MiniMax H3 Max is MiniMax and fal.ai's fast video model, and it renders 5 to 15 seconds at 768P or 480P with sound. Type your shot, or pin a first and last frame, and get the clip back in the same pass.
Sign in and write a line or two first — Enhance rewrites what you have.
You set the frames, the length and the shape, and the model fills in the motion and the sound. Every card below runs on a control you can reach in the composer above.
1344x768 · 7:4 · 10s
Pick your length and MiniMax H3 Max renders it faster than MiniMax H3, per MiniMax's own docs. You test ten ideas in the time one slow render takes.
Switch MiniMax H3 Max between 480P and 768P, and the credit estimate updates before you submit. You draft at 480P and spend more only on the keeper.
Upload a first frame, a last frame, or both, and MiniMax H3 Max fills in the motion between them. Reveals and transitions finish on the frame you chose.
Name the room tone, the foley and the line in your prompt, and the audio arrives composed with the picture. Your clip lands with sound that matches what is on screen. Ask for silence or no music and you get that too.
Set any whole number from 5 to 15 seconds, and the price moves with the seconds you keep. A short test costs least, and a long take tells the story.
Choose 21:9, 16:9, 4:3, 1:1, 3:4 or 9:16 before a text-to-video render starts. Native framing for each feed, with no crop job after. On image-to-video the shape follows the picture you upload.
You bring the product still, the daily short, or the transition you keep redoing. Start from a prompt or a frame, and post a clip that already has sound on it.
Run many prompt variants of the same shot quickly and cheaply, then spend real credits only on the version that works. You draft at 480P, change one line at a time, and watch the takes stack up in your history. Then you re-render the winner at 768P.
Animate an existing product photo into a vertical social cut that comes back with usable audio, without booking a shoot. You pin the product photo as the first frame, describe the move and the sound, and download a 9:16 MP4 ready for paid social.
Bridge two stills with motion that opens and ends exactly where the edit needs it, instead of hand-keying the move. You upload the opening frame and the closing frame, write the move between them, and drop the finished MP4 straight into your timeline.
Judge the model on real output before wiring anything, and see what a finished second actually costs. You sign in, write a prompt, and read the credit estimate beside Generate. Every take stays in your history as a downloadable MP4.
Sign up free for starter credits, then pay per second of MiniMax H3 Max video. A 480P second costs less than a 768P one, and the estimate updates before you submit.
You get the plain version here: what MiniMax H3 Max is, how fast it renders, what a clip costs, and when to pick MiniMax H3 instead.
MiniMax H3 Max is a video model released jointly by MiniMax and fal.ai on 2026-08-27. fal.ai post-trained it from MiniMax H3's open weights and tuned it for speed. It renders 5 to 15 seconds at 768P or 480P, with native audio in the same pass. On Seadanse you run it as text-to-video, or as MiniMax H3 Max image-to-video from a first and last frame.
Pick MiniMax H3 Max when you want many takes fast at 768P or 480P. Pick MiniMax H3 when the final cut needs sharper output, or when you want to feed the model more kinds of input. They are different jobs, not better and worse. Our MiniMax H3 page sits one click away in the same composer.
Not on its own. MiniMax H3 is the open-weight model, and MiniMax H3 Max is fal.ai's post-trained version of it. No separate weight release has been announced for the post-trained variant. Some platforms list the same model as Hailuo 3 Max.
It is built for speed. MiniMax says in its own docs that it generates faster than MiniMax H3, and fal.ai reports a five-second clip in under three seconds on its own stack. Your wait here also covers the queue and the file transfer, so treat those reports as a floor, not a promise.
You get 768P or 480P output at 24 FPS. Clips run 5 to 15 seconds, in whole seconds. Text-to-video offers 21:9, 16:9, 4:3, 1:1, 3:4 and 9:16. On image-to-video the aspect ratio follows the picture you upload.
Upload a first frame, a last frame, or both. One image on its own still works, and the model treats it as your opening frame. Add the closing frame too and you fix both ends, which is how reveals and transitions land where you planned.
Yes. Native audio comes back with the picture in the same pass, so room tone, foley, music and speech arrive already lined up. Name the sound in the same prompt as the shot, and ask for silence when you want none.
No. Sign in on this page, write a prompt, and render in the browser. There is nothing to install and no MiniMax H3 Max API key to manage. Your clip lands in your history and downloads as an MP4.
Sign-up is free, needs no credit card, and comes with starter credits. Those credits are a head start rather than a full render, because the shortest MiniMax H3 Max clip costs more than the grant. Top up with a credit pack or a plan, and the estimate shows before you submit.
You pay per second of video, and the rate depends on the resolution you pick. A 480P second costs less than a 768P one, and a short clip costs a fraction of a long one. The exact credit estimate sits beside Generate and updates on every setting change.
A render you pay for with a plan or a credit pack downloads clean. A render covered by free sign-up credits carries a visible watermark. Commercial use is included per plan terms, so paid clips can go into client work.
The credits come back automatically, so a failed job never costs you. The render also keeps going when you close the tab, and your history holds every finished clip for replay or download.
Every page here runs the real models in your browser — jump to the one that matches the clip you have in mind.
Sign up free, claim your starter credits, and send your first prompt to MiniMax H3 Max today. Your prompt and your settings survive the login round-trip.
