Merge branch 'comfyanonymous:master' into master

This commit is contained in:
Ruixi Fan 2024-05-23 16:55:45 -07:00 committed by GitHub
commit b2e2d37960
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
35 changed files with 723 additions and 240 deletions

View File

@ -7,7 +7,7 @@ on:
description: 'cuda version'
required: true
type: string
default: "121"
default: "124"
python_minor:
description: 'python minor version'
@ -19,7 +19,7 @@ on:
description: 'python patch version'
required: true
type: string
default: "2"
default: "3"
# push:
# branches:
# - master
@ -49,7 +49,7 @@ jobs:
echo 'import site' >> ./python3${{ inputs.python_minor }}._pth
curl https://bootstrap.pypa.io/get-pip.py -o get-pip.py
./python.exe get-pip.py
python -m pip wheel torch torchvision torchaudio mpmath==1.3.0 --pre --extra-index-url https://download.pytorch.org/whl/nightly/cu${{ inputs.cu }} -r ../ComfyUI/requirements.txt pygit2 -w ../temp_wheel_dir
python -m pip wheel torch torchvision torchaudio mpmath==1.3.0 numpy==1.26.4 --pre --extra-index-url https://download.pytorch.org/whl/nightly/cu${{ inputs.cu }} -r ../ComfyUI/requirements.txt pygit2 -w ../temp_wheel_dir
ls ../temp_wheel_dir
./python.exe -s -m pip install --pre ../temp_wheel_dir/*
sed -i '1i../ComfyUI' ./python3${{ inputs.python_minor }}._pth

View File

@ -41,29 +41,32 @@ Workflow examples can be found on the [Examples page](https://comfyanonymous.git
## Shortcuts
| Keybind | Explanation |
|---------------------------|--------------------------------------------------------------------------------------------------------------------|
| Ctrl + Enter | Queue up current graph for generation |
| Ctrl + Shift + Enter | Queue up current graph as first for generation |
| Ctrl + Z/Ctrl + Y | Undo/Redo |
| Ctrl + S | Save workflow |
| Ctrl + O | Load workflow |
| Ctrl + A | Select all nodes |
| Alt + C | Collapse/uncollapse selected nodes |
| Ctrl + M | Mute/unmute selected nodes |
| Ctrl + B | Bypass selected nodes (acts like the node was removed from the graph and the wires reconnected through) |
| Delete/Backspace | Delete selected nodes |
| Ctrl + Delete/Backspace | Delete the current graph |
| Space | Move the canvas around when held and moving the cursor |
| Ctrl/Shift + Click | Add clicked node to selection |
| Ctrl + C/Ctrl + V | Copy and paste selected nodes (without maintaining connections to outputs of unselected nodes) |
| Ctrl + C/Ctrl + Shift + V | Copy and paste selected nodes (maintaining connections from outputs of unselected nodes to inputs of pasted nodes) |
| Shift + Drag | Move multiple selected nodes at the same time |
| Ctrl + D | Load default graph |
| Q | Toggle visibility of the queue |
| H | Toggle visibility of history |
| R | Refresh graph |
| Double-Click LMB | Open node quick search palette |
| Keybind | Explanation |
|------------------------------------|--------------------------------------------------------------------------------------------------------------------|
| Ctrl + Enter | Queue up current graph for generation |
| Ctrl + Shift + Enter | Queue up current graph as first for generation |
| Ctrl + Z/Ctrl + Y | Undo/Redo |
| Ctrl + S | Save workflow |
| Ctrl + O | Load workflow |
| Ctrl + A | Select all nodes |
| Alt + C | Collapse/uncollapse selected nodes |
| Ctrl + M | Mute/unmute selected nodes |
| Ctrl + B | Bypass selected nodes (acts like the node was removed from the graph and the wires reconnected through) |
| Delete/Backspace | Delete selected nodes |
| Ctrl + Backspace | Delete the current graph |
| Space | Move the canvas around when held and moving the cursor |
| Ctrl/Shift + Click | Add clicked node to selection |
| Ctrl + C/Ctrl + V | Copy and paste selected nodes (without maintaining connections to outputs of unselected nodes) |
| Ctrl + C/Ctrl + Shift + V | Copy and paste selected nodes (maintaining connections from outputs of unselected nodes to inputs of pasted nodes) |
| Shift + Drag | Move multiple selected nodes at the same time |
| Ctrl + D | Load default graph |
| Alt + `+` | Canvas Zoom in |
| Alt + `-` | Canvas Zoom out |
| Ctrl + Shift + LMB + Vertical drag | Canvas Zoom in/out |
| Q | Toggle visibility of the queue |
| H | Toggle visibility of history |
| R | Refresh graph |
| Double-Click LMB | Open node quick search palette |
Ctrl can also be replaced with Cmd instead for macOS users
@ -133,7 +136,16 @@ After this you should have everything installed and can proceed to running Comfy
### Others:
#### [Intel Arc](https://github.com/comfyanonymous/ComfyUI/discussions/476)
#### Intel GPUs
Intel GPU support is available for all Intel GPUs supported by Intel's Extension for Pytorch (IPEX) with the support requirements listed in the [Installation](https://intel.github.io/intel-extension-for-pytorch/index.html#installation?platform=gpu) page. Choose your platform and method of install and follow the instructions. The steps are as follows:
1. Start by installing the drivers or kernel listed or newer in the Installation page of IPEX linked above for Windows and Linux if needed.
1. Follow the instructions to install [Intel's oneAPI Basekit](https://www.intel.com/content/www/us/en/developer/tools/oneapi/base-toolkit-download.html) for your platform.
1. Install the packages for IPEX using the instructions provided in the Installation page for your platform.
1. Follow the [ComfyUI manual installation](#manual-install-windows-linux) instructions for Windows and Linux and run ComfyUI normally as described above after everything is installed.
Additional discussion and help can be found [here](https://github.com/comfyanonymous/ComfyUI/discussions/476).
#### Apple Mac silicon
@ -195,20 +207,20 @@ To use a textual inversion concepts/embeddings in a text prompt put them in the
```embedding:embedding_filename.pt```
## How to increase generation speed?
Make sure you use the regular loaders/Load Checkpoint node to load checkpoints. It will auto pick the right settings depending on your GPU.
You can set this command line setting to disable the upcasting to fp32 in some cross attention operations which will increase your speed. Note that this will very likely give you black images on SD2.x models. If you use xformers or pytorch attention this option does not do anything.
```--dont-upcast-attention```
## How to show high-quality previews?
Use ```--preview-method auto``` to enable previews.
The default installation includes a fast latent preview method that's low-resolution. To enable higher-quality previews with [TAESD](https://github.com/madebyollin/taesd), download the [taesd_decoder.pth](https://github.com/madebyollin/taesd/raw/main/taesd_decoder.pth) (for SD1.x and SD2.x) and [taesdxl_decoder.pth](https://github.com/madebyollin/taesd/raw/main/taesdxl_decoder.pth) (for SDXL) models and place them in the `models/vae_approx` folder. Once they're installed, restart ComfyUI to enable high-quality previews.
## How to use TLS/SSL?
Generate a self-signed certificate (not appropriate for shared/production use) and key by running the command: `openssl req -x509 -newkey rsa:4096 -keyout key.pem -out cert.pem -sha256 -days 3650 -nodes -subj "/C=XX/ST=StateName/L=CityName/O=CompanyName/OU=CompanySectionName/CN=CommonNameOrHostname"`
Use `--tls-keyfile key.pem --tls-certfile cert.pem` to enable TLS/SSL, the app will now be accessible with `https://...` instead of `http://...`.
> Note: Windows users can use [alexisrolland/docker-openssl](https://github.com/alexisrolland/docker-openssl) or one of the [3rd party binary distributions](https://wiki.openssl.org/index.php/Binaries) to run the command example above.
<br/><br/>If you use a container, note that the volume mount `-v` can be a relative path so `... -v ".\:/openssl-certs" ...` would create the key & cert files in the current directory of your command prompt or powershell terminal.
## Support and dev channel
[Matrix space: #comfyui_space:matrix.org](https://app.element.io/#/room/%23comfyui_space%3Amatrix.org) (it's like discord but open source).

View File

@ -52,6 +52,7 @@ class ControlNet(nn.Module):
adm_in_channels=None,
transformer_depth_middle=None,
transformer_depth_output=None,
attn_precision=None,
device=None,
operations=comfy.ops.disable_weight_init,
**kwargs,
@ -202,7 +203,7 @@ class ControlNet(nn.Module):
SpatialTransformer(
ch, num_heads, dim_head, depth=num_transformers, context_dim=context_dim,
disable_self_attn=disabled_sa, use_linear=use_linear_in_transformer,
use_checkpoint=use_checkpoint, dtype=self.dtype, device=device, operations=operations
use_checkpoint=use_checkpoint, attn_precision=attn_precision, dtype=self.dtype, device=device, operations=operations
)
)
self.input_blocks.append(TimestepEmbedSequential(*layers))
@ -262,7 +263,7 @@ class ControlNet(nn.Module):
mid_block += [SpatialTransformer( # always uses a self-attn
ch, num_heads, dim_head, depth=transformer_depth_middle, context_dim=context_dim,
disable_self_attn=disable_middle_self_attn, use_linear=use_linear_in_transformer,
use_checkpoint=use_checkpoint, dtype=self.dtype, device=device, operations=operations
use_checkpoint=use_checkpoint, attn_precision=attn_precision, dtype=self.dtype, device=device, operations=operations
),
ResBlock(
ch,

