Viggle releases Viggle-Remake: LoRA models re-shoot existing video...

Viggle releases Viggle-Remake: LoRA models re-shoot existing video on top of MiniMax-H3

Viggle has released the Viggle-Remake repository on Hugging Face: a pair of LoRAs (rank 256, alpha 256) on top of the MiniMax-H3 base model (~20B parameters) re-shoots existing video, replacing characters (swap mode) or objects, background, weather, and lighting (edit mode) while preserving motion, camera, and audio.

Compiled by Sergey KostenchukPublished 2026-11-05Updated 2026-11-05

2026-11-05 Multimodal Anthropic

🎨 Viggle releases Viggle-Remake: Open video re-shooting

A pair of LoRAs (rank 256, alpha 256) on top of MiniMax-H3 (~20B) has been published on Hugging Face. Swap changes characters in an existing clip, while edit changes objects, background, weather, and lighting, preserving motion, camera, and audio. An MCP server is available for Claude Code and Codex agents: it automatically cuts video into windows, renders, and stitches them together.

🌍 Open weights make re-shooting accessible without paid APIs — a local pipeline with MCP. This is a step toward automating post-production: replacing actors, locations, and weather using coding agents rather than manual editing.

👤 Key files (~5.5 GB) run on cards with 12 GB VRAM or more via ComfyUI and GGUF. The render window is up to ~12 seconds; the model does not verify consent from people, so only use your own material.

Source 1: https://huggingface.co/Viggle/Viggle-Remake Source 2: https://comfyui-wiki.com/en/news/2026-11-04-viggle-remake

Sources