Short ads
Need a 10-second product motion by lunch? MiniMax H3 Max is built so a 15-second clip can come back in about 5 seconds. Write the packshot, generate, then keep the take that holds the logo still.
MiniMax H3 Max is fal's faster, prompt-following take on MiniMax H3. Write a sentence or upload a still, then generate a short clip in the browser.
Hero clip: MiniMax H3 image-to-video sample we host ourselves.
Open MiniMax H3 is the base. MiniMax H3 Max is the hosted version fal trained on top of it.
MiniMax H3 Max is a post-trained MiniMax H3 video model hosted by fal. It is not the open MiniMax H3 weights, and it is not a MiniMax official product page. The hook is speed plus following: fal's public demo made a 15-second Odysseus clip in about 5 seconds — faster than watching the result. You type a shot, or you upload a still, and MiniMax H3 Max turns that into a short video in this browser workspace. The hero is a MiniMax H3 image-to-video sample. The Odysseus speed clip sits in the section below.
| MiniMax H3 Max | MiniMax H3 Max video generator | H3 Max text to video | H3 Max image to video | 15 second video | generate in 5 seconds |
What one run can do
Speed is the MiniMax H3 Max story. The other clips are MiniMax H3 samples we already host — the family look Max is tuned from.
This Odysseus clip is MiniMax H3 Max: 15 seconds of video, generated in about 5 seconds on fal. That is the reason this page exists. You are not waiting out a long queue just to see if the shot works. Write the beat, generate, then run it again if the face or the camera is wrong.
Fal tuned MiniMax H3 Max for stronger prompt follow than base MiniMax H3. The clip here is a MiniMax H3 sample we already host: readable menu type, a named character, and a camera that stays on the beat you asked for. That is the kind of brief Max is trained to keep. If your shot names who moves, where the camera goes, and what must stay still, MiniMax H3 Max is the version to try.
H3 Max text to video starts from words only. Describe the subject, the light, and how the camera moves. The kitten clip is a MiniMax H3 playground text-to-video sample we host ourselves: a garden chase with gentle tracking. Use that look as a floor, then ask MiniMax H3 Max for the faster generate when you are iterating on the same sentence.
H3 Max image to video keeps the still you uploaded and adds motion. The horses clip is a MiniMax H3 playground image-to-video sample: two draft horses in flood water, almost no extra invention. Upload a product packshot, a character still, or a frame you already like, then tell MiniMax H3 Max how far the camera may move.
Four steps. No desktop app. MiniMax H3 Max runs in this browser workspace.
Say who is in the MiniMax H3 Max clip, what they do, and how the camera moves. Keep it in plain words. One clear sentence beats a pile of film jargon.
Choose a short MiniMax H3 Max run — typically 5 to 15 seconds — and a size such as 480p, 768p, or 2K when those options are on the model. Shorter drafts cost less time while you hunt for the right motion.
Press Generate. MiniMax H3 Max is the version built to return a 15-second clip in about 5 seconds on fal's demo path. Watch the Odysseus clip in the speed section if you want that speed in your head before you click.
Keep the MiniMax H3 Max file in your library, replay it, or run the same prompt again with a tighter camera note. Nothing here is a one-shot black box you cannot retry.
Adjectives do not help. These are the MiniMax H3 Max facts we will stand behind today.
Fal's MiniMax H3 Max demo produced a 15-second Odysseus clip in about 5 seconds. That is faster than playing the file. Use it when you need many drafts in one sitting.
MiniMax H3 Max sits on the MiniMax H3 length range: short shots, not a feature film. Plan one beat per generate, then stitch later if you need a longer story.
Hosts showing MiniMax H3 Max list 480p for cheap drafts, then 768p and 2K when you pay for a sharper file. Start small, then spend on the take you keep.
MiniMax H3 Max is fal's post-trained MiniMax H3. The Max weights are not public. Open MiniMax H3 is a different page and a different model. Do not expect to download Max and run it on your laptop.