View File

@ -35,6 +35,8 @@ parser = argparse.ArgumentParser()
parser.add_argument("--listen", type=str, default="127.0.0.1", metavar="IP", nargs="?", const="0.0.0.0", help="Specify the IP address to listen on (default: 127.0.0.1). If --listen is provided without an argument, it defaults to 0.0.0.0. (listens on all)")
parser.add_argument("--port", type=int, default=8188, help="Set the listen port.")
parser.add_argument("--tls-keyfile", type=str, help="Path to TLS (SSL) key file. Enables TLS, makes app accessible at https://... requires --tls-certfile to function")
parser.add_argument("--tls-certfile", type=str, help="Path to TLS (SSL) certificate file. Enables TLS, makes app accessible at https://... requires --tls-keyfile to function")
parser.add_argument("--enable-cors-header", type=str, default=None, metavar="ORIGIN", nargs="?", const="*", help="Enable CORS (Cross-Origin Resource Sharing) with optional origin or allow all with default '*'.")
parser.add_argument("--max-upload-size", type=float, default=100, help="Set the maximum upload size in MB.")
@ -49,7 +51,6 @@ cm_group = parser.add_mutually_exclusive_group()
cm_group.add_argument("--cuda-malloc", action="store_true", help="Enable cudaMallocAsync (enabled by default for torch 2.0 and up).")
cm_group.add_argument("--disable-cuda-malloc", action="store_true", help="Disable cudaMallocAsync.")
parser.add_argument("--dont-upcast-attention", action="store_true", help="Disable upcasting of attention. Can boost speed but increase the chances of black images.")
fp_group = parser.add_mutually_exclusive_group()
fp_group.add_argument("--force-fp32", action="store_true", help="Force fp32 (If this makes your GPU work better please report it).")
@ -94,6 +95,11 @@ attn_group.add_argument("--use-pytorch-cross-attention", action="store_true", he
parser.add_argument("--disable-xformers", action="store_true", help="Disable xformers.")
upcast = parser.add_mutually_exclusive_group()
upcast.add_argument("--force-upcast-attention", action="store_true", help="Force enable attention upcasting, please report if it fixes black images.")
upcast.add_argument("--dont-upcast-attention", action="store_true", help="Disable all upcasting of attention. Should be unnecessary except for debugging.")
vram_group = parser.add_mutually_exclusive_group()
vram_group.add_argument("--gpu-only", action="store_true", help="Store and run everything (text encoders/CLIP models, etc... on the GPU).")
vram_group.add_argument("--highvram", action="store_true", help="By default models will be unloaded to CPU memory after being used. This option keeps them in GPU memory.")

View File

@ -527,6 +527,9 @@ def sample_dpmpp_2s_ancestral(model, x, sigmas, extra_args=None, callback=None,
@torch.no_grad()
def sample_dpmpp_sde(model, x, sigmas, extra_args=None, callback=None, disable=None, eta=1., s_noise=1., noise_sampler=None, r=1 / 2):
"""DPM-Solver++ (stochastic)."""
if len(sigmas) <= 1:
return x
sigma_min, sigma_max = sigmas[sigmas > 0].min(), sigmas.max()
seed = extra_args.get("seed", None)
noise_sampler = BrownianTreeNoiseSampler(x, sigma_min, sigma_max, seed=seed, cpu=True) if noise_sampler is None else noise_sampler
@ -595,6 +598,8 @@ def sample_dpmpp_2m(model, x, sigmas, extra_args=None, callback=None, disable=No
@torch.no_grad()
def sample_dpmpp_2m_sde(model, x, sigmas, extra_args=None, callback=None, disable=None, eta=1., s_noise=1., noise_sampler=None, solver_type='midpoint'):
"""DPM-Solver++(2M) SDE."""
if len(sigmas) <= 1:
return x
if solver_type not in {'heun', 'midpoint'}:
raise ValueError('solver_type must be \'heun\' or \'midpoint\'')
@ -642,6 +647,9 @@ def sample_dpmpp_2m_sde(model, x, sigmas, extra_args=None, callback=None, disabl
def sample_dpmpp_3m_sde(model, x, sigmas, extra_args=None, callback=None, disable=None, eta=1., s_noise=1., noise_sampler=None):
"""DPM-Solver++(3M) SDE."""
if len(sigmas) <= 1:
return x
seed = extra_args.get("seed", None)
sigma_min, sigma_max = sigmas[sigmas > 0].min(), sigmas.max()
noise_sampler = BrownianTreeNoiseSampler(x, sigma_min, sigma_max, seed=seed, cpu=True) if noise_sampler is None else noise_sampler
@ -690,18 +698,27 @@ def sample_dpmpp_3m_sde(model, x, sigmas, extra_args=None, callback=None, disabl
@torch.no_grad()
def sample_dpmpp_3m_sde_gpu(model, x, sigmas, extra_args=None, callback=None, disable=None, eta=1., s_noise=1., noise_sampler=None):
if len(sigmas) <= 1:
return x
sigma_min, sigma_max = sigmas[sigmas > 0].min(), sigmas.max()
noise_sampler = BrownianTreeNoiseSampler(x, sigma_min, sigma_max, seed=extra_args.get("seed", None), cpu=False) if noise_sampler is None else noise_sampler
return sample_dpmpp_3m_sde(model, x, sigmas, extra_args=extra_args, callback=callback, disable=disable, eta=eta, s_noise=s_noise, noise_sampler=noise_sampler)
@torch.no_grad()
def sample_dpmpp_2m_sde_gpu(model, x, sigmas, extra_args=None, callback=None, disable=None, eta=1., s_noise=1., noise_sampler=None, solver_type='midpoint'):
if len(sigmas) <= 1:
return x
sigma_min, sigma_max = sigmas[sigmas > 0].min(), sigmas.max()
noise_sampler = BrownianTreeNoiseSampler(x, sigma_min, sigma_max, seed=extra_args.get("seed", None), cpu=False) if noise_sampler is None else noise_sampler
return sample_dpmpp_2m_sde(model, x, sigmas, extra_args=extra_args, callback=callback, disable=disable, eta=eta, s_noise=s_noise, noise_sampler=noise_sampler, solver_type=solver_type)
@torch.no_grad()
def sample_dpmpp_sde_gpu(model, x, sigmas, extra_args=None, callback=None, disable=None, eta=1., s_noise=1., noise_sampler=None, r=1 / 2):
if len(sigmas) <= 1:
return x
sigma_min, sigma_max = sigmas[sigmas > 0].min(), sigmas.max()
noise_sampler = BrownianTreeNoiseSampler(x, sigma_min, sigma_max, seed=extra_args.get("seed", None), cpu=False) if noise_sampler is None else noise_sampler
return sample_dpmpp_sde(model, x, sigmas, extra_args=extra_args, callback=callback, disable=disable, eta=eta, s_noise=s_noise, noise_sampler=noise_sampler, r=r)

