Wan2.7: Reference to Video
Upload a reference video and character image to generate a video with consistent character features.
About
This workflow turns a reference video into a new video featuring your chosen character while preserving the motion, framing, and timing of the original. It uses the Wan2.7 model via the Wan2ReferenceVideoApi node to track the reference video’s movement and composition, then conditions generation on a single character image you load with the LoadImage node. The output is a sequence of frames that maintain the target character’s facial features and overall look across time, which the SaveVideo node encodes into a final video file.
What you provide
- Reading the workflow
Runs in ComfyUI. This workflow can't run on this page. Open it in Comfy Cloud with everything loaded, or download it for your own ComfyUI.
Try in CloudQuestions & answers
You need a single character image (loaded with LoadImage) and a reference video (set inside Wan2ReferenceVideoApi). The node will output generated frames that SaveVideo encodes into a final video.
Use a high-resolution, front-facing, well-lit character image with minimal occlusions. Match the aspect ratio of the character crop to the reference video subject, and avoid reference clips with extreme motion blur, rapid cuts, or heavy occlusions.
For natural motion, set SaveVideo’s frame rate to match the reference video. Keep resolution moderate for stability and speed (matching the reference or the API node’s default). If results look unstable, try slightly lower resolution or a shorter clip.
Results depend on the exact inputs and settings. Keep the same character image, reference video, and node options to get consistent outputs. If the node exposes a seed parameter, reusing the same seed further improves reproducibility.