[Question] Add/Insert New Character via Edited First-Frame Reference + Masked Propagation for Motion Transfer

[phamkhactu]

Thank you for the amazing VACE project.

I'm trying to achieve a specific workflow that seems very close to what VACE's Masked V2V (MV2V) + Reference-Anything capabilities support, but I couldn't find clear guidance or examples in the README/docs/demo. I'd like to ask if this is already possible (and how to best set it up), or if it could be a nice feature enhancement.

Desired Workflow:

1. Start with an existing source video (src_video).
2. Extract the first frame of the source video.
3. Manually edit this first frame (using Photoshop, GIMP, or any image editor/AI tool) to insert/add a new person/character at a desired position (e.g., standing in an empty area, or replacing an existing one).
4. Use this edited first-frame image as the reference image (src_ref_images).
5. Run VACE in a masked/inpainting mode:

• Provide a mask (or bbox) that covers only the region where the new person should appear/move (e.g., the added area in the first frame, tracked across frames if possible).

• The goal: Generate a new video that keeps almost 100% of the original video intact (camera movement, background, lighting, timing, scene structure, other objects/people), only modifying the masked region to insert the new character from the reference.

• The new character should move/animate according to the motion/pose dynamics inferred from the source video

In short: Pixel-perfect preservation of the source video everywhere except the added new subject, with the new subject following realistic motion consistent with the scene.

One more time thanks a lot for your work and for considering this!