New set, same actor
Turn an ordinary corridor into an organic cave straight out of an 80s horror film, without touching the actor walking through it. One source video, one reference image, one locked prompt.
- Type
- ComfyUI workflow
- Status
- R&D
- Updated
- October 16, 2026
- Stack
- ComfyUIComfy CloudGemini Omni FlashReference image
The problem
Restyling a video with AI almost always damages what matters most: the character. The face changes, the silhouette drifts, the walk starts to float. For an ad or a series, the actor and the product must stay exactly as they were filmed.
The pipeline
A 12-node ComfyUI workflow, run on Comfy Cloud:
- The source video: a man walking down a corridor, shot simply.
- A reference image that sets the art direction: grade, light, textures, mood.
- A two-part prompt. The first part locks the actor (same face, same body, same clothes, same walk, same place in the frame, never regenerate him). The second describes the new set: walls swallowed by a blue-black organic mass, translucent veins pulsing with red fluid, breathing membranes.
- Lighting described as on a set: near-black, a single cold beam from above, volumetric rays in the smoke, a wet floor reflecting the blue, film grain.
- Format and duration injected automatically into the prompt (16:9, 6 s), then video editing by Gemini Omni Flash, which receives the video, the image and the text in a single pass.
What it changes
The shot keeps its original performance, rhythm and framing: only the world changes. A single shoot can be spun into several worlds, to test art directions or produce ad variants, without reshooting a single frame.