 * (value / s) \ No newline at end of file diff --git a/comfy/ldm/modules/attention.py b/comfy/ldm/modules/attention.py index 4f10bbc35..016795a59 100644 --- a/comfy/ldm/modules/attention.py +++ b/comfy/ldm/modules/attention.py @@ -160,32 +160,19 @@ def attention_sub_quad(query, key, value, heads, mask=None): mem_free_total, mem_free_torch = model_management.get_free_memory(query.device, True) - chunk_threshold_bytes = mem_free_torch * 0.5 #Using only this seems to work better on AMD - kv_chunk_size_min = None + kv_chunk_size = None + query_chunk_size = None - #not sure at all about the math here - #TODO: tweak this - if mem_free_total > 8192 * 1024 * 1024 * 1.3: - query_chunk_size_x = 1024 * 4 - elif mem_free_total > 4096 * 1024 * 1024 * 1.3: - query_chunk_size_x = 1024 * 2 - else: - query_chunk_size_x = 1024 - kv_chunk_size_min_x = None - kv_chunk_size_x = (int((chunk_threshold_bytes // (batch_x_heads * bytes_per_token * query_chunk_size_x)) * 2.0) // 1024) * 1024 - if kv_chunk_size_x < 1024: - kv_chunk_size_x = None + for x in [4096, 2048, 1024, 512, 256]: + count = mem_free_total / (batch_x_heads * bytes_per_token * x * 4.0) + if count >= k_tokens: + kv_chunk_size = k_tokens + query_chunk_size = x + break - if chunk_threshold_bytes is not None and qk_matmul_size_bytes <= chunk_threshold_bytes: - # the big matmul fits into our memory limit; do everything in 1 chunk, - # i.e. send it down the unchunked fast-path - query_chunk_size = q_tokens - kv_chunk_size = k_tokens - else: - query_chunk_size = query_chunk_size_x - kv_chunk_size = kv_chunk_size_x - kv_chunk_size_min = kv_chunk_size_min_x + if query_chunk_size is None: + query_chunk_size = 512 hidden_states = efficient_dot_product_attention( query, @@ -229,7 +216,7 @@ def attention_split(q, k, v, heads, mask=None): gb = 1024 ** 3 tensor_size = q.shape[0] * q.shape[1] * k.shape[1] * element_size - modifier = 3 if element_size == 2 else 2.5 + modifier = 3 mem_required = tensor_size * modifier steps = 1 @@ -257,10 +244,10 @@ def attention_split(q, k, v, heads, mask=None): s1 = einsum('b i d, b j d -> b i j', q[:, i:end].float(), k.float()) * scale else: s1 = einsum('b i d, b j d -> b i j', q[:, i:end], k) * scale - first_op_done = True s2 = s1.softmax(dim=-1).to(v.dtype) del s1 + first_op_done = True r1[:, i:end] = einsum('b i j, b j d -> b i d', s2, v) del s2 diff --git a/comfy/ldm/modules/diffusionmodules/openaimodel.py b/comfy/ldm/modules/diffusionmodules/openaimodel.py index 7dfdfc0a2..49c1e8cbb 100644 --- a/comfy/ldm/modules/diffusionmodules/openaimodel.py +++ b/comfy/ldm/modules/diffusionmodules/openaimodel.py @@ -251,6 +251,12 @@ class Timestep(nn.Module): def forward(self, t): return timestep_embedding(t, self.dim) +def apply_control(h, control, name): + if control is not None and name in control and len(control[name]) > 0: + ctrl = control[name].pop() + if ctrl is not None: + h += ctrl + return h class UNetModel(nn.Module): """ @@ -617,25 +623,17 @@ class UNetModel(nn.Module): for id, module in enumerate(self.input_blocks): transformer_options["block"] = ("input", id) h = forward_timestep_embed(module, h, emb, context, transformer_options) - if control is not None and 'input' in control and len(control['input']) > 0: - ctrl = control['input'].pop() - if ctrl is not None: - h += ctrl + h = apply_control(h, control, 'input') hs.append(h) + transformer_options["block"] = ("middle", 0) h = forward_timestep_embed(self.middle_block, h, emb, context, transformer_options) - if control is not None and 'middle' in control and len(control['middle']) > 0: - ctrl = control['middle'].pop() - if ctrl is not None: - h += ctrl + h = apply_control(h, control, 'middle') for id, module in enumerate(self.output_blocks): transformer_options["block"] = ("output", id) hsp = hs.pop() - if control is not None and 'output' in control and len(control['output']) > 0: - ctrl = control['output'].pop() - if ctrl is not None: - hsp += ctrl + hsp = apply_control(hsp, control, 'output') if "output_block_patch" in transformer_patches: patch = transformer_patches["output_block_patch"] diff --git a/comfy/ldm/modules/diffusionmodules/util.py b/comfy/ldm/modules/diffusionmodules/util.py index d890c8044..0298ca99d 100644 --- a/comfy/ldm/modules/diffusionmodules/util.py +++ b/comfy/ldm/modules/diffusionmodules/util.py @@ -170,8 +170,8 @@ def timestep_embedding(timesteps, dim, max_period=10000, repeat_only=False): if not repeat_only: half = dim // 2 freqs = torch.exp( - -math.log(max_period) * torch.arange(start=0, end=half, dtype=torch.float32) / half - ).to(device=timesteps.device) + -math.log(max_period) * torch.arange(start=0, end=half, dtype=torch.float32, device=timesteps.device) / half + ) args = timesteps[:, None].float() * freqs[None] embedding = torch.cat([torch.cos(args), torch.sin(args)], dim=-1) if dim % 2: diff --git a/comfy/lora.py b/comfy/lora.py index d4cf94c95..29c59d893 100644 --- a/comfy/lora.py +++ b/comfy/lora.py @@ -131,6 +131,18 @@ def load_lora(lora, to_load): loaded_keys.add(b_norm_name) patch_dict["{}.bias".format(to_load[x][:-len(".weight")])] = (b_norm,) + diff_name = "{}.diff".format(x) + diff_weight = lora.get(diff_name, None) + if diff_weight is not None: + patch_dict[to_load[x]] = (diff_weight,) + loaded_keys.add(diff_name) + + diff_bias_name = "{}.diff_b".format(x) + diff_bias = lora.get(diff_bias_name, None) + if diff_bias is not None: + patch_dict["{}.bias".format(to_load[x][:-len(".weight")])] = (diff_bias,) + loaded_keys.add(diff_bias_name) + for x in lora.keys(): if x not in loaded_keys: print("lora key not loaded", x) diff --git a/comfy/model_base.py b/comfy/model_base.py index ea3ea61f2..7ba253470 100644 --- a/comfy/model_base.py +++ b/comfy/model_base.py @@ -1,11 +1,9 @@ import torch from comfy.ldm.modules.diffusionmodules.openaimodel import UNetModel from comfy.ldm.modules.encoders.noise_aug_modules import CLIPEmbeddingNoiseAugmentation -from comfy.ldm.modules.diffusionmodules.util import make_beta_schedule from comfy.ldm.modules.diffusionmodules.openaimodel import Timestep import comfy.model_management import comfy.conds -import numpy as np from enum import Enum from . import utils @@ -13,6 +11,23 @@ class ModelType(Enum): EPS = 1 V_PREDICTION = 2 + +from comfy.model_sampling import EPS, V_PREDICTION, ModelSamplingDiscrete + +def model_sampling(model_config, model_type): + if model_type == ModelType.EPS: + c = EPS + elif model_type == ModelType.V_PREDICTION: + c = V_PREDICTION + + s = ModelSamplingDiscrete + + class ModelSampling(s, c): + pass + + return ModelSampling(model_config) + + class BaseModel(torch.nn.Module): def __init__(self, model_config, model_type=ModelType.EPS, device=None): super().__init__() @@ -20,10 +35,12 @@ class BaseModel(torch.nn.Module): unet_config = model_config.unet_config self.latent_format = model_config.latent_format self.model_config = model_config - self.register_schedule(given_betas=None, beta_schedule=model_config.beta_schedule, timesteps=1000, linear_start=0.00085, linear_end=0.012, cosine_s=8e-3) + if not unet_config.get("disable_unet_model_creation", False): self.diffusion_model = UNetModel(**unet_config, device=device) self.model_type = model_type + self.model_sampling = model_sampling(model_config, model_type) + self.adm_channels = unet_config.get("adm_in_channels", None) if self.adm_channels is None: self.adm_channels = 0 @@ -31,39 +48,25 @@ class BaseModel(torch.nn.Module): print("model_type", model_type.name) print("adm", self.adm_channels) - def register_schedule(self, given_betas=None, beta_schedule="linear", timesteps=1000, - linear_start=1e-4, linear_end=2e-2, cosine_s=8e-3): - if given_betas is not None: - betas = given_betas - else: - betas = make_beta_schedule(beta_schedule, timesteps, linear_start=linear_start, linear_end=linear_end, cosine_s=cosine_s) - alphas = 1. - betas - alphas_cumprod = np.cumprod(alphas, axis=0) - alphas_cumprod_prev = np.append(1., alphas_cumprod[:-1]) - - timesteps, = betas.shape - self.num_timesteps = int(timesteps) - self.linear_start = linear_start - self.linear_end = linear_end - - self.register_buffer('betas', torch.tensor(betas, dtype=torch.float32)) - self.register_buffer('alphas_cumprod', torch.tensor(alphas_cumprod, dtype=torch.float32)) - self.register_buffer('alphas_cumprod_prev', torch.tensor(alphas_cumprod_prev, dtype=torch.float32)) - def