Viggle Remake edits videos while preserving original movements

Viggle AI has released Viggle Remake, an AI video editing model that modifies subjects and environments while keeping original movements, direction, and audio intact. The tool is designed to reshoot entire footage by changing the style, characters, or visual context. The system preserves the camera work, timing, and soundtrack of the source video.

What was announced

Viggle Remake is the evolution of the previous Viggle-Animate model. While the previous version only replaced a single character, this one allows swapping clothes, adding or removing objects, and changing the weather, lighting, and setting. The model is released under the minimax-h3-community-license on Hugging Face. The base files weigh about 144 gigabytes, while the full repository reaches 210 gigabytes. Sources show concrete application examples: a La La Land scene reshoot in 22 different ways, and a Matrix Reloaded clip with the protagonists turned into a cat and Shiba Inu dogs. The tool is structured to handle entire videos, not just short isolated clips.

How it works

The system operates in two distinct modes to suit the user's technical skills. The first is driven by external AI agents compatible with the MCP protocol, a system that allows programs to communicate with each other: just provide a video and a text command. The agent finds the cuts, paints the keyframes, and renders the final result autonomously. The second mode is manual and requires direct intervention: you write a text prompt, select one to four reference frames, and the software generates the scene. The software works in windows of 90 to 345 frames at a resolution of 1280 by 736 pixels, keeping the original soundtrack.

What changes for businesses

Marketing agencies could use this tool to adapt international video campaigns to the local market. For example, by replacing protagonists or urban contexts without having to reshoot the commercials from scratch. An e-commerce store with a clothing catalog could take original runway shows and generate variants with models wearing items from different collections or with different seasonal backgrounds. Audiovisual productions and animation studios could also create animated storyboards or pre-visualizations of complex scenes starting from raw footage. This approach reduces initial production costs and speeds up the creative approval phases with clients, allowing them to test more visual variants for the same product.

Limitations and what to watch out for

Local installation requires very powerful hardware and is not within reach of all business computers. The minimum requirements indicated by sources vary based on the chosen configuration:

• no GPU for use via the official web application;
• 12 to 24 gigabytes of VRAM for manual configurations via ComfyUI;
• 32 gigabytes of VRAM and 60 gigabytes of RAM for the automated agent;
• 80 to 96 gigabytes for full configurations with Python scripts.

The model generates up to 345 frames at a time, so very long videos must be processed in sequential time windows. Sources do not specify the availability of European servers, actual processing times, or the exact costs of the web application.

How to try it

Those without dedicated hardware can try the tool directly from the official Viggle web application, which does not require local configurations or powerful GPUs. For those who want to run it locally, the code and model weights are downloadable from Hugging Face. Specific instructions are available for ComfyUI configurations or Python scripts, calibrated based on the available GPU power. Developers can use the sample.py file to generate single video windows while keeping the original soundtrack.