diff --git a/comfy/cli_args.py b/comfy/cli_args.py index b4bbfbfab..757fc245f 100644 --- a/comfy/cli_args.py +++ b/comfy/cli_args.py @@ -114,6 +114,9 @@ parser.add_argument("--disable-metadata", action="store_true", help="Disable sav parser.add_argument("--multi-user", action="store_true", help="Enables per-user storage.") +parser.add_argument("--verbose", action="store_true", help="Enables more debug prints.") + + if comfy.options.args_parsing: args = parser.parse_args() else: @@ -124,3 +127,10 @@ if args.windows_standalone_build: if args.disable_auto_launch: args.auto_launch = False + +import logging +logging_level = logging.WARNING +if args.verbose: + logging_level = logging.DEBUG + +logging.basicConfig(format="%(message)s", level=logging_level) diff --git a/comfy/clip_vision.py b/comfy/clip_vision.py index 8c77ee7a9..acc86be85 100644 --- a/comfy/clip_vision.py +++ b/comfy/clip_vision.py @@ -2,6 +2,7 @@ from .utils import load_torch_file, transformers_convert, state_dict_prefix_repl import os import torch import json +import logging import comfy.ops import comfy.model_patcher @@ -99,7 +100,7 @@ def load_clipvision_from_sd(sd, prefix="", convert_keys=False): clip = ClipVisionModel(json_config) m, u = clip.load_sd(sd) if len(m) > 0: - print("missing clip vision:", m) + logging.warning("missing clip vision: {}".format(m)) u = set(u) keys = list(sd.keys()) for k in keys: diff --git a/comfy/controlnet.py b/comfy/controlnet.py index f859a50d4..a5e7b23f7 100644 --- a/comfy/controlnet.py +++ b/comfy/controlnet.py @@ -1,6 +1,7 @@ import torch import math import os +import logging import comfy.utils import comfy.model_management import comfy.model_detection @@ -9,6 +10,7 @@ import comfy.ops import comfy.cldm.cldm import comfy.t2i_adapter.adapter +import comfy.ldm.cascade.controlnet def broadcast_image_to(tensor, target_batch_size, batched_number): @@ -37,6 +39,8 @@ class ControlBase: self.timestep_percent_range = (0.0, 1.0) self.global_average_pooling = False self.timestep_range = None + self.compression_ratio = 8 + self.upscale_algorithm = 'nearest-exact' if device is None: device = comfy.model_management.get_torch_device() @@ -77,6 +81,8 @@ class ControlBase: c.strength = self.strength c.timestep_percent_range = self.timestep_percent_range c.global_average_pooling = self.global_average_pooling + c.compression_ratio = self.compression_ratio + c.upscale_algorithm = self.upscale_algorithm def inference_memory_requirements(self, dtype): if self.previous_controlnet is not None: @@ -158,11 +164,11 @@ class ControlNet(ControlBase): dtype = self.manual_cast_dtype output_dtype = x_noisy.dtype - if self.cond_hint is None or x_noisy.shape[2] * 8 != self.cond_hint.shape[2] or x_noisy.shape[3] * 8 != self.cond_hint.shape[3]: + if self.cond_hint is None or x_noisy.shape[2] * self.compression_ratio != self.cond_hint.shape[2] or x_noisy.shape[3] * self.compression_ratio != self.cond_hint.shape[3]: if self.cond_hint is not None: del self.cond_hint self.cond_hint = None - self.cond_hint = comfy.utils.common_upscale(self.cond_hint_original, x_noisy.shape[3] * 8, x_noisy.shape[2] * 8, 'nearest-exact', "center").to(dtype).to(self.device) + self.cond_hint = comfy.utils.common_upscale(self.cond_hint_original, x_noisy.shape[3] * self.compression_ratio, x_noisy.shape[2] * self.compression_ratio, self.upscale_algorithm, "center").to(dtype).to(self.device) if x_noisy.shape[0] != self.cond_hint.shape[0]: self.cond_hint = broadcast_image_to(self.cond_hint, x_noisy.shape[0], batched_number) @@ -362,7 +368,7 @@ def load_controlnet(ckpt_path, model=None): leftover_keys = controlnet_data.keys() if len(leftover_keys) > 0: - print("leftover keys:", leftover_keys) + logging.warning("leftover keys: {}".format(leftover_keys)) controlnet_data = new_sd pth_key = 'control_model.zero_convs.0.0.weight' @@ -377,7 +383,7 @@ def load_controlnet(ckpt_path, model=None): else: net = load_t2i_adapter(controlnet_data) if net is None: - print("error checkpoint does not contain controlnet or t2i adapter data", ckpt_path) + logging.error("error checkpoint does not contain controlnet or t2i adapter data {}".format(ckpt_path)) return net if controlnet_config is None: @@ -412,7 +418,7 @@ def load_controlnet(ckpt_path, model=None): cd = controlnet_data[x] cd += model_sd[sd_key].type(cd.dtype).to(cd.device) else: - print("WARNING: Loaded a diff controlnet without a model. It will very likely not work.") + logging.warning("WARNING: Loaded a diff controlnet without a model. It will very likely not work.") class WeightsLoader(torch.nn.Module): pass @@ -421,7 +427,12 @@ def load_controlnet(ckpt_path, model=None): missing, unexpected = w.load_state_dict(controlnet_data, strict=False) else: missing, unexpected = control_model.load_state_dict(controlnet_data, strict=False) - print(missing, unexpected) + + if len(missing) > 0: + logging.warning("missing controlnet keys: {}".format(missing)) + + if len(unexpected) > 0: + logging.info("unexpected controlnet keys: {}".format(unexpected)) global_average_pooling = False filename = os.path.splitext(ckpt_path)[0] @@ -432,11 +443,13 @@ def load_controlnet(ckpt_path, model=None): return control class T2IAdapter(ControlBase): - def __init__(self, t2i_model, channels_in, device=None): + def __init__(self, t2i_model, channels_in, compression_ratio, upscale_algorithm, device=None): super().