apply_model(self, x, t, c_concat=None, c_crossattn=None, control=None, transformer_options={}, **kwargs): + sigma = t + xc = self.model_sampling.calculate_input(sigma, x) if c_concat is not None: - xc = torch.cat([x] + [c_concat], dim=1) - else: - xc = x + xc = torch.cat([xc] + [c_concat], dim=1) + context = c_crossattn dtype = self.get_dtype() xc = xc.to(dtype) - t = t.to(dtype) + t = self.model_sampling.timestep(t).float() context = context.to(dtype) extra_conds = {} for o in kwargs: - extra_conds[o] = kwargs[o].to(dtype) - return self.diffusion_model(xc, t, context=context, control=control, transformer_options=transformer_options, **extra_conds).float() + extra = kwargs[o] + if hasattr(extra, "to"): + extra = extra.to(dtype) + extra_conds[o] = extra + model_output = self.diffusion_model(xc, t, context=context, control=control, transformer_options=transformer_options, **extra_conds).float() + return self.model_sampling.calculate_denoised(sigma, model_output, x) def get_dtype(self): return self.diffusion_model.dtype diff --git a/comfy/model_patcher.py b/comfy/model_patcher.py index 50b725b86..6d7a61c41 100644 --- a/comfy/model_patcher.py +++ b/comfy/model_patcher.py @@ -6,11 +6,13 @@ import comfy.utils import comfy.model_management class ModelPatcher: - def __init__(self, model, load_device, offload_device, size=0, current_device=None): + def __init__(self, model, load_device, offload_device, size=0, current_device=None, weight_inplace_update=False): self.size = size self.model = model self.patches = {} self.backup = {} + self.object_patches = {} + self.object_patches_backup = {} self.model_options = {"transformer_options":{}} self.model_size() self.load_device = load_device @@ -20,6 +22,8 @@ class ModelPatcher: else: self.current_device = current_device + self.weight_inplace_update = weight_inplace_update + def model_size(self): if self.size > 0: return self.size @@ -38,6 +42,7 @@ class ModelPatcher: for k in self.patches: n.patches[k] = self.patches[k][:] + n.object_patches = self.object_patches.copy() n.model_options = copy.deepcopy(self.model_options) n.model_keys = self.model_keys return n @@ -91,6 +96,9 @@ class ModelPatcher: def set_model_output_block_patch(self, patch): self.set_model_patch(patch, "output_block_patch") + def add_object_patch(self, name, obj): + self.object_patches[name] = obj + def model_patches_to(self, device): to = self.model_options["transformer_options"] if "patches" in to: @@ -107,10 +115,10 @@ class ModelPatcher: for k in patch_list: if hasattr(patch_list[k], "to"): patch_list[k] = patch_list[k].to(device) - if "unet_wrapper_function" in self.model_options: - wrap_func = self.model_options["unet_wrapper_function"] + if "model_function_wrapper" in self.model_options: + wrap_func = self.model_options["model_function_wrapper"] if hasattr(wrap_func, "to"): - self.model_options["unet_wrapper_function"] = wrap_func.to(device) + self.model_options["model_function_wrapper"] = wrap_func.to(device) def model_dtype(self): if hasattr(self.model, "get_dtype"): @@ -128,6 +136,7 @@ class ModelPatcher: return list(p) def get_key_patches(self, filter_prefix=None): + comfy.model_management.unload_model_clones(self) model_sd = self.model_state_dict() p = {} for k in model_sd: @@ -150,6 +159,12 @@ class ModelPatcher: return sd def patch_model(self, device_to=None): + for k in self.object_patches: + old = getattr(self.model, k) + if k not in self.object_patches_backup: + self.object_patches_backup[k] = old + setattr(self.model, k, self.object_patches[k]) + model_sd = self.model_state_dict() for key in self.patches: if key not in model_sd: @@ -158,15 +173,20 @@ class ModelPatcher: weight = model_sd[key] + inplace_update = self.weight_inplace_update + if key not in self.backup: - self.backup[key] = weight.to(self.offload_device) + self.backup[key] = weight.to(device=device_to, copy=inplace_update) if device_to is not None: temp_weight = comfy.model_management.cast_to_device(weight, device_to, torch.float32, copy=True) else: temp_weight = weight.to(torch.float32, copy=True) out_weight = self.calculate_weight(self.patches[key], temp_weight, key).to(weight.dtype) - comfy.utils.set_attr(self.model, key, out_weight) + if inplace_update: + comfy.utils.copy_to_param(self.model, key, out_weight) + else: + comfy.utils.set_attr(self.model, key, out_weight) del temp_weight if device_to is not None: @@ -282,11 +302,21 @@ class ModelPatcher: def unpatch_model(self, device_to=None): keys = list(self.backup.keys()) - for k in keys: - comfy.utils.set_attr(self.model, k, self.backup[k]) + if self.weight_inplace_update: + for k in keys: + comfy.utils.copy_to_param(self.model, k, self.backup[k]) + else: + for k in keys: + comfy.utils.set_attr(self.model, k, self.backup[k]) self.backup = {} if device_to is not None: self.model.to(device_to) self.current_device = device_to + + keys = list(self.object_patches_backup.keys()) + for k in keys: + setattr(self.model, k, self.object_patches_backup[k]) + + self.object_patches_backup = {} diff --git a/comfy/model_sampling.py b/comfy/model_sampling.py new file mode 100644 index 000000000..a2935d47d --- /dev/null +++ b/comfy/model_sampling.py @@ -0,0 +1,80 @@ +import torch +import numpy as np +from comfy.ldm.modules.diffusionmodules.util import make_beta_schedule + + +class EPS: + def calculate_input(self, sigma, noise): + sigma = sigma.view(sigma.shape[:1] + (1,) * (noise.ndim - 1)) + return noise / (sigma ** 2 + self.sigma_data ** 2) ** 0.5 + + def calculate_denoised(self, sigma, model_output, model_input): + sigma = sigma.view(sigma.shape[:1] + (1,) * (model_output.ndim - 1)) + return model_input - model_output * sigma + + +class V_PREDICTION(EPS): + def calculate_denoised(self, sigma, model_output, model_input): + sigma = sigma.view(sigma.shape[:1] + (1,) * (model_output.ndim - 1)) + return model_input * self.sigma_data ** 2 / (sigma ** 2 + self.sigma_data ** 2) - model_output * sigma * self.sigma_data / (sigma ** 2 + self.sigma_data ** 2) ** 0.5 + + +class ModelSamplingDiscrete(torch.nn.Module): + def __init__(self, model_config=None): + super().__init__() + beta_schedule = "linear" + if model_config is not None: + beta_schedule = model_config.beta_schedule + self._register_schedule(given_betas=None, beta_schedule=beta_schedule, timesteps=1000, linear_start=0.00085, linear_end=0.012, cosine_s=8e-3) + self.sigma_data = 1.0 + + def _register_schedule(self, given_betas=None, beta_schedule="linear", timesteps=1000, + linear_start=1e-4, linear_end=2e-2, cosine_s=8e-3): + if given_betas is not None: + betas = given_betas + else: + betas = make_beta_schedule(beta_schedule, timesteps, linear_start=linear_start, linear_end=linear_end, cosine_s=cosine_s) + alphas = 1. - betas + alphas_cumprod = torch.tensor(np.cumprod(alphas, axis=0), dtype=torch.float32) + # alphas_cumprod_prev = np.append(1., alphas_cumprod[:-1]) + + timesteps, = betas.shape + self.num_timesteps = int(timesteps) + self.linear_start = linear_start + self.linear_end = linear_end + + # self.register_buffer('betas', torch.tensor(betas, dtype=torch.float32)) + # self.register_buffer('alphas_cumprod', torch.tensor(alphas_cumprod, dtype=torch.float32)) + # self.register_buffer('alphas_cumprod_prev', torch.tensor(alphas_cumprod_prev, dtype=torch.float32)) + + sigmas = ((1 - alphas_cumprod) / alphas_cumprod) ** 0.5 + self.set_sigmas(sigmas) + + def set_sigmas(self, sigmas): + self.register_buffer('sigmas', sigmas) + self.register_buffer('log_sigmas', sigmas.log()) + + @property + def sigma_min(self): + return self.sigmas[0] + + @property + def sigma_max(self): + return self.sigmas[-1] + + def timestep(self, sigma): + log_sigma = sigma.log() + dists = log_sigma.to(self.log_sigmas.device) - self.log_sigmas[:, None] + return dists.abs().argmin(dim=0).view(sigma.shape) + + def sigma(self, timestep): + t = torch.clamp(timestep.float(), min=0, max=(len(self.sigmas) - 1)) + low_idx = t.floor().long() + high_idx = t.ceil().long() + w = t.frac() + log_sigma = (1 - w) * self.log_sigmas[low_idx] + w * self.log_sigmas[high_idx] + return log_sigma.exp() + + def percent_to_sigma(self, percent): + return self.sigma(torch.tensor(percent * 999.0)) + diff --git a/comfy/ops.py b/comfy/ops.py index 610d54584..0bfb698aa 100644 --- a/comfy/ops.py +++ b/comfy/ops.py @@ -1,29 +1,23 @@ import torch from contextlib import contextmanager -class Linear(torch.nn.Module): - def __init__(self, in_features: int, out_features: int, bias: bool = True, - device=None, dtype=None) -> None: - factory_kwargs = {'device': device, 'dtype': dtype} - super().