Overview
reshot is an open-source video-processing tool that converts reference footage into structural control material for AI video generation. Rather than carrying the original actors into a new production, it extracts information about depth, movement, or contours so creators can reconstruct a shot with different characters, settings, and visual styles.
The tool is designed around controlled video-to-video work, including short dramas, character animation, and experimental visual projects. Its outputs are intended for workflows involving Seedance, MiniMax H3, Wan VACE, and ComfyUI.
Key Features
- Depth-map extraction captures foreground and background relationships, spatial layout, and camera movement.
- DWPose and OpenPose processing records human skeletal motion frame by frame.
- Canny edge extraction preserves contours, composition, object shapes, and architectural lines.
- Control-oriented output can be used as reference material in supported AI video and ControlNet workflows.
- Apache-2.0 licensing makes the project free and open source.
Workflow
Users load a source video, choose depth, pose, or Canny extraction, and process the footage into a control video or image sequence. Optional controls include frame sampling, line thickness, skeleton point size, and depth-output color or contrast.
The resulting material can then serve as structural input in a compatible generation workflow. The appropriate extraction mode depends on whether the project needs spatial consistency, precise movement, or preservation of shapes and framing.
Use Cases
reshot can preserve camera angles and performer blocking while a scene is rebuilt with new characters or costumes. Motion captured from ordinary footage can guide an animated character, while depth information can retain the structure of a location as lighting, weather, or style changes. Edge output is suited to stylization that keeps the original composition recognizable.