__init__(device) self.t2i_model = t2i_model self.channels_in = channels_in self.control_input = None + self.compression_ratio = compression_ratio + self.upscale_algorithm = upscale_algorithm def scale_image_to(self, width, height): unshuffle_amount = self.t2i_model.unshuffle_amount @@ -456,13 +469,13 @@ class T2IAdapter(ControlBase): else: return None - if self.cond_hint is None or x_noisy.shape[2] * 8 != self.cond_hint.shape[2] or x_noisy.shape[3] * 8 != self.cond_hint.shape[3]: + if self.cond_hint is None or x_noisy.shape[2] * self.compression_ratio != self.cond_hint.shape[2] or x_noisy.shape[3] * self.compression_ratio != self.cond_hint.shape[3]: if self.cond_hint is not None: del self.cond_hint self.control_input = None self.cond_hint = None - width, height = self.scale_image_to(x_noisy.shape[3] * 8, x_noisy.shape[2] * 8) - self.cond_hint = comfy.utils.common_upscale(self.cond_hint_original, width, height, 'nearest-exact', "center").float().to(self.device) + width, height = self.scale_image_to(x_noisy.shape[3] * self.compression_ratio, x_noisy.shape[2] * self.compression_ratio) + self.cond_hint = comfy.utils.common_upscale(self.cond_hint_original, width, height, self.upscale_algorithm, "center").float().to(self.device) if self.channels_in == 1 and self.cond_hint.shape[1] > 1: self.cond_hint = torch.mean(self.cond_hint, 1, keepdim=True) if x_noisy.shape[0] != self.cond_hint.shape[0]: @@ -481,11 +494,14 @@ class T2IAdapter(ControlBase): return self.control_merge(control_input, mid, control_prev, x_noisy.dtype) def copy(self): - c = T2IAdapter(self.t2i_model, self.channels_in) + c = T2IAdapter(self.t2i_model, self.channels_in, self.compression_ratio, self.upscale_algorithm) self.copy_to(c) return c def load_t2i_adapter(t2i_data): + compression_ratio = 8 + upscale_algorithm = 'nearest-exact' + if 'adapter' in t2i_data: t2i_data = t2i_data['adapter'] if 'adapter.body.0.resnets.0.block1.weight' in t2i_data: #diffusers format @@ -513,13 +529,22 @@ def load_t2i_adapter(t2i_data): if cin == 256 or cin == 768: xl = True model_ad = comfy.t2i_adapter.adapter.Adapter(cin=cin, channels=[channel, channel*2, channel*4, channel*4][:4], nums_rb=2, ksize=ksize, sk=True, use_conv=use_conv, xl=xl) + elif "backbone.0.0.weight" in keys: + model_ad = comfy.ldm.cascade.controlnet.ControlNet(c_in=t2i_data['backbone.0.0.weight'].shape[1], proj_blocks=[0, 4, 8, 12, 51, 55, 59, 63]) + compression_ratio = 32 + upscale_algorithm = 'bilinear' + elif "backbone.10.blocks.0.weight" in keys: + model_ad = comfy.ldm.cascade.controlnet.ControlNet(c_in=t2i_data['backbone.0.weight'].shape[1], bottleneck_mode="large", proj_blocks=[0, 4, 8, 12, 51, 55, 59, 63]) + compression_ratio = 1 + upscale_algorithm = 'nearest-exact' else: return None + missing, unexpected = model_ad.load_state_dict(t2i_data) if len(missing) > 0: - print("t2i missing", missing) + logging.warning("t2i missing {}".format(missing)) if len(unexpected) > 0: - print("t2i unexpected", unexpected) + logging.info("t2i unexpected {}".format(unexpected)) - return T2IAdapter(model_ad, model_ad.input_channels) + return T2IAdapter(model_ad, model_ad.input_channels, compression_ratio, upscale_algorithm) diff --git a/comfy/diffusers_convert.py b/comfy/diffusers_convert.py index eb561933a..18398cb32 100644 --- a/comfy/diffusers_convert.py +++ b/comfy/diffusers_convert.py @@ -1,5 +1,6 @@ import re import torch +import logging # conversion code from https://github.com/huggingface/diffusers/blob/main/scripts/convert_diffusers_to_original_stable_diffusion.py @@ -177,7 +178,7 @@ def convert_vae_state_dict(vae_state_dict): for k, v in new_state_dict.items(): for weight_name in weights_to_convert: if f"mid.attn_1.{weight_name}.weight" in k: - print(f"Reshaping {k} for SD format") + logging.info(f"Reshaping {k} for SD format") new_state_dict[k] = reshape_weight_for_sd(v) return new_state_dict diff --git a/comfy/ldm/cascade/controlnet.py b/comfy/ldm/cascade/controlnet.py new file mode 100644 index 000000000..5dac59394 --- /dev/null +++ b/comfy/ldm/cascade/controlnet.py @@ -0,0 +1,93 @@ +""" + This file is part of ComfyUI. + Copyright (C) 2024 Stability AI + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . +""" + +import torch +import torchvision +from torch import nn +from .common import LayerNorm2d_op + + +class CNetResBlock(nn.Module): + def __init__(self, c, dtype=None, device=None, operations=None): + super().