View File

@ -6,7 +6,7 @@ from einops import rearrange, repeat
from typing import Optional, Any
import logging
from .diffusionmodules.util import checkpoint, AlphaBlender, timestep_embedding
from .diffusionmodules.util import AlphaBlender, timestep_embedding
from .sub_quadratic_attention import efficient_dot_product_attention
from comfy import model_management
@ -19,13 +19,14 @@ from comfy.cli_args import args
import comfy.ops
ops = comfy.ops.disable_weight_init
# CrossAttn precision handling
if args.dont_upcast_attention:
logging.info("disabling upcasting of attention")
_ATTN_PRECISION = "fp16"
else:
_ATTN_PRECISION = "fp32"
FORCE_UPCAST_ATTENTION_DTYPE = model_management.force_upcast_attention_dtype()
def get_attn_precision(attn_precision):
if args.dont_upcast_attention:
return None
if FORCE_UPCAST_ATTENTION_DTYPE is not None:
return FORCE_UPCAST_ATTENTION_DTYPE
return attn_precision
def exists(val):
return val is not None
@ -85,7 +86,9 @@ class FeedForward(nn.Module):
def Normalize(in_channels, dtype=None, device=None):
return torch.nn.GroupNorm(num_groups=32, num_channels=in_channels, eps=1e-6, affine=True, dtype=dtype, device=device)
def attention_basic(q, k, v, heads, mask=None):
def attention_basic(q, k, v, heads, mask=None, attn_precision=None):
attn_precision = get_attn_precision(attn_precision)
b, _, dim_head = q.shape
dim_head //= heads
scale = dim_head ** -0.5
@ -101,7 +104,7 @@ def attention_basic(q, k, v, heads, mask=None):
)
# force cast to fp32 to avoid overflowing
if _ATTN_PRECISION =="fp32":
if attn_precision == torch.float32:
sim = einsum('b i d, b j d -> b i j', q.float(), k.float()) * scale
else:
sim = einsum('b i d, b j d -> b i j', q, k) * scale
@ -135,7 +138,9 @@ def attention_basic(q, k, v, heads, mask=None):
return out
def attention_sub_quad(query, key, value, heads, mask=None):
def attention_sub_quad(query, key, value, heads, mask=None, attn_precision=None):
attn_precision = get_attn_precision(attn_precision)
b, _, dim_head = query.shape
dim_head //= heads
@ -146,7 +151,7 @@ def attention_sub_quad(query, key, value, heads, mask=None):
key = key.unsqueeze(3).reshape(b, -1, heads, dim_head).permute(0, 2, 3, 1).reshape(b * heads, dim_head, -1)
dtype = query.dtype
upcast_attention = _ATTN_PRECISION =="fp32" and query.dtype != torch.float32
upcast_attention = attn_precision == torch.float32 and query.dtype != torch.float32
if upcast_attention:
bytes_per_token = torch.finfo(torch.float32).bits//8
else:
@ -195,7 +200,9 @@ def attention_sub_quad(query, key, value, heads, mask=None):
hidden_states = hidden_states.unflatten(0, (-1, heads)).transpose(1,2).flatten(start_dim=2)
return hidden_states
def attention_split(q, k, v, heads, mask=None):
def attention_split(q, k, v, heads, mask=None, attn_precision=None):
attn_precision = get_attn_precision(attn_precision)
b, _, dim_head = q.shape
dim_head //= heads
scale = dim_head ** -0.5
@ -214,10 +221,12 @@ def attention_split(q, k, v, heads, mask=None):
mem_free_total = model_management.get_free_memory(q.device)
if _ATTN_PRECISION =="fp32":
if attn_precision == torch.float32:
element_size = 4
upcast = True
else:
element_size = q.element_size()
upcast = False
gb = 1024 ** 3
tensor_size = q.shape[0] * q.shape[1] * k.shape[1] * element_size
@ -251,7 +260,7 @@ def attention_split(q, k, v, heads, mask=None):
slice_size = q.shape[1] // steps if (q.shape[1] % steps) == 0 else q.shape[1]
for i in range(0, q.shape[1], slice_size):
end = i + slice_size
if _ATTN_PRECISION =="fp32":
if upcast:
with torch.autocast(enabled=False, device_type = 'cuda'):
s1 = einsum('b i d, b j d -> b i j', q[:, i:end].float(), k.float()) * scale
else:
@ -297,24 +306,30 @@ def attention_split(q, k, v, heads, mask=None):
BROKEN_XFORMERS = False
try:
x_vers = xformers.__version__
#I think 0.0.23 is also broken (q with bs bigger than 65535 gives CUDA error)
BROKEN_XFORMERS = x_vers.startswith("0.0.21") or x_vers.startswith("0.0.22") or x_vers.startswith("0.0.23")
# XFormers bug confirmed on all versions from 0.0.21 to 0.0.26 (q with bs bigger than 65535 gives CUDA error)
BROKEN_XFORMERS = x_vers.startswith("0.0.2") and not x_vers.startswith("0.0.20")
except:
pass
def attention_xformers(q, k, v, heads, mask=None):
def attention_xformers(q, k, v, heads, mask=None, attn_precision=None):
b, _, dim_head = q.shape
dim_head //= heads
disabled_xformers = False
if BROKEN_XFORMERS:
if b * heads > 65535:
return attention_pytorch(q, k, v, heads, mask)
disabled_xformers = True
if not disabled_xformers:
if torch.jit.is_tracing() or torch.jit.is_scripting():
disabled_xformers = True
if disabled_xformers:
return attention_pytorch(q, k, v, heads, mask)
q, k, v = map(
lambda t: t.unsqueeze(3)
.reshape(b, -1, heads, dim_head)
.permute(0, 2, 1, 3)
.reshape(b * heads, -1, dim_head)
.contiguous(),
lambda t: t.reshape(b, -1, heads, dim_head),
(q, k, v),
)
@ -327,14 +342,11 @@ def attention_xformers(q, k, v, heads, mask=None):
out = xformers.ops.memory_efficient_attention(q, k, v, attn_bias=mask)
out = (
out.unsqueeze(0)
.reshape(b, heads, -1, dim_head)
.permute(0, 2, 1, 3)
.reshape(b, -1, heads * dim_head)
out.reshape(b, -1, heads * dim_head)
)
return out
def attention_pytorch(q, k, v, heads, mask=None):
def attention_pytorch(q, k, v, heads, mask=None, attn_precision=None):
b, _, dim_head = q.shape
dim_head //= heads
q, k, v = map(
@ -384,10 +396,11 @@ def optimized_attention_for_device(device, mask=False, small_input=False):
class CrossAttention(nn.Module):
def __init__(self, query_dim, context_dim=None, heads=8, dim_head=64, dropout=0., dtype=None, device=None, operations=ops):
def __init__(self, query_dim, context_dim=None, heads=8, dim_head=64, dropout=0., attn_precision=None, dtype=None, device=None, operations=ops):
super().__init__()
inner_dim = dim_head * heads
context_dim = default(context_dim, query_dim)
self.attn_precision = attn_precision
self.heads = heads
self.dim_head = dim_head
@ -409,15 +422,15 @@ class CrossAttention(nn.Module):
v = self.to_v(context)
if mask is None:
out = optimized_attention(q, k, v, self.heads)
out = optimized_attention(q, k, v, self.heads, attn_precision=self.attn_precision)
else:
out = optimized_attention_masked(q, k, v, self.heads, mask)
out = optimized_attention_masked(q, k, v, self.heads, mask, attn_precision=self.attn_precision)
return self.to_out(out)
class BasicTransformerBlock(nn.Module):
def __init__(self, dim, n_heads, d_head, dropout=0., context_dim=None, gated_ff=True, checkpoint=True, ff_in=False, inner_dim=None,
disable_self_attn=False, disable_temporal_crossattention=False, switch_temporal_ca_to_sa=False, dtype=None, device=None, operations=ops):
disable_self_attn=False, disable_temporal_crossattention=False, switch_temporal_ca_to_sa=False, attn_precision=None, dtype=None, device=None, operations=ops):
super().__init__()
self.ff_in = ff_in or inner_dim is not None
@ -425,6 +438,7 @@ class BasicTransformerBlock(nn.Module):
inner_dim = dim
self.is_res = inner_dim == dim
self.attn_precision = attn_precision
if self.ff_in:
self.norm_in = operations.LayerNorm(dim, dtype=dtype, device=device)
@ -432,7 +446,7 @@ class BasicTransformerBlock(nn.Module):
self.disable_self_attn = disable_self_attn
self.attn1 = CrossAttention(query_dim=inner_dim, heads=n_heads, dim_head=d_head, dropout=dropout,
context_dim=context_dim if self.disable_self_attn else None, dtype=dtype, device=device, operations=operations) # is a self-attention if not self.disable_self_attn
context_dim=context_dim if self.disable_self_attn else None, attn_precision=self.attn_precision, dtype=dtype, device=device, operations=operations) # is a self-attention if not self.disable_self_attn
self.ff = FeedForward(inner_dim, dim_out=dim, dropout=dropout, glu=gated_ff, dtype=dtype, device=device, operations=operations)
if disable_temporal_crossattention:
@ -446,20 +460,16 @@ class BasicTransformerBlock(nn.Module):
context_dim_attn2 = context_dim
self.attn2 = CrossAttention(query_dim=inner_dim, context_dim=context_dim_attn2,
heads=n_heads, dim_head=d_head, dropout=dropout, dtype=dtype, device=device, operations=operations) # is self-attn if context is none
heads=n_heads, dim_head=d_head, dropout=dropout, attn_precision=self.attn_precision, dtype=dtype, device=device, operations=operations) # is self-attn if context is none
self.norm2 = operations.LayerNorm(inner_dim, dtype=dtype, device=device)
self.norm1 = operations.LayerNorm(inner_dim, dtype=dtype, device=device)
self.norm3 = operations.LayerNorm(inner_dim, dtype=dtype, device=device)
self.checkpoint = checkpoint
self.n_heads = n_heads
self.d_head = d_head
self.switch_temporal_ca_to_sa = switch_temporal_ca_to_sa
def forward(self, x, context=None, transformer_options={}):
return checkpoint(self._forward, (x, context, transformer_options), self.parameters(), self.checkpoint)
def _forward(self, x, context=None, transformer_options={}):
extra_options = {}
block = transformer_options.get("block", None)
block_index = transformer_options.get("block_index", 0)
@ -476,6 +486,7 @@ class BasicTransformerBlock(nn.Module):
extra_options["n_heads"] = self.n_heads
extra_options["dim_head"] = self.d_head
extra_options["attn_precision"] = self.attn_precision
if self.ff_in:
x_skip = x
@ -586,7 +597,7 @@ class SpatialTransformer(nn.Module):
def __init__(self, in_channels, n_heads, d_head,
depth=1, dropout=0., context_dim=None,
disable_self_attn=False, use_linear=False,
use_checkpoint=True, dtype=None, device=None, operations=ops):
use_checkpoint=True, attn_precision=None, dtype=None, device=None, operations=ops):
super().__init__()
if exists(context_dim) and not isinstance(context_dim, list):
context_dim = [context_dim] * depth
@ -604,7 +615,7 @@ class SpatialTransformer(nn.Module):
self.transformer_blocks = nn.ModuleList(
[BasicTransformerBlock(inner_dim, n_heads, d_head, dropout=dropout, context_dim=context_dim[d],
disable_self_attn=disable_self_attn, checkpoint=use_checkpoint, dtype=dtype, device=device, operations=operations)
disable_self_attn=disable_self_attn, checkpoint=use_checkpoint, attn_precision=attn_precision, dtype=dtype, device=device, operations=operations)
for d in range(depth)]
)
if not use_linear:
@ -625,7 +636,7 @@ class SpatialTransformer(nn.Module):
x = self.norm(x)
if not self.use_linear:
x = self.proj_in(x)
x = rearrange(x, 'b c h w -> b (h w) c').contiguous()
x = x.movedim(1, 3).flatten(1, 2).contiguous()
if self.use_linear:
x = self.proj_in(x)
for i, block in enumerate(self.transformer_blocks):
@ -633,7 +644,7 @@ class SpatialTransformer(nn.Module):
x = block(x, context=context[i], transformer_options=transformer_options)
if self.use_linear:
x = self.proj_out(x)
x = rearrange(x, 'b (h w) c -> b c h w', h=h, w=w).contiguous()
x = x.reshape(x.shape[0], h, w, x.shape[-1]).movedim(3, 1).contiguous()
if not self.use_linear:
x = self.proj_out(x)
return x + x_in
@ -660,6 +671,7 @@ class SpatialVideoTransformer(SpatialTransformer):
disable_self_attn=False,
disable_temporal_crossattention=False,
max_time_embed_period: int = 10000,
attn_precision=None,
dtype=None, device=None, operations=ops
):
super().__init__(
@ -672,6 +684,7 @@ class SpatialVideoTransformer(SpatialTransformer):
context_dim=context_dim,
use_linear=use_linear,
disable_self_attn=disable_self_attn,
attn_precision=attn_precision,
dtype=dtype, device=device, operations=operations
)
self.time_depth = time_depth
@ -701,6 +714,7 @@ class SpatialVideoTransformer(SpatialTransformer):
inner_dim=time_mix_inner_dim,
disable_self_attn=disable_self_attn,
disable_temporal_crossattention=disable_temporal_crossattention,
attn_precision=attn_precision,
dtype=dtype, device=device, operations=operations
)
for _ in range(self.depth)

View File

@ -3,7 +3,6 @@ import math
import torch
import torch.nn as nn
import numpy as np
from einops import rearrange
from typing import Optional, Any
import logging

