mirrors/ComfyUI
1
0
Fork 0
mirror of https://github.com/comfyanonymous/ComfyUI.git synced 2025-08-16 01:44:22 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'master' into worksplit-multigpu

Browse Source
This commit is contained in:
Jedrzej Kosinski 2025-08-13 14:06:09 -07:00
commit 3677943fa5
16 changed files with 149 additions and 23 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

17
.github/workflows/stable-release.yml vendored
View File

@ -12,17 +12,17 @@ on:
description: 'CUDA version'
required: true
type: string
default: "128"
default: "129"
python_minor:
description: 'Python minor version'
required: true
type: string
default: "12"
default: "13"
python_patch:
description: 'Python patch version'
required: true
type: string
default: "10"
default: "6"
jobs:
@ -66,8 +66,13 @@ jobs:
curl https://bootstrap.pypa.io/get-pip.py -o get-pip.py
./python.exe get-pip.py
./python.exe -s -m pip install ../cu${{ inputs.cu }}_python_deps/*
sed -i '1i../ComfyUI' ./python3${{ inputs.python_minor }}._pth
cd ..
sed -i '1i../ComfyUI' ./python3${{ inputs.python_minor }}._pth
rm ./Lib/site-packages/torch/lib/dnnl.lib #I don't think this is actually used and I need the space
rm ./Lib/site-packages/torch/lib/libprotoc.lib
rm ./Lib/site-packages/torch/lib/libprotobuf.lib
cd ..
git clone --depth 1 https://github.com/comfyanonymous/taesd
cp taesd/*.safetensors ./ComfyUI_copy/models/vae_approx/
@ -85,7 +90,7 @@ jobs:
cd ..
"C:\Program Files\7-Zip\7z.exe" a -t7z -m0=lzma2 -mx=9 -mfb=128 -md=512m -ms=on -mf=BCJ2 ComfyUI_windows_portable.7z ComfyUI_windows_portable
"C:\Program Files\7-Zip\7z.exe" a -t7z -m0=lzma2 -mx=9 -mfb=128 -md=768m -ms=on -mf=BCJ2 ComfyUI_windows_portable.7z ComfyUI_windows_portable
mv ComfyUI_windows_portable.7z ComfyUI/ComfyUI_windows_portable_nvidia.7z
cd ComfyUI_windows_portable

6
.github/workflows/windows_release_dependencies.yml vendored
View File

@ -17,19 +17,19 @@ on:
description: 'cuda version'
required: true
type: string
default: "128"
default: "129"
python_minor:
description: 'python minor version'
required: true
type: string
default: "12"
default: "13"
python_patch:
description: 'python patch version'
required: true
type: string
default: "10"
default: "6"
# push:
# branches:
# - master

12
.github/workflows/windows_release_package.yml vendored
View File

@ -7,19 +7,19 @@ on:
description: 'cuda version'
required: true
type: string
default: "128"
default: "129"
python_minor:
description: 'python minor version'
required: true
type: string
default: "12"
default: "13"
python_patch:
description: 'python patch version'
required: true
type: string
default: "10"
default: "6"
# push:
# branches:
# - master
@ -64,6 +64,10 @@ jobs:
./python.exe get-pip.py
./python.exe -s -m pip install ../cu${{ inputs.cu }}_python_deps/*
sed -i '1i../ComfyUI' ./python3${{ inputs.python_minor }}._pth
rm ./Lib/site-packages/torch/lib/dnnl.lib #I don't think this is actually used and I need the space
rm ./Lib/site-packages/torch/lib/libprotoc.lib
rm ./Lib/site-packages/torch/lib/libprotobuf.lib
cd ..
git clone --depth 1 https://github.com/comfyanonymous/taesd
@ -82,7 +86,7 @@ jobs:
cd ..
"C:\Program Files\7-Zip\7z.exe" a -t7z -m0=lzma2 -mx=9 -mfb=128 -md=512m -ms=on -mf=BCJ2 ComfyUI_windows_portable.7z ComfyUI_windows_portable
"C:\Program Files\7-Zip\7z.exe" a -t7z -m0=lzma2 -mx=9 -mfb=128 -md=768m -ms=on -mf=BCJ2 ComfyUI_windows_portable.7z ComfyUI_windows_portable
mv ComfyUI_windows_portable.7z ComfyUI/new_ComfyUI_windows_portable_nvidia_cu${{ inputs.cu }}_or_cpu.7z
cd ComfyUI_windows_portable

