Media Lab

Media Lab is a self-hosted studio for generative media. One web app puts today's best open models behind a single prompt box: photoreal and typographic images, edits with up to ten references, cinematic video with synchronized audio, full songs, instrumental music and sound effects, and text-to-speech. Everything runs on your own GPUs through ComfyUI workers; nothing is sent to a cloud generator. It is built for a small circle — you and your friends — with private accounts, two-factor login and per-account quotas.

Creators and tinkerers with their own GPUs who want a private, self-hosted AI media studio without learning ComfyUI.

Running open image, video and audio models yourself usually means wiring ComfyUI node graphs, hunting for the right weights and settings for each model, and juggling separate tools for pictures, clips, songs and voices. Hosted AI studios avoid that, but your prompts and files leave your machine and you pay per generation. Media Lab puts it all behind one prompt box on your own GPUs: images and edits, video with sound, songs, sound effects, voices, character swaps, upscaling, and a Director that turns a song into a full music video. It picks the model and settings for you, runs one worker per GPU so jobs never split across cards, and keeps every file on your server. A guided installer detects your GPUs and downloads verified models. You can also drive it from your own AI agent over MCP.