__init__() + self.blocks = nn.Sequential( + LayerNorm2d_op(operations)(c, dtype=dtype, device=device), + nn.GELU(), + operations.Conv2d(c, c, kernel_size=3, padding=1), + LayerNorm2d_op(operations)(c, dtype=dtype, device=device), + nn.GELU(), + operations.Conv2d(c, c, kernel_size=3, padding=1), + ) + + def forward(self, x): + return x + self.blocks(x) + + +class ControlNet(nn.Module): + def __init__(self, c_in=3, c_proj=2048, proj_blocks=None, bottleneck_mode=None, dtype=None, device=None, operations=nn): + super().__init__() + if bottleneck_mode is None: + bottleneck_mode = 'effnet' + self.proj_blocks = proj_blocks + if bottleneck_mode == 'effnet': + embd_channels = 1280 + self.backbone = torchvision.models.efficientnet_v2_s().features.eval() + if c_in != 3: + in_weights = self.backbone[0][0].weight.data + self.backbone[0][0] = operations.Conv2d(c_in, 24, kernel_size=3, stride=2, bias=False, dtype=dtype, device=device) + if c_in > 3: + # nn.init.constant_(self.backbone[0][0].weight, 0) + self.backbone[0][0].weight.data[:, :3] = in_weights[:, :3].clone() + else: + self.backbone[0][0].weight.data = in_weights[:, :c_in].clone() + elif bottleneck_mode == 'simple': + embd_channels = c_in + self.backbone = nn.Sequential( + operations.Conv2d(embd_channels, embd_channels * 4, kernel_size=3, padding=1, dtype=dtype, device=device), + nn.LeakyReLU(0.2, inplace=True), + operations.Conv2d(embd_channels * 4, embd_channels, kernel_size=3, padding=1, dtype=dtype, device=device), + ) + elif bottleneck_mode == 'large': + self.backbone = nn.Sequential( + operations.Conv2d(c_in, 4096 * 4, kernel_size=1, dtype=dtype, device=device), + nn.LeakyReLU(0.2, inplace=True), + operations.Conv2d(4096 * 4, 1024, kernel_size=1, dtype=dtype, device=device), + *[CNetResBlock(1024, dtype=dtype, device=device, operations=operations) for _ in range(8)], + operations.Conv2d(1024, 1280, kernel_size=1, dtype=dtype, device=device), + ) + embd_channels = 1280 + else: + raise ValueError(f'Unknown bottleneck mode: {bottleneck_mode}') + self.projections = nn.ModuleList() + for _ in range(len(proj_blocks)): + self.projections.append(nn.Sequential( + operations.Conv2d(embd_channels, embd_channels, kernel_size=1, bias=False, dtype=dtype, device=device), + nn.LeakyReLU(0.2, inplace=True), + operations.Conv2d(embd_channels, c_proj, kernel_size=1, bias=False, dtype=dtype, device=device), + )) + # nn.init.constant_(self.projections[-1][-1].weight, 0) # zero output projection + self.xl = False + self.input_channels = c_in + self.unshuffle_amount = 8 + + def forward(self, x): + x = self.backbone(x) + proj_outputs = [None for _ in range(max(self.proj_blocks) + 1)] + for i, idx in enumerate(self.proj_blocks): + proj_outputs[idx] = self.projections[i](x) + return proj_outputs diff --git a/comfy/ldm/cascade/stage_c.py b/comfy/ldm/cascade/stage_c.py index 08e33aded..67c1e52b6 100644 --- a/comfy/ldm/cascade/stage_c.py +++ b/comfy/ldm/cascade/stage_c.py @@ -194,10 +194,10 @@ class StageC(nn.Module): hasattr(block, '_fsdp_wrapped_module') and isinstance(block._fsdp_wrapped_module, ResBlock)): if cnet is not None: - next_cnet = cnet() + next_cnet = cnet.pop() if next_cnet is not None: x = x + nn.functional.interpolate(next_cnet, size=x.shape[-2:], mode='bilinear', - align_corners=True) + align_corners=True).to(x.dtype) x = block(x) elif isinstance(block, AttnBlock) or ( hasattr(block, '_fsdp_wrapped_module') and isinstance(block._fsdp_wrapped_module, @@ -228,10 +228,10 @@ class StageC(nn.Module): x = torch.nn.functional.interpolate(x, skip.shape[-2:], mode='bilinear', align_corners=True) if cnet is not None: - next_cnet = cnet() + next_cnet = cnet.pop() if next_cnet is not None: x = x + nn.functional.interpolate(next_cnet, size=x.shape[-2:], mode='bilinear', - align_corners=True) + align_corners=True).to(x.dtype) x = block(x, skip) elif isinstance(block, AttnBlock) or ( hasattr(block, '_fsdp_wrapped_module') and isinstance(block._fsdp_wrapped_module, @@ -248,7 +248,7 @@ class StageC(nn.Module): x = upscaler(x) return x - def forward(self, x, r, clip_text, clip_text_pooled, clip_img, cnet=None, **kwargs): + def forward(self, x, r, clip_text, clip_text_pooled, clip_img, control=None, **kwargs): # Process the conditioning embeddings r_embed = self.gen_r_embedding(r).to(dtype=x.dtype) for c in self.t_conds: @@ -256,10 +256,13 @@ class StageC(nn.Module): r_embed = torch.cat([r_embed, self.gen_r_embedding(t_cond).to(dtype=x.dtype)], dim=1) clip = self.gen_c_embeddings(clip_text, clip_text_pooled, clip_img) + if control is not None: + cnet = control.get("input") + else: + cnet = None + # Model Blocks x = self.embedding(x) - if cnet is not None: - cnet = ControlNetDeliverer(cnet) level_outputs = self._down_encode(x, r_embed, clip, cnet) x = self._up_decode(level_outputs, r_embed, clip, cnet) return self.clf(x) diff --git a/comfy/lora.py