2
comfy/ldm/hunyuan3d/vae.py
View File

@ -178,7 +178,7 @@ class FourierEmbedder(nn.Module):
class CrossAttentionProcessor:
def __call__(self, attn, q, k, v):
out = F.scaled_dot_product_attention(q, k, v)
out = comfy.ops.scaled_dot_product_attention(q, k, v)
return out

4
comfy/ldm/modules/attention.py
View File

@ -448,7 +448,7 @@ def attention_pytorch(q, k, v, heads, mask=None, attn_precision=None, skip_resha
mask = mask.unsqueeze(1)
if SDP_BATCH_LIMIT >= b:
out = torch.nn.functional.scaled_dot_product_attention(q, k, v, attn_mask=mask, dropout_p=0.0, is_causal=False)
out = comfy.ops.scaled_dot_product_attention(q, k, v, attn_mask=mask, dropout_p=0.0, is_causal=False)
if not skip_output_reshape:
out = (
out.transpose(1, 2).reshape(b, -1, heads * dim_head)
@ -461,7 +461,7 @@ def attention_pytorch(q, k, v, heads, mask=None, attn_precision=None, skip_resha
if mask.shape[0] > 1:
m = mask[i : i + SDP_BATCH_LIMIT]
out[i : i + SDP_BATCH_LIMIT] = torch.nn.functional.scaled_dot_product_attention(
out[i : i + SDP_BATCH_LIMIT] = comfy.ops.scaled_dot_product_attention(
q[i : i + SDP_BATCH_LIMIT],
k[i : i + SDP_BATCH_LIMIT],
v[i : i + SDP_BATCH_LIMIT],

2
comfy/ldm/modules/diffusionmodules/model.py
View File

@ -285,7 +285,7 @@ def pytorch_attention(q, k, v):
)
try:
out = torch.nn.functional.scaled_dot_product_attention(q, k, v, attn_mask=None, dropout_p=0.0, is_causal=False)
out = comfy.ops.scaled_dot_product_attention(q, k, v, attn_mask=None, dropout_p=0.0, is_causal=False)
out = out.transpose(2, 3).reshape(orig_shape)
except model_management.OOM_EXCEPTION:
logging.warning("scaled_dot_product_attention OOMed: switched to slice attention")

19
comfy/ldm/wan/model.py
View File

@ -391,6 +391,7 @@ class WanModel(torch.nn.Module):
cross_attn_norm=True,
eps=1e-6,
flf_pos_embed_token_number=None,
in_dim_ref_conv=None,
image_model=None,
device=None,
dtype=None,
@ -484,6 +485,11 @@ class WanModel(torch.nn.Module):
else:
self.img_emb = None
if in_dim_ref_conv is not None:
self.ref_conv = operations.Conv2d(in_dim_ref_conv, dim, kernel_size=patch_size[1:], stride=patch_size[1:], device=operation_settings.get("device"), dtype=operation_settings.get("dtype"))
else:
self.ref_conv = None
def forward_orig(
self,
x,
@ -526,6 +532,13 @@ class WanModel(torch.nn.Module):
e = e.reshape(t.shape[0], -1, e.shape[-1])
e0 = self.time_projection(e).unflatten(2, (6, self.dim))
full_ref = None
if self.ref_conv is not None:
full_ref = kwargs.get("reference_latent", None)
if full_ref is not None:
full_ref = self.ref_conv(full_ref).flatten(2).transpose(1, 2)
x = torch.concat((full_ref, x), dim=1)
# context
context = self.text_embedding(context)
@ -552,6 +565,9 @@ class WanModel(torch.nn.Module):
# head
x = self.head(x, e)
if full_ref is not None:
x = x[:, full_ref.shape[1]:]
# unpatchify
x = self.unpatchify(x, grid_sizes)
return x
@ -570,6 +586,9 @@ class WanModel(torch.nn.Module):
x = torch.cat([x, time_dim_concat], dim=2)
t_len = ((x.shape[2] + (patch_size[0] // 2)) // patch_size[0])
if self.ref_conv is not None and "reference_latent" in kwargs:
t_len += 1
img_ids = torch.zeros((t_len, h_len, w_len, 3), device=x.device, dtype=x.dtype)
img_ids[:, :, :, 0] = img_ids[:, :, :, 0] + torch.linspace(0, t_len - 1, steps=t_len, device=x.device, dtype=x.dtype).reshape(-1, 1, 1)
img_ids[:, :, :, 1] = img_ids[:, :, :, 1] + torch.linspace(0, h_len - 1, steps=h_len, device=x.device, dtype=x.dtype).reshape(1, -1, 1)

