Self-hosted · five open models · CPU, your GPU or API
Wan 2.2, HunyuanVideo 1.5, LTX-2.5, MiniMax H3 and Kandinsky 6.0 behind one composer — plus the pay-per-use leaders when you want them. Every option shows its device, ETA and cost before you press Render.
Honest by design: no GPU online means amber “CPU” badges and real ETAs — never a fake “fast”.
The models
Ready, downloading or GPU-only — each card tells you what it will do on the hardware we have right now, and which licence applies.
How it works
Prompt, presets, a reference image if you want motion from a still. Pick any of the models.
Every place the job can run — server CPU, your GPU, an API — with its ETA and cost, before you press Render.
Auto picks the fastest included option; or choose yourself. CPU drafts are free; APIs are priced per second in USD and SAR.
Toasts, the queue badge and browser notifications. Scrub, download, duplicate with changes, re-run on a GPU.
Three places a job can run
The portal never presents a slow render as fast or a paid one as free. Each option shows its device, ETA and cost side by side; refusals say why and offer the alternative.
64 threads and 128 GB on our own box, capped so the live product never starves. Small variants, short clips, ETAs in the tens of minutes — measured, never guessed.
Run the Studio Agent on your own PC or Mac and your GPU appears as an option — behind your router, nothing to open. Rented GPU nodes register the same way.
Veo, Kling, Seedance, Wan 3.0, MiniMax H3 at 2K, Nano Banana, FLUX.3 — through an admin-held provider key. Every price carries a verification date and a source.