H3 Max vs H3
MiniMax H3 Max is the faster, fal-tuned hosted version. MiniMax H3 is the open-weights base. Same family, different job.
| Base model MiniMax H3 Open weights · MiniMax base | ||
|---|---|---|
| What it is | Fal's post-trained MiniMax H3. Hosted, not a new MiniMax open drop. | MiniMax's open-weights video model. The base Max is trained on. |
| Who trained it | Fal, on top of MiniMax H3 | MiniMax |
| Weights | Not public. You run MiniMax H3 Max on a host. | Public. You can download MiniMax H3 and run it yourself. |
| Speed (published demo) | 15-second clip in about 5 seconds on fal's MiniMax H3 Max demo — faster than watching it. | Slower on the same hosts. A 5-second 480p MiniMax H3 clip on fal was often under 10 seconds, not faster than playback. |
| Prompt follow | Tuned to keep picky briefs: who moves, where the camera goes, what must stay still. | Strong base follow. Max exists because fal wanted this tighter. |
| How you start a clip | Write a MiniMax H3 Max prompt, or upload a still, in this browser. | Text to video or image to video on the MiniMax H3 model card. |
| Clip length | Short shots, typically 5 to 15 seconds — one beat per generate. | Same 5 to 15 second family range. |
| On this website | This page. Generate MiniMax H3 Max in /video. | Not this page. Open MiniMax H3 is a different site. |
Who it is for
If you already know the shot and you are tired of waiting, MiniMax H3 Max is the version to open.
Need a 10-second product motion by lunch? MiniMax H3 Max is built so a 15-second clip can come back in about 5 seconds. Write the packshot, generate, then keep the take that holds the logo still.
Upload a pack photo and ask MiniMax H3 Max for a slow orbit or a gentle push-in. Image to video keeps the label; you only add the move. Retry until the reflection and the turn feel honest.
Directors who storyboard in stills can turn each board into a MiniMax H3 Max clip. Keep one action per generate. The Odysseus speed clip on this page is that kind of beat: one look, one camera push, done.
If you post every day, waiting minutes per take kills the habit. MiniMax H3 Max is the faster MiniMax H3 so you can throw away bad takes. The anime rooftop sample shows how much motion one short prompt can hold.
Pitch a world before you hire a crew. Write the MiniMax H3 Max prompt with the one thing that must be true — a kitten that actually chases, a menu that stays readable, a hero who turns to the island — then generate until that thing holds.
You already have a face or a costume photo. MiniMax H3 Max image to video should keep that person and only add motion. Start with a tiny head turn or a walk, not a full stunt, so the identity does not drift.
MiniMax H3 Max is fal's post-trained version of MiniMax H3. It is a hosted video model, not a new MiniMax open-weights drop. The public hook is speed: fal showed a 15-second Odysseus clip generated in about 5 seconds. This site is an independent browser workspace for MiniMax H3 Max. It is not MiniMax and it is not fal.
MiniMax H3 is the open-weights base. MiniMax H3 Max is the fal-tuned hosted variant: stronger prompt follow and a faster generate path. Max weights are not public. If you want the open model, that is a different site. If you want the 5-second 15-second demo, you want MiniMax H3 Max.
On fal's demo, MiniMax H3 Max made 15 seconds of video in about 5 seconds. That is the Odysseus clip in the speed section. Real queue time on a busy day can be slower. Treat 5 seconds as the published demo, not a legal SLA, and still much faster than waiting out a long MiniMax H3 run.
Yes. H3 Max text to video starts from a prompt. H3 Max image to video starts from a still plus a motion note. Both live in /video on this site. Family-look clips on this page are MiniMax H3 samples we already host, plus a few playground files copied onto our own storage — not hot-linked.
Hosts that already show MiniMax H3 Max list 480p, 768p, and 2K. Use 480p while you hunt for motion, then spend on 768p or 2K for the take you keep. Exact buttons on this workspace follow the live model card once the fal MiniMax H3 Max endpoint is wired.
No. MiniMax H3 Max is fal-hosted and the Max weights are not released. MiniMax H3 (the base) is the open-weights model. Do not mix the two when you search for a download.
Fal trained MiniMax H3 Max on top of MiniMax H3. MiniMax trained the base. This website did not train either model. We only run a browser UI and store our own copies of demo files.
No fal API key. Sign in with Google on this site, then generate in /video. We do not ask you to paste a fal key. Payment is off for now; when checkout returns, it will be a later pass, not a requirement to open the page.
Write a shot or upload a still. MiniMax H3 Max is the faster MiniMax H3 — 15 seconds of video, about 5 seconds to make on the published demo.