UPDATE!
download my ComfyUI Build https://huggingface.co/datasets/StefanFalkok/ComfyUI_portable_torch_2.10.0_cu130_cp313_sageattention_triton/tree/main , also u need download and install CUDA 13.0 (https://developer.nvidia.com/cuda-13-0-0-download-archive) and VS Code (https://visualstudio.microsoft.com/downloads/)
My TG Channel - https://t.me/StefanFalkokAI
My TG Chat - https://t.me/+y4R5JybDZcFjMjFi
Also big thanks to
for helping with workflows in past!
Hi! I introduce my working workflows with Wan 2.2 generation video for ComfyUI
I have included 4 workflows such as t2v, i2v, v2v, flf2v (firstlastframe) and foley audio
You need to have Wan 2.2 10 steps models (https://huggingface.co/StefanFalkok/Wan_2.2_10steps/tree/main), clip (https://huggingface.co/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/blob/main/split_files/text_encoders/umt5_xxl_fp16.safetensors) and vae (wan 2.1 vae)
GGUF Wan 2.2 10 steps models
https://huggingface.co/StefanFalkok/Wan_2.2_10steps_GGUF
Leave comments if you have trouble or you found the problem with workflows
Description
FAQ
Comments (30)
Hello,
Thank you very much for sharing this. Everything works well and the result is very pleasing, but in my case I had to disable the Sage node because it was causing everything to crash. I have everything up to date and I have an RTX 4070 Super, so normally there shouldn't be any issues, but after trying several different solutions, nothing worked, so I disabled it and everything works fine now.
Hi, thanks for the review! Try installing Sageattention 1.0.6 instead 2.2.0 - maybe it helps
I run nsfw version without any cuts and cuts from image to video. But it doesn't do any sexy moves, it just walks, I didn't even change the prompt.
Did you generate nsfw content via i2v workflow or t2v ? Because t2v with nsfw doesn't work correctly
or maybe you need picture with nsfw content
Hello,
I'm trying to run your workflow on a Runpod instance, but I'm encountering a persistent issue with the WAS Node Suite. I've already installed the RES4LYF node to fix the samplers, and that part seems to be working. However, the WAS Node Suite itself is giving me an error. The problem is a cv2 incompatibility, and I get the following error message on startup: AttributeError: module 'cv2' has no attribute 'getBuildInformation'
I've already tried to uninstall and reinstall different versions of opencv-python and opencv-python-headless, but the error persists.
Could this be a known issue, or is there a specific installation step or dependency I might be missing for this node on Runpod? Any help would be appreciated.
Hi. What kind of GPU do you use ? What kind of torch version has it been installed in your comfy?
Maybe you're using another models wan 2.2, and wan torch node gives errors, maybe Clip conflicts with your wan 2.2 model and u just need change another clip from umt5_xxl_bf16 to fp8
falkok21z My GPU is an NVIDIA RTX A5000 on a Runpod instance, with 24GB of VRAM. The PyTorch version is 2.2.0, and it has support for CUDA version 12.1
falkok21z I uninstalled and reinstalled the wan 2.2 models, and I changed the umt5_xxl_bf16 clip to fp8, but I'm still getting the same result.
Every time I open the workflow, it says there was an error installing the WAS Node Suite, which is why the seed node is not running. Maybe the problem is with my ComfyUI...
meovvenom try use Wan 2.2 e5m2 fp8 models by Kijai (link in description), because your GPU doesn't support e4m3fn models. U need update pytorch version to 2.7.x, try install CUDA 12.8 - maybe it helps
i had a Ram error after first ksampler finish.
