Pick your model
One prompt box, five frontier video models behind it. Each has a different character — pick per clip, not per subscription. Models appear in the picker as they're available; you never configure a provider key yourself.
AI · Video
The clip you need, described in a sentence.
Type a prompt, pick a model, and a performance-ready mp4 lands in your library. Seed it with an image for image-to-video, flag it as a loop for seamless VJ playback, and keep performing while it renders — the job runs server-side and reports into the Tasks panel.
Runway Gen-4.5
High-fidelity cinematic generation with excellent prompt adherence. Up to 10 seconds.
Kling 3.0 Pro + O3 Pro
Two Kling tiers — strong physical realism and camera movement. Up to 10 seconds.
Seedance 2
ByteDance’s generator — stylized motion and dance-adjacent prompts land especially well. Up to 10 seconds.
Google Veo 3.1
Text plus audio generation — clips arrive with a generated soundtrack. Up to 8 seconds.
Prompt, length, loop, seed image
Set the clip length up to the model's max, toggle loop for a seam-free cycle built for clip slots, and optionally attach a seed image to run image-to-video — hand the model a still from your set and it animates in that visual world.
Generate while you perform
Submitting never blocks the deck. The job queues server-side, progress streams into the Tasks panel alongside every other AI generation, and the finished mp4 lands in your library as a regular video clip — loop, reverse, bounce, cue points and all.
Fair metering
Video is metered per second of output, per model, charged when you submit. If a job fails before it renders, every token comes back. Cancel while it's still queued and you get a full refund; cancel mid-render and half the tokens return.
Tokens: metered per second of video, per model, on submit. Failed jobs refund in full; queued cancels refund in full, mid-render cancels refund half.