View File

@ -258,7 +258,7 @@ class ResBlock(TimestepBlock):
else:
if emb_out is not None:
if self.exchange_temb_dims:
emb_out = rearrange(emb_out, "b t c ... -> b c t ...")
emb_out = emb_out.movedim(1, 2)
h = h + emb_out
h = self.out_layers(h)
return self.skip_connection(x) + h
@ -431,6 +431,7 @@ class UNetModel(nn.Module):
video_kernel_size=None,
disable_temporal_crossattention=False,
max_ddpm_temb_period=10000,
attn_precision=None,
device=None,
operations=ops,
):
@ -550,13 +551,14 @@ class UNetModel(nn.Module):
disable_self_attn=disable_self_attn,
disable_temporal_crossattention=disable_temporal_crossattention,
max_time_embed_period=max_ddpm_temb_period,
attn_precision=attn_precision,
dtype=self.dtype, device=device, operations=operations
)
else:
return SpatialTransformer(
ch, num_heads, dim_head, depth=depth, context_dim=context_dim,
disable_self_attn=disable_self_attn, use_linear=use_linear_in_transformer,
use_checkpoint=use_checkpoint, dtype=self.dtype, device=device, operations=operations
use_checkpoint=use_checkpoint, attn_precision=attn_precision, dtype=self.dtype, device=device, operations=operations
)
def get_resblock(

View File

@ -162,7 +162,7 @@ class BaseModel(torch.nn.Module):
c_concat = kwargs.get("noise_concat", None)
if c_concat is not None:
out['c_concat'] = comfy.conds.CONDNoiseShape(data)
out['c_concat'] = comfy.conds.CONDNoiseShape(c_concat)
return out

View File

@ -5,6 +5,7 @@ from comfy.cli_args import args
import comfy.utils
import torch
import sys
import platform
class VRAMState(Enum):
DISABLED = 0 #No vram present: no need to move models to vram
@ -83,7 +84,7 @@ def get_torch_device():
return torch.device("cpu")
else:
if is_intel_xpu():
return torch.device("xpu")
return torch.device("xpu", torch.xpu.current_device())
else:
return torch.device(torch.cuda.current_device())
@ -102,8 +103,8 @@ def get_total_memory(dev=None, torch_total_too=False):
elif is_intel_xpu():
stats = torch.xpu.memory_stats(dev)
mem_reserved = stats['reserved_bytes.all.current']
mem_total = torch.xpu.get_device_properties(dev).total_memory
mem_total_torch = mem_reserved
mem_total = torch.xpu.get_device_properties(dev).total_memory
else:
stats = torch.cuda.memory_stats(dev)
mem_reserved = stats['reserved_bytes.all.current']
@ -119,10 +120,11 @@ 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)
logging.info("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:
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:
logging.info("pytorch version: {}".format(torch.version.__version__))
except:
pass
try:
OOM_EXCEPTION = torch.cuda.OutOfMemoryError
@ -285,7 +287,7 @@ class LoadedModel:
else:
return self.model_memory()
def model_load(self, lowvram_model_memory=0):
def model_load(self, lowvram_model_memory=0, force_patch_weights=False):
patch_model_to = self.device
self.model.model_patches_to(self.device)
@ -295,7 +297,7 @@ class LoadedModel:
try:
if lowvram_model_memory > 0 and load_weights:
self.real_model = self.model.patch_model_lowvram(device_to=patch_model_to, lowvram_model_memory=lowvram_model_memory)
self.real_model = self.model.patch_model_lowvram(device_to=patch_model_to, lowvram_model_memory=lowvram_model_memory, force_patch_weights=force_patch_weights)
else:
self.real_model = self.model.patch_model(device_to=patch_model_to, patch_weights=load_weights)
except Exception as e:
@ -304,11 +306,16 @@ class LoadedModel:
raise e
if is_intel_xpu() and not args.disable_ipex_optimize:
self.real_model = torch.xpu.optimize(self.real_model.eval(), inplace=True, auto_kernel_selection=True, graph_mode=True)
self.real_model = ipex.optimize(self.real_model.eval(), graph_mode=True, concat_linear=True)
self.weights_loaded = True
return self.real_model
def should_reload_model(self, force_patch_weights=False):
if force_patch_weights and self.model.lowvram_patch_counter > 0:
return True
return False
def model_unload(self, unpatch_weights=True):
self.model.unpatch_model(self.model.offload_device, unpatch_weights=unpatch_weights)
self.model.model_patches_to(self.model.offload_device)
@ -379,7 +386,7 @@ def free_memory(memory_required, device, keep_loaded=[]):
if mem_free_torch > mem_free_total * 0.25:
soft_empty_cache()
def load_models_gpu(models, memory_required=0):
def load_models_gpu(models, memory_required=0, force_patch_weights=False):
global vram_state
inference_memory = minimum_inference_memory()
@ -391,12 +398,22 @@ def load_models_gpu(models, memory_required=0):
models_already_loaded = []
for x in models:
loaded_model = LoadedModel(x)
loaded = None
if loaded_model in current_loaded_models:
index = current_loaded_models.index(loaded_model)
current_loaded_models.insert(0, current_loaded_models.pop(index))
models_already_loaded.append(loaded_model)
else:
try:
loaded_model_index = current_loaded_models.index(loaded_model)
except:
loaded_model_index = None
if loaded_model_index is not None:
loaded = current_loaded_models[loaded_model_index]
if loaded.should_reload_model(force_patch_weights=force_patch_weights): #TODO: cleanup this model reload logic
current_loaded_models.pop(loaded_model_index).model_unload(unpatch_weights=True)
loaded = None
else:
models_already_loaded.append(loaded)
if loaded is None:
if hasattr(x, "model"):
logging.info(f"Requested to load {x.model.__class__.__name__}")
models_to_load.append(loaded_model)
@ -436,15 +453,13 @@ def load_models_gpu(models, memory_required=0):
model_size = loaded_model.model_memory_required(torch_dev)
current_free_mem = get_free_memory(torch_dev)
lowvram_model_memory = int(max(64 * (1024 * 1024), (current_free_mem - 1024 * (1024 * 1024)) / 1.3 ))
if model_size > (current_free_mem - inference_memory): #only switch to lowvram if really necessary
vram_set_state = VRAMState.LOW_VRAM
else:
if model_size <= (current_free_mem - inference_memory): #only switch to lowvram if really necessary
lowvram_model_memory = 0
if vram_set_state == VRAMState.NO_VRAM:
lowvram_model_memory = 64 * 1024 * 1024
cur_loaded_model = loaded_model.model_load(lowvram_model_memory)
cur_loaded_model = loaded_model.model_load(lowvram_model_memory, force_patch_weights=force_patch_weights)
current_loaded_models.insert(0, loaded_model)
return
@ -552,8 +567,6 @@ def text_encoder_device():
if args.gpu_only:
return get_torch_device()
elif vram_state == VRAMState.HIGH_VRAM or vram_state == VRAMState.NORMAL_VRAM:
if is_intel_xpu():
return torch.device("cpu")
if should_use_fp16(prioritize_performance=False):
return get_torch_device()
else:
@ -617,8 +630,14 @@ def supports_dtype(device, dtype): #TODO
def device_supports_non_blocking(device):
if is_device_mps(device):
return False #pytorch bug? mps doesn't support non blocking
return True
def device_should_use_non_blocking(device):
if not device_supports_non_blocking(device):
return False
return False
# return True #TODO: figure out why this causes issues
# return True #TODO: figure out why this causes memory issues on Nvidia and possibly others
def cast_to_device(tensor, device, dtype, copy=False):
device_supports_cast = False
@ -630,7 +649,7 @@ def cast_to_device(tensor, device, dtype, copy=False):
elif is_intel_xpu():
device_supports_cast = True
non_blocking = device_supports_non_blocking(device)
non_blocking = device_should_use_non_blocking(device)
if device_supports_cast:
if copy:
@ -673,6 +692,18 @@ def pytorch_attention_flash_attention():
return True
return False
def force_upcast_attention_dtype():
upcast = args.force_upcast_attention
try:
if platform.mac_ver()[0] in ['14.5']: #black image bug on OSX Sonoma 14.5
upcast = True
except:
pass
if upcast:
return torch.float32
else:
return None
def get_free_memory(dev=None, torch_free_too=False):
global directml_enabled
if dev is None:
@ -688,10 +719,10 @@ def get_free_memory(dev=None, torch_free_too=False):
elif is_intel_xpu():
stats = torch.xpu.memory_stats(dev)
mem_active = stats['active_bytes.all.current']
mem_allocated = stats['allocated_bytes.all.current']
mem_reserved = stats['reserved_bytes.all.current']
mem_free_torch = mem_reserved - mem_active
mem_free_total = torch.xpu.get_device_properties(dev).total_memory - mem_allocated
mem_free_xpu = torch.xpu.get_device_properties(dev).total_memory - mem_reserved
mem_free_total = mem_free_xpu + mem_free_torch
else:
stats = torch.cuda.memory_stats(dev)
mem_active = stats['active_bytes.all.current']