0x00007FFD08BBC1D1, C:\1\ComfyUI_windows_portable\python_embeded\python313.dll(0x00007FFD08B00000) + 0xBC1D1 byte(s), PyObjectMakeTpCall() + 0x81 byte(s)0x00007FFD08D9D678, C:\1\ComfyUI_windows_portable\python_embeded\python313.dll(0x00007FFD08B00000) + 0x29D678 byte(s), PyContextNewHamtForTests() + 0x80 byte(s)0x00007FFD08D9D954, C:\1\ComfyUI_windows_portable\python_embeded\python313.dll(0x00007FFD08B00000) + 0x29D954 byte(s), PyContextNewHamtForTests() + 0x35C byte(s)0x00007FFD08B248F3, C:\1\ComfyUI_windows_portable\python_embeded\python313.dll(0x00007FFD08B00000) + 0x248F3 byte(s), PyArg_ParseTupleAndKeywords() + 0x2C23 byte(s)0x00007FFD08B90571, C:\1\ComfyUI_windows_portable\python_embeded\python313.dll(0x00007FFD08B00000) + 0x90571 byte(s), PyObject_Call() + 0x125 byte(s)0x00007FFD08B904BB, C:\1\ComfyUI_windows_portable\python_embeded\python313.dll(0x00007FFD08B00000) + 0x904BB byte(s), PyObject_Call() + 0x6F byte(s)0x00007FFD08B33EDC, C:\1\ComfyUI_windows_portable\python_embeded\python313.dll(0x00007FFD08B00000) + 0x33EDC byte(s), PyEvalEvalFrameDefault() + 0x5E7C byte(s)0x00007FFD08B2CF87, C:\1\ComfyUI_windows_portable\python_embeded\python313.dll(0x00007FFD08B00000) + 0x2CF87 byte(s), PyObject_Vectorcall() + 0xD97 byte(s)0x00007FFD08B90E25, C:\1\ComfyUI_windows_portable\python_embeded\python313.dll(0x00007FFD08B00000) + 0x90E25 byte(s), PyObject_CallNoArgs() + 0x329 byte(s)0x00007FFD08B90571, C:\1\ComfyUI_windows_portable\python_embeded\python313.dll(0x00007FFD08B00000) + 0x90571 byte(s), PyObject_Call() + 0x125 byte(s)0x00007FFD08B904BB, C:\1\ComfyUI_windows_portable\python_embeded\python313.dll(0x00007FFD08B00000) + 0x904BB byte(s), PyObject_Call() + 0x6F byte(s)0x00007FFD08C32338, C:\1\ComfyUI_windows_portable\python_embeded\python313.dll(0x00007FFD08B00000) + 0x132338 byte(s), PyThreadState_Clear() + 0x3D0 byte(s)0x00007FFD08C3215E, C:\1\ComfyUI_windows_portable\python_embeded\python313.dll(0x00007FFD08B00000) + 0x13215E byte(s), PyThreadState_Clear() + 0x1F6 byte(s)
It's a "Run out of memory" U need to increase the RAM with large volume, at least 64 GB
i just change my pagfile.sys ram to 50gb and its running well 'pagfile.sys is increase your ram with your storage' u guys can google it. Works for me
tomrisoyaturan336 i see. I'm glad you have solved the problem
What should i do with this?
KSamplerAdvanced
AttributeError: type object 'CompiledKernel' has no attribute 'launch_enter_hook' Set TORCHDYNAMO_VERBOSE=1 for the internal stack trace (please do this especially if you're reporting a bug to PyTorch). For even more developer context, set TORCH_LOGS="+dynamo"
what kind of gpu and what the pytorch version? Maybe you haven't installed torch to your comfy
@falkok21z 3080ti pytorch ver 2.7.0+cu128
@mrbluegear908 try use e5m2 model - maybe it works, because e4m3fn models doesn't support rtx 3000 and older gpus
@falkok21z i tried and it didnt help
@mrbluegear908 try use without torch, or find compatible version of torch for 3080 ti. Many people complain abouy many errors on 3000 rtx and older series
@falkok21z after the removed the torchcompile everything working fine now thanks you! But only thing that the animation is so least and not dynamic is there something tweaking that?
@mrbluegear908 try increase cfg from 1.3 to 2, but generation will take longer time. Also u can change sampler like unipc - the most universal sampler. Or maybe you use lightx 4steps, I recommend use lightx from wan 2.1 like rank 128-256 bf16 and set 10 steps - you get more dynamic video
@falkok21z okay now which way to speed up greneration without sacrifice so much quality?
@mrbluegear908 use lightx Lora for wan 2.1 and set 8 steps, use sageattention and torch (or leave only sageattention)
another question why it does have weird color mask on video ? something like it doesn't have a clear image
@mrbluegear908 hi. Soon I'll release a new correct workflows with with bug fixes and hints, because I have tested workflows and I made a seriously mistake early
Hello, I have mapped the model and LORA to their full names according to the workflow, but it is not functioning properly. It prompts 'mat1 and mat2 shapes cannot be multiplied'. How can I resolve this? I look forward to your generous reply.
Hi, maybe u need change clip model (another umt5 xxl version between bf16 and fp8 e4m3fn, or use wan 2.1 vae instead wan 2.2 vae (because 2.2 vae works with 5B model only), or use wan 2.2 models from https://huggingface.co/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/tree/main/split_files/diffusion_models instead Kijai. Maybe try generate videos withot Torch Compile (because many users who has 4000, 3000 rtx and older complain about this moment with e4m3fn model and download e5m2 wan models)
@falkok21z Thank you very much for your reply. I will try the solution you provided later. If there are still issues, I will contact you. Thank you again for your generous and quick response.