Frames and references do not mix
A request is either frame-led or reference-led. If you need both a fixed first frame and several references, decide which one defines the look and let the other go into the prompt text.
176Credits
Remaining credits：
Public
BestAIVideo
Shoot MiniMax H3 Clips from Text, Frames, or Your Own Audio
MiniMax H3 turns a written brief into a 4 to 15 second video and generates stereo sound with the picture. Here you can start from plain text, anchor a shot with a first and last frame, or hand the model reference images and audio clips, then check the credit cost before you submit.
What MiniMax H3 Gives You to Control
MiniMax H3 reads text, images and audio together, so the same generator covers a blank-page idea, a storyboard frame and a mood built from your own sound. These are the controls that matter in practice.
Three ways to start a shot
Pick text to video, frame generation with a required first frame and an optional last frame, or Omni Reference with one to five images. Each mode shows only the inputs it can use.
768P for tests, 2K for the keeper
Run quick compositions at 768P, then switch the same brief to 2K when fur, fabric or lettering needs to hold up on a large screen. 2K is the default selection.
Whole-second lengths from 4 to 15
Choose any integer duration between 4 and 15 seconds, so a product reveal, a transition or a short scene gets exactly the room it needs.
Sound that arrives with the picture
Dialogue, room tone, effects and music are generated alongside the video in stereo. There is no audio switch to hunt for, so the sound direction belongs in your prompt.
How to Brief a MiniMax H3 Shot
MiniMax says prompt structure matters more on H3 than on most video models, because a multimodal language model reads your brief. Treat it like a short shooting note rather than a list of tags.
Say what the shot is
Open with the subject and setting in one sentence, then the single action that should happen. A shot with one subject and one move is easier to judge than a crowded scene.
Give each reference one job
In Omni Reference, name what every image contributes: the character, the product, the location or the style. Audio files work the same way, whether they supply a voice, music or ambience.
Add camera notes after the description
MiniMax documents short bracketed camera cues such as [pan], [zoom] and [static] placed after the scene text. Keep to one camera move per clip.
Write the soundtrack into the brief
Place each dialogue line, effect or music cue next to the action it belongs to. Prompts can run to 7,000 characters, which is plenty for a timed sound plan.
Three MiniMax H3 Shots Worth Studying Before You Generate
A 2K macro portrait of a crested gecko
Held against a black background, the clip shows how 2K keeps individual scales, a translucent eyelid and a reflective iris sharp while the animal breathes, blinks and flicks its tongue. Its prompt names the lighting, the focus depth and the single natural motion.
A vertical move up a retro-futurist tower
The camera climbs past repeating windows and hanging lamps while green and magenta light slides across metal walls. The prompt writes sound as a path, from the elevator hum to cables rattling across the stereo field, which is the kind of direction H3 can follow.
A fifteen second neo-noir title sequence
A spinning record, a rain-soaked train and a red evidence folder are cut together with moving type and halftone texture. It is a useful template for a clip that depends on rhythm and match cuts rather than a single continuous action.
MiniMax H3 Production Notes
Plan duration around the action
A 4 second clip fits one gesture. Past 10 seconds you can afford a second beat or a cut, but write it in order so the model is not guessing what comes next.
Review sound as its own pass
Mute the clip once to judge motion, then listen with the picture hidden. Problems in timing and balance are easier to hear when you separate the two.
Where MiniMax H3 Earns Its Credits
Title sequences and mood films
Build a 15 second opening with type, cuts and a musical pulse for a pitch deck or a festival teaser, using one prompt for picture and sound.
Product pages that move
Animate a packshot from a first frame, or add a last frame so the product settles into the position your layout expects.
Vertical posters and key art
Use the 9:16 ratio to give character art a slow camera drift, a breathing loop or a gentle change of light.
Campaign variations
Feed the same reference images through several briefs to test colour, camera and music direction before committing to one.
Game and animation previs
Rough out a cutscene beat, an interface moment or a creature intro at 768P, then finish only the approved shots at 2K.
Explainers with narration cues
Pair a clear visual step with a spoken line or an effect so a short concept lesson arrives with its audio already in place.
MiniMax H3 Questions, Answered
MiniMax H3, also called Hailuo 3.0 or Hailuo 03, is MiniMax's multimodal video model. It reads text, images and audio as one context and returns video with stereo sound generated at the same time.
Text to video, frame generation and Omni Reference. Frame generation needs a first frame and accepts an optional last frame. Omni Reference needs at least one image and takes up to five, plus up to three optional audio files.
Use 768P while you are still deciding on composition and motion, since it is the cheaper way to iterate. Move to 2K for the final pass, especially when texture, small text or product detail matters.
Any whole number of seconds from 4 to 15. The default is 6 seconds, which suits a single action; use the longer lengths for sequences with cuts or a sound build-up.
Yes. Voices, ambience, effects and music are generated in stereo with the video, so there is nothing to upload unless you want to steer the sound with your own audio references.
Yes. Upload a first frame to start the shot, or add a last frame to make the video travel between two defined looks. The aspect ratio follows the frame you upload.
One to five images in JPG, PNG, WebP, HEIC or HEIF, and up to three WAV or MP3 files. An image is required, and the ratio can be set to adaptive so the output follows your references.
Text generation offers 21:9, 16:9, 4:3, 1:1, 3:4 and 9:16, with 16:9 as the default. Reference mode adds an adaptive option, and frame mode follows the uploaded image.
Describe the subject and setting, then one action, then the camera move, then the sound. Short bracketed cues like [static] or [pan] can follow the scene text, and every reference should have a stated purpose.
Use clean, well-lit images where the subject is not cut off, and keep the style consistent across them. Trim silence from audio so a voice, a music cue or an ambience bed is easy to identify.
Check identity and composition first, then whether the action finishes inside the chosen length. After that look at camera continuity, hands and fine detail, and finally dialogue timing and the stereo balance.
Start Your Next Shot with MiniMax H3
Choose text, frames or references, pick 768P or 2K, and see the credit cost before you submit.