View File

@ -6,6 +6,8 @@ import uuid
import comfy.utils
import comfy.model_management
from comfy.types import UnetWrapperFunction
def apply_weight_decompose(dora_scale, weight):
weight_norm = (
@ -16,7 +18,7 @@ def apply_weight_decompose(dora_scale, weight):
.transpose(0, 1)
)
return weight * (dora_scale / weight_norm)
return weight * (dora_scale / weight_norm).type(weight.dtype)
def set_model_options_patch_replace(model_options, patch, name, block_name, number, transformer_index=None):
to = model_options["transformer_options"].copy()
@ -58,6 +60,7 @@ class ModelPatcher:
self.weight_inplace_update = weight_inplace_update
self.model_lowvram = False
self.lowvram_patch_counter = 0
self.patches_uuid = uuid.uuid4()
def model_size(self):
@ -116,7 +119,7 @@ class ModelPatcher:
if disable_cfg1_optimization:
self.model_options["disable_cfg1_optimization"] = True
def set_model_unet_function_wrapper(self, unet_wrapper_function):
def set_model_unet_function_wrapper(self, unet_wrapper_function: UnetWrapperFunction):
self.model_options["model_function_wrapper"] = unet_wrapper_function
def set_model_denoise_mask_function(self, denoise_mask_function):
@ -272,7 +275,7 @@ class ModelPatcher:
return self.model
def patch_model_lowvram(self, device_to=None, lowvram_model_memory=0):
def patch_model_lowvram(self, device_to=None, lowvram_model_memory=0, force_patch_weights=False):
self.patch_model(device_to, patch_weights=False)
logging.info("loading in lowvram mode {}".format(lowvram_model_memory/(1024 * 1024)))
@ -284,6 +287,7 @@ class ModelPatcher:
return self.model_patcher.calculate_weight(self.model_patcher.patches[self.key], weight, self.key)
mem_counter = 0
patch_counter = 0
for n, m in self.model.named_modules():
lowvram_weight = False
if hasattr(m, "comfy_cast_weights"):
@ -296,9 +300,17 @@ class ModelPatcher:
if lowvram_weight:
if weight_key in self.patches:
m.weight_function = LowVramPatch(weight_key, self)
if force_patch_weights:
self.patch_weight_to_device(weight_key)
else:
m.weight_function = LowVramPatch(weight_key, self)
patch_counter += 1
if bias_key in self.patches:
m.bias_function = LowVramPatch(bias_key, self)
if force_patch_weights:
self.patch_weight_to_device(bias_key)
else:
m.bias_function = LowVramPatch(bias_key, self)
patch_counter += 1
m.prev_comfy_cast_weights = m.comfy_cast_weights
m.comfy_cast_weights = True
@ -311,6 +323,7 @@ class ModelPatcher:
logging.debug("lowvram: loaded module regularly {}".format(m))
self.model_lowvram = True
self.lowvram_patch_counter = patch_counter
return self.model
def calculate_weight(self, patches, weight, key):
@ -462,6 +475,7 @@ class ModelPatcher:
m.bias_function = None
self.model_lowvram = False
self.lowvram_patch_counter = 0
keys = list(self.backup.keys())

View File

@ -21,7 +21,7 @@ import comfy.model_management
def cast_bias_weight(s, input):
bias = None
non_blocking = comfy.model_management.device_supports_non_blocking(input.device)
non_blocking = comfy.model_management.device_should_use_non_blocking(input.device)
if s.bias is not None:
bias = s.bias.to(device=input.device, dtype=input.dtype, non_blocking=non_blocking)
if s.bias_function is not None:

View File

@ -539,6 +539,9 @@ class KSAMPLER(Sampler):
def ksampler(sampler_name, extra_options={}, inpaint_options={}):
if sampler_name == "dpm_fast":
def dpm_fast_function(model, noise, sigmas, extra_args, callback, disable):
if len(sigmas) <= 1:
return noise
sigma_min = sigmas[-1]
if sigma_min == 0:
sigma_min = sigmas[-2]
@ -547,6 +550,9 @@ def ksampler(sampler_name, extra_options={}, inpaint_options={}):
sampler_function = dpm_fast_function
elif sampler_name == "dpm_adaptive":
def dpm_adaptive_function(model, noise, sigmas, extra_args, callback, disable, **extra_options):
if len(sigmas) <= 1:
return noise
sigma_min = sigmas[-1]
if sigma_min == 0:
sigma_min = sigmas[-2]

View File

@ -418,6 +418,8 @@ def load_gligen(ckpt_path):
return comfy.model_patcher.ModelPatcher(model, load_device=model_management.get_torch_device(), offload_device=model_management.unet_offload_device())
def load_checkpoint(config_path=None, ckpt_path=None, output_vae=True, output_clip=True, embedding_directory=None, state_dict=None, config=None):
logging.warning("Warning: The load checkpoint with config function is deprecated and will eventually be removed, please use the other one.")
model, clip, vae, _ = load_checkpoint_guess_config(ckpt_path, output_vae=output_vae, output_clip=output_clip, output_clipvision=False, embedding_directory=embedding_directory, output_model=True)
#TODO: this function is a mess and should be removed eventually
if config is None:
with open(config_path, 'r') as stream:
@ -425,81 +427,20 @@ def load_checkpoint(config_path=None, ckpt_path=None, output_vae=True, output_cl
model_config_params = config['model']['params']
clip_config = model_config_params['cond_stage_config']
scale_factor = model_config_params['scale_factor']
vae_config = model_config_params['first_stage_config']
fp16 = False
if "unet_config" in model_config_params:
if "params" in model_config_params["unet_config"]:
unet_config = model_config_params["unet_config"]["params"]
if "use_fp16" in unet_config:
fp16 = unet_config.pop("use_fp16")
if fp16:
unet_config["dtype"] = torch.float16
noise_aug_config = None
if "noise_aug_config" in model_config_params:
noise_aug_config = model_config_params["noise_aug_config"]
model_type = model_base.ModelType.EPS
if "parameterization" in model_config_params:
if model_config_params["parameterization"] == "v":
model_type = model_base.ModelType.V_PREDICTION
m = model.clone()
class ModelSamplingAdvanced(comfy.model_sampling.ModelSamplingDiscrete, comfy.model_sampling.V_PREDICTION):
pass
m.add_object_patch("model_sampling", ModelSamplingAdvanced(model.model.model_config))
model = m
clip = None
vae = None
layer_idx = clip_config.get("params", {}).get("layer_idx", None)
if layer_idx is not None:
clip.clip_layer(layer_idx)
class WeightsLoader(torch.nn.Module):
pass
if state_dict is None:
state_dict = comfy.utils.load_torch_file(ckpt_path)
class EmptyClass:
pass
model_config = comfy.supported_models_base.BASE({})
from . import latent_formats
model_config.latent_format = latent_formats.SD15(scale_factor=scale_factor)
model_config.unet_config = model_detection.convert_config(unet_config)
if config['model']["target"].endswith("ImageEmbeddingConditionedLatentDiffusion"):
model = model_base.SD21UNCLIP(model_config, noise_aug_config["params"], model_type=model_type)
else:
model = model_base.BaseModel(model_config, model_type=model_type)
if config['model']["target"].endswith("LatentInpaintDiffusion"):
model.set_inpaint()
if fp16:
model = model.half()
offload_device = model_management.unet_offload_device()
model = model.to(offload_device)
model.load_model_weights(state_dict, "model.diffusion_model.")
if output_vae:
vae_sd = comfy.utils.state_dict_prefix_replace(state_dict, {"first_stage_model.": ""}, filter_keys=True)
vae = VAE(sd=vae_sd, config=vae_config)
if output_clip:
w = WeightsLoader()
clip_target = EmptyClass()
clip_target.params = clip_config.get("params", {})
if clip_config["target"].endswith("FrozenOpenCLIPEmbedder"):
clip_target.clip = sd2_clip.SD2ClipModel
clip_target.tokenizer = sd2_clip.SD2Tokenizer
clip = CLIP(clip_target, embedding_directory=embedding_directory)
w.cond_stage_model = clip.cond_stage_model.clip_h
elif clip_config["target"].endswith("FrozenCLIPEmbedder"):
clip_target.clip = sd1_clip.SD1ClipModel
clip_target.tokenizer = sd1_clip.SD1Tokenizer
clip = CLIP(clip_target, embedding_directory=embedding_directory)
w.cond_stage_model = clip.cond_stage_model.clip_l
load_clip_weights(w, state_dict)
return (comfy.model_patcher.ModelPatcher(model, load_device=model_management.get_torch_device(), offload_device=offload_device), clip, vae)
return (model, clip, vae)
def load_checkpoint_guess_config(ckpt_path, output_vae=True, output_clip=True, output_clipvision=False, embedding_directory=None, output_model=True):
sd = comfy.utils.load_torch_file(ckpt_path)
@ -545,7 +486,11 @@ 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:
logging.warning("clip missing: {}".format(m))
m_filter = list(filter(lambda a: ".logit_scale" not in a and ".transformer.text_projection.weight" not in a, m))
if len(m_filter) > 0:
logging.warning("clip missing: {}".format(m))
else:
logging.debug("clip missing: {}".format(m))
if len(u) > 0:
logging.debug("clip unexpected {}:".format(u))
@ -617,7 +562,7 @@ def save_checkpoint(output_path, model, clip=None, vae=None, clip_vision=None, m
load_models.append(clip.load_model())
clip_sd = clip.get_sd()
model_management.load_models_gpu(load_models)
model_management.load_models_gpu(load_models, force_patch_weights=True)
clip_vision_sd = clip_vision.get_sd() if clip_vision is not None else None
sd = model.model.state_dict_for_saving(clip_sd, vae.get_sd(), clip_vision_sd)
for k in extra_keys:

View File

@ -65,6 +65,12 @@ class SD20(supported_models_base.BASE):
"use_temporal_attention": False,
}
unet_extra_config = {
"num_heads": -1,
"num_head_channels": 64,
"attn_precision": torch.float32,
}
latent_format = latent_formats.SD15
def model_type(self, state_dict, prefix=""):
@ -276,6 +282,12 @@ class SVD_img2vid(supported_models_base.BASE):
"use_temporal_resblock": True
}
unet_extra_config = {
"num_heads": -1,
"num_head_channels": 64,
"attn_precision": torch.float32,
}
clip_vision_prefix = "conditioner.embedders.0.open_clip.model.visual."
latent_format = latent_formats.SD15

32
comfy/types.py Normal file
View File

@ -0,0 +1,32 @@
import torch
from typing import Callable, Protocol, TypedDict, Optional, List
class UnetApplyFunction(Protocol):
"""Function signature protocol on comfy.model_base.BaseModel.apply_model"""
def __call__(self, x: torch.Tensor, t: torch.Tensor, **kwargs) -> torch.Tensor:
pass
class UnetApplyConds(TypedDict):
"""Optional conditions for unet apply function."""
c_concat: Optional[torch.Tensor]
c_crossattn: Optional[torch.Tensor]
control: Optional[torch.Tensor]
transformer_options: Optional[dict]
class UnetParams(TypedDict):
# Tensor of shape [B, C, H, W]
input: torch.Tensor
# Tensor of shape [B]
timestep: torch.Tensor
c: UnetApplyConds
# List of [0, 1], [0], [1], ...
# 0 means unconditional, 1 means conditional
cond_or_uncond: List[int]
UnetWrapperFunction = Callable[[UnetApplyFunction, UnetParams], torch.Tensor]

