Powering the future of world models

Infrastructure for generative media and world models

Request early accessExplore the engines
Offline

Inference engine for offline video and image generation

High-throughput serving for batch video and image workloads, behind a simple job API.

Real-time

Inference engine for world models and live interactive video

Low-latency serving for world models and interactive video streams that respond as they generate.

MiniMax-H3 — Drama 1
MiniMax-H3 — Drama 2
MiniMax-H3 — Drama 3
MiniMax-H3 — Cinematic 1
MiniMax-H3 — Cinematic 2
MiniMax-H3 — Cinematic 3
MiniMax-H3 — Scenery 1
MiniMax-H3 — Scenery 2
MiniMax-H3 — Scenery 3

Open-source video models, self-hosted

We run open-weight generative models on our own GPU fleet — no black-box vendor API in the loop.

MiniMax-H3

Live

Open-weight video+audio generation, served via SGLang across our GPU fleet. Compare outputs in the Video Arena.

$0.06/sec at 768p · $0.08/sec at 2K

LTX-2.3 / 2.5

Coming soon

Lightricks' open-weight video model. LTX-2.3 is confirmed compatible with our serving stack and is next in line to onboard.

FLUX 3

Watching for open weights

Black Forest Labs' image/video/audio model. Currently API-only upstream — we'll self-host as soon as open weights ship.

Try the Video Arena →

How it works

01

Get an API token

Sign up and generate a scoped API token from your dashboard in seconds.

02

Send video generation jobs

Call the API with your prompt and model of choice to queue a video generation job.

03

Pull status and download

Poll the job endpoint for progress, then grab the finished video URL once it's ready.

Get early access

We're onboarding teams building with AI agents. Join the waitlist and we'll be in touch.