Viggle Remake: open H3 LoRAs that re-shoot a clip (swap or edit) plus an agent MCP
On Nov 4 Viggle shipped Remake: a video transformation agent with open weights. Per Viggle, two MiniMax-H3 LoRAs (swap and edit) keep the original motion, camera, timing and soundtrack while you recast people, swap objects, or change background, weather, light or style. Hosted app from about $0.96 per 10 s at 720p (their figure); ComfyUI and MCP paths for local/agent runs.
What it is, per Viggle. Remake is the follow-on to the Viggle-Animate preview: Animate replaced a character; Remake also swaps, removes and inserts objects and changes setting, weather or light. Clips can run up to about two minutes in the app by windowing. The launch post pitched conversational fixes ("keep chatting to fix any shot") and compared hosted price to Higgsfield Genjutsu ("less than half," their claim).
Open weights. The card at huggingface.co/Viggle/Viggle-Remake releases swap and edit LoRAs on MiniMax H3, a reference sampler (sample.py), ComfyUI nodes/workflows with int8/GGUF down to about 12 GB VRAM per their docs, and an mcp/ package so Claude Code or Codex can cut windows, paint keyframes, render and splice with the source soundtrack. Full bf16 sampling wants a large GPU (they cite 80 GB+); smaller ranks (128/64) and quantized packs are documented. Weights are a MiniMax H3 Model Derivative under that community licence; inference/MCP code is Apache 2.0. Do not git-clone the whole repo with LFS if you only need the LoRAs or MCP — Viggle warns the full tree is ~210 GB.
Signals (HOT). Official @ViggleAI announce on Nov 4 took off (~619 likes, ~168 reposts, ~499 bookmarks, ~562k impressions when we checked). Creator demos kept circulating through Nov 6 (recasts, restyles, ComfyUI walkthroughs). That is official announcement plus an author post that took off inside 72 hours, so we flag it hot. Listed ReelStack curators had not covered it in this window, and it was not on GitHub Trending today.
Consent and limits, in their own words. Identity comes from the frames you paint; the model does not verify consent. Viggle's card says use it only on footage and likenesses you have the right to change, and to label AI-generated output. Known limits they list: ~12 s per render window, six-section H3 prompts required, sound is carried not generated, and non-humanoid long takes or unpinned multi-shot clips can drift. We have not run Remake ourselves.
Our take: for production teams that already sit on MiniMax H3, Remake is the open path to plate-level recast/edit without sending every change through a closed generator — next to OPENjutsu-style swaps and Body Double. Start on a consent-only clip of your own footage: hosted app for a smoke test, then ComfyUI quantized packs or the MCP worker if you have GPUs. Treat Genjutsu price comparisons as vendor marketing until you meter your own job.
Related tools on ReelStack
Skills & repos in our directory
- Viggle Remake · Repo, VFX & compositing
- Comfy Previs · Repo, Generation
- Seam · ComfyUI, VFX & compositing
- OpenField · Repo, Generation
- Kitty Open Studio · MCP, Generation
- Henge · ComfyUI, Generation
More from the Lab
- HyperFlow H3 Flash: Video Rebirth's 14B, 6-step MiniMax-H3 that makes video and sound on one consumer GPU
- ComfyUI lands in ChatGPT, and v0.44 adds ProRes saving and MiniMax H3 edit nodes
- P-H3-Image: Pruna turns MiniMax H3 into an 8-step image model
- Pika H3 Flex: open six-step adapters for MiniMax H3, plus audio stems
Topics: ComfyUI · Open-weight video & audio models · Video MCP servers