View File

@ -107,8 +107,7 @@ class SDTurboScheduler:
def get_sigmas(self, model, steps, denoise):
start_step = 10 - int(10 * denoise)
timesteps = torch.flip(torch.arange(1, 11) * 100 - 1, (0,))[start_step:start_step + steps]
comfy.model_management.load_models_gpu([model])
sigmas = model.model.model_sampling.sigma(timesteps)
sigmas = model.get_model_object("model_sampling").sigma(timesteps)
sigmas = torch.cat([sigmas, sigmas.new_zeros([1])])
return (sigmas, )
@ -140,6 +139,7 @@ class SplitSigmas:
}
}
RETURN_TYPES = ("SIGMAS","SIGMAS")
RETURN_NAMES = ("high_sigmas", "low_sigmas")
CATEGORY = "sampling/custom_sampling/sigmas"
FUNCTION = "get_sigmas"
@ -149,6 +149,27 @@ class SplitSigmas:
sigmas2 = sigmas[step:]
return (sigmas1, sigmas2)
class SplitSigmasDenoise:
@classmethod
def INPUT_TYPES(s):
return {"required":
{"sigmas": ("SIGMAS", ),
"denoise": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 1.0, "step": 0.01}),
}
}
RETURN_TYPES = ("SIGMAS","SIGMAS")
RETURN_NAMES = ("high_sigmas", "low_sigmas")
CATEGORY = "sampling/custom_sampling/sigmas"
FUNCTION = "get_sigmas"
def get_sigmas(self, sigmas, denoise):
steps = max(sigmas.shape[-1] - 1, 0)
total_steps = round(steps * denoise)
sigmas1 = sigmas[:-(total_steps)]
sigmas2 = sigmas[-(total_steps + 1):]
return (sigmas1, sigmas2)
class FlipSigmas:
@classmethod
def INPUT_TYPES(s):
@ -600,6 +621,7 @@ NODE_CLASS_MAPPINGS = {
"SamplerDPMPP_SDE": SamplerDPMPP_SDE,
"SamplerDPMAdaptative": SamplerDPMAdaptative,
"SplitSigmas": SplitSigmas,
"SplitSigmasDenoise": SplitSigmasDenoise,
"FlipSigmas": FlipSigmas,
"CFGGuider": CFGGuider,

View File

@ -42,7 +42,7 @@ class FreeU:
on_cpu_devices = {}
def output_block_patch(h, hsp, transformer_options):
scale = scale_dict.get(h.shape[1], None)
scale = scale_dict.get(int(h.shape[1]), None)
if scale is not None:
h[:,:h.shape[1] // 2] = h[:,:h.shape[1] // 2] * scale[0]
if hsp.device not in on_cpu_devices:
@ -81,7 +81,7 @@ class FreeU_V2:
on_cpu_devices = {}
def output_block_patch(h, hsp, transformer_options):
scale = scale_dict.get(h.shape[1], None)
scale = scale_dict.get(int(h.shape[1]), None)
if scale is not None:
hidden_mean = h.mean(1).unsqueeze(1)
B = hidden_mean.shape[0]

View File

@ -20,8 +20,9 @@ class PatchModelAddDownscale:
CATEGORY = "_for_testing"
def patch(self, model, block_number, downscale_factor, start_percent, end_percent, downscale_after_skip, downscale_method, upscale_method):
sigma_start = model.model.model_sampling.percent_to_sigma(start_percent)
sigma_end = model.model.model_sampling.percent_to_sigma(end_percent)
model_sampling = model.get_model_object("model_sampling")
sigma_start = model_sampling.percent_to_sigma(start_percent)
sigma_end = model_sampling.percent_to_sigma(end_percent)
def input_block_patch(h, transformer_options):
if transformer_options["block"][1] == block_number:

View File

@ -175,9 +175,14 @@ def save_checkpoint(model, clip=None, vae=None, clip_vision=None, filename_prefi
enable_modelspec = True
if isinstance(model.model, comfy.model_base.SDXL):
metadata["modelspec.architecture"] = "stable-diffusion-xl-v1-base"
if isinstance(model.model, comfy.model_base.SDXL_instructpix2pix):
metadata["modelspec.architecture"] = "stable-diffusion-xl-v1-edit"
else:
metadata["modelspec.architecture"] = "stable-diffusion-xl-v1-base"
elif isinstance(model.model, comfy.model_base.SDXLRefiner):
metadata["modelspec.architecture"] = "stable-diffusion-xl-v1-refiner"
elif isinstance(model.model, comfy.model_base.SVD_img2vid):
metadata["modelspec.architecture"] = "stable-video-diffusion-img2vid-v1"
else:
enable_modelspec = False
@ -262,7 +267,7 @@ class CLIPSave:
for x in extra_pnginfo:
metadata[x] = json.dumps(extra_pnginfo[x])
comfy.model_management.load_models_gpu([clip.load_model()])
comfy.model_management.load_models_gpu([clip.load_model()], force_patch_weights=True)
clip_sd = clip.get_sd()
for prefix in ["clip_l.", "clip_g.", ""]:

View File

@ -5,12 +5,12 @@ import math
from einops import rearrange, repeat
import os
from comfy.ldm.modules.attention import optimized_attention, _ATTN_PRECISION
from comfy.ldm.modules.attention import optimized_attention
import comfy.samplers
# from comfy/ldm/modules/attention.py
# but modified to return attention scores as well as output
def attention_basic_with_sim(q, k, v, heads, mask=None):
def attention_basic_with_sim(q, k, v, heads, mask=None, attn_precision=None):
b, _, dim_head = q.shape
dim_head //= heads
scale = dim_head ** -0.5
@ -26,7 +26,7 @@ def attention_basic_with_sim(q, k, v, heads, mask=None):
)
# force cast to fp32 to avoid overflowing
if _ATTN_PRECISION =="fp32":
if attn_precision == torch.float32:
sim = einsum('b i d, b j d -> b i j', q.float(), k.float()) * scale
else:
sim = einsum('b i d, b j d -> b i j', q, k) * scale
@ -121,13 +121,13 @@ class SelfAttentionGuidance:
if 1 in cond_or_uncond:
uncond_index = cond_or_uncond.index(1)
# do the entire attention operation, but save the attention scores to attn_scores
(out, sim) = attention_basic_with_sim(q, k, v, heads=heads)
(out, sim) = attention_basic_with_sim(q, k, v, heads=heads, attn_precision=extra_options["attn_precision"])
# when using a higher batch size, I BELIEVE the result batch dimension is [uc1, ... ucn, c1, ... cn]
n_slices = heads * b
attn_scores = sim[n_slices * uncond_index:n_slices * (uncond_index+1)]
return out
else:
return optimized_attention(q, k, v, heads=heads)
return optimized_attention(q, k, v, heads=heads, attn_precision=extra_options["attn_precision"])
def post_cfg_function(args):
nonlocal attn_scores

View File

@ -0,0 +1,33 @@
import nodes
import folder_paths
MAX_RESOLUTION = nodes.MAX_RESOLUTION
class WebcamCapture(nodes.LoadImage):
@classmethod
def INPUT_TYPES(s):
return {
"required": {
"image": ("WEBCAM", {}),
"width": ("INT", {"default": 0, "min": 0, "max": MAX_RESOLUTION, "step": 1}),
"height": ("INT", {"default": 0, "min": 0, "max": MAX_RESOLUTION, "step": 1}),
"capture_on_queue": ("BOOLEAN", {"default": True}),
}
}
RETURN_TYPES = ("IMAGE",)
FUNCTION = "load_capture"
CATEGORY = "image"
def load_capture(s, image, **kwargs):
return super().load_image(folder_paths.get_annotated_filepath(image))
NODE_CLASS_MAPPINGS = {
"WebcamCapture": WebcamCapture,
}
NODE_DISPLAY_NAME_MAPPINGS = {
"WebcamCapture": "Webcam Capture",
}

View File

@ -622,8 +622,27 @@ def full_type_name(klass):
def validate_prompt(prompt):
outputs = set()
for x in prompt:
class_ = nodes.NODE_CLASS_MAPPINGS[prompt[x]['class_type']]
if hasattr(class_, 'OUTPUT_NODE') and class_.OUTPUT_NODE == True:
if 'class_type' not in prompt[x]:
error = {
"type": "invalid_prompt",
"message": f"Cannot execute because a node is missing the class_type property.",
"details": f"Node ID '#{x}'",
"extra_info": {}
}
return (False, error, [], [])
class_type = prompt[x]['class_type']
class_ = nodes.NODE_CLASS_MAPPINGS.get(class_type, None)
if class_ is None:
error = {
"type": "invalid_prompt",
"message": f"Cannot execute because node {class_type} does not exist.",
"details": f"Node ID '#{x}'",
"extra_info": {}
}
return (False, error, [], [])
if hasattr(class_, 'OUTPUT_NODE') and class_.OUTPUT_NODE is True:
outputs.add(x)
if len(outputs) == 0:

View File

@ -258,7 +258,7 @@ def get_save_image_path(filename_prefix, output_dir, image_width=0, image_height
raise Exception(err)
try:
counter = max(filter(lambda a: a[1][:-1] == filename and a[1][-1] == "_", map(map_filename, os.listdir(full_output_folder))))[0] + 1
counter = max(filter(lambda a: os.path.normcase(a[1][:-1]) == os.path.normcase(filename) and a[1][-1] == "_", map(map_filename, os.listdir(full_output_folder))))[0] + 1
except ValueError:
counter = 1
except FileNotFoundError:

