MINIMAX H3 / MULTIMODAL VIDEO GENERATION
Upload references.
Describe the scene.
Generate.
Seconds of video.
Image + sound + motion.
Connect.
Your key stays in memory only.
Warm the GPU.
Reserve a worker so generation starts in seconds instead of minutes. One worker costs ~$0.04/hr.
Generation works without a warm GPU — the first job auto-warms a worker for 10 minutes. Reserve time above to keep a worker standing by for instant starts.
Build the context.
Images
0 / 9Character, product, style.
PNG / JPG / WEBP · UP TO 9Audio
0 / 3Voice, ambience, rhythm.
MP3 / WAV / M4A · UP TO 3Video
0 / 3Movement, framing, pacing.
MP4 / WEBM · UP TO 312 combined max. At least one image or video required. Clips: 2–15 s each · ≤15 s total per category · 25 MiB per file.
Select references above. Nothing uploads until you click Upload selected.
Describe the scene.
Uploads references, creates the job, dispatches to GPU, and polls until done.
Track progress.
Waiting
Connect and upload references to get started.
The details.
Ready when you are.
Connect to view API responses.