__init__() - self.in_features = in_features - self.out_features = out_features - self.weight = torch.nn.Parameter(torch.empty((out_features, in_features), **factory_kwargs)) - if bias: - self.bias = torch.nn.Parameter(torch.empty(out_features, **factory_kwargs)) - else: - self.register_parameter('bias', None) - - def forward(self, input): - return torch.nn.functional.linear(input, self.weight, self.bias) +class Linear(torch.nn.Linear): + def reset_parameters(self): + return None class Conv2d(torch.nn.Conv2d): def reset_parameters(self): return None +class Conv3d(torch.nn.Conv3d): + def reset_parameters(self): + return None + def conv_nd(dims, *args, **kwargs): if dims == 2: return Conv2d(*args, **kwargs) + elif dims == 3: + return Conv3d(*args, **kwargs) else: raise ValueError(f"unsupported dimensions: {dims}") diff --git a/comfy/samplers.py b/comfy/samplers.py index f930aa39b..a839ee9e2 100644 --- a/comfy/samplers.py +++ b/comfy/samplers.py @@ -1,11 +1,8 @@ from .k_diffusion import sampling as k_diffusion_sampling -from .k_diffusion import external as k_diffusion_external from .extra_samplers import uni_pc import torch import enum from comfy import model_management -from .ldm.models.diffusion.ddim import DDIMSampler -from .ldm.modules.diffusionmodules.util import make_ddim_timesteps import math from comfy import model_base import comfy.utils @@ -13,7 +10,7 @@ import comfy.conds #The main sampling function shared by all the samplers -#Returns predicted noise +#Returns denoised def sampling_function(model_function, x, timestep, uncond, cond, cond_scale, model_options={}, seed=None): def get_area_and_mult(conds, x_in, timestep_in): area = (x_in.shape[2], x_in.shape[3], 0, 0) @@ -139,10 +136,10 @@ def sampling_function(model_function, x, timestep, uncond, cond, cond_scale, mod def calc_cond_uncond_batch(model_function, cond, uncond, x_in, timestep, max_total_area, model_options): out_cond = torch.zeros_like(x_in) - out_count = torch.ones_like(x_in)/100000.0 + out_count = torch.ones_like(x_in) * 1e-37 out_uncond = torch.zeros_like(x_in) - out_uncond_count = torch.ones_like(x_in)/100000.0 + out_uncond_count = torch.ones_like(x_in) * 1e-37 COND = 0 UNCOND = 1 @@ -242,7 +239,6 @@ def sampling_function(model_function, x, timestep, uncond, cond, cond_scale, mod del out_count out_uncond /= out_uncond_count del out_uncond_count - return out_cond, out_uncond @@ -252,29 +248,20 @@ def sampling_function(model_function, x, timestep, uncond, cond, cond_scale, mod cond, uncond = calc_cond_uncond_batch(model_function, cond, uncond, x, timestep, max_total_area, model_options) if "sampler_cfg_function" in model_options: - args = {"cond": cond, "uncond": uncond, "cond_scale": cond_scale, "timestep": timestep} - return model_options["sampler_cfg_function"](args) + args = {"cond": x - cond, "uncond": x - uncond, "cond_scale": cond_scale, "timestep": timestep, "input": x, "sigma": timestep} + return x - model_options["sampler_cfg_function"](args) else: return uncond + (cond - uncond) * cond_scale - -class CompVisVDenoiser(k_diffusion_external.DiscreteVDDPMDenoiser): - def __init__(self, model, quantize=False, device='cpu'): - super().__init__(model, model.alphas_cumprod, quantize=quantize) - - def get_v(self, x, t, cond, **kwargs): - return self.inner_model.apply_model(x, t, cond, **kwargs) - - class CFGNoisePredictor(torch.nn.Module): def __init__(self, model): super().__init__() self.inner_model = model - self.alphas_cumprod = model.alphas_cumprod def apply_model(self, x, timestep, cond, uncond, cond_scale, model_options={}, seed=None): out = sampling_function(self.inner_model.apply_model, x, timestep, uncond, cond, cond_scale, model_options=model_options, seed=seed) return out - + def forward(self, *args, **kwargs): + return self.apply_model(*args, **kwargs) class KSamplerX0Inpaint(torch.nn.Module): def __init__(self, model): @@ -293,32 +280,40 @@ class KSamplerX0Inpaint(torch.nn.Module): return out def simple_scheduler(model, steps): + s = model.model_sampling sigs = [] - ss = len(model.sigmas) / steps + ss = len(s.sigmas) / steps for x in range(steps): - sigs += [float(model.sigmas[-(1 + int(x * ss))])] + sigs += [float(s.sigmas[-(1 + int(x * ss))])] sigs += [0.0] return torch.FloatTensor(sigs) def ddim_scheduler(model, steps): + s = model.model_sampling sigs = [] - ddim_timesteps = make_ddim_timesteps(ddim_discr_method="uniform", num_ddim_timesteps=steps, num_ddpm_timesteps=model.inner_model.inner_model.num_timesteps, verbose=False) - for x in range(len(ddim_timesteps) - 1, -1, -1): - ts = ddim_timesteps[x] - if ts > 999: - ts = 999 - sigs.append(model.t_to_sigma(torch.tensor(ts))) + ss = len(s.sigmas) // steps + x = 1 + while x < len(s.sigmas): + sigs += [float(s.sigmas[x])] + x += ss + sigs = sigs[::-1] sigs += [0.0] return torch.FloatTensor(sigs) -def sgm_scheduler(model, steps): +def normal_scheduler(model, steps, sgm=False, floor=False): + s = model.model_sampling + start = s.timestep(s.sigma_max) + end = s.timestep(s.sigma_min) + + if sgm: + timesteps = torch.linspace(start, end, steps + 1)[:-1] + else: + timesteps = torch.linspace(start, end, steps) + sigs = [] - timesteps = torch.linspace(model.inner_model.inner_model.num_timesteps - 1, 0, steps + 1)[:-1].type(torch.int) for x in range(len(timesteps)): ts = timesteps[x] - if ts > 999: - ts = 999 - sigs.append(model.t_to_sigma(torch.tensor(ts))) + sigs.append(s.sigma(ts)) sigs += [0.0] return torch.FloatTensor(sigs) @@ -418,15 +413,16 @@ def create_cond_with_same_area_if_none(conds, c): conds += [out] def calculate_start_end_timesteps(model, conds): + s = model.model_sampling for t in range(len(conds)): x = conds[t] timestep_start = None timestep_end = None if 'start_percent' in x: - timestep_start = model.sigma_to_t(model.t_to_sigma(torch.tensor(x['start_percent'] * 999.0))) + timestep_start = s.percent_to_sigma(x['start_percent']) if 'end_percent' in x: - timestep_end = model.sigma_to_t(model.t_to_sigma(torch.tensor(x['end_percent'] * 999.0))) + timestep_end = s.percent_to_sigma(x['end_percent']) if (timestep_start is not None) or (timestep_end is not None): n = x.copy() @@ -437,14 +433,15 @@ def calculate_start_end_timesteps(model, conds): conds[t] = n def pre_run_control(model, conds): + s = model.model_sampling for t in range(len(conds)): x = conds[t] timestep_start = None timestep_end = None - percent_to_timestep_function = lambda a: model.sigma_to_t(model.t_to_sigma(torch.tensor(a) * 999.0)) + percent_to_timestep_function = lambda a: s.percent_to_sigma(a) if 'control' in x: - x['control'].pre_run(model.inner_model.inner_model, percent_to_timestep_function) + x['control'].pre_run(model, percent_to_timestep_function) def apply_empty_x_to_equal_area(conds, uncond, name, uncond_fill_func): cond_cnets = [] @@ -508,42 +505,9 @@ class Sampler: pass def max_denoise(self, model_wrap, sigmas): - return math.isclose(float(model_wrap.sigma_max), float(sigmas[0]), rel_tol=1e-05) - -class DDIM(Sampler): - def sample(self, model_wrap, sigmas, extra_args, callback, noise, latent_image=None, denoise_mask=None, disable_pbar=False): - timesteps = [] - for s in range(sigmas.shape[0]): - timesteps.insert(0, model_wrap.sigma_to_discrete_timestep(sigmas[s])) - noise_mask = None - if denoise_mask is not None: - noise_mask = 1.0 - denoise_mask - - ddim_callback = None - if callback is not None: - total_steps = len(timesteps) - 1 - ddim_callback = lambda pred_x0, i: callback(i, pred_x0, None, total_steps) - - max_denoise = self.max_denoise(model_wrap, sigmas) - - ddim_sampler = DDIMSampler(model_wrap.inner_model.inner_model, device=noise.device) - ddim_sampler.make_schedule_timesteps(ddim_timesteps=timesteps, verbose=False) - z_enc = ddim_sampler.stochastic_encode(latent_image, torch.tensor([len(timesteps) - 1] * noise.shape[0]).to(noise.device), noise=noise, max_denoise=max_denoise) - samples, _ = ddim_sampler.sample_custom(ddim_timesteps=timesteps, - batch_size=noise.shape[0], - shape=noise.shape[1:], - verbose=False, - eta=0.0, - x_T=z_enc, - x0=latent_image, - img_callback=ddim_callback, - denoise_function=model_wrap.predict_eps_discrete_timestep, - extra_args=extra_args, - mask=noise_mask, - to_zero=sigmas[-1]==0, - end_step=sigmas.shape[0] - 1, - disable_pbar=disable_pbar) - return samples + max_sigma = float(model_wrap.inner_model.model_sampling.sigma_max) + sigma = float(sigmas[0]) + return math.isclose(max_sigma, sigma, rel_tol=1e-05) or sigma > max_sigma class UNIPC(Sampler): def sample(self, model_wrap, sigmas, extra_args, callback, noise, latent_image=None, denoise_mask=None, disable_pbar=False): @@ -555,15 +519,19 @@ class UNIPCBH2(Sampler): KSAMPLER_NAMES = ["euler", "euler_ancestral", "heun", "dpm_2", "dpm_2_ancestral", "lms", "dpm_fast", "dpm_adaptive", "dpmpp_2s_ancestral", "dpmpp_sde", "dpmpp_sde_gpu", - "dpmpp_2m", "dpmpp_2m_sde", "dpmpp_2m_sde_gpu", "dpmpp_3m_sde", "dpmpp_3m_sde_gpu", "ddpm"] + "dpmpp_2m", "dpmpp_2m_sde", "dpmpp_2m_sde_gpu", "dpmpp_3m_sde", "dpmpp_3m_sde_gpu", "ddpm", "lcm"] -def ksampler(sampler_name, extra_options={}): +def ksampler(sampler_name, extra_options={}, inpaint_options={}): class KSAMPLER(Sampler): def sample(self, model_wrap, sigmas, extra_args, callback, noise, latent_image=None, denoise_mask=None, disable_pbar=False): extra_args["denoise_mask"] = denoise_mask model_k = KSamplerX0Inpaint(model_wrap) model_k.latent_image = latent_image - model_k.noise = noise + if inpaint_options.get("random", False): #TODO: Should this be the default? + generator = torch.manual_seed(extra_args.get("seed", 41) + 1) + model_k.noise = torch.randn(noise.shape, generator=generator, device="cpu").to(noise.dtype).to(noise.device) + else: + model_k.noise = noise if self.max_denoise(model_wrap, sigmas): noise = noise * torch.sqrt(1.0 + sigmas[0] ** 2.0) @@ -592,11 +560,7 @@ def ksampler(sampler_name, extra_options={}): def wrap_model(model): model_denoise = CFGNoisePredictor(model) - if model.model_type == model_base.ModelType.V_PREDICTION: - model_wrap = CompVisVDenoiser(model_denoise, quantize=True) - else: - model_wrap = k_diffusion_external.CompVisDenoiser(model_denoise, quantize=True) - return model_wrap + return model_denoise def sample(model, noise, positive, negative, cfg, device, sampler, sigmas, model_options={}, latent_image=None, denoise_mask=None, callback=None, disable_pbar=False, seed=None): positive = positive[:] @@ -607,8 +571,8 @@ def sample(model, noise, positive, negative, cfg, device, sampler, sigmas, model model_wrap = wrap_model(model) - calculate_start_end_timesteps(model_wrap, negative) - calculate_start_end_timesteps(model_wrap, positive) + calculate_start_end_timesteps(model, negative) + calculate_start_end_timesteps(model, positive) #make sure each cond area has an opposite one with the same area for c in positive: @@ -616,7 +580,7 @@ def sample(model, noise, positive, negative, cfg, device, sampler, sigmas, model for c in negative: create_cond_with_same_area_if_none(positive, c) - pre_run_control(model_wrap, negative + positive) + pre_run_control(model, negative + positive) apply_empty_x_to_equal_area(list(filter(lambda c: c.get('control_apply_to_uncond', False) == True, positive)), negative, 'control', lambda cond_cnets, x: cond_cnets[x]) apply_empty_x_to_equal_area(positive, negative, 'gligen', lambda cond_cnets, x: cond_cnets[x]) @@ -637,19 +601,18 @@ SCHEDULER_NAMES = ["normal", "karras", "exponential", "sgm_uniform", "simple", " SAMPLER_NAMES = KSAMPLER_NAMES + ["ddim", "uni_pc", "uni_pc_bh2"] def calculate_sigmas_scheduler(model, scheduler_name, steps): - model_wrap = wrap_model(model) if scheduler_name == "karras": - sigmas = k_diffusion_sampling.get_sigmas_karras(n=steps, sigma_min=float(model_wrap.sigma_min), sigma_max=float(model_wrap.sigma_max)) + sigmas = k_diffusion_sampling.get_sigmas_karras(n=steps, sigma_min=float(model.model_sampling.sigma_min), sigma_max=float(model.model_sampling.sigma_max)) elif scheduler_name == "exponential": - sigmas = k_diffusion_sampling.get_sigmas_exponential(n=steps, sigma_min=float(model_wrap.sigma_min), sigma_max=float(model_wrap.sigma_max)) + sigmas = k_diffusion_sampling.get_sigmas_exponential(n=steps, sigma_min=float(model.model_sampling.sigma_min), sigma_max=float(model.model_sampling.sigma_max)) elif scheduler_name == "normal": - sigmas = model_wrap.get_sigmas(steps) + sigmas = normal_scheduler(model, steps) elif scheduler_name == "simple": - sigmas = simple_scheduler(model_wrap, steps) + sigmas = simple_scheduler(model, steps) elif scheduler_name == "ddim_uniform": - sigmas = ddim_scheduler(model_wrap, steps) + sigmas = ddim_scheduler(model, steps) elif scheduler_name == "sgm_uniform": - sigmas = sgm_scheduler(model_wrap, steps) + sigmas = normal_scheduler(model, steps, sgm=True) else: print("error invalid scheduler", self.scheduler) return sigmas @@ -660,7 +623,7 @@ def sampler_class(name): elif name == "uni_pc_bh2": sampler = UNIPCBH2 elif name == "ddim": - sampler = DDIM + sampler = ksampler("euler", inpaint_options={"random": True}) else: sampler = ksampler(name) return sampler diff --git a/comfy/sd.py b/comfy/sd.py index 4a2823c9d..65d94f46e 100644 --- a/comfy/sd.py +++ b/comfy/sd.py @@ -55,13 +55,26 @@ def load_clip_weights(model, sd): def load_lora_for_models(model, clip, lora, strength_model, strength_clip): - key_map = comfy.lora.model_lora_keys_unet(model.model) - key_map = comfy.lora.model_lora_keys_clip(clip.cond_stage_model, key_map) + key_map = {} + if model is not None: + key_map = comfy.lora.model_lora_keys_unet(model.model, key_map) + if clip is not None: + key_map = comfy.lora.model_lora_keys_clip(clip.cond_stage_model, key_map) + loaded = comfy.lora.load_lora(lora, key_map) - new_modelpatcher = model.clone() - k = new_modelpatcher.add_patches(loaded, strength_model) - new_clip = clip.clone() - k1 = new_clip.add_patches(loaded, strength_clip) + if model is not None: + new_modelpatcher = model.clone() + k = new_modelpatcher.add_patches(loaded, strength_model) + else: + k = () + new_modelpatcher = None + + if clip is not None: + new_clip = clip.clone() + k1 = new_clip.add_patches(loaded, strength_clip) + else: + k1 = () + new_clip = None k = set(k) k1 = set(k1) for x in loaded: @@ -483,6 +496,9 @@ def load_unet(unet_path): #load unet in diffusers format model = model_config.get_model(new_sd, "") model = model.to(offload_device) model.load_model_weights(new_sd, "") + left_over = sd.keys() + if len(left_over) > 0: + print("left over keys in unet:", left_over) return comfy.model_patcher.ModelPatcher(model, load_device=model_management.get_torch_device(), offload_device=offload_device) def save_checkpoint(output_path, model, clip, vae, metadata=None): diff --git a/comfy/sd1_clip.py b/comfy/sd1_clip.py index fdaa1e6c7..7db7ee0f4 100644 --- a/comfy/sd1_clip.py +++ b/comfy/sd1_clip.py @@ -8,32 +8,54 @@ import zipfile from . import model_management import contextlib +def gen_empty_tokens(special_tokens, length): + start_token = special_tokens.get("start", None) + end_token = special_tokens.get("end", None) + pad_token = special_tokens.get("pad") + output = [] + if start_token is not None: + output.append(start_token) + if end_token is not None: + output.append(end_token) + output += [pad_token] * (length - len(output)) + return output + class ClipTokenWeightEncoder: def encode_token_weights(self, token_weight_pairs): - to_encode = list(self.empty_tokens) + to_encode = list() + max_token_len = 0 + has_weights = False for x in token_weight_pairs: tokens = list(map(lambda a: a[0], x)) + max_token_len = max(len(tokens), max_token_len) + has_weights = has_weights or not all(map(lambda a: a[1] == 1.0, x)) to_encode.append(tokens) + sections = len(to_encode) + if has_weights or sections == 0: + to_encode.append(gen_empty_tokens(self.special_tokens, max_token_len)) + out, pooled = self.encode(to_encode) - z_empty = out[0:1] - if pooled.shape[0] > 1: - first_pooled = pooled[1:2] + if pooled is not None: + first_pooled = pooled[0:1].cpu() else: - first_pooled = pooled[0:1] + first_pooled = pooled output = [] - for k in range(1, out.shape[0]): + for k in range(0, sections): z = out[k:k+1] - for i in range(len(z)): - for