How it works
Start from a still — either a photo you upload or a product image generated in the production studio, where your product stays pixel-locked. Pick an engine, describe the motion, and generate. The same pipeline covers the finishing work: text-to-speech voice-overs, generated sound effects, and upscaling of images and video at ×4 or ×8. One approved clip or render becomes a full multi-format delivery pack in one action.
Bring your own key — what that actually means
Keou does not run video models itself, and does not resell compute. It is the orchestration layer: your provider key lives in your browser's localStorage, rides each request in an X-Provider-Key header, is used for that single provider call, and is never written to the server's disk, database or logs. You can grep the source for that header and follow the whole path.
Honest limits
- Cost is per clip, at your provider's rates. Video is more expensive than images. The platform adds nothing, but the provider bill is real — check your provider's pricing before a large batch.
- The hosted no-account studio watermarks every video and publishes creations to a public community gallery, with a shared queue: 3 generations running at once, 60 waiting, 20 per network address. A free account keeps work private; a self-hosted install has no watermark, no gallery and no shared queue.
- Engine quality varies by engine and prompt. These are third-party models; Keou routes to them, it does not improve their physics. Generate, review, re-run — that is the workflow.
Open source, end to end
The full studio — video, images, polish, remix, adapt, packs, voice, sound, upscaling, teams, share links — ships in the public repository under MIT. There is no paid tier holding features back. The documentation covers the API (POST /api/video generates a clip from a still) and the self-hosting guide covers Docker and Railway.