MiniMax H3 (Hailuo 03) Launch: A Guide to Generating 2K Audio-Visual Videos

MiniMax H3 (Hailuo 03) is now integrated with Genvizu. Currently, it supports text-to-video and single-frame-to-video generation, outputting 2K resolution videos. You can choose durations from 4 to 15 seconds and generate native stereo sound. Whether you need to quickly create a complete audio-visual piece from a description or bring a static image to life, H3 is ready to use.

Key Features of MiniMax H3

MiniMax defines H3 as a universal multimodal generation model. According to the official MiniMax announcement, the model can jointly understand text, image, video, and audio contexts to generate videos up to 2K resolution, 15 seconds in length, with native stereo sound. While the official documentation mentions H3's multi-shot modeling capabilities, this doesn't mean complex prompts will automatically result in stable shot compositions. For practical creation, explicit control over shot sequence, actions, and sound sources is still recommended.

Genvizu's initial release streamlines the workflow into two primary use cases:

• Text-to-Video: Create shots from text prompts, with options for 21:9, 16:9, 4:3, 1:1, 3:4, or 9:16;

• Single Frame to Video: Upload an image as the starting frame, and the model will generate dynamic content by continuing the subject, scene, and composition.

Currently, Reference-to-Video is not supported, nor can video or audio reference materials be used. Although the H3 model itself has broader multimodal capabilities, the parameters available on the Genvizu interface represent the currently usable range.

Text-to-Video vs. Image-to-Video

GoalRecommended ModeReasonTrying out ads, short films, or ambianceText-to-VideoComposition, aspect ratio, and actions can all be defined by text.Animating character photos, product images, or illustrationsSingle Frame to VideoThe initial frame helps lock the subject's appearance and initial composition.Quickly verifying prompt direction4–6 secondsLower trial cost, easier to check a single action.Expressing a narrative with a beginning, middle, and end8–15 secondsSufficient time to arrange shots and sound changes.

For image-to-video generation, the source image should ideally have a clear subject, complete composition, and minimal compression artifacts. Focus your prompt on describing "what happens next," without repeating visual details already evident in the image.

MiniMax H3 Credit Pricing

H3 consistently outputs 2K resolution and is billed based on the selected duration, at a rate of 46 credits/second:

DurationCredits Required4 seconds1846 seconds27610 seconds46015 seconds690

Both text-to-video and single frame-to-video use the same output pricing. For a complete price comparison, please refer to Genvizu Model Pricing. For your first test, starting with 4 seconds is recommended to confirm the subject, action, and camera direction before increasing the duration.

Crafting Effective MiniMax H3 Prompts

An effective H3 prompt can be structured in the following order:

[subject and scene] + [actions in time order] + [camera movement] + [lighting changes] + [sound sources] + [consistency constraints]

Text-to-Video Examples

A neon-lit intersection on a rainy night, a courier in a silver trench coat rides a bike through standing water.
For the first two seconds, a low-angle tracking shot follows the wheels splashing water, then the camera rises and follows from behind and to the side,
a tram passes in the distance, red and blue lights sweep across the wet road.
Sound: close rain, tires cutting through water, a distant tram bell, clear spatial positioning, no narration.
Keep the character's clothing, the bike's color, and the night scene direction consistent.

Single Frame to Video Examples

The character first keeps looking out the window, then slowly turns to look at the camera and smiles softly.
The camera pushes in slowly and steadily, curtains move in a light breeze, warm sunset light gradually sweeps across the side of the face.
Sound: soft wind, distant birdsong outside the window, faint rustle of clothing, no music.
Keep the face, hairstyle, clothing, and interior furnishings from the first frame unchanged, do not add new characters.

When writing audio-visual prompts, clearly specify "what the sound is, where it comes from, and when it occurs." For actions, arrange them chronologically. In short films, defining one main action and a primary camera movement usually yields more controllable results than stacking multiple turns, runs, costume changes, and shot transitions.

Generating Your First 2K Video with MiniMax H3

Start by selecting a 4-second duration with a clear action and specific sound to validate your direction. Once satisfied, extend it to a full shot: Open the MiniMax H3 model page to start generating