ComfyUI: Generate character-consistent video with Wan 2.6 R2V
ComfyUI
Demo summary
The creator demonstrates a ComfyUI workflow using the Wan 2.6 API node to transform a 5-second reference video of himself into a new scene while maintaining character consistency and native audio lip-sync.
Step-by-step
- Open the ComfyUI workflow containing the Load Video, Wan 2.6 API, and Video Combine nodes.
- Upload a short reference video of your subject into the Load Video node.
- Enter a prompt describing the new scene, setting, action, and mood.
- Specify the dialogue in the prompt if you want the character to say specific words.
- Run the workflow to generate the 5 or 10-second clip.
Options
- Input up to three reference videos to include multiple subjects or specific objects
- Let the model generate random speech or provide a specific script for lip-syncing
- Reference people, animals, characters, or objects
Watch out for
- Using the Wan 2.6 API node incurs a cost (approximately 50 cents per video)
- Reference videos are limited to 5 or 10-second clips at up to 1080p resolution
Tips
- Cram as much information as possible into the reference video by capturing different angles and expressions.
- Move the camera around the subject (like a 360° walk-around) to help the model understand the subject's form and movement.
- Include talking in the reference video to enable the native audio lip-sync and voice replication features.
Highlights
“So we have a very simple workflow to work with in Comfy here... this should be fairly simple, right?”
All demos from “Wan 2.6 is HERE! R2V is AWESOME!”
- 1:003:01Generate character-consistent video with Wan 2.6 R2VCurrentThe creator demonstrates a ComfyUI workflow using the Wan 2.6 API node to transform a 5-second reference video of himself into a new scene while maintaining character consistency and native audio lip-sync.ComfyUI· Video to Video
- 5:170:33Multi-reference video generation in Wan 2.6A demonstration of using multiple input videos (a subject, a motorcycle, and a location) within the Wan 2.6 workflow to generate a complex scene of the subject riding a specific bike in a specific setting.WWan· Video to Video
- Watch “Wan 2.6 is HERE! R2V is AWESOME!” →
Video to Video
8:392:29Comparing LTX 2.3 Vanilla vs. TensNodes for Facial ExpressionsA side-by-side demonstration shows the TensNodes workflow maintaining a consistent facial structure during a smile and wave, whereas the vanilla LTX model drifts from the reference image.SOTAI
15:344:59Character replacement with Wan AnimateThe creator demonstrates using Wan Animate and SAM 3 to mask a character in a reference video and replace them with a new character from a static image while maintaining the original motion.pixaroma- 21:072:43Cartoon animation with Wan SCAILThe user shows how to animate a cartoon ballerina using a real-person video as a motion reference via the Wan SCAIL workflow in ComfyUI.pixaroma
4:031:00Generating Guide Videos with BFS NodeThe demo shows how to use the BFS Node extension to combine a reference face image and a driving video into a green-screen guide video required for the LTX sampling process.Veteran AI- 5:030:47Injecting Latent Constraints with Add Guide MultiThe video demonstrates using the 'Add Guide Multi' node from the KJ Node extension to inject the guide video constraints and VAE-encoded latents into the sampling pipeline.Veteran AI
15:352:08Final video generation with Wan 2.1The demonstration shows how to use the Wan 2.1 V2V model to combine the generated 3D motion and a reference character image into a final, high-quality AI video animation.Benji’s AI Playground
8:521:04Replacing a dancing character with MochaA concrete demonstration of replacing a woman dancing in a video with a new character using the Mocha workflow in ComfyUI.SOTAI- 12:450:58Replacing a human with an animated werewolfThe video demonstrates replacing a human actor with a stylized werewolf character while maintaining consistent textures and movement.SOTAI
- 1:003:01Generate character-consistent video with Wan 2.6 R2VCurrentThe creator demonstrates a ComfyUI workflow using the Wan 2.6 API node to transform a 5-second reference video of himself into a new scene while maintaining character consistency and native audio lip-sync.Sebastian Kamph
1:551:00Adding reference images for character consistencyThe video shows how to use the MSR custom node to input up to four reference images and a background image to maintain character facial consistency across video frames.bigboss97
2:280:44Swap human character for 3D Pixar model in MochaThe creator demonstrates swapping a real person in a video with a 3D Pixar-style character using Mocha, showing the transfer of lip movements and hand gestures.AI Search- 20:110:27Generating final character swap videoThe video shows the K-Sampler process and the final rendered output of a character replacement task within ComfyUI.AI Search