Update 4-3-2025:
city96 has published a full set of ggufs for this model you can find here if you are looking for a different size.
https://huggingface.co/city96/Wan2.1-Fun-14B-Control-gguf/tree/main
____
This is a q5_k_m of the fun control model that you can find here:
https://huggingface.co/alibaba-pai/Wan2.1-Fun-14B-Control/blob/main/README_en.md
This model allows you to use pose, depth, or other info to guide your video (see the hugging face page for details.)
I made it using instructions from here:
https://github.com/city96/ComfyUI-GGUF/tree/convert_refactor_new
I've tested it on a few videos and it seems to work and is much faster on my 3090 than using the much larger fp8 quantized by kijai that you can find here:
https://huggingface.co/Kijai/WanVideo_comfy/blob/main/Wan2.1-Fun-Control-14B_fp8_e4m3fn.safetensors
You'll need to use the WanFunControlToVideo node in comfy core.
Description
FAQ
Comments (15)
Impressive that you managed to get that rather convoluted procedure (unless it's been simplified in the 3 days since I last checked...), but know that city's been converting the latest models to gguf days after they're released.
https://huggingface.co/city96/Wan2.1-Fun-14B-InP-gguf/tree/main
I saw they got the inp done but i was looking for the control one. I'm sure they'll get to it but I wanted it now and thought I'd share.
@o0oradaro0o Oh, my bad. Didn't even notice these were separate models.
@o0oradaro0o May I ask what's the difference between InP and Control version? I want to use it like controlnet from sd.
@bbeggs234 my understanding is the inp model is for start/end frames (provide 2 images one for how you want the video to start and one for how you want it to end and it will interpolate between them). the model i uploaded is like sd controlnet where for each frame you want it to generate you provide a guiding image like a pose or a depth map. (see my sample video, if that's the thing you wanted this is the model you need.)
hope that helps.
Nice! thanks
city96 is slow with FunControl somewhy
Very useful, thank you
Is there repository anywhere for motion poses animations. ie. for dances or anything like that? I know the other pose repositories are stills.
download a video of someone moving, load video --> preprocessor of choice --> image resize if you want --> video combine (set same or wan/hun fps), you have a pose video. Also found this, might be good still https://civitai.com/models/56307/character-walking-and-running-animation-poses-8-directions
I’m getting this error.
WanFunControlToVideo
Calculated padded input size per channel: (0 x 64 x 64). Kernel size: (1 x 1 x 1). Kernel size can't be greater than actual input size
Both the input image and video are 512x512. I double-checked that every node uses those dimensions. I also tested with 480x480, but still no luck.
Update/reinstall. Dependencies back end error.
Any idea how to lower the strength of ControlNet or configure it? Sometimes, OpenPose doesn't work and gives black frames. And if my character in the input image has short hair, but the input video has long hair, DepthAnything will make it that long(it also applies to clothing or the character's build)
how can i get the workflow，i try to test it，please，3u
you download an image here and you open the webp on comfyui
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.