j in range(len(z[i])): - weight = token_weight_pairs[k - 1][j][1] - z[i][j] = (z[i][j] - z_empty[0][j]) * weight + z_empty[0][j] + if has_weights: + z_empty = out[-1] + for i in range(len(z)): + for j in range(len(z[i])): + weight = token_weight_pairs[k][j][1] + if weight != 1.0: + z[i][j] = (z[i][j] - z_empty[j]) * weight + z_empty[j] output.append(z) if (len(output) == 0): - return z_empty.cpu(), first_pooled.cpu() - return torch.cat(output, dim=-2).cpu(), first_pooled.cpu() + return out[-1:].cpu(), first_pooled + return torch.cat(output, dim=-2).cpu(), first_pooled class SDClipModel(torch.nn.Module, ClipTokenWeightEncoder): """Uses the CLIP transformer encoder for text (from huggingface)""" @@ -43,37 +65,43 @@ class SDClipModel(torch.nn.Module, ClipTokenWeightEncoder): "hidden" ] def __init__(self, version="openai/clip-vit-large-patch14", device="cpu", max_length=77, - freeze=True, layer="last", layer_idx=None, textmodel_json_config=None, textmodel_path=None, dtype=None): # clip-vit-base-patch32 + freeze=True, layer="last", layer_idx=None, textmodel_json_config=None, textmodel_path=None, dtype=None, + special_tokens={"start": 49406, "end": 49407, "pad": 49407},layer_norm_hidden_state=True, config_class=CLIPTextConfig, + model_class=CLIPTextModel, inner_name="text_model"): # clip-vit-base-patch32 super().__init__() assert layer in self.LAYERS self.num_layers = 12 if textmodel_path is not None: - self.transformer = CLIPTextModel.from_pretrained(textmodel_path) + self.transformer = model_class.from_pretrained(textmodel_path) else: if textmodel_json_config is None: textmodel_json_config = os.path.join(os.path.dirname(os.path.realpath(__file__)), "sd1_clip_config.json") - config = CLIPTextConfig.from_json_file(textmodel_json_config) + config = config_class.from_json_file(textmodel_json_config) self.num_layers = config.num_hidden_layers with comfy.ops.use_comfy_ops(device, dtype): with modeling_utils.no_init_weights(): - self.transformer = CLIPTextModel(config) + self.transformer = model_class(config) + self.inner_name = inner_name if dtype is not None: self.transformer.to(dtype) - self.transformer.text_model.embeddings.token_embedding.to(torch.float32) - self.transformer.text_model.embeddings.position_embedding.to(torch.float32) + inner_model = getattr(self.transformer, self.inner_name) + if hasattr(inner_model, "embeddings"): + inner_model.embeddings.to(torch.float32) + else: + self.transformer.set_input_embeddings(self.transformer.get_input_embeddings().to(torch.float32)) self.max_length = max_length if freeze: self.freeze() self.layer = layer self.layer_idx = None - self.empty_tokens = [[49406] + [49407] * 76] + self.special_tokens = special_tokens self.text_projection = torch.nn.Parameter(torch.eye(self.transformer.get_input_embeddings().weight.shape[1])) self.logit_scale = torch.nn.Parameter(torch.tensor(4.6055)) self.enable_attention_masks = False - self.layer_norm_hidden_state = True + self.layer_norm_hidden_state = layer_norm_hidden_state if layer == "hidden": assert layer_idx is not None assert abs(layer_idx) <= self.num_layers @@ -117,7 +145,7 @@ class SDClipModel(torch.nn.Module, ClipTokenWeightEncoder): else: print("WARNING: shape mismatch when trying to apply embedding, embedding will be ignored", y.shape[0], current_embeds.weight.shape[1]) while len(tokens_temp) < len(x): - tokens_temp += [self.empty_tokens[0][-1]] + tokens_temp += [self.special_tokens["pad"]] out_tokens += [tokens_temp] n = token_dict_size @@ -142,7 +170,7 @@ class SDClipModel(torch.nn.Module, ClipTokenWeightEncoder): tokens = self.set_up_textual_embeddings(tokens, backup_embeds) tokens = torch.LongTensor(tokens).to(device) - if self.transformer.text_model.final_layer_norm.weight.dtype != torch.float32: + if getattr(self.transformer, self.inner_name).final_layer_norm.weight.dtype != torch.float32: precision_scope = torch.autocast else: precision_scope = lambda a, b: contextlib.nullcontext(a) @@ -168,12 +196,16 @@ class SDClipModel(torch.nn.Module, ClipTokenWeightEncoder): else: z = outputs.hidden_states[self.layer_idx] if self.layer_norm_hidden_state: - z = self.transformer.text_model.final_layer_norm(z) + z = getattr(self.transformer, self.inner_name).final_layer_norm(z) - pooled_output = outputs.pooler_output - if self.text_projection is not None: + if hasattr(outputs, "pooler_output"): + pooled_output = outputs.pooler_output.float() + else: + pooled_output = None + + if self.text_projection is not None and pooled_output is not None: pooled_output = pooled_output.float().to(self.text_projection.device) @ self.text_projection.float() - return z.float(), pooled_output.float() + return z.float(), pooled_output def encode(self, tokens): return self(tokens) @@ -343,17 +375,24 @@ def load_embed(embedding_name, embedding_directory, embedding_size, embed_key=No return embed_out class SDTokenizer: - def __init__(self, tokenizer_path=None, max_length=77, pad_with_end=True, embedding_directory=None, embedding_size=768, embedding_key='clip_l'): + def __init__(self, tokenizer_path=None, max_length=77, pad_with_end=True, embedding_directory=None, embedding_size=768, embedding_key='clip_l', tokenizer_class=CLIPTokenizer, has_start_token=True, pad_to_max_length=True): if tokenizer_path is None: tokenizer_path = os.path.join(os.path.dirname(os.path.realpath(__file__)), "sd1_tokenizer") - self.tokenizer = CLIPTokenizer.from_pretrained(tokenizer_path) + self.tokenizer = tokenizer_class.from_pretrained(tokenizer_path) self.max_length = max_length - self.max_tokens_per_section = self.max_length - 2 empty = self.tokenizer('')["input_ids"] - self.start_token = empty[0] - self.end_token = empty[1] + if has_start_token: + self.tokens_start = 1 + self.start_token = empty[0] + self.end_token = empty[1] + else: + self.tokens_start = 0 + self.start_token = None + self.end_token = empty[0] self.pad_with_end = pad_with_end + self.pad_to_max_length = pad_to_max_length + vocab = self.tokenizer.get_vocab() self.inv_vocab = {v: k for k, v in vocab.items()} self.embedding_directory = embedding_directory @@ -414,11 +453,13 @@ class SDTokenizer: else: continue #parse word - tokens.append([(t, weight) for t in self.tokenizer(word)["input_ids"][1:-1]]) + tokens.append([(t, weight) for t in self.tokenizer(word)["input_ids"][self.tokens_start:-1]]) #reshape token array to CLIP input size batched_tokens = [] - batch = [(self.start_token, 1.0, 0)] + batch = [] + if self.start_token is not None: + batch.append((self.start_token, 1.0, 0)) batched_tokens.append(batch) for i, t_group in enumerate(tokens): #determine if we're going to try and keep the tokens in a single batch @@ -435,16 +476,21 @@ class SDTokenizer: #add end token and pad else: batch.append((self.end_token, 1.0, 0)) - batch.extend([(pad_token, 1.0, 0)] * (remaining_length)) + if self.pad_to_max_length: + batch.extend([(pad_token, 1.0, 0)] * (remaining_length)) #start new batch - batch = [(self.start_token, 1.0, 0)] + batch = [] + if self.start_token is not None: + batch.append((self.start_token, 1.0, 0)) batched_tokens.append(batch) else: batch.extend([(t,w,i+1) for t,w in t_group]) t_group = [] #fill last batch - batch.extend([(self.end_token, 1.0, 0)] + [(pad_token, 1.0, 0)] * (self.max_length - len(batch) - 1)) + batch.append((self.end_token, 1.0, 0)) + if self.pad_to_max_length: + batch.extend([(pad_token, 1.0, 0)] * (self.max_length - len(batch))) if not return_word_ids: batched_tokens = [[(t, w) for t, w,_ in x] for x in batched_tokens] diff --git a/comfy/sd2_clip.py b/comfy/sd2_clip.py index ebabf7ccd..2ee0ca055 100644 --- a/comfy/sd2_clip.py +++ b/comfy/sd2_clip.py @@ -9,8 +9,7 @@ class SD2ClipHModel(sd1_clip.SDClipModel): layer_idx=23 textmodel_json_config = os.path.join(os.path.dirname(os.path.realpath(__file__)), "sd2_clip_config.json") - super().__init__(device=device, freeze=freeze, layer=layer, layer_idx=layer_idx, textmodel_json_config=textmodel_json_config, textmodel_path=textmodel_path, dtype=dtype) - self.empty_tokens = [[49406] + [49407] + [0] * 75] + super().