b/comfy/lora.py index 21b9897a6..637380d54 100644 --- a/comfy/lora.py +++ b/comfy/lora.py @@ -1,4 +1,5 @@ import comfy.utils +import logging LORA_CLIP_MAP = { "mlp.fc1": "mlp_fc1", @@ -156,7 +157,7 @@ def load_lora(lora, to_load): for x in lora.keys(): if x not in loaded_keys: - print("lora key not loaded", x) + logging.warning("lora key not loaded: {}".format(x)) return patch_dict def model_lora_keys_clip(model, key_map={}): diff --git a/comfy/model_base.py b/comfy/model_base.py index a9de13667..a2514ca5e 100644 --- a/comfy/model_base.py +++ b/comfy/model_base.py @@ -1,4 +1,5 @@ import torch +import logging from comfy.ldm.modules.diffusionmodules.openaimodel import UNetModel, Timestep from comfy.ldm.cascade.stage_c import StageC from comfy.ldm.cascade.stage_b import StageB @@ -66,8 +67,8 @@ class BaseModel(torch.nn.Module): if self.adm_channels is None: self.adm_channels = 0 self.inpaint_model = False - print("model_type", model_type.name) - print("adm", self.adm_channels) + logging.warning("model_type {}".format(model_type.name)) + logging.info("adm {}".format(self.adm_channels)) def apply_model(self, x, t, c_concat=None, c_crossattn=None, control=None, transformer_options={}, **kwargs): sigma = t @@ -183,10 +184,10 @@ class BaseModel(torch.nn.Module): to_load = self.model_config.process_unet_state_dict(to_load) m, u = self.diffusion_model.load_state_dict(to_load, strict=False) if len(m) > 0: - print("unet missing:", m) + logging.warning("unet missing: {}".format(m)) if len(u) > 0: - print("unet unexpected:", u) + logging.warning("unet unexpected: {}".format(u)) del to_load return self diff --git a/comfy/model_detection.py b/comfy/model_detection.py index 07ee85708..b7c3be309 100644 --- a/comfy/model_detection.py +++ b/comfy/model_detection.py @@ -1,5 +1,6 @@ import comfy.supported_models import comfy.supported_models_base +import logging def count_blocks(state_dict_keys, prefix_string): count = 0 @@ -186,7 +187,7 @@ def model_config_from_unet_config(unet_config): if model_config.matches(unet_config): return model_config(unet_config) - print("no match", unet_config) + logging.error("no match {}".format(unet_config)) return None def model_config_from_unet(state_dict, unet_key_prefix, use_base_if_no_match=False): diff --git a/comfy/model_management.py b/comfy/model_management.py index 9f9248831..dd262e260 100644 --- a/comfy/model_management.py +++ b/comfy/model_management.py @@ -1,4 +1,5 @@ import psutil +import logging from enum import Enum from comfy.cli_args import args import comfy.utils @@ -29,7 +30,7 @@ lowvram_available = True xpu_available = False if args.deterministic: - print("Using deterministic algorithms for pytorch") + logging.warning("Using deterministic algorithms for pytorch") torch.use_deterministic_algorithms(True, warn_only=True) directml_enabled = False @@ -41,7 +42,7 @@ if args.directml is not None: directml_device = torch_directml.device() else: directml_device = torch_directml.device(device_index) - print("Using directml with device:", torch_directml.device_name(device_index)) + logging.warning("Using directml with device: {}".format(torch_directml.device_name(device_index))) # torch_directml.disable_tiled_resources(True) lowvram_available = False #TODO: need to find a way to get free memory in directml before this can be enabled by default. @@ -117,10 +118,10 @@ def get_total_memory(dev=None, torch_total_too=False): total_vram = get_total_memory(get_torch_device()) / (1024 * 1024) total_ram = psutil.virtual_memory().total / (1024 * 1024) -print("Total VRAM {:0.0f} MB, total RAM {:0.0f} MB".format(total_vram, total_ram)) +logging.warning("Total VRAM {:0.0f} MB, total RAM {:0.0f} MB".format(total_vram, total_ram)) if not args.normalvram and not args.cpu: if lowvram_available and total_vram <= 4096: - print("Trying to enable lowvram mode because your GPU seems to have 4GB or less. If you don't want this use: --normalvram") + logging.warning("Trying to enable lowvram mode because your GPU seems to have 4GB or less. If you don't want this use: --normalvram") set_vram_to = VRAMState.LOW_VRAM try: @@ -143,12 +144,10 @@ else: pass try: XFORMERS_VERSION = xformers.version.__version__ - print("xformers version:", XFORMERS_VERSION) + logging.warning("xformers version: {}".format(XFORMERS_VERSION)) if XFORMERS_VERSION.startswith("0.0.18"): - print() - print("WARNING: This version of xformers has a major bug where you will get black images when generating high resolution images.") - print("Please downgrade or upgrade xformers to a different version.") - print() + logging.warning("\nWARNING: This version of xformers has a major bug where you will get black images when generating high resolution images.") + logging.warning("Please downgrade or upgrade xformers to a different version.