MiniMax H3 AI Video Generator: Write a Scene, Get a 2K Clip
From a written prompt to a finished shot in one step.
MiniMax H3 is a video model made by MiniMax. On Banana AI it runs in text to video mode: describe the subject, the action, the camera and the light, choose a length from 4 to 15 seconds, and pick 768P or 2K. The credit total appears before you generate, and a failed run gives the credits back.
- Text to video from a plain-language prompt
- Any length from 4 to 15 seconds
- Six ratios, from 21:9 to 9:16
- 768P drafts or 2K finals, from 8 credits a second
Sample Videos
More Generators That Share Your Credit Balance
Animate your own pictures with Kling 4.0, design stills in an image model, or compare engines on the same idea. Every tool below runs on one account and one balance.
MiniMax H3: Text to Video with a 2K Finish
MiniMax describes H3 as a general-purpose model that can read written prompts, pictures, footage and sound recordings together and produce picture and sound in one pass. This page uses its text to video mode, which keeps the workflow simple: one prompt, one clip. Draft ads, product moments, animated posters and short story beats at 768P, then rerun the version you like at 2K for extra detail.
How to Make a Video with MiniMax H3
Three steps, all in the browser. On this page the prompt carries the whole brief, so the effort goes into describing the shot clearly.
Describe the Shot in Order
Start with the subject, then the place, the action, the camera move and the lighting. Say how the scene should change from the first second to the last, for example a slow push toward a sneaker as droplets lift off its sole. If sound matters, write it down too: footsteps, rain, one spoken line. MiniMax says H3 is built to create sound together with the picture, so check what you get before you publish.
Set Ratio, Length and Resolution
Choose 21:9, 16:9, 4:3, 1:1, 3:4 or 9:16 to match where the clip will play, and pick a length between 4 and 15 seconds. 768P costs 8 credits per second and 2K costs 13, so a 10-second draft is 80 credits at 768P and 130 at 2K. The form opens at 15 seconds, 2K and 21:9 (195 credits); lower these for a cheaper first test.
Review, Then Rerun at 2K
Watch the draft at full speed and check the motion, the framing and any lettering or logos. Change one phrase at a time so you learn what each edit does. Once timing and composition work, render the keeper at 2K. Finished clips are saved in your library, and failed runs return their credits automatically.
Four Things to Know About MiniMax H3
What MiniMax says about the model, and how each point plays out when you work from a text prompt on Banana AI. The sample clips show the kind of shot these prompts aim for.
One Prompt Carries the Whole Brief
MiniMax built H3 to accept several kinds of input, from written prompts and pictures to footage and voice recordings. On Banana AI the form uses the text to video mode, so your words do all the directing. Treat the prompt like a shot list: subject first, then setting, action, camera and light, each in a short plain sentence. A pencil castle coming alive on textured paper, or ink lines spreading across a page, are good examples of ideas that work from words alone. Upload-based modes are not offered on this page.
Sound Planned Alongside the Picture
According to MiniMax, H3 creates 32 kHz stereo audio at the same time as the visuals instead of laying a track on top afterwards, and its dialogue support is described as stable in 11 languages. In practice, the sound you want belongs in the prompt: liquid splashing against glass, a low hum, one short spoken line. Keep any dialogue brief and put the exact words in quotes. Before a clip goes public, have a fluent listener check timing, pronunciation and everything said aloud.
768P Drafts, 2K Finals
MiniMax explains that H3 first makes a base clip whose shorter edge is 768 pixels and then rebuilds fine textures, edges and lettering for its 2K version, rather than just enlarging the frame. On Banana AI those are your two resolution choices: 768P at 8 credits per second for testing composition and timing, and 2K at 13 credits per second for the file you deliver. A puppy nosing through wet grass shows why this matters, since fur and dew are where low resolution falls apart first.
Framing That Fits Every Screen
Six aspect ratios cover most placements: 21:9 for cinematic banners, 16:9 for players and websites, 4:3 and 1:1 for feeds, 3:4 and 9:16 for vertical stories. MiniMax lists 24 frames per second for H3 and says it can plan several shots inside one clip. Write camera moves in plain terms, such as a slow aerial glide over a lake or a low tracking shot beside a runner, and give each move a clear start and finish within the 4 to 15 seconds you choose.