__init__(device=device, freeze=freeze, layer=layer, layer_idx=layer_idx, textmodel_json_config=textmodel_json_config, textmodel_path=textmodel_path, dtype=dtype, special_tokens={"start": 49406, "end": 49407, "pad": 0}) class SD2ClipHTokenizer(sd1_clip.SDTokenizer): def __init__(self, tokenizer_path=None, embedding_directory=None): diff --git a/comfy/sdxl_clip.py b/comfy/sdxl_clip.py index 4c508a0ea..673399e22 100644 --- a/comfy/sdxl_clip.py +++ b/comfy/sdxl_clip.py @@ -9,9 +9,8 @@ class SDXLClipG(sd1_clip.SDClipModel): layer_idx=-2 textmodel_json_config = os.path.join(os.path.dirname(os.path.realpath(__file__)), "clip_config_bigg.json") - super().__init__(device=device, freeze=freeze, layer=layer, layer_idx=layer_idx, textmodel_json_config=textmodel_json_config, textmodel_path=textmodel_path, dtype=dtype) - self.empty_tokens = [[49406] + [49407] + [0] * 75] - self.layer_norm_hidden_state = False + super().__init__(device=device, freeze=freeze, layer=layer, layer_idx=layer_idx, textmodel_json_config=textmodel_json_config, textmodel_path=textmodel_path, dtype=dtype, + special_tokens={"start": 49406, "end": 49407, "pad": 0}, layer_norm_hidden_state=False) def load_sd(self, sd): return super().load_sd(sd) @@ -38,8 +37,7 @@ class SDXLTokenizer: class SDXLClipModel(torch.nn.Module): def __init__(self, device="cpu", dtype=None): super().__init__() - self.clip_l = sd1_clip.SDClipModel(layer="hidden", layer_idx=11, device=device, dtype=dtype) - self.clip_l.layer_norm_hidden_state = False + self.clip_l = sd1_clip.SDClipModel(layer="hidden", layer_idx=11, device=device, dtype=dtype, layer_norm_hidden_state=False) self.clip_g = SDXLClipG(device=device, dtype=dtype) def clip_layer(self, layer_idx): diff --git a/comfy/utils.py b/comfy/utils.py index 6a0c54e80..4b484d07a 100644 --- a/comfy/utils.py +++ b/comfy/utils.py @@ -261,6 +261,14 @@ def set_attr(obj, attr, value): setattr(obj, attrs[-1], torch.nn.Parameter(value)) del prev +def copy_to_param(obj, attr, value): + # inplace update tensor instead of replacing it + attrs = attr.split(".") + for name in attrs[:-1]: + obj = getattr(obj, name) + prev = getattr(obj, attrs[-1]) + prev.data.copy_(value) + def get_attr(obj, attr): attrs = attr.split(".") for name in attrs: diff --git a/comfy_extras/nodes_custom_sampler.py b/comfy_extras/nodes_custom_sampler.py index b52ad8fbd..154ecd0d2 100644 --- a/comfy_extras/nodes_custom_sampler.py +++ b/comfy_extras/nodes_custom_sampler.py @@ -188,7 +188,7 @@ class SamplerCustom: {"model": ("MODEL",), "add_noise": ("BOOLEAN", {"default": True}), "noise_seed": ("INT", {"default": 0, "min": 0, "max": 0xffffffffffffffff}), - "cfg": ("FLOAT", {"default": 8.0, "min": 0.0, "max": 100.0, "step":0.5, "round": 0.01}), + "cfg": ("FLOAT", {"default": 8.0, "min": 0.0, "max": 100.0, "step":0.1, "round": 0.01}), "positive": ("CONDITIONING", ), "negative": ("CONDITIONING", ), "sampler": ("SAMPLER", ), diff --git a/comfy_extras/nodes_model_advanced.py b/comfy_extras/nodes_model_advanced.py new file mode 100644 index 000000000..399123eaa --- /dev/null +++ b/comfy_extras/nodes_model_advanced.py @@ -0,0 +1,168 @@ +import folder_paths +import comfy.sd +import comfy.model_sampling +import torch + +class LCM(comfy.model_sampling.EPS): + def calculate_denoised(self, sigma, model_output, model_input): + timestep = self.timestep(sigma).view(sigma.shape[:1] + (1,) * (model_output.ndim - 1)) + sigma = sigma.view(sigma.shape[:1] + (1,) * (model_output.ndim - 1)) + x0 = model_input - model_output * sigma + + sigma_data = 0.5 + scaled_timestep = timestep * 10.0 #timestep_scaling + + c_skip = sigma_data**2 / (scaled_timestep**2 + sigma_data**2) + c_out = scaled_timestep / (scaled_timestep**2 + sigma_data**2) ** 0.5 + + return c_out * x0 + c_skip * model_input + +class ModelSamplingDiscreteLCM(torch.nn.Module): + def __init__(self): + super().__init__() + self.sigma_data = 1.0 + timesteps = 1000 + beta_start = 0.00085 + beta_end = 0.012 + + betas = torch.linspace(beta_start**0.5, beta_end**0.5, timesteps, dtype=torch.float32) ** 2 + alphas = 1.0 - betas + alphas_cumprod = torch.cumprod(alphas, dim=0) + + original_timesteps = 50 + self.skip_steps = timesteps // original_timesteps + + + alphas_cumprod_valid = torch.zeros((original_timesteps), dtype=torch.float32) + for x in range(original_timesteps): + alphas_cumprod_valid[original_timesteps - 1 - x] = alphas_cumprod[timesteps - 1 - x * self.skip_steps] + + sigmas = ((1 - alphas_cumprod_valid) / alphas_cumprod_valid) ** 0.5 + self.set_sigmas(sigmas) + + def set_sigmas(self, sigmas): + self.register_buffer('sigmas', sigmas) + self.register_buffer('log_sigmas', sigmas.log()) + + @property + def sigma_min(self): + return self.sigmas[0] + + @property + def sigma_max(self): + return self.sigmas[-1] + + def timestep(self, sigma): + log_sigma = sigma.log() + dists = log_sigma.to(self.log_sigmas.device) - self.log_sigmas[:, None] + return dists.abs().argmin(dim=0).view(sigma.shape) * self.skip_steps + (self.skip_steps - 1) + + def sigma(self, timestep): + t = torch.clamp(((timestep - (self.skip_steps - 1)) / self.skip_steps).float(), min=0, max=(len(self.sigmas) - 1)) + low_idx = t.floor().long() + high_idx = t.ceil().long() + w = t.frac() + log_sigma = (1 - w) * self.log_sigmas[low_idx] + w * self.log_sigmas[high_idx] + return log_sigma.exp() + + def percent_to_sigma(self, percent): + return self.sigma(torch.tensor(percent * 999.0)) + + +def rescale_zero_terminal_snr_sigmas(sigmas): + alphas_cumprod = 1 / ((sigmas * sigmas) + 1) + alphas_bar_sqrt = alphas_cumprod.sqrt() + + # Store old values. + alphas_bar_sqrt_0 = alphas_bar_sqrt[0].clone() + alphas_bar_sqrt_T = alphas_bar_sqrt[-1].clone() + + # Shift so the last timestep is zero. + alphas_bar_sqrt -= (alphas_bar_sqrt_T) + + # Scale so the first timestep is back to the old value. + alphas_bar_sqrt *= alphas_bar_sqrt_0 / (alphas_bar_sqrt_0 - alphas_bar_sqrt_T) + + # Convert alphas_bar_sqrt to betas + alphas_bar = alphas_bar_sqrt**2 # Revert sqrt + alphas_bar[-1] = 4.8973451890853435e-08 + return ((1 - alphas_bar) / alphas_bar) ** 0.5 + +class ModelSamplingDiscrete: + @classmethod + def INPUT_TYPES(s): + return {"required": { "model": ("MODEL",), + "sampling": (["eps", "v_prediction", "lcm"],), + "zsnr": ("BOOLEAN", {"default": False}), + }} + + RETURN_TYPES = ("MODEL",) + FUNCTION = "patch" + + CATEGORY = "advanced/model" + + def patch(self, model, sampling, zsnr): + m = model.clone() + + sampling_base = comfy.model_sampling.ModelSamplingDiscrete + if sampling == "eps": + sampling_type = comfy.model_sampling.EPS + elif sampling == "v_prediction": + sampling_type = comfy.model_sampling.V_PREDICTION + elif sampling == "lcm": + sampling_type = LCM + sampling_base = ModelSamplingDiscreteLCM + + class ModelSamplingAdvanced(sampling_base, sampling_type): + pass + + model_sampling = ModelSamplingAdvanced() + if zsnr: + model_sampling.set_sigmas(rescale_zero_terminal_snr_sigmas(model_sampling.sigmas)) + + m.add_object_patch("model_sampling", model_sampling) + return (m, ) + +class RescaleCFG: + @classmethod + def INPUT_TYPES(s): + return {"required": { "model": ("MODEL",), + "multiplier": ("FLOAT", {"default": 0.7, "min": 0.0, "max": 1.0, "step": 0.01}), + }} + RETURN_TYPES = ("MODEL",) + FUNCTION = "patch" + + CATEGORY = "advanced/model" + + def patch(self, model, multiplier): + def rescale_cfg(args): + cond = args["cond"] + uncond = args["uncond"] + cond_scale = args["cond_scale"] + sigma = args["sigma"] + sigma = sigma.view(sigma.shape[:1] + (1,) * (cond.ndim - 1)) + x_orig = args["input"] + + #rescale cfg has to be done on v-pred model output + x = x_orig / (sigma * sigma + 1.0) + cond = ((x - (x_orig - cond)) * (sigma ** 2 + 1.0) ** 0.5) / (sigma) + uncond = ((x - (x_orig - uncond)) * (sigma ** 2 + 1.0) ** 0.5) / (sigma) + + #rescalecfg + x_cfg = uncond + cond_scale * (cond - uncond) + ro_pos = torch.std(cond, dim=(1,2,3), keepdim=True) + ro_cfg = torch.std(x_cfg, dim=(1,2,3), keepdim=True) + + x_rescaled = x_cfg * (ro_pos / ro_cfg) + x_final = multiplier * x_rescaled + (1.0 - multiplier) * x_cfg + + return x_orig - (x - x_final * sigma / (sigma * sigma + 1.0) ** 0.5) + + m = model.clone() + m.set_model_sampler_cfg_function(rescale_cfg) + return (m, ) + +NODE_CLASS_MAPPINGS = { + "ModelSamplingDiscrete": ModelSamplingDiscrete, + "RescaleCFG": RescaleCFG, +} diff --git a/comfy_extras/nodes_post_processing.py b/comfy_extras/nodes_post_processing.py index 3f651e594..12704f545 100644 --- a/comfy_extras/nodes_post_processing.py +++ b/comfy_extras/nodes_post_processing.py @@ -23,7 +23,7 @@ class Blend: "max": 1.0, "step": 0.01 }), - "blend_mode": (["normal", "multiply", "screen", "overlay", "soft_light"],), + "blend_mode": (["normal", "multiply", "screen", "overlay", "soft_light", "difference"],), }, } @@ -54,6 +54,8 @@ class Blend: return torch.where(img1 <= 0.5, 2 * img1 * img2, 1 - 2 * (1 - img1) * (1 - img2)) elif mode == "soft_light": return torch.where(img2 <= 0.5, img1 - (1 - 2 * img2) * img1 * (1 - img1), img1 + (2 * img2 - 1) * (self.g(img1) - img1)) + elif mode == "difference": + return img1 - img2 else: raise ValueError(f"Unsupported blend mode: {mode}") @@ -126,7 +128,7 @@ class Quantize: "max": 256, "step": 1 }), - "dither": (["none", "floyd-steinberg"],), + "dither": (["none", "floyd-steinberg", "bayer-2", "bayer-4", "bayer-8", "bayer-16"],), }, } @@ -135,19 +137,47 @@ class Quantize: CATEGORY = "image/postprocessing" - def quantize(self, image: torch.Tensor, colors: int = 256, dither: str = "FLOYDSTEINBERG"): + def bayer(im, pal_im, order): + def normalized_bayer_matrix(n): + if n == 0: + return np.zeros((1,1), "float32") + else: + q = 4 ** n + m = q * normalized_bayer_matrix(n - 1) + return np.bmat(((m-1.5, m+0.5), (m+1.5, m-0.5))) / q + + num_colors = len(pal_im.getpalette()) // 3 + spread = 2 * 256 / num_colors + bayer_n = int(math.log2(order)) + bayer_matrix = torch.from_numpy(spread * normalized_bayer_matrix(bayer_n) + 0.5) + + result = torch.from_numpy(np.array(im).astype(np.float32)) + tw = math.ceil(result.shape[0] / bayer_matrix.shape[0]) + th = math.ceil(result.shape[1] / bayer_matrix.shape[1]) + tiled_matrix = bayer_matrix.tile(tw, th).unsqueeze(-1) + result.add_(tiled_matrix[:result.shape[0],:result.shape[1]]).clamp_(0, 255) + result = result.to(dtype=torch.uint8) + + im = Image.fromarray(result.cpu().numpy()) + im = im.quantize(palette=pal_im, dither=Image.Dither.NONE) + return im + + def quantize(self, image: torch.Tensor, colors: int, dither: str): batch_size, height, width, _ = image.shape result = torch.zeros_like(image) - dither_option = Image.Dither.FLOYDSTEINBERG if dither == "floyd-steinberg" else Image.Dither.NONE - for b in range(batch_size): - tensor_image = image[b] - img = (tensor_image * 255).to(torch.uint8).numpy() - pil_image = Image.fromarray(img, mode='RGB') + im = Image.fromarray((image[b] * 255).to(torch.uint8).numpy(), mode='RGB') - palette = pil_image.quantize(colors=colors) # Required as described in https://github.com/python-pillow/Pillow/issues/5836 - quantized_image = pil_image.quantize(colors=colors, palette=palette, dither=dither_option) + pal_im = im.quantize(colors=colors) # Required as described in https://github.com/python-pillow/Pillow/issues/5836 + + if dither == "none": + quantized_image = im.quantize(palette=pal_im, dither=Image.Dither.NONE) + elif dither == "floyd-steinberg": + quantized_image = im.quantize(palette=pal_im, dither=Image.Dither.FLOYDSTEINBERG) + elif dither.startswith("bayer"): + order = int(dither.split('-')[-1]) + quantized_image = Quantize.bayer(im, pal_im, order) quantized_array = torch.tensor(np.array(quantized_image.convert("RGB"))).float() / 255 result[b] = quantized_array diff --git a/comfy_extras/nodes_rebatch.py b/comfy_extras/nodes_rebatch.py index 0a9daf272..88a4ebe29 100644 --- a/comfy_extras/nodes_rebatch.py +++ b/comfy_extras/nodes_rebatch.py @@ -4,7 +4,7 @@ class LatentRebatch: @classmethod def INPUT_TYPES(s): return {"required": { "latents": ("LATENT",), - "batch_size": ("INT", {"default": 1, "min": 1, "max": 64}), + "batch_size": ("INT", {"default": 1, "min": 1, "max": 4096}), }} RETURN_TYPES = ("LATENT",) INPUT_IS_LIST = True diff --git a/latent_preview.py b/latent_preview.py index e1553c85c..6e758a1a9 100644 --- a/latent_preview.py +++ b/latent_preview.py @@ -22,7 +22,7 @@ class TAESDPreviewerImpl(LatentPreviewer): self.taesd = taesd def decode_latent_to_preview(self, x0): - x_sample = self.taesd.decoder(x0)[0].detach() + x_sample = self.taesd.decoder(x0[:1])[0].detach() # x_sample = self.taesd.unscale_latents(x_sample).div(4).add(0.5) # returns value in [-2, 2] x_sample = x_sample.sub(0.5).mul(2) diff --git a/nodes.py b/nodes.py index 61ebbb8b4..2bbfd8fe8 100644 --- a/nodes.py +++ b/nodes.py @@ -1218,7 +1218,7 @@ class KSampler: {"model": ("MODEL",), "seed": ("INT", {"default": 0, "min": 0, "max": 0xffffffffffffffff}), "steps": ("INT", {"default": 20, "min": 1, "max": 10000}), - "cfg": ("FLOAT", {"default": 8.0, "min": 0.0, "max": 100.0, "step":0.5, "round": 0.01}), + "cfg": ("FLOAT", {"default": 8.0, "min": 0.0, "max": 100.0, "step":0.1, "round": 0.01}), "sampler_name": (comfy.samplers.KSampler.SAMPLERS, ), "scheduler": (comfy.samplers.KSampler.SCHEDULERS, ), "positive": ("CONDITIONING", ), @@ -1244,7 +1244,7 @@ class KSamplerAdvanced: "add_noise": (["enable", "disable"], ), "noise_seed": ("INT", {"default": 0, "min": 0, "max": 0xffffffffffffffff}), "steps": ("INT", {"default": 20, "min": 1, "max": 10000}), - "cfg": ("FLOAT", {"default": 8.0, "min": 0.0, "max": 100.0, "step":0.5, "round": 0.01}), + "cfg": ("FLOAT", {"default": 8.0, "min": 0.0, "max": 100.0, "step":0.1, "round": 0.01}), "sampler_name": (comfy.samplers.KSampler.SAMPLERS, ), "scheduler": (comfy.samplers.KSampler.SCHEDULERS, ), "positive": ("CONDITIONING", ), @@ -1798,6 +1798,7 @@ def init_custom_nodes(): "nodes_freelunch.py", "nodes_custom_sampler.py", "nodes_hypertile.py", + "nodes_model_advanced.py", ] for node_file in extras_files: diff --git a/web/extensions/core/contextMenuFilter.js b/web/extensions/core/contextMenuFilter.js index 152cd7043..0a305391a 100644 --- a/web/extensions/core/contextMenuFilter.js +++ b/web/extensions/core/contextMenuFilter.js @@ -25,7 +25,7 @@ const ext = { requestAnimationFrame(() => { const currentNode = LGraphCanvas.active_canvas.current_node; const clickedComboValue = currentNode.widgets - .filter(w => w.type === "combo" && w.options.values.length === values.length) + ?.filter(w => w.type === "combo" && w.options.values.length === values.length) .find(w => w.options.values.every((v, i) => v === values[i])) ?.value; diff --git a/web/extensions/core/nodeTemplates.js b/web/extensions/core/nodeTemplates.js index 434491075..b6479f454 100644 --- a/web/extensions/core/nodeTemplates.js +++ b/web/extensions/core/nodeTemplates.js @@ -14,6 +14,9 @@ import { ComfyDialog, $el } from "../../scripts/ui.js"; // To delete/rename: // Right click the canvas // Node templates -> Manage +// +// To rearrange: +// Open the manage dialog and Drag and drop elements using the "Name:" label as handle const id = "Comfy.NodeTemplates"; @@ -22,6 +25,10 @@ class ManageTemplates extends ComfyDialog { super(); this.element.classList.add("comfy-manage-templates"); this.templates = this.load(); + this.draggedEl = null; + this.saveVisualCue = null; + this.emptyImg = new Image(); + this.emptyImg.src = 'data:image/gif;base64,R0lGODlhAQABAIAAAAUEBAAAACwAAAAAAQABAAACAkQBADs='; this.importInput = $el("input", { type: "file", @@ -35,14 +42,11 @@ class ManageTemplates extends ComfyDialog { createButtons() { const btns = super.createButtons(); - btns[0].textContent = "Cancel"; - btns.unshift( - $el("button", { - type: "button", - textContent: "Save", - onclick: () => this.save(), - }) - ); + btns[0].textContent = "Close"; + btns[0].onclick = (e) => { + clearTimeout(this.saveVisualCue); + this.close(); + }; btns.unshift( $el("button", { type: "button", @@ -71,25 +75,6 @@ class ManageTemplates extends ComfyDialog { } } - save() { - // Find all visible inputs and save them as our new list - const inputs = this.element.querySelectorAll("input"); - const updated = []; - - for (let i = 0; i < inputs.length; i++) { - const input = inputs[i]; - if (input.parentElement.style.display !