\n") XFORMERS_ENABLED_VAE = False except: pass @@ -213,11 +212,11 @@ elif args.highvram or args.gpu_only: FORCE_FP32 = False FORCE_FP16 = False if args.force_fp32: - print("Forcing FP32, if this improves things please report it.") + logging.warning("Forcing FP32, if this improves things please report it.") FORCE_FP32 = True if args.force_fp16: - print("Forcing FP16.") + logging.warning("Forcing FP16.") FORCE_FP16 = True if lowvram_available: @@ -231,12 +230,12 @@ if cpu_state != CPUState.GPU: if cpu_state == CPUState.MPS: vram_state = VRAMState.SHARED -print(f"Set vram state to: {vram_state.name}") +logging.warning(f"Set vram state to: {vram_state.name}") DISABLE_SMART_MEMORY = args.disable_smart_memory if DISABLE_SMART_MEMORY: - print("Disabling smart memory management") + logging.warning("Disabling smart memory management") def get_torch_device_name(device): if hasattr(device, 'type'): @@ -254,11 +253,11 @@ def get_torch_device_name(device): return "CUDA {}: {}".format(device, torch.cuda.get_device_name(device)) try: - print("Device:", get_torch_device_name(get_torch_device())) + logging.warning("Device: {}".format(get_torch_device_name(get_torch_device()))) except: - print("Could not pick default device.") + logging.warning("Could not pick default device.") -print("VAE dtype:", VAE_DTYPE) +logging.warning("VAE dtype: {}".format(VAE_DTYPE)) current_loaded_models = [] @@ -301,7 +300,7 @@ class LoadedModel: raise e if lowvram_model_memory > 0: - print("loading in lowvram mode", lowvram_model_memory/(1024 * 1024)) + logging.warning("loading in lowvram mode {}".format(lowvram_model_memory/(1024 * 1024))) mem_counter = 0 for m in self.real_model.modules(): if hasattr(m, "comfy_cast_weights"): @@ -314,7 +313,7 @@ class LoadedModel: elif hasattr(m, "weight"): #only modules with comfy_cast_weights can be set to lowvram mode m.to(self.device) mem_counter += module_size(m) - print("lowvram: loaded module regularly", m) + logging.warning("lowvram: loaded module regularly {}".format(m)) self.model_accelerated = True @@ -348,7 +347,7 @@ def unload_model_clones(model): to_unload = [i] + to_unload for i in to_unload: - print("unload clone", i) + logging.warning("unload clone {}".format(i)) current_loaded_models.pop(i).model_unload() def free_memory(memory_required, device, keep_loaded=[]): @@ -390,7 +389,7 @@ def load_models_gpu(models, memory_required=0): models_already_loaded.append(loaded_model) else: if hasattr(x, "model"): - print(f"Requested to load {x.model.__class__.__name__}") + logging.warning(f"Requested to load {x.model.__class__.__name__}") models_to_load.append(loaded_model) if len(models_to_load) == 0: @@ -400,7 +399,7 @@ def load_models_gpu(models, memory_required=0): free_memory(extra_mem, d, models_already_loaded) return - print(f"Loading {len(models_to_load)} new model{'s' if len(models_to_load) > 1 else ''}") + logging.warning(f"Loading {len(models_to_load)} new model{'s' if len(models_to_load) > 1 else ''}") total_memory_required = {} for loaded_model in models_to_load: diff --git a/comfy/model_patcher.py b/comfy/model_patcher.py index 4a5d42b03..5e578dffc 100644 --- a/comfy/model_patcher.py +++ b/comfy/model_patcher.py @@ -1,6 +1,7 @@ import torch import copy import inspect +import logging import comfy.utils import comfy.model_management @@ -187,7 +188,7 @@ class ModelPatcher: model_sd = self.model_state_dict() for key in self.patches: if key not in model_sd: - print("could not patch. key doesn't exist in model:", key) + logging.warning("could not patch. key doesn't exist in model: {}".format(key)) continue weight = model_sd[key] @@ -236,7 +237,7 @@ class ModelPatcher: w1 = v[0] if alpha != 0.0: if w1.shape != weight.shape: - print("WARNING SHAPE MISMATCH {} WEIGHT NOT MERGED {} != {}".format(key, w1.shape, weight.shape)) + logging.warning("WARNING SHAPE MISMATCH {} WEIGHT NOT MERGED {} != {}".format(key, w1.shape, weight.shape)) else: weight += alpha * comfy.model_management.cast_to_device(w1, weight.device, weight.dtype) elif patch_type == "lora": #lora/locon @@ -252,7 +253,7 @@ class ModelPatcher: try: weight += (alpha * torch.mm(mat1.flatten(start_dim=1), mat2.flatten(start_dim=1))).reshape(weight.shape).type(weight.dtype) except Exception as e: - print("ERROR", key, e) + logging.error("ERROR {} {} {}".format(patch_type, key, e)) elif patch_type == "lokr": w1 = v[0] w2 = v[1] @@ -291,7 +292,7 @@ class ModelPatcher: try: weight += alpha * torch.kron(w1, w2).reshape(weight.shape).type(weight.dtype) except Exception as e: - print("ERROR", key, e) + logging.error("ERROR {} {} {}".format(patch_type, key, e)) elif patch_type == "loha": w1a = v[0] w1b = v[1] @@ -320,7 +321,7 @@ class ModelPatcher: try: weight += (alpha * m1 * m2).reshape(weight.shape).type(weight.dtype) except Exception as e: - print("ERROR", key, e) + logging.error("ERROR {} {} {}".format(patch_type, key, e)) elif patch_type == "glora": if v[4] is not None: alpha *= v[4] / v[0].shape[0] @@ -330,9 +331,12 @@ class ModelPatcher: b1 = comfy.model_management.cast_to_device(v[2].flatten(start_dim=1), weight.device, torch.float32) b2 = comfy.model_management.cast_to_device(v[3].flatten(start_dim=1), weight.device, torch.float32) - weight += ((torch.mm(b2, b1) + torch.mm(torch.mm(weight.flatten(start_dim=1), a2), a1)) * alpha).reshape(weight.shape).type(weight.dtype) + try: + weight += ((torch.mm(b2, b1) + torch.mm(torch.mm(weight.flatten(start_dim=1), a2), a1)) * alpha).reshape(weight.shape).type(weight.dtype) + except Exception as e: + logging.error("ERROR {} {} {}".format(patch_type, key, e)) else: - print("patch type not recognized", patch_type, key) + logging.warning("patch type not recognized {} {}".format(patch_type, key)) return weight diff --git a/comfy/sd.py b/comfy/sd.py index fd5d604e0..3e4b9e47f 100644 --- a/comfy/sd.py +++ b/comfy/sd.py @@ -1,5 +1,6 @@ import torch from enum import Enum +import logging from comfy import model_management from .ldm.models.autoencoder import AutoencoderKL, AutoencodingEngine @@ -37,7 +38,7 @@ def load_model_weights(model, sd): w = sd.pop(x) del w if len(m) > 0: - print("missing", m) + logging.warning("missing {}".format(m)) return model def load_clip_weights(model, sd): @@ -81,7 +82,7 @@ def load_lora_for_models(model, clip, lora, strength_model, strength_clip): k1 = set(k1) for x in loaded: if (x not in k) and (x not in k1): - print("NOT LOADED", x) + logging.warning("NOT LOADED {}".format(x)) return (new_modelpatcher, new_clip) @@ -225,10 +226,10 @@ class VAE: m, u = self.first_stage_model.load_state_dict(sd, strict=False) if len(m) > 0: - print("Missing VAE keys", m) + logging.warning("Missing VAE keys {}".format(m)) if len(u) > 0: - print("Leftover VAE keys", u) + logging.info("Leftover VAE keys {}".format(u)) if device is None: device = model_management.vae_device() @@ -291,7 +292,7 @@ class VAE: samples = samples_in[x:x+batch_number].to(self.vae_dtype).to(self.device) pixel_samples[x:x+batch_number] = self.process_output(self.first_stage_model.decode(samples).to(self.output_device).float()) except model_management.OOM_EXCEPTION as e: - print("Warning: Ran out of memory when regular VAE decoding, retrying with tiled VAE decoding.") + logging.warning("Warning: Ran out of memory when regular VAE decoding, retrying with tiled VAE decoding.") pixel_samples = self.decode_tiled_(samples_in) pixel_samples = pixel_samples.to(self.output_device).movedim(1,-1) @@ -317,7 +318,7 @@ class VAE: samples[x:x+batch_number] = self.first_stage_model.encode(pixels_in).to(self.output_device).float() except model_management.OOM_EXCEPTION as e: - print("Warning: Ran out of memory when regular VAE encoding, retrying with tiled VAE encoding.") + logging.warning("Warning: Ran out of memory when regular VAE encoding, retrying with tiled VAE encoding.") samples = self.encode_tiled_(pixel_samples) return samples @@ -393,10 +394,10 @@ def load_clip(ckpt_paths, embedding_directory=None, clip_type=CLIPType.STABLE_DI for c in clip_data: m, u = clip.load_sd(c) if len(m) > 0: - print("clip missing:", m) + logging.warning("clip missing: {}".format(m)) if len(u) > 0: - print("clip unexpected:", u) + logging.info("clip unexpected: {}".format(u)) return clip def load_gligen(ckpt_path): @@ -534,21 +535,21 @@ def load_checkpoint_guess_config(ckpt_path, output_vae=True, output_clip=True, o clip = CLIP(clip_target, embedding_directory=embedding_directory) m, u = clip.load_sd(clip_sd, full_model=True) if len(m) > 0: - print("clip missing:", m) + logging.warning("clip missing: {}".format(m)) if len(u) > 0: - print("clip unexpected:", u) + logging.info("clip unexpected {}:".format(u)) else: - print("no CLIP/text encoder weights in checkpoint, the text encoder model will not be loaded.") + logging.warning("no CLIP/text encoder weights in checkpoint, the text encoder model will not be loaded.") left_over = sd.keys() if len(left_over) > 0: - print("left over keys:", left_over) + logging.info("left over keys: {}".format(left_over)) if output_model: model_patcher = comfy.model_patcher.ModelPatcher(model, load_device=load_device, offload_device=model_management.unet_offload_device(), current_device=inital_load_device) if inital_load_device != torch.device("cpu"): - print("loaded straight to GPU") + logging.warning("loaded straight to GPU") model_management.load_model_gpu(model_patcher) return (model_patcher, clip, vae, clipvision) @@ -577,7 +578,7 @@ def load_unet_state_dict(sd): #load unet in diffusers format if k in sd: new_sd[diffusers_keys[k]] = sd.pop(k) else: - print(diffusers_keys[k], k) + logging.warning("{} {}".format(diffusers_keys[k], k)) offload_device = model_management.unet_offload_device() unet_dtype = model_management.unet_dtype(model_params=parameters, supported_dtypes=model_config.supported_inference_dtypes) @@ -588,14 +589,14 @@ def load_unet_state_dict(sd): #load