Simple Wan wrapper released
Requires Comfyui version v0.3.27 or higher.
https://github.com/Flow-two/flow2-wan-video
(OPTIONAL) Triton & SageAttention Installation
https://civarchive.com/articles/12851/easy-installation-triton-and-sageattention
Installation
option 1
Download archive zip.
flow2-wan-video.zip extract to
ComfyUI\custom_nodesGo to
ComfyUI_windows_portable/folder and Run command topython_embeded\python.exe -m pip install -r "ComfyUI\custom_nodes\flow2-wan-video\requirements.txt"
option 2
Go to comfyUI custom_nodes folder,
ComfyUI/custom_nodes/Go to
ComfyUI_windows_portable/folder and Run command topython_embeded\python.exe -m pip install -r "ComfyUI\custom_nodes\flow2-wan-video\requirements.txt"
Upscale Model
https://huggingface.co/mixfox/Upscale-Models/blob/main/4x_foolhardy_Remacri_ExtraSmoother.pth
to ComfyUI/models/upscale_models
Wrapper Features
Various image settings (blur, saturate, noise, quality)
Various model patcher integrated into one
Support for frame interpolation
Support for upscaler
Support high quality sampling preview
Support teacache retention mode
Support model auto download
Version 1.0
Features
Easy and simple parameter setup
Supports various parameters to minimize artifacts
Supports Skip Layer Guidance (SLG) to minimize artifacts
Two-step sampling process for faster sampling speed and relatively less noise
Faster sampling speed with TeaCache support
Supports various samplers
Easy upscaling control
Supports various frame rates for frame interpolation
Supports lower VRAM usage with the introduction of gguf
Nodes
ComfyUI-Custom-Scripts
ComfyUI_LayerStyle
rgthree-comfy
ComfyUI-KJNodes
ComfyUI-VideoHelperSuite
ComfyUI-Frame-Interpolation
ComfyUI-mxToolkit
Warning: You are fully responsible for any legal or ethical violations that may occur in the production of this video. It is important to comply with all relevant laws and regulations when creating and distributing the video, and to be cautious not to infringe on the rights of others.
Description
fixed error on download comfy model
FAQ
Comments (11)
Dumb bug but mentioning anyway, any image pasted into the flow2 Resize Image node won't be detected and throws an error. Images that exist as a file (not clipboard) work fine. Hope it can be fixed.
When generating using WAN start/end images, there is a high probability that the generated video will have a stronger purple tint than the loaded image.
LTX・HunyuanVideo start/end video generation also generates videos with a similar tendency, so this may be an unavoidable problem, but does anyone know the cause and a workaround?
(*No color adjustment nodes are used.)
safetensors files throwing this error in all 3 demo workflows, gguf is fine:
WanVideoModelLoader_F2
download_huggingface_model() takes 3 positional arguments but 62 were given
I’m curious what you think about OptimalSteps. Are you planning to use it too for faster generation? I tried it with your old workflow, and it worked great for both T2V and I2V(Euler only). It’s fast, and the quality looks good. What do you think?
Im getting ASS results with Image to Video- making blotchy colour spots all over the image no matter what I try. The Text to Video is perfect.
I was able to get your very old workflow for Image to video before you deleted it, and that still works great, but this new one just gives Bad results
It's much faster and more convenient than before. Thank you.
However, when I move my hand while creating the video, only the hand part is too blurry.
What should I do?
I'm having a problem initializing the video. I get this error: WanVideoModelLoader_F2
No module named 'flow2-wan-video.gguf.tools'
And I've already updated to the new version 1.10
I'm just curious, but in your preview image enhance_strength, cfg_zero_steps and skip_layer are either 0 or disabled.
From what I remember browsing is that skip_layer optimizes the movement especially with loras, enhance_video enhances specific object representations and their movements...and I forgot what cfg_zero_steps does >.<
Do you have a recommendation on what to activate/set if using a Q6 quant?
I can't find the model 'fp8_e5m2' in the list in version 1.1. Any plans to add it?
Upscale用了反而报错了....
SamplerCustom
mat1 and mat2 shapes cannot be multiplied (769x4863 and 5120x5120)
i think v1 is better than v1.10.. idk why. too much face changing with i2v in v1.10 than v1
Looks like we don't have an active mirror for this file right now.
CivArchive is a community-maintained index — we catalog mirrors that volunteers upload to HuggingFace, torrents, and other public hosts. Looks like no one has uploaded a copy of this file yet.
Some files do get recovered over time through contributions. If you're looking for this one, feel free to ask in Discord, or help preserve it if you have a copy.