== "none") { - const t = this.templates[i]; - t.name = input.value.trim() || input.getAttribute("data-name"); - updated.push(t); - } - } - - this.templates = updated; - this.store(); - this.close(); - } - store() { localStorage.setItem(id, JSON.stringify(this.templates)); } @@ -145,71 +130,155 @@ class ManageTemplates extends ComfyDialog { super.show( $el( "div", - { - style: { - display: "grid", - gridTemplateColumns: "1fr auto", - gap: "5px", - }, - }, - this.templates.flatMap((t) => { + {}, + this.templates.flatMap((t,i) => { let nameInput; return [ $el( - "label", + "div", { - textContent: "Name: ", + dataset: { id: i }, + className: "tempateManagerRow", + style: { + display: "grid", + gridTemplateColumns: "1fr auto", + border: "1px dashed transparent", + gap: "5px", + backgroundColor: "var(--comfy-menu-bg)" + }, + ondragstart: (e) => { + this.draggedEl = e.currentTarget; + e.currentTarget.style.opacity = "0.6"; + e.currentTarget.style.border = "1px dashed yellow"; + e.dataTransfer.effectAllowed = 'move'; + e.dataTransfer.setDragImage(this.emptyImg, 0, 0); + }, + ondragend: (e) => { + e.target.style.opacity = "1"; + e.currentTarget.style.border = "1px dashed transparent"; + e.currentTarget.removeAttribute("draggable"); + + // rearrange the elements in the localStorage + this.element.querySelectorAll('.tempateManagerRow').forEach((el,i) => { + var prev_i = el.dataset.id; + + if ( el == this.draggedEl && prev_i != i ) { + [this.templates[i], this.templates[prev_i]] = [this.templates[prev_i], this.templates[i]]; + } + el.dataset.id = i; + }); + this.store(); + }, + ondragover: (e) => { + e.preventDefault(); + if ( e.currentTarget == this.draggedEl ) + return; + + let rect = e.currentTarget.getBoundingClientRect(); + if (e.clientY > rect.top + rect.height / 2) { + e.currentTarget.parentNode.insertBefore(this.draggedEl, e.currentTarget.nextSibling); + } else { + e.currentTarget.parentNode.insertBefore(this.draggedEl, e.currentTarget); + } + } }, [ - $el("input", { - value: t.name, - dataset: { name: t.name }, - $: (el) => (nameInput = el), - }), + $el( + "label", + { + textContent: "Name: ", + style: { + cursor: "grab", + }, + onmousedown: (e) => { + // enable dragging only from the label + if (e.target.localName == 'label') + e.currentTarget.parentNode.draggable = 'true'; + } + }, + [ + $el("input", { + value: t.name, + dataset: { name: t.name }, + style: { + transitionProperty: 'background-color', + transitionDuration: '0s', + }, + onchange: (e) => { + clearTimeout(this.saveVisualCue); + var el = e.target; + var row = el.parentNode.parentNode; + this.templates[row.dataset.id].name = el.value.trim() || 'untitled'; + this.store(); + el.style.backgroundColor = 'rgb(40, 95, 40)'; + el.style.transitionDuration = '0s'; + this.saveVisualCue = setTimeout(function () { + el.style.transitionDuration = '.7s'; + el.style.backgroundColor = 'var(--comfy-input-bg)'; + }, 15); + }, + onkeypress: (e) => { + var el = e.target; + clearTimeout(this.saveVisualCue); + el.style.transitionDuration = '0s'; + el.style.backgroundColor = 'var(--comfy-input-bg)'; + }, + $: (el) => (nameInput = el), + }) + ] + ), + $el( + "div", + {}, + [ + $el("button", { + textContent: "Export", + style: { + fontSize: "12px", + fontWeight: "normal", + }, + onclick: (e) => { + const json = JSON.stringify({templates: [t]}, null, 2); // convert the data to a JSON string + const blob = new Blob([json], {type: "application/json"}); + const url = URL.createObjectURL(blob); + const a = $el("a", { + href: url, + download: (nameInput.value || t.name) + ".json", + style: {display: "none"}, + parent: document.body, + }); + a.click(); + setTimeout(function () { + a.remove(); + window.URL.revokeObjectURL(url); + }, 0); + }, + }), + $el("button", { + textContent: "Delete", + style: { + fontSize: "12px", + color: "red", + fontWeight: "normal", + }, + onclick: (e) => { + const item = e.target.parentNode.parentNode; + item.parentNode.removeChild(item); + this.templates.splice(item.dataset.id*1, 1); + this.store(); + // update the rows index, setTimeout ensures that the list is updated + var that = this; + setTimeout(function (){ + that.element.querySelectorAll('.tempateManagerRow').forEach((el,i) => { + el.dataset.id = i; + }); + }, 0); + }, + }), + ] + ), ] - ), - $el( - "div", - {}, - [ - $el("button", { - textContent: "Export", - style: { - fontSize: "12px", - fontWeight: "normal", - }, - onclick: (e) => { - const json = JSON.stringify({templates: [t]}, null, 2); // convert the data to a JSON string - const blob = new Blob([json], {type: "application/json"}); - const url = URL.createObjectURL(blob); - const a = $el("a", { - href: url, - download: (nameInput.value || t.name) + ".json", - style: {display: "none"}, - parent: document.body, - }); - a.click(); - setTimeout(function () { - a.remove(); - window.URL.revokeObjectURL(url); - }, 0); - }, - }), - $el("button", { - textContent: "Delete", - style: { - fontSize: "12px", - color: "red", - fontWeight: "normal", - }, - onclick: (e) => { - nameInput.value = ""; - e.target.parentElement.style.display = "none"; - e.target.parentElement.previousElementSibling.style.display = "none"; - }, - }), - ] - ), + ) ]; }) ) diff --git a/web/scripts/app.js b/web/scripts/app.js index 41ea4796f..be24c309b 100644 --- a/web/scripts/app.js +++ b/web/scripts/app.js @@ -5,6 +5,22 @@ import { api } from "./api.js"; import { defaultGraph } from "./defaultGraph.js"; import { getPngMetadata, getWebpMetadata, importA1111, getLatentMetadata } from "./pnginfo.js"; + +function sanitizeNodeName(string) { + let entityMap = { + '&': '', + '<': '', + '>': '', + '"': '', + "'": '', + '`': '', + '=': '' + }; + return String(string).replace(/[&<>"'`=]/g, function fromEntityMap (s) { + return entityMap[s]; + }); +} + /** * @typedef {import("types/comfy").ComfyExtension} ComfyExtension */ @@ -1453,6 +1469,17 @@ export class ComfyApp { localStorage.setItem("litegrapheditor_clipboard", old); } + showMissingNodesError(missingNodeTypes, hasAddedNodes = true) { + this.ui.dialog.show( + `When loading the graph, the following node types were not found: ${hasAddedNodes ? "Nodes that have failed to load will show as red on the graph." : ""}` + ); + this.logging.addEntry("Comfy.App", "warn", { + MissingNodes: missingNodeTypes, + }); + } + /** * Populates the graph with the specified workflow data * @param {*} graphData A serialized graph object @@ -1480,6 +1507,7 @@ export class ComfyApp { // Find missing node types if (!(n.type in LiteGraph.registered_node_types)) { + n.type = sanitizeNodeName(n.type); missingNodeTypes.push(n.type); } } @@ -1570,14 +1598,7 @@ export class ComfyApp { } if (missingNodeTypes.length) { - this.ui.dialog.show( - `When loading the graph, the following node types were not found: Nodes that have failed to load will show as red on the graph.` - ); - this.logging.addEntry("Comfy.App", "warn", { - MissingNodes: missingNodeTypes, - }); + this.showMissingNodesError(missingNodeTypes); } } @@ -1808,9 +1829,11 @@ export class ComfyApp { } else if (file.type === "application/json" || file.name?.endsWith(".json")) { const reader = new FileReader(); reader.onload = () => { - var jsonContent = JSON.parse(reader.result); + const jsonContent = JSON.parse(reader.result); if (jsonContent?.templates) { this.loadTemplateData(jsonContent); + } else if(this.isApiJson(jsonContent)) { + this.loadApiJson(jsonContent); } else { this.loadGraphData(jsonContent); } @@ -1824,6 +1847,51 @@ export class ComfyApp { } } + isApiJson(data) { + return Object.values(data).every((v) => v.class_type); + } + + loadApiJson(apiData) { + const missingNodeTypes = Object.values(apiData).filter((n) => !LiteGraph.registered_node_types[n.class_type]); + if (missingNodeTypes.length) { + this.showMissingNodesError(missingNodeTypes.map(t => t.class_type), false); + return; + } + + const ids = Object.keys(apiData); + app.graph.clear(); + for (const id of ids) { + const data = apiData[id]; + const node = LiteGraph.createNode(data.class_type); + node.id = id; + graph.add(node); + } + + for (const id of ids) { + const data = apiData[id]; + const node = app.graph.getNodeById(id); + for (const input in data.inputs ?? {}) { + const value = data.inputs[input]; + if (value instanceof Array) { + const [fromId, fromSlot] = value; + const fromNode = app.graph.getNodeById(fromId); + const toSlot = node.inputs?.findIndex((inp) => inp.name === input); + if (toSlot !== -1) { + fromNode.connect(fromSlot, node, toSlot); + } + } else { + const widget = node.widgets?.find((w) => w.name === input); + if (widget) { + widget.value = value; + widget.callback?.(value); + } + } + } + } + + app.graph.arrange(); + } + /** * Registers a Comfy web extension with the app * @param {ComfyExtension} extension