PickViggle-Remake: An agent pipeline that remakes an entire video with a single-sentence instruction
Viggle/Viggle-Remake
About the project
Given only a video and a one-line prompt, the agent automatically handles everything from shot segmentation to keyframe painting, rendering, and final compositing. It integrates with agents supporting the MCP protocol, such as Claude Code or Codex, to modify content while preserving the original movement, camera work, timing, and sound.
It offers two LoRA modes based on MiniMax-H3. The swap mode replaces characters or figures, while the edit mode performs object addition/deletion, background/weather/lighting changes, and style transfer. It takes 1 to 4 painted frames and a 6-section format prompt as input and renders in windows of up to 345 frames (approximately 12 seconds).
Whereas the existing Viggle-Animate v2 focused on character replacement, Remake expands the scope to include object insertion/deletion and environment changes. It requires a 144GB scale base model and an 80GB or larger GPU, but it provides int8/GGUF files for ComfyUI (minimum 12GB), enabling execution even in low-spec environments.
It is suitable for producers who want to achieve reshoot effects, such as actor replacement, prop changes, and weather modifications, on videos they have the rights to. The model does not verify consent; therefore, consent procedures and AI-generated content labeling are the user's responsibility when using likenesses of real people.
Viggle/Viggle-Remake
The original page has no description.
video-to-video
This introduction was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report errors, attribution issues, or removal requests via Contact.