Viggle Remake reshoots video with two open MiniMax H3 LoRAs

Viggle released Viggle Remake, an open release that re-shoots existing footage on MiniMax H3, on November 4, 2026. The ComfyUI Wiki dates the release, and Viggle's announcement of it, to that day. The package is two LoRAs on H3's base transformer, a small ComfyUI node pack, ready-made graphs, quantized builds, and an MCP server that remakes a whole film shot by shot.

The model card calls Remake the "Viggle-Animate v2" previewed on the earlier Viggle-Animate card. Viggle-Animate shipped in late August and replaced a character in a shot from one repainted frame. The ComfyUI Wiki reports Remake is a LoRA on the stock H3 transformer rather than a finetune, and edits more than people.

Two modes ship. The swap LoRA replaces people and characters, one or several at once. The edit LoRA does the rest: swap, remove or insert objects, or change the background, the weather or the light. Both take the source clip plus one to four painted reference frames, each of which can be pinned to the source frame it was painted from. Identity comes from the painted frames, not a face encoder or a segmentation mask, which is why the pipeline needs no pose estimation or tracking.

Running it yourself needs the MiniMax H3 base model, about 144 GB of files downloaded separately from MiniMaxAI/MiniMax-H3. The LoRA download is about 5.5 GB, against roughly 210 GB for the entire repository. The reference sampler runs as a script with flags including --pin, --frames, --short-edge, --seed and --lora-repo, and at bf16 the card says it needs one 80 GB or larger GPU. Each render covers about 12 seconds, and --short-edge 736 renders 1280x736.

For ComfyUI, the card repacks the two LoRAs as viggle_remake_swap_comfyui.safetensors and viggle_remake_edit_comfyui.safetensors, with three nodes and sigma presets tested on ComfyUI 0.42.0. The card also ships merged int8 and GGUF transformers that it says run the model on smaller cards, down to 12 GB.

The license is stated plainly. The card says the weights, including the merged int8 and GGUF files, are a Model Derivative of MiniMax H3, so the MiniMax H3 Community License applies to them, while the code in inference/, examples/, comfyui/ and mcp/ is Apache 2.0. The card warns the model cannot verify identity or consent, since identity is decided by the frames a user paints. Viggle's own site loaded but does not name Remake.

Viggle Remake runs in the Viggle app with no setup, or from the LoRAs and base weights on Hugging Face.