10
comfy/model_base.py
View File

@ -1124,7 +1124,11 @@ class WAN21(BaseModel):
mask = mask.repeat(1, 4, 1, 1, 1)
mask = utils.resize_to_batch_size(mask, noise.shape[0])
return torch.cat((mask, image), dim=1)
concat_mask_index = kwargs.get("concat_mask_index", 0)
if concat_mask_index != 0:
return torch.cat((image[:, :concat_mask_index], mask, image[:, concat_mask_index:]), dim=1)
else:
return torch.cat((mask, image), dim=1)
def extra_conds(self, **kwargs):
out = super().extra_conds(**kwargs)
@ -1140,6 +1144,10 @@ class WAN21(BaseModel):
if time_dim_concat is not None:
out['time_dim_concat'] = comfy.conds.CONDRegular(self.process_latent_in(time_dim_concat))
reference_latents = kwargs.get("reference_latents", None)
if reference_latents is not None:
out['reference_latent'] = comfy.conds.CONDRegular(self.process_latent_in(reference_latents[-1])[:, :, 0])
return out

5
comfy/model_detection.py
View File

@ -373,6 +373,11 @@ def detect_unet_config(state_dict, key_prefix, metadata=None):
flf_weight = state_dict.get('{}img_emb.emb_pos'.format(key_prefix))
if flf_weight is not None:
dit_config["flf_pos_embed_token_number"] = flf_weight.shape[1]
ref_conv_weight = state_dict.get('{}ref_conv.weight'.format(key_prefix))
if ref_conv_weight is not None:
dit_config["in_dim_ref_conv"] = ref_conv_weight.shape[1]
return dit_config
if '{}latent_in.weight'.format(key_prefix) in state_dict_keys: # Hunyuan 3D

23
comfy/ops.py
View File

@ -24,6 +24,29 @@ import comfy.float
import comfy.rmsnorm
import contextlib
def scaled_dot_product_attention(q, k, v, *args, **kwargs):
return torch.nn.functional.scaled_dot_product_attention(q, k, v, *args, **kwargs)
try:
if torch.cuda.is_available():
from torch.nn.attention import SDPBackend, sdpa_kernel
SDPA_BACKEND_PRIORITY = [
SDPBackend.FLASH_ATTENTION,
SDPBackend.EFFICIENT_ATTENTION,
SDPBackend.MATH,
]
SDPA_BACKEND_PRIORITY.insert(0, SDPBackend.CUDNN_ATTENTION)
@sdpa_kernel(backends=SDPA_BACKEND_PRIORITY, set_priority=True)
def scaled_dot_product_attention(q, k, v, *args, **kwargs):
return torch.nn.functional.scaled_dot_product_attention(q, k, v, *args, **kwargs)
except (ModuleNotFoundError, TypeError):
logging.warning("Could not set sdpa backend priority.")
cast_to = comfy.model_management.cast_to #TODO: remove once no more references
def cast_to_input(weight, input, non_blocking=False, copy=True):

2
comfy_api/latest/_ui.py
View File

@ -12,7 +12,7 @@ import torch
try:
import torchaudio
TORCH_AUDIO_AVAILABLE = True
except ImportError:
except:
TORCH_AUDIO_AVAILABLE = False
from PIL import Image as PILImage
from PIL.PngImagePlugin import PngInfo

6
comfy_api_nodes/nodes_kling.py
View File

@ -1690,7 +1690,11 @@ class KlingImageGenerationNode(KlingImageGenerationBase):
):
self.validate_prompt(prompt, negative_prompt)
if image is not None:
if image is None:
image_type = None
elif model_name == KlingImageGenModelName.kling_v1:
raise ValueError(f"The model {KlingImageGenModelName.kling_v1.value} does not support reference images.")
else:
image = tensor_to_base64_string(image)
initial_operation = SynchronousOperation(

58
comfy_extras/nodes_wan.py
View File

@ -103,6 +103,63 @@ class WanFunControlToVideo:
out_latent["samples"] = latent
return (positive, negative, out_latent)
class Wan22FunControlToVideo:
@classmethod
def INPUT_TYPES(s):
return {"required": {"positive": ("CONDITIONING", ),
"negative": ("CONDITIONING", ),
"vae": ("VAE", ),
"width": ("INT", {"default": 832, "min": 16, "max": nodes.MAX_RESOLUTION, "step": 16}),
"height": ("INT", {"default": 480, "min": 16, "max": nodes.MAX_RESOLUTION, "step": 16}),
"length": ("INT", {"default": 81, "min": 1, "max": nodes.MAX_RESOLUTION, "step": 4}),
"batch_size": ("INT", {"default": 1, "min": 1, "max": 4096}),
},
"optional": {"ref_image": ("IMAGE", ),
"control_video": ("IMAGE", ),
# "start_image": ("IMAGE", ),
}}
RETURN_TYPES = ("CONDITIONING", "CONDITIONING", "LATENT")
RETURN_NAMES = ("positive", "negative", "latent")
FUNCTION = "encode"
CATEGORY = "conditioning/video_models"
def encode(self, positive, negative, vae, width, height, length, batch_size, ref_image=None, start_image=None, control_video=None):
latent = torch.zeros([batch_size, 16, ((length - 1) // 4) + 1, height // 8, width // 8], device=comfy.model_management.intermediate_device())
concat_latent = torch.zeros([batch_size, 16, ((length - 1) // 4) + 1, height // 8, width // 8], device=comfy.model_management.intermediate_device())
concat_latent = comfy.latent_formats.Wan21().process_out(concat_latent)
concat_latent = concat_latent.repeat(1, 2, 1, 1, 1)
mask = torch.ones((1, 1, latent.shape[2] * 4, latent.shape[-2], latent.shape[-1]))
if start_image is not None:
start_image = comfy.utils.common_upscale(start_image[:length].movedim(-1, 1), width, height, "bilinear", "center").movedim(1, -1)
concat_latent_image = vae.encode(start_image[:, :, :, :3])
concat_latent[:,16:,:concat_latent_image.shape[2]] = concat_latent_image[:,:,:concat_latent.shape[2]]
mask[:, :, :start_image.shape[0] + 3] = 0.0
ref_latent = None
if ref_image is not None:
ref_image = comfy.utils.common_upscale(ref_image[:1].movedim(-1, 1), width, height, "bilinear", "center").movedim(1, -1)
ref_latent = vae.encode(ref_image[:, :, :, :3])
if control_video is not None:
control_video = comfy.utils.common_upscale(control_video[:length].movedim(-1, 1), width, height, "bilinear", "center").movedim(1, -1)
concat_latent_image = vae.encode(control_video[:, :, :, :3])
concat_latent[:,:16,:concat_latent_image.shape[2]] = concat_latent_image[:,:,:concat_latent.shape[2]]
mask = mask.view(1, mask.shape[2] // 4, 4, mask.shape[3], mask.shape[4]).transpose(1, 2)
positive = node_helpers.conditioning_set_values(positive, {"concat_latent_image": concat_latent, "concat_mask": mask, "concat_mask_index": 16})
negative = node_helpers.conditioning_set_values(negative, {"concat_latent_image": concat_latent, "concat_mask": mask, "concat_mask_index": 16})
if ref_latent is not None:
positive = node_helpers.conditioning_set_values(positive, {"reference_latents": [ref_latent]}, append=True)
negative = node_helpers.conditioning_set_values(negative, {"reference_latents": [ref_latent]}, append=True)
out_latent = {}
out_latent["samples"] = latent
return (positive, negative, out_latent)
class WanFirstLastFrameToVideo:
@classmethod
def INPUT_TYPES(s):
@ -733,6 +790,7 @@ NODE_CLASS_MAPPINGS = {
"WanTrackToVideo": WanTrackToVideo,
"WanImageToVideo": WanImageToVideo,
"WanFunControlToVideo": WanFunControlToVideo,
"Wan22FunControlToVideo": Wan22FunControlToVideo,
"WanFunInpaintToVideo": WanFunInpaintToVideo,
"WanFirstLastFrameToVideo": WanFirstLastFrameToVideo,
"WanVaceToVideo": WanVaceToVideo,

2
comfyui_version.py
View File

@ -1,3 +1,3 @@
# This file is automatically generated by the build process when version is
# updated in pyproject.toml.
__version__ = "0.3.49"
__version__ = "0.3.50"

2
pyproject.toml
View File

@ -1,6 +1,6 @@
[project]
name = "ComfyUI"
version = "0.3.49"
version = "0.3.50"
readme = "README.md"
license = { file = "LICENSE" }
requires-python = ">=3.9"

2
requirements.txt
View File

@ -1,5 +1,5 @@
comfyui-frontend-package==1.24.4
comfyui-workflow-templates==0.1.53
comfyui-workflow-templates==0.1.59
comfyui-embedded-docs==0.2.6
torch
torchsde