View File

@ -4,6 +4,7 @@ import struct
import numpy as np
from comfy.cli_args import args, LatentPreviewMethod
from comfy.taesd.taesd import TAESD
import comfy.model_management
import folder_paths
import comfy.utils
import logging
@ -37,12 +38,13 @@ class Latent2RGBPreviewer(LatentPreviewer):
self.latent_rgb_factors = torch.tensor(latent_rgb_factors, device="cpu")
def decode_latent_to_preview(self, x0):
latent_image = x0[0].permute(1, 2, 0).cpu() @ self.latent_rgb_factors
self.latent_rgb_factors = self.latent_rgb_factors.to(dtype=x0.dtype, device=x0.device)
latent_image = x0[0].permute(1, 2, 0) @ self.latent_rgb_factors
latents_ubyte = (((latent_image + 1) / 2)
.clamp(0, 1) # change scale from -1..1 to 0..1
.mul(0xFF) # to 0..255
.byte()).cpu()
).to(device="cpu", dtype=torch.uint8, non_blocking=comfy.model_management.device_supports_non_blocking(latent_image.device))
return Image.fromarray(latents_ubyte.numpy())
@ -63,8 +65,6 @@ def get_previewer(device, latent_format):
if method == LatentPreviewMethod.Auto:
method = LatentPreviewMethod.Latent2RGB
if taesd_decoder_path:
method = LatentPreviewMethod.TAESD
if method == LatentPreviewMethod.TAESD:
if taesd_decoder_path:

View File

@ -243,11 +243,11 @@ if __name__ == "__main__":
call_on_start = None
if args.auto_launch:
def startup_server(address, port):
def startup_server(scheme, address, port):
import webbrowser
if os.name == 'nt' and address == '0.0.0.0':
address = '127.0.0.1'
webbrowser.open(f"http://{address}:{port}")
webbrowser.open(f"{scheme}://{address}:{port}")
call_on_start = startup_server
try:

View File

@ -1,3 +1,4 @@
from PIL import ImageFile, UnidentifiedImageError
def conditioning_set_values(conditioning, values={}):
c = []
@ -8,3 +9,16 @@ def conditioning_set_values(conditioning, values={}):
c.append(n)
return c
def pillow(fn, arg):
prev_value = None
try:
x = fn(arg)
except (OSError, UnidentifiedImageError, ValueError): #PIL issues #4472 and #2445, also fixes ComfyUI issue #3416
prev_value = ImageFile.LOAD_TRUNCATED_IMAGES
ImageFile.LOAD_TRUNCATED_IMAGES = True
x = fn(arg)
finally:
if prev_value is not None:
ImageFile.LOAD_TRUNCATED_IMAGES = prev_value
return x

View File

@ -10,14 +10,14 @@ import time
import random
import logging
from PIL import Image, ImageOps, ImageSequence
from PIL import Image, ImageOps, ImageSequence, ImageFile
from PIL.PngImagePlugin import PngInfo
import numpy as np
import safetensors.torch
sys.path.insert(0, os.path.join(os.path.dirname(os.path.realpath(__file__)), "comfy"))
import comfy.diffusers_load
import comfy.samplers
import comfy.sample
@ -1456,14 +1456,29 @@ class LoadImage:
FUNCTION = "load_image"
def load_image(self, image):
image_path = folder_paths.get_annotated_filepath(image)
img = Image.open(image_path)
img = node_helpers.pillow(Image.open, image_path)
output_images = []
output_masks = []
w, h = None, None
excluded_formats = ['MPO']
for i in ImageSequence.Iterator(img):
i = ImageOps.exif_transpose(i)
i = node_helpers.pillow(ImageOps.exif_transpose, i)
if i.mode == 'I':
i = i.point(lambda i: i * (1 / 255))
image = i.convert("RGB")
if len(output_images) == 0:
w = image.size[0]
h = image.size[1]
if image.size[0] != w or image.size[1] != h:
continue
image = np.array(image).astype(np.float32) / 255.0
image = torch.from_numpy(image)[None,]
if 'A' in i.getbands():
@ -1474,7 +1489,7 @@ class LoadImage:
output_images.append(image)
output_masks.append(mask.unsqueeze(0))
if len(output_images) > 1:
if len(output_images) > 1 and img.format not in excluded_formats:
output_image = torch.cat(output_images, dim=0)
output_mask = torch.cat(output_masks, dim=0)
else:
@ -1515,8 +1530,8 @@ class LoadImageMask:
FUNCTION = "load_image"
def load_image(self, image, channel):
image_path = folder_paths.get_annotated_filepath(image)
i = Image.open(image_path)
i = ImageOps.exif_transpose(i)
i = node_helpers.pillow(Image.open, image_path)
i = node_helpers.pillow(ImageOps.exif_transpose, i)
if i.getbands() != ("R", "G", "B", "A"):
if i.mode == 'I':
i = i.point(lambda i: i * (1 / 255))
@ -1946,6 +1961,7 @@ def init_custom_nodes():
"nodes_align_your_steps.py",
"nodes_attention_multiply.py",
"nodes_advanced_samplers.py",
"nodes_webcam.py",
]
import_failed = []

View File

@ -13,6 +13,7 @@ import urllib
import json
import glob
import struct
import ssl
from PIL import Image, ImageOps
from PIL.PngImagePlugin import PngInfo
from io import BytesIO
@ -734,14 +735,22 @@ class PromptServer():
async def start(self, address, port, verbose=True, call_on_start=None):
runner = web.AppRunner(self.app, access_log=None)
await runner.setup()
site = web.TCPSite(runner, address, port)
ssl_ctx = None
scheme = "http"
if args.tls_keyfile and args.tls_certfile:
ssl_ctx = ssl.SSLContext(protocol=ssl.PROTOCOL_TLS_SERVER, verify_mode=ssl.CERT_NONE)
ssl_ctx.load_cert_chain(certfile=args.tls_certfile,
keyfile=args.tls_keyfile)
scheme = "https"
site = web.TCPSite(runner, address, port, ssl_context=ssl_ctx)
await site.start()
if verbose:
logging.info("Starting server\n")
logging.info("To see the GUI go to: http://{}:{}".format(address, port))
logging.info("To see the GUI go to: {}://{}:{}".format(scheme, address, port))
if call_on_start is not None:
call_on_start(address, port)
call_on_start(scheme, address, port)
def add_on_prompt_handler(self, handler):
self.on_prompt_handlers.append(handler)

View File

@ -21,7 +21,6 @@ app.registerExtension({
s: "#comfy-save-button",
o: "#comfy-file-input",
Backspace: "#comfy-clear-button",
Delete: "#comfy-clear-button",
d: "#comfy-load-default-button",
};

View File

@ -164,6 +164,41 @@ class MaskEditorDialog extends ComfyDialog {
return divElement;
}
createOpacitySlider(self, name, callback) {
const divElement = document.createElement('div');
divElement.id = "maskeditor-opacity-slider";
divElement.style.cssFloat = "left";
divElement.style.fontFamily = "sans-serif";
divElement.style.marginRight = "4px";
divElement.style.color = "var(--input-text)";
divElement.style.backgroundColor = "var(--comfy-input-bg)";
divElement.style.borderRadius = "8px";
divElement.style.borderColor = "var(--border-color)";
divElement.style.borderStyle = "solid";
divElement.style.fontSize = "15px";
divElement.style.height = "21px";
divElement.style.padding = "1px 6px";
divElement.style.display = "flex";
divElement.style.position = "relative";
divElement.style.top = "2px";
divElement.style.pointerEvents = "auto";
self.opacity_slider_input = document.createElement('input');
self.opacity_slider_input.setAttribute('type', 'range');
self.opacity_slider_input.setAttribute('min', '0.1');
self.opacity_slider_input.setAttribute('max', '1.0');
self.opacity_slider_input.setAttribute('step', '0.01')
self.opacity_slider_input.setAttribute('value', '0.7');
const labelElement = document.createElement("label");
labelElement.textContent = name;
divElement.appendChild(labelElement);
divElement.appendChild(self.opacity_slider_input);
self.opacity_slider_input.addEventListener("input", callback);
return divElement;
}
setlayout(imgCanvas, maskCanvas) {
const self = this;
@ -203,6 +238,13 @@ class MaskEditorDialog extends ComfyDialog {
self.updateBrushPreview(self, null, null);
});
this.brush_opacity_slider = this.createOpacitySlider(self, "Opacity", (event) => {
self.brush_opacity = event.target.value;
if (self.brush_color_mode !== "negative") {
self.maskCanvas.style.opacity = self.brush_opacity;
}
});
this.colorButton = this.createLeftButton(this.getColorButtonText(), () => {
if (self.brush_color_mode === "black") {
self.brush_color_mode = "white";
@ -237,6 +279,7 @@ class MaskEditorDialog extends ComfyDialog {
bottom_panel.appendChild(this.saveButton);
bottom_panel.appendChild(cancelButton);
bottom_panel.appendChild(this.brush_size_slider);
bottom_panel.appendChild(this.brush_opacity_slider);
bottom_panel.appendChild(this.colorButton);
imgCanvas.style.position = "absolute";
@ -472,7 +515,7 @@ class MaskEditorDialog extends ComfyDialog {
else {
return {
mixBlendMode: "initial",
opacity: "0.7",
opacity: this.brush_opacity,
};
}
}
@ -538,6 +581,7 @@ class MaskEditorDialog extends ComfyDialog {
this.maskCtx.putImageData(maskData, 0, 0);
}
brush_opacity = 0.7;
brush_size = 10;
brush_color_mode = "black";
drawing_mode = false;

