CogVideoX-Fun Sampler for Video to Video

This is CogVideoX-Fun's video-to-video and control-conditioned sampler, pairing with the -Pose and -Control checkpoints from LoadCogVideoXFunModel. It's the same job Wan2_2FunV2VSampler does for the Wan side of this pack - restyle an existing clip, or drive generation with a preprocessed conditioning video - just against the older, lighter architecture. When this feature first shipped, one early tester on the -Pose variant described it plainly: it "accepts a validation video (openpose samples) and a control video," and results were noticeably livelier and more motion-willing than the earlier plain InP checkpoints of the same generation.

How it works

Give it a source video and a denoise_strength below 1.0, and it partially renoises and resamples rather than generating from scratch - the video equivalent of img2img denoise. Feed it a control_video instead (or alongside) for pose/canny/depth-driven conditioning, using whichever scheduler you prefer from the classic diffusion set this model family uses.

The inputs that matter

• cogvideoxfun_model - from LoadCogVideoXFunModel, a -Pose or -Control checkpoint.
• prompt / negative_prompt - STRING_PROMPT type; wire from FunTextBox.
• video_length - default 49, range 5-85, step 4.
• base_resolution - 512/768/960/1024, default 768.
• denoise_strength - default 0.7, range 0.05-1.0. The main creative knob: low keeps you close to the source, high gives the model more room to diverge from it.
• steps - default 50, up to 200.
• cfg - default 6, range 1-20.
• scheduler - Euler, Euler A, DPM++, PNDM, or DDIM, default DDIM.
• validation_video (optional, IMAGE) - the source clip for straight restyling.
• control_video (optional, IMAGE) - a preprocessed pose/canny/depth video matching what the checkpoint was trained on.

Output: images - a frame batch.

Installing it

Part of the base VideoX-Fun pack. Search ComfyUI Manager for VideoX-Fun; if it's not indexed:

cd ComfyUI/custom_nodes
git clone https://github.com/aigc-apps/VideoX-Fun
pip install -r VideoX-Fun/requirements.txt

Restart ComfyUI.

Common issues

No control video and unsure how to make one. For pose specifically, this pack's own VideoToOpenpose node handles it. For Canny or Depth conditioning against a -Control checkpoint, you'll need ComfyUI's separate controlnet-aux preprocessors - nothing else in this pack extracts those.

Output doesn't follow the control video at all. Confirm you're on a -Pose or -Control checkpoint and not an -InP one loaded through the wrong model_type - an Inpaint-trained checkpoint doesn't know what to do with conditioning it was never trained to use.

Barely changes from the source video. denoise_strength too low - push it toward 0.8-1.0 for a heavier restyle.

This feels less commonly used than the Wan equivalent, and it is. By the time this pack added Wan support, most of the community's attention on video-to-video and control-conditioned generation had already moved to Wan plus VACE. This node still works and does exactly what it says, but if you're choosing fresh and don't have a specific reason to be on CogVideoX-Fun, Wan2_2FunV2VSampler is where you'll find more workflows, more LoRAs, and a bigger community to ask when something goes wrong.