Where MiniMax H3 Fits Your Work
Four kinds of project where a prompt-only clip moves the conversation forward faster than a written brief.
Ads and Launch Teasers
Test three openings for a campaign before anyone books a shoot. Write the product, the mood and the one action that should catch the eye, render each at 768P, and let the team choose from footage rather than descriptions. Rerun the winner at 2K in the ratio each channel needs, and review logos, claims and on-screen text by hand before publishing.
Product and Store Visuals
Show a sneaker splashing through a puddle, a serum dropping onto glass, or a watch catching the light as it turns. Describe the shape, color and material precisely, because this page works from text, not uploaded photos. Treat results as concept footage, and confirm that colors, proportions and printed details match the real item before you use them.
Film, Music and Story Concepts
Sketch the opening of a short film, a visualizer for a track, or the key beat of a pitch. Sound and picture can be described in the same prompt, which makes a concept easier to judge than a storyboard with no sound. Use 21:9 for a cinematic feel, keep each clip to a single beat, and assemble the sequence in your editor.
Games, Creatures and Motion Tests
Put a character, a creature or a whole world into motion for a trailer test or an internal prototype. Describe the look in detail and the movement in sequence, then compare a few versions side by side. Art, narrative and marketing can all react to the same clip, which shortens the feedback loop before real production begins.
MiniMax H3 on Banana AI: Common Questions
Modes, lengths, prices and what to check before you share a clip. Each answer describes how the model works on Banana AI today.
A video model built by MiniMax. The company describes it as multimodal, able to take written prompts, pictures, footage and sound recordings and to produce picture and stereo sound together. On Banana AI you can use its text to video mode in the browser with the same credits as every other model. Banana AI is independent and not affiliated with MiniMax.
Not on this page. MiniMax H3 on Banana AI is text to video only, so image, first-and-last-frame and reference modes are not available here. To animate a picture you already have, use Kling 4.0, which accepts your images as keyframes.
Any whole number of seconds from 4 to 15. Short clips are cheaper to test, while longer ones leave room for a camera move or a change of scene within the same generation.
Two: 768P and 2K. MiniMax says its 2K stage rebuilds detail from the original request instead of simply upscaling, so fine textures and small lettering tend to hold up better. Use 768P to check composition and timing, then 2K for the final file.
Six: 21:9, 16:9, 4:3, 1:1, 3:4 and 9:16. That covers cinematic banners, video players and websites, square and near-square feeds, and vertical stories.
It costs 8 credits per second at 768P and 13 per second at 2K. A 4-second clip at 768P is 32 credits, a 10-second clip is 80 at 768P or 130 at 2K, and the longest option, 15 seconds at 2K, is 195. The exact total is shown before you generate, and credit packs start at $4.9 for 60 credits.
Every new account receives 10 credits at no charge, valid for 30 days. That is enough to explore the image models, but the cheapest MiniMax H3 clip costs 32 credits, so you will need a credit pack to render video.
MiniMax says H3 generates stereo sound together with the picture and describes its dialogue as stable in 11 languages, English, Spanish, French, German and Chinese among them. Describe the sound you want in the prompt, and check every result, especially spoken lines, before you share it.
Give the clip one main action and describe it in order: subject, setting, movement, camera, light, then sound. Concrete nouns and verbs beat lists of style words. When a result misses, change a single phrase and rerun at 768P, so you can tell which edit made the difference before paying for 2K.
The credits return to your balance automatically, so you can adjust the prompt and try again without paying twice.
Nothing to install: sign in, write a prompt and generate in the browser on a laptop or a phone. Each clip appears next to the form and is saved to the asset library in your account, where you can compare versions and download the one you want.
Paid plans download without a watermark and include commercial usage rights, so clips can go into ads, product pages and client work. You remain responsible for trademarks, likenesses, spoken claims and any text that appears on screen.
Pick MiniMax H3 when you are starting from words alone and want a 2K option for clips of 4 to 15 seconds. Pick Kling 4.0 when you already have images to use as keyframes or need up to 30 seconds at 720p or 1080p. Both use the same credit balance.
Start Your First MiniMax H3 Clip with One Sentence
Describe the shot you have in mind, draft it at 768P, and move to 2K once the framing and timing feel right. The cost of every MiniMax H3 run is shown before you press generate, so you can try a few directions before choosing the one to finish.