View File

@ -0,0 +1,126 @@
import { app } from "../../scripts/app.js";
import { api } from "../../scripts/api.js";
const WEBCAM_READY = Symbol();
app.registerExtension({
name: "Comfy.WebcamCapture",
getCustomWidgets(app) {
return {
WEBCAM(node, inputName) {
let res;
node[WEBCAM_READY] = new Promise((resolve) => (res = resolve));
const container = document.createElement("div");
container.style.background = "rgba(0,0,0,0.25)";
container.style.textAlign = "center";
const video = document.createElement("video");
video.style.height = video.style.width = "100%";
const loadVideo = async () => {
try {
const stream = await navigator.mediaDevices.getUserMedia({ video: true, audio: false });
container.replaceChildren(video);
setTimeout(() => res(video), 500); // Fallback as loadedmetadata doesnt fire sometimes?
video.addEventListener("loadedmetadata", () => res(video), false);
video.srcObject = stream;
video.play();
} catch (error) {
const label = document.createElement("div");
label.style.color = "red";
label.style.overflow = "auto";
label.style.maxHeight = "100%";
label.style.whiteSpace = "pre-wrap";
if (window.isSecureContext) {
label.textContent = "Unable to load webcam, please ensure access is granted:\n" + error.message;
} else {
label.textContent = "Unable to load webcam. A secure context is required, if you are not accessing ComfyUI on localhost (127.0.0.1) you will have to enable TLS (https)\n\n" + error.message;
}
container.replaceChildren(label);
}
};
loadVideo();
return { widget: node.addDOMWidget(inputName, "WEBCAM", container) };
},
};
},
nodeCreated(node) {
if ((node.type, node.constructor.comfyClass !== "WebcamCapture")) return;
let video;
const camera = node.widgets.find((w) => w.name === "image");
const w = node.widgets.find((w) => w.name === "width");
const h = node.widgets.find((w) => w.name === "height");
const captureOnQueue = node.widgets.find((w) => w.name === "capture_on_queue");
const canvas = document.createElement("canvas");
const capture = () => {
canvas.width = w.value;
canvas.height = h.value;
const ctx = canvas.getContext("2d");
ctx.drawImage(video, 0, 0, w.value, h.value);
const data = canvas.toDataURL("image/png");
const img = new Image();
img.onload = () => {
node.imgs = [img];
app.graph.setDirtyCanvas(true);
requestAnimationFrame(() => {
node.setSizeForImage?.();
});
};
img.src = data;
};
const btn = node.addWidget("button", "waiting for camera...", "capture", capture);
btn.disabled = true;
btn.serializeValue = () => undefined;
camera.serializeValue = async () => {
if (captureOnQueue.value) {
capture();
} else if (!node.imgs?.length) {
const err = `No webcam image captured`;
alert(err);
throw new Error(err);
}
// Upload image to temp storage
const blob = await new Promise((r) => canvas.toBlob(r));
const name = `${+new Date()}.png`;
const file = new File([blob], name);
const body = new FormData();
body.append("image", file);
body.append("subfolder", "webcam");
body.append("type", "temp");
const resp = await api.fetchApi("/upload/image", {
method: "POST",
body,
});
if (resp.status !== 200) {
const err = `Error uploading camera image: ${resp.status} - ${resp.statusText}`;
alert(err);
throw new Error(err);
}
return `webcam/${name} [temp]`;
};
node[WEBCAM_READY].then((v) => {
video = v;
// If width isnt specified then use video output resolution
if (!w.value) {
w.value = video.videoWidth || 640;
h.value = video.videoHeight || 480;
}
btn.disabled = false;
btn.label = "capture";
});
},
});

View File

@ -262,6 +262,36 @@ export class ComfyApp {
})
);
}
#addRestoreWorkflowView() {
const serialize = LGraph.prototype.serialize;
const self = this;
LGraph.prototype.serialize = function() {
const workflow = serialize.apply(this, arguments);
// Store the drag & scale info in the serialized workflow if the setting is enabled
if (self.enableWorkflowViewRestore.value) {
if (!workflow.extra) {
workflow.extra = {};
}
workflow.extra.ds = {
scale: self.canvas.ds.scale,
offset: self.canvas.ds.offset,
};
} else if (workflow.extra?.ds) {
// Clear any old view data
delete workflow.extra.ds;
}
return workflow;
}
this.enableWorkflowViewRestore = this.ui.settings.addSetting({
id: "Comfy.EnableWorkflowViewRestore",
name: "Save and restore canvas position and zoom level in workflows",
type: "boolean",
defaultValue: true
});
}
/**
* Adds special context menu handling for nodes
@ -953,6 +983,12 @@ export class ComfyApp {
const origProcessMouseDown = LGraphCanvas.prototype.processMouseDown;
LGraphCanvas.prototype.processMouseDown = function(e) {
// prepare for ctrl+shift drag: zoom start
if(e.ctrlKey && e.shiftKey && e.buttons) {
self.zoom_drag_start = [e.x, e.y, this.ds.scale];
return;
}
const res = origProcessMouseDown.apply(this, arguments);
this.selected_group_moving = false;
@ -973,6 +1009,26 @@ export class ComfyApp {
const origProcessMouseMove = LGraphCanvas.prototype.processMouseMove;
LGraphCanvas.prototype.processMouseMove = function(e) {
// handle ctrl+shift drag
if(e.ctrlKey && e.shiftKey && self.zoom_drag_start) {
// stop canvas zoom action
if(!e.buttons) {
self.zoom_drag_start = null;
return;
}
// calculate delta
let deltaY = e.y - self.zoom_drag_start[1];
let startScale = self.zoom_drag_start[2];
let scale = startScale - deltaY/100;
this.ds.changeScale(scale, [this.ds.element.width/2, this.ds.element.height/2]);
this.graph.change();
return;
}
const orig_selected_group = this.selected_group;
if (this.selected_group && !this.selected_group_resizing && !this.selected_group_moving) {
@ -1059,6 +1115,20 @@ export class ComfyApp {
// Trigger onPaste
return true;
}
if((e.key === '+') && e.altKey) {
block_default = true;
let scale = this.ds.scale * 1.1;
this.ds.changeScale(scale, [this.ds.element.width/2, this.ds.element.height/2]);
this.graph.change();
}
if((e.key === '-') && e.altKey) {
block_default = true;
let scale = this.ds.scale * 1 / 1.1;
this.ds.changeScale(scale, [this.ds.element.width/2, this.ds.element.height/2]);
this.graph.change();
}
}
this.graph.change();
@ -1465,6 +1535,7 @@ export class ComfyApp {
this.#addProcessKeyHandler();
this.#addConfigureHandler();
this.#addApiUpdateHandlers();
this.#addRestoreWorkflowView();
this.graph = new LGraph();
@ -1765,6 +1836,10 @@ export class ComfyApp {
try {
this.graph.configure(graphData);
if (this.enableWorkflowViewRestore.value && graphData.extra?.ds) {
this.canvas.ds.offset = graphData.extra.ds.offset;
this.canvas.ds.scale = graphData.extra.ds.scale;
}
} catch (error) {
let errorHint = [];
// Try extracting filename to see if it was caused by an extension script
@ -2082,6 +2157,14 @@ export class ComfyApp {
api.dispatchEvent(new CustomEvent("promptQueued", { detail: { number, batchCount } }));
}
showErrorOnFileLoad(file) {
this.ui.dialog.show(
$el("div", [
$el("p", {textContent: `Unable to find workflow in ${file.name}`})
]).outerHTML
);
}
/**
* Loads workflow data from the specified file
* @param {File} file
@ -2089,27 +2172,27 @@ export class ComfyApp {
async handleFile(file) {
if (file.type === "image/png") {
const pngInfo = await getPngMetadata(file);
if (pngInfo) {
if (pngInfo.workflow) {
await this.loadGraphData(JSON.parse(pngInfo.workflow));
} else if (pngInfo.prompt) {
this.loadApiJson(JSON.parse(pngInfo.prompt));
} else if (pngInfo.parameters) {
importA1111(this.graph, pngInfo.parameters);
}
if (pngInfo?.workflow) {
await this.loadGraphData(JSON.parse(pngInfo.workflow));
} else if (pngInfo?.prompt) {
this.loadApiJson(JSON.parse(pngInfo.prompt));
} else if (pngInfo?.parameters) {
importA1111(this.graph, pngInfo.parameters);
} else {
this.showErrorOnFileLoad(file);
}
} else if (file.type === "image/webp") {
const pngInfo = await getWebpMetadata(file);
if (pngInfo) {
if (pngInfo.workflow) {
this.loadGraphData(JSON.parse(pngInfo.workflow));
} else if (pngInfo.Workflow) {
this.loadGraphData(JSON.parse(pngInfo.Workflow)); // Support loading workflows from that webp custom node.
} else if (pngInfo.prompt) {
this.loadApiJson(JSON.parse(pngInfo.prompt));
} else if (pngInfo.Prompt) {
this.loadApiJson(JSON.parse(pngInfo.Prompt)); // Support loading prompts from that webp custom node.
}
// Support loading workflows from that webp custom node.
const workflow = pngInfo?.workflow || pngInfo?.Workflow;
const prompt = pngInfo?.prompt || pngInfo?.Prompt;
if (workflow) {
this.loadGraphData(JSON.parse(workflow));
} else if (prompt) {
this.loadApiJson(JSON.parse(prompt));
} else {
this.showErrorOnFileLoad(file);
}
} else if (file.type === "application/json" || file.name?.endsWith(".json")) {
const reader = new FileReader();
@ -2130,7 +2213,11 @@ export class ComfyApp {
await this.loadGraphData(JSON.parse(info.workflow));
} else if (info.prompt) {
this.loadApiJson(JSON.parse(info.prompt));
} else {
this.showErrorOnFileLoad(file);
}
} else {
this.showErrorOnFileLoad(file);
}
}
@ -2238,6 +2325,12 @@ export class ComfyApp {
await this.#invokeExtensionsAsync("refreshComboInNodes", defs);
}
resetView() {
app.canvas.ds.scale = 1;
app.canvas.ds.offset = [0, 0]
app.graph.setDirtyCanvas(true, true);
}
/**
* Clean current state
*/

View File

@ -597,16 +597,23 @@ export class ComfyUI {
if (!confirmClear.value || confirm("Clear workflow?")) {
app.clean();
app.graph.clear();
app.resetView();
}
}
}),
$el("button", {
id: "comfy-load-default-button", textContent: "Load Default", onclick: async () => {
if (!confirmClear.value || confirm("Load default workflow?")) {
app.resetView();
await app.loadGraphData()
}
}
}),
$el("button", {
id: "comfy-reset-view-button", textContent: "Reset View", onclick: async () => {
app.resetView();
}
}),
]);
const devMode = this.settings.addSetting({

View File

@ -229,7 +229,11 @@ function createIntWidget(node, inputName, inputData, app, isSeedInput) {
val,
function (v) {
const s = this.options.step / 10;
this.value = Math.round(v / s) * s;
let sh = this.options.min % s;
if (isNaN(sh)) {
sh = 0;
}
this.value = Math.round((v - sh) / s) * s + sh;
},
config
),