Short ads
A 10-second product spot in one sitting: five angles of the same shot in under a minute of rendering, keep the one where the logo holds still, export with sound.
Write one sentence or drop a still. MiniMax H3 Max returns a 5–15 second clip with sound before you can refill your coffee — and it follows the brief better than any model on the boards right now.
Sample clip made with MiniMax H3 Max.
A 5-second shot comes back in under 3 seconds. Up to 15 seconds, with synced sound, from text or a still.
MiniMax H3 Max is the speed-tuned edition of MiniMax H3, hosted here so you generate in your browser. A 5-second shot comes back in under 3 seconds — Design Arena measured MiniMax H3 quality at more than 50× the speed (Aug 2026). Type a shot or drop a photo, pick 5 to 15 seconds, and keep the take you like. Then do it again, because the second try costs less time than reading this paragraph.
What one run can do
Four MiniMax H3 Max sample clips, four jobs: a picky brief, a text prompt, a still that starts moving, and a long take with sound.
A 5-second 768p clip in under 3 seconds (August 2026 benchmarks). Write the beat, generate, look, fix one word, generate again — ten drafts in the minute one used to cost.
H3 Max ranks #1 for prompt understanding in Design Arena and Artificial Analysis head-to-heads (Aug 2026). This clip keeps a whole brief at once: a named person, a top-down bed, readable type, and a change under the blanket. Name who moves, where the camera sits, and what must stay put — it listens.
Type one sentence and MiniMax H3 Max returns a 5–15 second clip at 480p, 768p or 1080P with natively synced audio. Pick a frame — 21:9 for a wide shot, 9:16 for a phone — and the model handles motion, light, and sound in one pass.
Upload a photo as the first frame, add an optional last frame, and describe the move. Product shots and character stills come out steadiest: label, face, and layout hold; only the motion is new.
No install. Write, pick, generate, keep.
Say who is in the frame, what they do, and how the camera moves. Plain words work best for MiniMax H3 Max: “a barista pours latte art, slow push-in, morning light.”
Choose 5 to 15 seconds and 480p, 768p or 1080P. Short 480p drafts are the fastest way to find the right motion; switch to 768p or 1080P for the take you keep.
Press Generate. A 5-second shot returns in under 3 seconds, so you can judge it, change one word, and run it again without losing your train of thought.
Save the MiniMax H3 Max clip to your library, replay it with sound, or rerun the same prompt with a tighter camera note. Every take stays.
Every MiniMax H3 Max figure below has a source and a date.
August 2026 benchmarks: a 5-second 768p clip in under 3 seconds — Design Arena puts it at more than 50× the speed of MiniMax H3 at equal quality.
Design Arena, Aug 2026: Elo 1,341, first place on the image-to-video leaderboard, with and without audio.
Whole-second lengths from 5 to 15. 768p renders at 1344×768, 24 FPS, with audio synced in the same pass.
21:9, 16:9, 4:3, 1:1, 3:4, 9:16. Image to video keeps the shape of the still you upload.
H3 Max vs H3
Same family. H3 Max is the speed-tuned, hosted edition; MiniMax H3 is the open base with a 2K tier on MiniMax's own service.
| Base model MiniMax H3 Open base · MiniMax | ||
|---|---|---|
| What you get | The speed-tuned edition, running in this browser workspace | MiniMax's base video model; H3-Base weights are open |
| Speed | 5-second clip in under 3 seconds; more than 50× the speed of MiniMax H3 at equal quality (Design Arena, Aug 2026) | Official endpoint pace; Design Arena puts H3 Max at 50× its speed at equal quality |
| Quality ranking | #1 overall, prompt understanding, aesthetics (Design Arena + Artificial Analysis, Aug 2026) | Elo 1,333 on Design Arena image-to-video, just behind H3 Max |
| Prompt follow | Post-trained so MiniMax H3 Max keeps a picky brief: who moves, where the camera sits, what stays still | Base-model adherence; no speed post-training |
| How you start | Text, or a still as first frame with an optional last frame — right here | Text, first/last frame, and reference inputs on MiniMax's platform |
| Clip length | 5–15 seconds | Up to 15 seconds |
| Resolution | 480p / 768p / 1080P with synced audio (768p: 1344×768, 24 FPS) | 768p / 2K on MiniMax's hosted service |
| Where to run it | This site — sign up and try it free | MiniMax platform, or your own GPUs with the open weights |
Who it is for
If you know the shot and hate the wait: under 3 seconds per 5-second draft, sound included.
A 10-second product spot in one sitting: five angles of the same shot in under a minute of rendering, keep the one where the logo holds still, export with sound.
Turn a flat catalog photo into a slow orbit or a gentle push-in for the product page. The packaging text stays readable through the whole move.
Turn each storyboard frame into a clip — one action per generate. The vase-to-ship sample above is exactly that kind of beat.
Post every day without the wait killing the habit. MiniMax H3 Max lets you throw away weak takes, keep the good one, add your caption, and ship it while the idea is still warm.
Show a client three moods of the same scene in the time one render used to take: same prompt, three frames, three lighting notes. Side by side, the choice makes itself.
Drop a character still as the first frame and describe the move. MiniMax H3 Max keeps the face and outfit consistent while the scene changes around them — a walk, a turn, a reaction shot.
A 5-second 768p clip renders in under 3 seconds — roughly 35× the throughput of the official MiniMax H3 endpoint, and about 15× faster than models of comparable quality (August 2026 benchmarks; Design Arena measured more than 50× the speed of MiniMax H3). In practice that means you judge a take, fix one word, and have the next one before you finish the thought.
Short clips with sound, 5 to 15 seconds long: product spots, ads, story beats, social posts, concept tests, and character shots. Start MiniMax H3 Max from one sentence or from a still image; pick the frame shape that matches where the clip will play.
Yes. Text to video takes a prompt and a frame size. Image to video takes a first frame (required), an optional last frame, and a prompt for the motion; the MiniMax H3 Max output keeps the shape of your still, so a product photo stays a product photo — just moving.
480p, 768p or 1080P, with 480p selected by default and 768p rendering at 1344×768 at 24 FPS. Lengths run from 5 to 15 whole seconds, default 5. Output costs 7 credits per second at 480p, 15 at 768p and 30 at 1080P. Audio is generated in the same pass and synced to the picture, so a door closing sounds like a door closing on the frame it closes.
In August 2026 head-to-head evaluations by Design Arena and Artificial Analysis, MiniMax H3 Max ranked first for overall quality, prompt understanding, and aesthetics. On Design Arena's image-to-video board it holds Elo 1,341, first place, with and without audio.
Yes. Sign up, spend your starter credits on a short 480p MiniMax H3 Max draft, and watch the result with sound. Paid plans are for daily volume; see the pricing page for the tiers.
No. Everything runs in this browser workspace: write, generate, replay, download. Your MiniMax H3 Max clips stay in your library, ready to rerun with a new note.
H3 Max is a post-trained edition of MiniMax H3 tuned for speed and tighter prompt following. Open MiniMax H3 weights exist under the MiniMax H3 Community License; H3 Max is the hosted, faster edition you generate with here.
Write a shot or drop a still. 5 to 15 seconds of video with sound, back in under 3 seconds. Free to start.