Describe what you want. Your coding agent writes the design as HTML, Envie renders it frame by frame with a virtualized clock, and six verification gates inspect the result before you get a file. Free. No watermark. No account.
Works with Claude Code, Cursor, and any MCP client
No presets. No stock footage. Your AI writes each composition as HTML — any layout, any motion, landscape, portrait, or square. The same canvas it uses for interfaces, now filmed frame by frame. CSS animations, WebGL, Canvas 2D, animated GIFs, Web Workers: all frame-locked, all seeded for determinism.
No templates, no library, no presets. The AI writes the design from scratch. Nothing it makes looks like anyone else's content.
Every render is frame-locked with a virtualized clock. Same composition, same video, every time. No timing flakiness, no dropped frames.
Six machine gates inspect every render before it reaches you. A failing video is never handed over — your AI gets the report and fixes it.
Renders on your machine, in your IDE, with your AI. No upload, no cloud processing. What you render is yours — no attribution required.
AI generates with confidence even when the output is broken. Envie assumes nothing: every render passes six gates before it reaches you. Fail one, your AI gets the full report and tries again — the same loop you already use for code.
The file is real, complete, and playable. No zero-byte “successes”.
Video and audio both present. Silent movies don’t ship.
The audio carries real signal, not an empty track.
No dead black segments anywhere in the timeline.
No frozen stretches, individual or cumulative. Motion persists.
The ending is alive. Videos that die before they finish get rejected.
Envie runs locally. Nothing is uploaded. The requirements are the runtime dependencies it needs to do the work.
ENVIE_CHROME to your binary path.--narration "text" (Windows TTS) or --audio file.wav (all platforms). A video with no audio fails G2.One install. A blank canvas. Verified output, every time. Free.