unet in diffusers format model.load_model_weights(new_sd, "") left_over = sd.keys() if len(left_over) > 0: - print("left over keys in unet:", left_over) + logging.warning("left over keys in unet: {}".format(left_over)) return comfy.model_patcher.ModelPatcher(model, load_device=load_device, offload_device=offload_device) def load_unet(unet_path): sd = comfy.utils.load_torch_file(unet_path) model = load_unet_state_dict(sd) if model is None: - print("ERROR UNSUPPORTED UNET", unet_path) + logging.error("ERROR UNSUPPORTED UNET {}".format(unet_path)) raise RuntimeError("ERROR: Could not detect model type of: {}".format(unet_path)) return model diff --git a/comfy/sd1_clip.py b/comfy/sd1_clip.py index 87e3eaa4d..ff6db0d20 100644 --- a/comfy/sd1_clip.py +++ b/comfy/sd1_clip.py @@ -8,6 +8,7 @@ import zipfile from . import model_management import comfy.clip_model import json +import logging def gen_empty_tokens(special_tokens, length): start_token = special_tokens.get("start", None) @@ -137,7 +138,7 @@ class SDClipModel(torch.nn.Module, ClipTokenWeightEncoder): tokens_temp += [next_new_token] next_new_token += 1 else: - print("WARNING: shape mismatch when trying to apply embedding, embedding will be ignored", y.shape[0], current_embeds.weight.shape[1]) + logging.warning("WARNING: shape mismatch when trying to apply embedding, embedding will be ignored {} != {}".format(y.shape[0], current_embeds.weight.shape[1])) while len(tokens_temp) < len(x): tokens_temp += [self.special_tokens["pad"]] out_tokens += [tokens_temp] @@ -329,9 +330,7 @@ def load_embed(embedding_name, embedding_directory, embedding_size, embed_key=No else: embed = torch.load(embed_path, map_location="cpu") except Exception as e: - print(traceback.format_exc()) - print() - print("error loading embedding, skipping loading:", embedding_name) + logging.warning("{}\n\nerror loading embedding, skipping loading: {}".format(traceback.format_exc(), embedding_name)) return None if embed_out is None: @@ -422,7 +421,7 @@ class SDTokenizer: embedding_name = word[len(self.embedding_identifier):].strip('\n') embed, leftover = self._try_get_embedding(embedding_name) if embed is None: - print(f"warning, embedding:{embedding_name} does not exist, ignoring") + logging.warning(f"warning, embedding:{embedding_name} does not exist, ignoring") else: if len(embed.shape) == 1: tokens.append([(embed, weight)]) diff --git a/comfy/utils.py b/comfy/utils.py index 5deb14cd2..8caecd866 100644 --- a/comfy/utils.py +++ b/comfy/utils.py @@ -5,6 +5,7 @@ import comfy.checkpoint_pickle import safetensors.torch import numpy as np from PIL import Image +import logging def load_torch_file(ckpt, safe_load=False, device=None): if device is None: @@ -14,14 +15,14 @@ def load_torch_file(ckpt, safe_load=False, device=None): else: if safe_load: if not 'weights_only' in torch.load.__code__.co_varnames: - print("Warning torch.load doesn't support weights_only on this pytorch version, loading unsafely.") + logging.warning("Warning torch.load doesn't support weights_only on this pytorch version, loading unsafely.") safe_load = False if safe_load: pl_sd = torch.load(ckpt, map_location=device, weights_only=True) else: pl_sd = torch.load(ckpt, map_location=device, pickle_module=comfy.checkpoint_pickle) if "global_step" in pl_sd: - print(f"Global Step: {pl_sd['global_step']}") + logging.info(f"Global Step: {pl_sd['global_step']}") if "state_dict" in pl_sd: sd = pl_sd["state_dict"] else: diff --git a/comfy_extras/nodes_custom_sampler.py b/comfy_extras/nodes_custom_sampler.py index 99f9ea7dc..0ad1246a6 100644 --- a/comfy_extras/nodes_custom_sampler.py +++ b/comfy_extras/nodes_custom_sampler.py @@ -228,6 +228,23 @@ class SamplerDPMPP_SDE: sampler = comfy.samplers.ksampler(sampler_name, {"eta": eta, "s_noise": s_noise, "r": r}) return (sampler, ) +class SamplerEulerAncestral: + @classmethod + def INPUT_TYPES(s): + return {"required": + {"eta": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 100.0, "step":0.01, "round": False}), + "s_noise": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 100.0, "step":0.01, "round": False}), + } + } + RETURN_TYPES = ("SAMPLER",) + CATEGORY = "sampling/custom_sampling/samplers" + + FUNCTION = "get_sampler" + + def get_sampler(self, eta, s_noise): + sampler = comfy.samplers.ksampler("euler_ancestral", {"eta": eta, "s_noise": s_noise}) + return (sampler, ) + class SamplerCustom: @classmethod def INPUT_TYPES(s): @@ -288,6 +305,7 @@ NODE_CLASS_MAPPINGS = { "VPScheduler": VPScheduler, "SDTurboScheduler": SDTurboScheduler, "KSamplerSelect": KSamplerSelect, + "SamplerEulerAncestral": SamplerEulerAncestral, "SamplerDPMPP_2M_SDE": SamplerDPMPP_2M_SDE, "SamplerDPMPP_SDE": SamplerDPMPP_SDE, "SplitSigmas": SplitSigmas, diff --git a/comfy_extras/nodes_model_merging.py b/comfy_extras/nodes_model_merging.py index d594cf490..a25b73ca7 100644 --- a/comfy_extras/nodes_model_merging.py +++ b/comfy_extras/nodes_model_merging.py @@ -87,6 +87,50 @@ class CLIPMergeSimple: m.add_patches({k: kp[k]}, 1.0 - ratio, ratio) return (m, ) + +class CLIPSubtract: + @classmethod + def INPUT_TYPES(s): + return {"required": { "clip1": ("CLIP",), + "clip2": ("CLIP",), + "multiplier": ("FLOAT", {"default": 1.0, "min": -10.0, "max": 10.0, "step": 0.01}), + }} + RETURN_TYPES = ("CLIP",) + FUNCTION = "merge" + + CATEGORY = "advanced/model_merging" + + def merge(self, clip1, clip2, multiplier): + m = clip1.clone() + kp = clip2.get_key_patches() + for k in kp: + if k.endswith(".position_ids") or k.endswith(".logit_scale"): + continue + m.add_patches({k: kp[k]}, - multiplier, multiplier) + return (m, ) + + +class CLIPAdd: + @classmethod + def INPUT_TYPES(s): + return {"required": { "clip1": ("CLIP",), + "clip2": ("CLIP",), + }} + RETURN_TYPES = ("CLIP",) + FUNCTION = "merge" + + CATEGORY = "advanced/model_merging" + + def merge(self, clip1, clip2): + m = clip1.clone() + kp = clip2.get_key_patches() + for k in kp: + if k.endswith(".position_ids") or k.endswith(".logit_scale"): + continue + m.add_patches({k: kp[k]}, 1.0, 1.0) + return (m, ) + + class ModelMergeBlocks: @classmethod def INPUT_TYPES(s): @@ -279,6 +323,8 @@ NODE_CLASS_MAPPINGS = { "ModelMergeAdd": ModelAdd, "CheckpointSave": CheckpointSave, "CLIPMergeSimple": CLIPMergeSimple, + "CLIPMergeSubtract": CLIPSubtract, + "CLIPMergeAdd": CLIPAdd, "CLIPSave": CLIPSave, "VAESave": VAESave, } diff --git a/comfy_extras/nodes_stable_cascade.py b/comfy_extras/nodes_stable_cascade.py index b795d0083..7e2d37d22 100644 --- a/comfy_extras/nodes_stable_cascade.py +++ b/comfy_extras/nodes_stable_cascade.py @@ -37,7 +37,7 @@ class StableCascade_EmptyLatentImage: RETURN_NAMES = ("stage_c", "stage_b") FUNCTION = "generate" - CATEGORY = "_for_testing/stable_cascade" + CATEGORY = "latent/stable_cascade" def generate(self, width, height, compression, batch_size=1): c_latent = torch.zeros([batch_size, 16, height // compression, width // compression]) @@ -63,7 +63,7 @@ class StableCascade_StageC_VAEEncode: RETURN_NAMES = ("stage_c", "stage_b") FUNCTION = "generate" - CATEGORY = "_for_testing/stable_cascade" + CATEGORY = "latent/stable_cascade" def generate(self, image, vae, compression): width = image.shape[-2] @@ -91,7 +91,7 @@ class StableCascade_StageB_Conditioning: FUNCTION = "set_prior" - CATEGORY = "_for_testing/stable_cascade" + CATEGORY = "conditioning/stable_cascade" def set_prior(self, conditioning, stage_c): c = [] @@ -102,8 +102,39 @@ class StableCascade_StageB_Conditioning: c.append(n) return (c, ) +class StableCascade_SuperResolutionControlnet: + def __init__(self, device="cpu"): + self.device = device + + @classmethod + def INPUT_TYPES(s): + return {"required": { + "image": ("IMAGE",), + "vae": ("VAE", ), + }} + RETURN_TYPES = ("IMAGE", "LATENT", "LATENT") + RETURN_NAMES = ("controlnet_input", "stage_c", "stage_b") + FUNCTION = "generate" + + CATEGORY = "_for_testing/stable_cascade" + + def generate(self, image, vae): + width = image.shape[-2] + height = image.shape[-3] + batch_size = image.shape[0] + controlnet_input = vae.encode(image[:,:,:,:3]).movedim(1, -1) + + c_latent = torch.zeros([batch_size, 16, height // 16, width // 16]) + b_latent = torch.zeros([batch_size, 4, height // 2, width // 2]) + return (controlnet_input, { + "samples": c_latent, + }, { + "samples": b_latent, + }) + NODE_CLASS_MAPPINGS = { "StableCascade_EmptyLatentImage": StableCascade_EmptyLatentImage, "StableCascade_StageB_Conditioning": StableCascade_StageB_Conditioning, "StableCascade_StageC_VAEEncode": StableCascade_StageC_VAEEncode, + "StableCascade_SuperResolutionControlnet": StableCascade_SuperResolutionControlnet, } diff --git a/execution.py b/execution.py index 3e9d53b0e..5f8ea835c 100644 --- a/execution.py +++ b/execution.py @@ -35,8 +35,7 @@ def get_input_data(inputs, class_def, unique_id, outputs={}, prompt={}, extra_da if h[x] == "PROMPT": input_data_all[x] = [prompt] if h[x] == "EXTRA_PNGINFO": - if "extra_pnginfo" in extra_data: - input_data_all[x] = [extra_data['extra_pnginfo']] + input_data_all[x] = [extra_data.get('extra_pnginfo', None)] if h[x] == "UNIQUE_ID": input_data_all[x] = [unique_id] return input_data_all diff --git a/nodes.py b/nodes.py index ea49d5490..c334cdb95 100644 --- a/nodes.py +++ b/nodes.py @@ -1976,7 +1976,7 @@ def init_custom_nodes(): print("WARNING: some comfy_extras/ nodes did not import correctly. This may be because they are missing some dependencies.\n") for node in import_failed: print("IMPORT FAILED: {}".format(node)) - print("\nThis issue might be caused by missing dependencies.") + print("\nThis issue might be caused by missing newly dependencies added the last time you updated ComfyUI.") if args.windows_standalone_build: print("Please run the update script: update/update_comfyui.bat") else: