Merge branch 'master' into save-images

.ci/nightly/update_windows/update.py

@@ -0,0 +1,65 @@
+import pygit2
+from datetime import datetime
+import sys
+
+def pull(repo, remote_name='origin', branch='master'):
+    for remote in repo.remotes:
+        if remote.name == remote_name:
+            remote.fetch()
+            remote_master_id = repo.lookup_reference('refs/remotes/origin/%s' % (branch)).target
+            merge_result, _ = repo.merge_analysis(remote_master_id)
+            # Up to date, do nothing
+            if merge_result & pygit2.GIT_MERGE_ANALYSIS_UP_TO_DATE:
+                return
+            # We can just fastforward
+            elif merge_result & pygit2.GIT_MERGE_ANALYSIS_FASTFORWARD:
+                repo.checkout_tree(repo.get(remote_master_id))
+                try:
+                    master_ref = repo.lookup_reference('refs/heads/%s' % (branch))
+                    master_ref.set_target(remote_master_id)
+                except KeyError:
+                    repo.create_branch(branch, repo.get(remote_master_id))
+                repo.head.set_target(remote_master_id)
+            elif merge_result & pygit2.GIT_MERGE_ANALYSIS_NORMAL:
+                repo.merge(remote_master_id)
+
+                if repo.index.conflicts is not None:
+                    for conflict in repo.index.conflicts:
+                        print('Conflicts found in:', conflict[0].path)
+                    raise AssertionError('Conflicts, ahhhhh!!')
+
+                user = repo.default_signature
+                tree = repo.index.write_tree()
+                commit = repo.create_commit('HEAD',
+                                            user,
+                                            user,
+                                            'Merge!',
+                                            tree,
+                                            [repo.head.target, remote_master_id])
+                # We need to do this or git CLI will think we are still merging.
+                repo.state_cleanup()
+            else:
+                raise AssertionError('Unknown merge analysis result')
+
+
+repo = pygit2.Repository(str(sys.argv[1]))
+ident = pygit2.Signature('comfyui', 'comfy@ui')
+try:
+    print("stashing current changes")
+    repo.stash(ident)
+except KeyError:
+    print("nothing to stash")
+backup_branch_name = 'backup_branch_{}'.format(datetime.today().strftime('%Y-%m-%d_%H_%M_%S'))
+print("creating backup branch: {}".format(backup_branch_name))
+repo.branches.local.create(backup_branch_name, repo.head.peel())
+
+print("checking out master branch")
+branch = repo.lookup_branch('master')
+ref = repo.lookup_reference(branch.name)
+repo.checkout(ref)
+
+print("pulling latest changes")
+pull(repo)
+
+print("Done!")
+

.ci/nightly/update_windows/update_comfyui.bat

@@ -0,0 +1,2 @@
+..\python_embeded\python.exe .\update.py ..\ComfyUI\
+pause

.ci/nightly/update_windows/update_comfyui_and_python_dependencies.bat

@@ -0,0 +1,3 @@
+..\python_embeded\python.exe .\update.py ..\ComfyUI\
+..\python_embeded\python.exe -s -m pip install --upgrade --pre torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu118 -r ../ComfyUI/requirements.txt pygit2
+pause

.ci/nightly/windows_base_files/README_VERY_IMPORTANT.txt

@@ -0,0 +1,27 @@
+HOW TO RUN:
+
+if you have a NVIDIA gpu:
+
+run_nvidia_gpu.bat
+
+
+
+To run it in slow CPU mode:
+
+run_cpu.bat
+
+
+
+IF YOU GET A RED ERROR IN THE UI MAKE SURE YOU HAVE A MODEL/CHECKPOINT IN: ComfyUI\models\checkpoints
+
+You can download the stable diffusion 1.5 one from: https://huggingface.co/runwayml/stable-diffusion-v1-5/blob/main/v1-5-pruned-emaonly.ckpt
+
+
+
+RECOMMENDED WAY TO UPDATE:
+To update the ComfyUI code: update\update_comfyui.bat
+
+
+
+To update ComfyUI with the python dependencies:
+update\update_comfyui_and_python_dependencies.bat

.ci/nightly/windows_base_files/run_cpu.bat

@@ -0,0 +1,2 @@
+.\python_embeded\python.exe -s ComfyUI\main.py --cpu --windows-standalone-build
+pause

.ci/nightly/windows_base_files/run_nvidia_gpu.bat

@@ -0,0 +1,2 @@
+.\python_embeded\python.exe -s ComfyUI\main.py --windows-standalone-build --use-pytorch-cross-attention
+pause

.ci/update_windows/update.py

@@ -0,0 +1,65 @@
+import pygit2
+from datetime import datetime
+import sys
+
+def pull(repo, remote_name='origin', branch='master'):
+    for remote in repo.remotes:
+        if remote.name == remote_name:
+            remote.fetch()
+            remote_master_id = repo.lookup_reference('refs/remotes/origin/%s' % (branch)).target
+            merge_result, _ = repo.merge_analysis(remote_master_id)
+            # Up to date, do nothing
+            if merge_result & pygit2.GIT_MERGE_ANALYSIS_UP_TO_DATE:
+                return
+            # We can just fastforward
+            elif merge_result & pygit2.GIT_MERGE_ANALYSIS_FASTFORWARD:
+                repo.checkout_tree(repo.get(remote_master_id))
+                try:
+                    master_ref = repo.lookup_reference('refs/heads/%s' % (branch))
+                    master_ref.set_target(remote_master_id)
+                except KeyError:
+                    repo.create_branch(branch, repo.get(remote_master_id))
+                repo.head.set_target(remote_master_id)
+            elif merge_result & pygit2.GIT_MERGE_ANALYSIS_NORMAL:
+                repo.merge(remote_master_id)
+
+                if repo.index.conflicts is not None:
+                    for conflict in repo.index.conflicts:
+                        print('Conflicts found in:', conflict[0].path)
+                    raise AssertionError('Conflicts, ahhhhh!!')
+
+                user = repo.default_signature
+                tree = repo.index.write_tree()
+                commit = repo.create_commit('HEAD',
+                                            user,
+                                            user,
+                                            'Merge!',
+                                            tree,
+                                            [repo.head.target, remote_master_id])
+                # We need to do this or git CLI will think we are still merging.
+                repo.state_cleanup()
+            else:
+                raise AssertionError('Unknown merge analysis result')
+
+
+repo = pygit2.Repository(str(sys.argv[1]))
+ident = pygit2.Signature('comfyui', 'comfy@ui')
+try:
+    print("stashing current changes")
+    repo.stash(ident)
+except KeyError:
+    print("nothing to stash")
+backup_branch_name = 'backup_branch_{}'.format(datetime.today().strftime('%Y-%m-%d_%H_%M_%S'))
+print("creating backup branch: {}".format(backup_branch_name))
+repo.branches.local.create(backup_branch_name, repo.head.peel())
+
+print("checking out master branch")
+branch = repo.lookup_branch('master')
+ref = repo.lookup_reference(branch.name)
+repo.checkout(ref)
+
+print("pulling latest changes")
+pull(repo)
+
+print("Done!")
+

.ci/update_windows/update_comfyui.bat

@@ -0,0 +1,2 @@
+..\python_embeded\python.exe .\update.py ..\ComfyUI\
+pause

.ci/update_windows/update_comfyui_and_python_dependencies.bat

@@ -0,0 +1,3 @@
+..\python_embeded\python.exe .\update.py ..\ComfyUI\
+..\python_embeded\python.exe -s -m pip install --upgrade torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu117 xformers -r ../ComfyUI/requirements.txt pygit2
+pause

.ci/windows_base_files/README_VERY_IMPORTANT.txt

@@ -0,0 +1,27 @@
+HOW TO RUN:
+
+if you have a NVIDIA gpu:
+
+run_nvidia_gpu.bat
+
+
+
+To run it in slow CPU mode:
+
+run_cpu.bat
+
+
+
+IF YOU GET A RED ERROR IN THE UI MAKE SURE YOU HAVE A MODEL/CHECKPOINT IN: ComfyUI\models\checkpoints
+
+You can download the stable diffusion 1.5 one from: https://huggingface.co/runwayml/stable-diffusion-v1-5/blob/main/v1-5-pruned-emaonly.ckpt
+
+
+RECOMMENDED WAY TO UPDATE:
+To update the ComfyUI code: update\update_comfyui.bat
+
+
+
+To update ComfyUI with the python dependencies:
+update\update_comfyui_and_python_dependencies.bat
+

.ci/windows_base_files/run_cpu.bat

@@ -0,0 +1,2 @@
+.\python_embeded\python.exe -s ComfyUI\main.py --cpu --windows-standalone-build
+pause

.ci/windows_base_files/run_nvidia_gpu.bat

@@ -0,0 +1,2 @@
+.\python_embeded\python.exe -s ComfyUI\main.py --windows-standalone-build
+pause

.github/workflows/windows_release.yml

@@ -0,0 +1,64 @@
+name: "Windows Release"
+
+on:
+  workflow_dispatch:
+#  push:
+#    branches:
+#      - master
+
+jobs:
+  build:
+    permissions:
+        contents: "write"
+        packages: "write"
+        pull-requests: "read"
+    runs-on: windows-latest
+    steps:
+        - uses: actions/checkout@v3
+          with:
+            fetch-depth: 0
+
+        - shell: bash
+          run: |
+            cd ..
+            cp -r ComfyUI ComfyUI_copy
+            curl https://www.python.org/ftp/python/3.10.9/python-3.10.9-embed-amd64.zip -o python_embeded.zip
+            unzip python_embeded.zip -d python_embeded
+            cd python_embeded
+            echo 'import site' >> ./python310._pth
+            curl https://bootstrap.pypa.io/get-pip.py -o get-pip.py
+            ./python.exe get-pip.py
+            ./python.exe -s -m pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu117 xformers -r ../ComfyUI/requirements.txt pygit2
+            sed -i '1i../ComfyUI' ./python310._pth
+            cd ..
+
+
+            mkdir ComfyUI_windows_portable
+            mv python_embeded ComfyUI_windows_portable
+            mv ComfyUI_copy ComfyUI_windows_portable/ComfyUI
+
+            cd ComfyUI_windows_portable
+
+            mkdir update
+            cp -r ComfyUI/.ci/update_windows/* ./update/
+            cp -r ComfyUI/.ci/windows_base_files/* ./
+
+            cd ..
+
+            "C:\Program Files\7-Zip\7z.exe" a -t7z -m0=lzma -mx=8 -mfb=64 -md=32m -ms=on ComfyUI_windows_portable.7z ComfyUI_windows_portable
+            mv ComfyUI_windows_portable.7z ComfyUI/ComfyUI_windows_portable_nvidia_or_cpu.7z
+
+            cd ComfyUI_windows_portable
+            python_embeded/python.exe -s ComfyUI/main.py --quick-test-for-ci --cpu
+
+            ls
+
+
+        - name: Upload binaries to release
+          uses: svenstaro/upload-release-action@v2
+          with:
+                repo_token: ${{ secrets.GITHUB_TOKEN }}
+                file: ComfyUI_windows_portable_nvidia_or_cpu.7z
+                tag: "latest"
+                overwrite: true
+

.github/workflows/windows_release_cu118.yml

@@ -0,0 +1,63 @@
+name: "Windows Release cu118"
+
+on:
+  workflow_dispatch:
+#  push:
+#    branches:
+#      - master
+
+jobs:
+  build:
+    permissions:
+        contents: "write"
+        packages: "write"
+        pull-requests: "read"
+    runs-on: windows-latest
+    steps:
+        - uses: actions/checkout@v3
+          with:
+            fetch-depth: 0
+
+        - shell: bash
+          run: |
+            cd ..
+            cp -r ComfyUI ComfyUI_copy
+            curl https://www.python.org/ftp/python/3.10.9/python-3.10.9-embed-amd64.zip -o python_embeded.zip
+            unzip python_embeded.zip -d python_embeded
+            cd python_embeded
+            echo 'import site' >> ./python310._pth
+            curl https://bootstrap.pypa.io/get-pip.py -o get-pip.py
+            ./python.exe get-pip.py
+            ./python.exe -s -m pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu118 xformers -r ../ComfyUI/requirements.txt pygit2
+            sed -i '1i../ComfyUI' ./python310._pth
+            cd ..
+
+
+            mkdir ComfyUI_windows_portable
+            mv python_embeded ComfyUI_windows_portable
+            mv ComfyUI_copy ComfyUI_windows_portable/ComfyUI
+
+            cd ComfyUI_windows_portable
+
+            mkdir update
+            cp -r ComfyUI/.ci/update_windows/* ./update/
+            cp -r ComfyUI/.ci/windows_base_files/* ./
+
+            cd ..
+
+            "C:\Program Files\7-Zip\7z.exe" a -t7z -m0=lzma -mx=8 -mfb=64 -md=32m -ms=on ComfyUI_windows_portable.7z ComfyUI_windows_portable
+            mv ComfyUI_windows_portable.7z ComfyUI/ComfyUI_windows_portable_nvidia_cu118_or_cpu.7z
+
+            cd ComfyUI_windows_portable
+            python_embeded/python.exe -s ComfyUI/main.py --quick-test-for-ci --cpu
+
+            ls
+
+        - name: Upload binaries to release
+          uses: svenstaro/upload-release-action@v2
+          with:
+                repo_token: ${{ secrets.GITHUB_TOKEN }}
+                file: ComfyUI_windows_portable_nvidia_cu118_or_cpu.7z
+                tag: "latest"
+                overwrite: true
+

.github/workflows/windows_release_nightly_pytorch.yml

@@ -0,0 +1,66 @@
+name: "Windows Release Nightly pytorch"
+
+on:
+  workflow_dispatch:
+#  push:
+#    branches:
+#      - master
+
+jobs:
+  build:
+    permissions:
+        contents: "write"
+        packages: "write"
+        pull-requests: "read"
+    runs-on: windows-latest
+    steps:
+        - uses: actions/checkout@v3
+          with:
+            fetch-depth: 0
+        - uses: actions/setup-python@v4
+          with:
+            python-version: '3.10.9'
+        - shell: bash
+          run: |
+            cd ..
+            cp -r ComfyUI ComfyUI_copy
+            curl https://www.python.org/ftp/python/3.10.9/python-3.10.9-embed-amd64.zip -o python_embeded.zip
+            unzip python_embeded.zip -d python_embeded
+            cd python_embeded
+            echo 'import site' >> ./python310._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 --pre --extra-index-url https://download.pytorch.org/whl/nightly/cu118 -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' ./python310._pth
+            cd ..
+
+
+            mkdir ComfyUI_windows_portable_nightly_pytorch
+            mv python_embeded ComfyUI_windows_portable_nightly_pytorch
+            mv ComfyUI_copy ComfyUI_windows_portable_nightly_pytorch/ComfyUI
+
+            cd ComfyUI_windows_portable_nightly_pytorch
+
+            mkdir update
+            cp -r ComfyUI/.ci/update_windows/* ./update/
+            cp -r ComfyUI/.ci/windows_base_files/* ./
+
+            cd ..
+
+            "C:\Program Files\7-Zip\7z.exe" a -t7z -m0=lzma -mx=8 -mfb=64 -md=32m -ms=on ComfyUI_windows_portable_nightly_pytorch.7z ComfyUI_windows_portable_nightly_pytorch
+            mv ComfyUI_windows_portable_nightly_pytorch.7z ComfyUI/ComfyUI_windows_portable_nvidia_or_cpu_nightly_pytorch.7z
+
+            cd ComfyUI_windows_portable_nightly_pytorch
+            python_embeded/python.exe -s ComfyUI/main.py --quick-test-for-ci --cpu
+
+            ls
+
+        - name: Upload binaries to release
+          uses: svenstaro/upload-release-action@v2
+          with:
+                repo_token: ${{ secrets.GITHUB_TOKEN }}
+                file: ComfyUI_windows_portable_nvidia_or_cpu_nightly_pytorch.7z
+                tag: "latest"
+                overwrite: true

README.md

@@ -31,6 +31,20 @@ Workflow examples can be found on the [Examples page](https://comfyanonymous.git
 
 # Installing
 
+## Windows
+
+There is a portable standalone build for Windows that should work for running on Nvidia GPUs or for running on your CPU only on the [releases page](https://github.com/comfyanonymous/ComfyUI/releases).
+
+### [Direct link to download](https://github.com/comfyanonymous/ComfyUI/releases/download/latest/ComfyUI_windows_portable_nvidia_or_cpu.7z)
+
+Just download, extract and run. Make sure you put your Stable Diffusion checkpoints/models (the huge ckpt/safetensors files) in: ComfyUI\models\checkpoints
+
+## Colab Notebook
+
+To run it on colab or paperspace you can use my [Colab Notebook](notebooks/comfyui_colab.ipynb) here: [Link to open with google colab](https://colab.research.google.com/github/comfyanonymous/ComfyUI/blob/master/notebooks/comfyui_colab.ipynb)
+
+## Manual Install (Windows, Linux)
+
 Git clone this repo.
 
 Put your SD checkpoints (the huge ckpt/safetensors files) in: models/checkpoints
@@ -39,16 +53,17 @@ Put your VAE in: models/vae
 
 At the time of writing this pytorch has issues with python versions higher than 3.10 so make sure your python/pip versions are 3.10.
 
-### AMD
-AMD users can install rocm and pytorch with pip if you don't have it already installed:
+### AMD (Linux only)
+AMD users can install rocm and pytorch with pip if you don't have it already installed, this is the command to install the stable version:
+
+```pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/rocm5.4.2```
 
-```pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/rocm5.2```
 
 ### NVIDIA
 
 Nvidia users should install torch using this command:
 
-```pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu117```
+```pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu118```
 
 Nvidia users should also install Xformers for a speed boost but can still run the software without it.
 
@@ -112,10 +127,6 @@ To use a textual inversion concepts/embeddings in a text prompt put them in the
 
 ```embedding:embedding_filename.pt```
 
-### Colab Notebook
-
-To run it on colab or paperspace you can use my [Colab Notebook](notebooks/comfyui_colab.ipynb) here: [Link to open with google colab](https://colab.research.google.com/github/comfyanonymous/ComfyUI/blob/master/notebooks/comfyui_colab.ipynb)
-
 ### Fedora
 
 To get python 3.10 on fedora:

comfy/cldm/cldm.py

@@ -59,9 +59,9 @@ class ControlNet(nn.Module):
 
         if context_dim is not None:
             assert use_spatial_transformer, 'Fool!! You forgot to use the spatial transformer for your cross-attention conditioning...'
-            from omegaconf.listconfig import ListConfig
-            if type(context_dim) == ListConfig:
-                context_dim = list(context_dim)
+            # from omegaconf.listconfig import ListConfig
+            # if type(context_dim) == ListConfig:
+            #     context_dim = list(context_dim)
 
         if num_heads_upsample == -1:
             num_heads_upsample = num_heads

comfy/ldm/models/diffusion/ddpm.py

@@ -18,7 +18,6 @@ import itertools
 from tqdm import tqdm
 from torchvision.utils import make_grid
 # from pytorch_lightning.utilities.distributed import rank_zero_only
-from omegaconf import ListConfig
 
 from ldm.util import log_txt_as_img, exists, default, ismap, isimage, mean_flat, count_params, instantiate_from_config
 from ldm.modules.ema import LitEma
@@ -1124,8 +1123,8 @@ class LatentDiffusion(DDPM):
     def get_unconditional_conditioning(self, batch_size, null_label=None):
         if null_label is not None:
             xc = null_label
-            if isinstance(xc, ListConfig):
-                xc = list(xc)
+            # if isinstance(xc, ListConfig):
+            #     xc = list(xc)
             if isinstance(xc, dict) or isinstance(xc, list):
                 c = self.get_learned_conditioning(xc)
             else:

comfy/ldm/modules/attention.py

@@ -11,12 +11,10 @@ from .sub_quadratic_attention import efficient_dot_product_attention
 
 import model_management
 
-try:
+
+if model_management.xformers_enabled():
     import xformers
     import xformers.ops
-    XFORMERS_IS_AVAILBLE = True
-except:
-    XFORMERS_IS_AVAILBLE = False
 
 # CrossAttn precision handling
 import os
@@ -481,23 +479,19 @@ class CrossAttentionPytorch(nn.Module):
         return self.to_out(out)
 
 import sys
-if XFORMERS_IS_AVAILBLE == False or "--disable-xformers" in sys.argv:
+if model_management.xformers_enabled():
+    print("Using xformers cross attention")
+    CrossAttention = MemoryEfficientCrossAttention
+elif model_management.pytorch_attention_enabled():
+    print("Using pytorch cross attention")
+    CrossAttention = CrossAttentionPytorch
+else:
     if "--use-split-cross-attention" in sys.argv:
         print("Using split optimization for cross attention")
         CrossAttention = CrossAttentionDoggettx
     else:
-        if "--use-pytorch-cross-attention" in sys.argv:
-            print("Using pytorch cross attention")
-            torch.backends.cuda.enable_math_sdp(False)
-            torch.backends.cuda.enable_flash_sdp(True)
-            torch.backends.cuda.enable_mem_efficient_sdp(True)
-            CrossAttention = CrossAttentionPytorch
-        else:
-            print("Using sub quadratic optimization for cross attention, if you have memory or speed issues try using: --use-split-cross-attention")
-            CrossAttention = CrossAttentionBirchSan
-else:
-    print("Using xformers cross attention")
-    CrossAttention = MemoryEfficientCrossAttention
+        print("Using sub quadratic optimization for cross attention, if you have memory or speed issues try using: --use-split-cross-attention")
+        CrossAttention = CrossAttentionBirchSan
 
 
 class BasicTransformerBlock(nn.Module):

comfy/ldm/modules/diffusionmodules/model.py

@@ -9,13 +9,9 @@ from typing import Optional, Any
 from ldm.modules.attention import MemoryEfficientCrossAttention
 import model_management
 
-try:
+if model_management.xformers_enabled():
     import xformers
     import xformers.ops
-    XFORMERS_IS_AVAILBLE = True
-except:
-    XFORMERS_IS_AVAILBLE = False
-    print("No module 'xformers'. Proceeding without it.")
 
 try:
     OOM_EXCEPTION = torch.cuda.OutOfMemoryError
@@ -303,6 +299,64 @@ class MemoryEfficientAttnBlock(nn.Module):
         out = self.proj_out(out)
         return x+out
 
+class MemoryEfficientAttnBlockPytorch(nn.Module):
+    def __init__(self, in_channels):
+        super().__init__()
+        self.in_channels = in_channels
+
+        self.norm = Normalize(in_channels)
+        self.q = torch.nn.Conv2d(in_channels,
+                                 in_channels,
+                                 kernel_size=1,
+                                 stride=1,
+                                 padding=0)
+        self.k = torch.nn.Conv2d(in_channels,
+                                 in_channels,
+                                 kernel_size=1,
+                                 stride=1,
+                                 padding=0)
+        self.v = torch.nn.Conv2d(in_channels,
+                                 in_channels,
+                                 kernel_size=1,
+                                 stride=1,
+                                 padding=0)
+        self.proj_out = torch.nn.Conv2d(in_channels,
+                                        in_channels,
+                                        kernel_size=1,
+                                        stride=1,
+                                        padding=0)
+        self.attention_op: Optional[Any] = None
+
+    def forward(self, x):
+        h_ = x
+        h_ = self.norm(h_)
+        q = self.q(h_)
+        k = self.k(h_)
+        v = self.v(h_)
+
+        # compute attention
+        B, C, H, W = q.shape
+        q, k, v = map(lambda x: rearrange(x, 'b c h w -> b (h w) c'), (q, k, v))
+
+        q, k, v = map(
+            lambda t: t.unsqueeze(3)
+            .reshape(B, t.shape[1], 1, C)
+            .permute(0, 2, 1, 3)
+            .reshape(B * 1, t.shape[1], C)
+            .contiguous(),
+            (q, k, v),
+        )
+        out = torch.nn.functional.scaled_dot_product_attention(q, k, v, attn_mask=None, dropout_p=0.0, is_causal=False)
+
+        out = (
+            out.unsqueeze(0)
+            .reshape(B, 1, out.shape[1], C)
+            .permute(0, 2, 1, 3)
+            .reshape(B, out.shape[1], C)
+        )
+        out = rearrange(out, 'b (h w) c -> b c h w', b=B, h=H, w=W, c=C)
+        out = self.proj_out(out)
+        return x+out
 
 class MemoryEfficientCrossAttentionWrapper(MemoryEfficientCrossAttention):
     def forward(self, x, context=None, mask=None):
@@ -315,8 +369,10 @@ class MemoryEfficientCrossAttentionWrapper(MemoryEfficientCrossAttention):
 
 def make_attn(in_channels, attn_type="vanilla", attn_kwargs=None):
     assert attn_type in ["vanilla", "vanilla-xformers", "memory-efficient-cross-attn", "linear", "none"], f'attn_type {attn_type} unknown'
-    if XFORMERS_IS_AVAILBLE and attn_type == "vanilla":
+    if model_management.xformers_enabled() and attn_type == "vanilla":
         attn_type = "vanilla-xformers"
+    if model_management.pytorch_attention_enabled() and attn_type == "vanilla":
+        attn_type = "vanilla-pytorch"
     print(f"making attention of type '{attn_type}' with {in_channels} in_channels")
     if attn_type == "vanilla":
         assert attn_kwargs is None
@@ -324,6 +380,8 @@ def make_attn(in_channels, attn_type="vanilla", attn_kwargs=None):
     elif attn_type == "vanilla-xformers":
         print(f"building MemoryEfficientAttnBlock with {in_channels} in_channels...")
         return MemoryEfficientAttnBlock(in_channels)
+    elif attn_type == "vanilla-pytorch":
+        return MemoryEfficientAttnBlockPytorch(in_channels)
     elif type == "memory-efficient-cross-attn":
         attn_kwargs["query_dim"] = in_channels
         return MemoryEfficientCrossAttentionWrapper(**attn_kwargs)

comfy/ldm/modules/diffusionmodules/openaimodel.py

@@ -477,9 +477,9 @@ class UNetModel(nn.Module):
 
         if context_dim is not None:
             assert use_spatial_transformer, 'Fool!! You forgot to use the spatial transformer for your cross-attention conditioning...'
-            from omegaconf.listconfig import ListConfig
-            if type(context_dim) == ListConfig:
-                context_dim = list(context_dim)
+            # from omegaconf.listconfig import ListConfig
+            # if type(context_dim) == ListConfig:
+            #     context_dim = list(context_dim)
 
         if num_heads_upsample == -1:
             num_heads_upsample = num_heads

comfy/model_management.py

@@ -31,8 +31,25 @@ try:
 except:
     pass
 
-if "--cpu" in sys.argv:
-    vram_state = CPU
+if "--disable-xformers" in sys.argv:
+    XFORMERS_IS_AVAILBLE = False
+else:
+    try:
+        import xformers
+        import xformers.ops
+        XFORMERS_IS_AVAILBLE = True
+    except:
+        XFORMERS_IS_AVAILBLE = False
+
+ENABLE_PYTORCH_ATTENTION = False
+if "--use-pytorch-cross-attention" in sys.argv:
+    torch.backends.cuda.enable_math_sdp(True)
+    torch.backends.cuda.enable_flash_sdp(True)
+    torch.backends.cuda.enable_mem_efficient_sdp(True)
+    ENABLE_PYTORCH_ATTENTION = True
+    XFORMERS_IS_AVAILBLE = False
+
+
 if "--lowvram" in sys.argv:
     set_vram_to = LOW_VRAM
 if "--novram" in sys.argv:
@@ -54,6 +71,8 @@ if set_vram_to == LOW_VRAM or set_vram_to == NO_VRAM:
     total_vram_available_mb = (total_vram - 1024) // 2
     total_vram_available_mb = int(max(256, total_vram_available_mb))
 
+if "--cpu" in sys.argv:
+    vram_state = CPU
 
 print("Set vram state to:", ["CPU", "NO VRAM", "LOW VRAM", "NORMAL VRAM", "HIGH VRAM"][vram_state])
 
@@ -159,6 +178,14 @@ def get_autocast_device(dev):
         return dev.type
     return "cuda"
 
+def xformers_enabled():
+    if vram_state == CPU:
+        return False
+    return XFORMERS_IS_AVAILBLE
+
+def pytorch_attention_enabled():
+    return ENABLE_PYTORCH_ATTENTION
+
 def get_free_memory(dev=None, torch_free_too=False):
     if dev is None:
         dev = get_torch_device()

comfy/sd.py

@@ -6,7 +6,7 @@ import sd2_clip
 import model_management
 from .ldm.util import instantiate_from_config
 from .ldm.models.autoencoder import AutoencoderKL
-from omegaconf import OmegaConf
+import yaml
 from .cldm import cldm
 from .t2i_adapter import adapter
 
@@ -726,12 +726,19 @@ def load_clip(ckpt_path, embedding_directory=None):
     return clip
 
 def load_checkpoint(config_path, ckpt_path, output_vae=True, output_clip=True, embedding_directory=None):
-    config = OmegaConf.load(config_path)
+    with open(config_path, 'r') as stream:
+        config = yaml.safe_load(stream)
     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"]:
+            if "use_fp16" in model_config_params["unet_config"]["params"]:
+                fp16 = model_config_params["unet_config"]["params"]["use_fp16"]
+
     clip = None
     vae = None
 
@@ -750,9 +757,13 @@ def load_checkpoint(config_path, ckpt_path, output_vae=True, output_clip=True, e
         w.cond_stage_model = clip.cond_stage_model
         load_state_dict_to = [w]
 
-    model = instantiate_from_config(config.model)
+    model = instantiate_from_config(config["model"])
     sd = load_torch_file(ckpt_path)
     model = load_model_weights(model, sd, verbose=False, load_state_dict_to=load_state_dict_to)
+
+    if fp16:
+        model = model.half()
+
     return (ModelPatcher(model), clip, vae)
 
 
@@ -853,4 +864,7 @@ def load_checkpoint_guess_config(ckpt_path, output_vae=True, output_clip=True, e
     model = instantiate_from_config(model_config)
     model = load_model_weights(model, sd, verbose=False, load_state_dict_to=load_state_dict_to)
 
+    if fp16:
+        model = model.half()
+
     return (ModelPatcher(model), clip, vae)

main.py

@@ -1,5 +1,6 @@
 import os
 import sys
+import shutil
 
 import threading
 import asyncio
@@ -8,9 +9,6 @@ if os.name == "nt":
     import logging
     logging.getLogger("xformers").addFilter(lambda record: 'A matching Triton is not available' not in record.getMessage())
 
-import execution
-import server
-
 if __name__ == "__main__":
     if '--help' in sys.argv:
         print("Valid Command line Arguments:")
@@ -18,6 +16,8 @@ if __name__ == "__main__":
         print("\t--port 8188\t\t\tSet the listen port.")
         print("\t--dont-upcast-attention\t\tDisable upcasting of attention \n\t\t\t\t\tcan boost speed but increase the chances of black images.\n")
         print("\t--use-split-cross-attention\tUse the split cross attention optimization instead of the sub-quadratic one.\n\t\t\t\t\tIgnored when xformers is used.")
+        print("\t--use-pytorch-cross-attention\tUse the new pytorch 2.0 cross attention function.")
+        print("\t--disable-xformers\t\tdisables xformers")
         print()
         print("\t--highvram\t\t\tBy default models will be unloaded to CPU memory after being used.\n\t\t\t\t\tThis option keeps them in GPU memory.\n")
         print("\t--normalvram\t\t\tUsed to force normal vram use if lowvram gets automatically enabled.")
@@ -31,6 +31,9 @@ if __name__ == "__main__":
         print("disabling upcasting of attention")
         os.environ['ATTN_PRECISION'] = "fp16"
 
+import execution
+import server
+
 def prompt_worker(q, server):
     e = execution.PromptExecutor(server)
     while True:
@@ -38,8 +41,8 @@ def prompt_worker(q, server):
         e.execute(item[-2], item[-1])
         q.task_done(item_id, e.outputs)
 
-async def run(server, address='', port=8188, verbose=True):
-    await asyncio.gather(server.start(address, port, verbose), server.publish_loop())
+async def run(server, address='', port=8188, verbose=True, call_on_start=None):
+    await asyncio.gather(server.start(address, port, verbose, call_on_start), server.publish_loop())
 
 def hijack_progress(server):
     from tqdm.auto import tqdm
@@ -51,7 +54,14 @@ def hijack_progress(server):
         return v
     setattr(tqdm, "update", wrapped_func)
 
+def cleanup_temp():
+    temp_dir = os.path.join(os.path.dirname(os.path.realpath(__file__)), "temp")
+    if os.path.exists(temp_dir):
+        shutil.rmtree(temp_dir, ignore_errors=True)
+
 if __name__ == "__main__":
+    cleanup_temp()
+
     loop = asyncio.new_event_loop()
     asyncio.set_event_loop(loop)
     server = server.PromptServer(loop)
@@ -76,11 +86,22 @@ if __name__ == "__main__":
     except:
         pass
 
+    if '--quick-test-for-ci' in sys.argv:
+        exit(0)
+
+    call_on_start = None
+    if "--windows-standalone-build" in sys.argv:
+        def startup_server(address, port):
+            import webbrowser
+            webbrowser.open("http://{}:{}".format(address, port))
+        call_on_start = startup_server
+
     if os.name == "nt":
         try:
-            loop.run_until_complete(run(server, address=address, port=port, verbose=not dont_print))
+            loop.run_until_complete(run(server, address=address, port=port, verbose=not dont_print, call_on_start=call_on_start))
         except KeyboardInterrupt:
             pass
     else:
-        loop.run_until_complete(run(server, address=address, port=port, verbose=not dont_print))
+        loop.run_until_complete(run(server, address=address, port=port, verbose=not dont_print, call_on_start=call_on_start))
 
+    cleanup_temp()

nodes.py

@@ -775,6 +775,7 @@ class KSamplerAdvanced:
 class SaveImage:
     def __init__(self):
         self.output_dir = os.path.join(os.path.dirname(os.path.realpath(__file__)), "output")
+        self.url_suffix = ""
 
     @classmethod
     def INPUT_TYPES(s):
@@ -818,6 +819,9 @@ class SaveImage:
             os.makedirs(full_output_folder, exist_ok=True)
             counter = 1
 
+        if not os.path.exists(self.output_dir):
+            os.makedirs(self.output_dir)
+
         paths = list()
         for image in images:
             i = 255. * image.cpu().numpy()
@@ -828,12 +832,25 @@ class SaveImage:
             if extra_pnginfo is not None:
                 for x in extra_pnginfo:
                     metadata.add_text(x, json.dumps(extra_pnginfo[x]))
+
             file = f"{filename}_{counter:05}_.png"
             img.save(os.path.join(full_output_folder, file), pnginfo=metadata, optimize=True)
-            paths.append(os.path.join(subfolder, file))
+            paths.append(os.path.join(subfolder, file + self.url_suffix))
             counter += 1
         return { "ui": { "images": paths } }
 
+class PreviewImage(SaveImage):
+    def __init__(self):
+        self.output_dir = os.path.join(os.path.dirname(os.path.realpath(__file__)), "temp")
+        self.url_suffix = "?type=temp"
+
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required":
+                    {"images": ("IMAGE", ), },
+                "hidden": {"prompt": "PROMPT", "extra_pnginfo": "EXTRA_PNGINFO"},
+                }
+
 class LoadImage:
     input_dir = os.path.join(os.path.dirname(os.path.realpath(__file__)), "input")
     @classmethod
@@ -954,6 +971,7 @@ NODE_CLASS_MAPPINGS = {
     "EmptyLatentImage": EmptyLatentImage,
     "LatentUpscale": LatentUpscale,
     "SaveImage": SaveImage,
+    "PreviewImage": PreviewImage,
     "LoadImage": LoadImage,
     "LoadImageMask": LoadImageMask,
     "ImageScale": ImageScale,

requirements.txt

@@ -1,7 +1,6 @@
 torch
 torchdiffeq
 torchsde
-omegaconf
 einops
 open-clip-torch
 transformers
@@ -9,3 +8,4 @@ safetensors
 pytorch_lightning
 aiohttp
 accelerate
+pyyaml

server.py

@@ -113,7 +113,7 @@ class PromptServer():
         async def view_image(request):
             if "file" in request.rel_url.query:
                 type = request.rel_url.query.get("type", "output")
-                if type != "output" and type != "input":
+                if type not in ["output", "input", "temp"]:
                     return web.Response(status=400)
 
                 output_dir = os.path.join(os.path.dirname(os.path.realpath(__file__)), type)
@@ -267,7 +267,7 @@ class PromptServer():
             msg = await self.messages.get()
             await self.send(*msg)
 
-    async def start(self, address, port, verbose=True):
+    async def start(self, address, port, verbose=True, call_on_start=None):
         runner = web.AppRunner(self.app)
         await runner.setup()
         site = web.TCPSite(runner, address, port)
@@ -278,3 +278,6 @@ class PromptServer():
         if verbose:
             print("Starting server\n")
             print("To see the GUI go to: http://{}:{}".format(address, port))
+        if call_on_start is not None:
+            call_on_start(address, port)
+

web/scripts/app.js

@@ -144,7 +144,14 @@ class ComfyApp {
 					if (numImages === 1 && !imageIndex) {
 						this.imageIndex = imageIndex = 0;
 					}
-					let shiftY = this.type === "SaveImage" ? 55 : this.imageOffset || 0;
+
+					let shiftY;
+					if (this.imageOffset != null) {
+						shiftY = this.imageOffset;
+					} else {
+						shiftY = this.computeSize()[1];
+					}
+
 					let dw = this.size[0];
 					let dh = this.size[1];
 					dh -= shiftY;
@@ -400,6 +407,15 @@ class ComfyApp {
 		api.init();
 	}
 
+	#addKeyboardHandler() {
+		window.addEventListener("keydown", (e) => {
+			// Queue prompt using ctrl or command + enter
+			if ((e.ctrlKey || e.metaKey) && (e.key === "Enter" || e.keyCode === 13 || e.keyCode === 10)) {
+				this.queuePrompt(e.shiftKey ? -1 : 0);
+			}
+		});
+	}
+
 	/**
 	 * Loads all extensions from the API into the window
 	 */
@@ -466,6 +482,7 @@ class ComfyApp {
 		this.#addApiUpdateHandlers();
 		this.#addDropHandler();
 		this.#addPasteHandler();
+		this.#addKeyboardHandler();
 
 		await this.#invokeExtensionsAsync("setup");
 	}
@@ -499,7 +516,11 @@ class ComfyApp {
 
 						if (Array.isArray(type)) {
 							// Enums e.g. latent rotation
-							this.addWidget("combo", inputName, type[0], () => {}, { values: type });
+							let defaultValue = type[0];
+							if (inputData[1] && inputData[1].default) {
+								defaultValue = inputData[1].default;
+							}
+							this.addWidget("combo", inputName, defaultValue, () => {}, { values: type });
 						} else if (`${type}:${inputName}` in widgets) {
 							// Support custom widgets by Type:Name
 							Object.assign(config, widgets[`${type}:${inputName}`](this, inputName, inputData, app) || {});