Always there. Draft quality. Included.
- 64 threads and 128 GB, capped so the live product never starves
- Wan 2.2, HunyuanVideo 1.5 and Kandinsky 6.0 in draft profiles
- MiniMax H3 and LTX-2.5 say “GPU required” instead of pretending
Live Five open video models · CPU, GPU or API
Wan 2.2, HunyuanVideo 1.5, LTX-2.5, MiniMax H3 and Kandinsky 6.0 on AlGraphy's own server, a cloud H100 or your GPU — plus the pay-per-use leaders — in one composer that shows the device, time and cost before you render.
Measured times · progress shown as a time-lapse
Self-hosted: Wan 2.2, HunyuanVideo 1.5, LTX-2.5, MiniMax H3, Kandinsky 6.0. Pay per use: Veo 3.1, Kling 4.0, Seedance 2.5, Wan 3.0, Gemini Omni Flash, Nano Banana Pro, FLUX.3.
Powered by MiniMax H3 where H3 is used · community-licence terms for LTX-2.5 and HunyuanVideo 1.5 on the licences page
Live from the studio
Real numbers from the API every 15 seconds. When it can't reach the server it shows nothing — never a stale figure.
How it works
A sentence, a still, a storyboard or a PDF. Templates and your brand presets fill in the rest; the composer suggests the model that fits the prompt.
Every place the job can run — our server CPU, a cloud H100, your own GPU, an API — with its measured time and cost, before you press Render.
Auto picks the fastest included option. Or choose yourself: CPU drafts cost nothing, GPU time is metered to the cent, API prices carry their source.
Every render becomes a conversation: revise it, extend it seamlessly, trim, reframe to 9:16, join the takes, export WebM, GIF or MP3.
The models
Ready, downloading or GPU-only — each card says what it will do on the hardware online right now, and which licence applies.
Benchmarks
Wall-clock time per clip, rendered by us on a cloud H100 in October 2026. The estimate you see before rendering comes from the same measurements.
Extend a clip
+2 seconds, seamless, on the H100.
Server CPU, honestly
Wan 2.2 draft, 49 frames at 640×384 — slow, draft quality, and labelled as such.
Three places a job can run