Pro FeatureMiniMax H3 delivers native 2K video with synchronized stereo sound, up to 15 seconds—combine text, start/end frames, and tagged image, video, and audio references in one context.
12Ratio
Resolution
Duration
3Watch a short clip to get a feel for MiniMax H3 before you plan your own video with text, references, and sound.
Six MiniMax H3 prompts covering first and last frames, references, selective transforms, text to video, motion transfer, and instruction edits.
First & last frame: dusk rack-focus onto the pavilion entrance—different beat from the hero daytime search.
Keep the twin circular binocular mask fixed; rack focus onto a glowing outdoor brand pavilion entrance at dusk with warm string lights and a red flag accent; Wes Anderson 35mm grain; soft stereo wind and distant chatter.
Reference to video: lock one modern identity across a bright urban beat.
Use Image 2 as the locked character reference—face, camel trench coat, hair. She walks a bright sunlit Tokyo crosswalk; keep identity stable; natural camera; soft stereo traffic and footsteps.
Selective edit: keep the street real; voxelize only trees and cars.
Preserve buildings and pedestrians in Video 1 as photoreal live action. Transform only trees and cars into voxel-block objects using Image 1 as style; keep motion and shadows physically correct.
Text to video: bright laundromat + cute hand-drawn cartoon cat, slice-of-life mix.
Bright sunny laundromat with warm window light; a cute orange hand-drawn cartoon cat perched on a washer; casual phone-camera feel; soft stereo washers and street murmur—cozy whimsical mix.
Motion reference: same action path, new photoreal subjects.
Match the action in Video 1 from a locked-off wide camera. Replace the three people with three photoreal capybaras; preserve the movement path exactly; keep lighting and shadows realistic.
Video editing: swap plate background and rematch character light.
Remove the green screen in Video 1 and replace it with a fairy-tale background like Video 2. Match background motion to the characters and relight them to fit the new world.
What MiniMax H3 is built for: native 2K detail, 24fps cinematic motion, longer clips, character consistency, chat-style edits, and synced stereo audio.
MiniMax H3 renders at full 2K pixel density natively—fine edges on feathers, petals, skin, and product detail stay sharp for large screens, not a soft preview that needs a separate upscaler.
H3 targets a fixed 24fps film cadence with stronger temporal consistency—action like a motorcycle lean holds together across the take instead of warping mid-shot.
MiniMax H3 can generate up to 15 seconds in a single pass—enough for a multi-beat beauty or storyboard take without stitching several short clips together.
Lock face, styling, motion, and voice across generations with multimodal references—up to nine images, three videos, and three audio tracks in one context.
Edit with plain instructions—change wardrobe, retarget a background, or retime an action—while the rest of the shot stays the same.
Every generation can return synced stereo—speech, effects, and room tone timed to the cut—so trench ambience, footsteps, and dialogue land with the picture in one pass.
Choose H3 when you need native 2K, multimodal references, motion transfer, or chat-style edits with synced stereo—for brand films, product sites, and commercial shorts.
Label start-frame, image, video, and audio jobs before the action line so H3 can resolve relationships.
Say whether you are creating a new cinematic clip or restaging a product, background, or sound bed.
Call out what must not change—logo, hands, table, face—beside what should transform.
Write vocal, ambience, and music energy with the same care as camera moves.
Pick 5–15s and social or cinematic ratios before you spend a generation.
Route logo- and text-critical shots through the higher-detail 2K regenerate path when available.
See how MiniMax H3 moves beyond earlier short clips toward unified multimodal generation and editing—plus a higher-detail 2K refine pass.
| Features | Hailuo 02 | MiniMax H3 Base | MiniMax H3 2K |
|---|---|---|---|
| Core job | Short text/image to video clips. | Multimodal generation and precise AV editing in one system. | Same H3 brief, regenerated at higher detail. |
| Multimodal context | Prompt plus basic image refs. | Text + ≤9 images + ≤3 videos + ≤3 audio as one context. | Original context reused during 2K regenerate. |
| Reference transfer | Start-frame / style-led. | Character, motion, camera, and voice cues transfer into new scenes. | Same cues with sharper micro-detail recovery. |
| Editing | Mostly regenerate the whole clip. | Targeted people/object/background/dialogue edits. | Edit then refine detail without plain upscaling. |
| Native audio | Often host-dependent. | Native stereo with the video. | Same AV pair refined at 2K. |
| Resolution path | 720p–1080p-era positioning. | ~768p Base / host defaults. | In-context 2K regenerate for text and brand marks. |
Three steps: choose text or image, write your scene and add references if you need them, then generate, preview, and download.
Use text when you want a new scene. Use an image when you already have a starting frame—or add more photos, clips, and audio when the person, motion, or voice must stay consistent.
Say what happens, how the camera moves, and what you want to hear. If you upload references, say what each one should control—and what should stay the same.
Generate the clip, check the look and the sound, then download it—or edit only the part that looks off and try again.
Built for longer ad creatives, filmmaking previs, character-driven shorts, and commercial production—e-commerce, brand social, game trailers, and UI motion.
Ship 5–15s vertical stories with cinematic motion and native stereo without silent placeholders.
Restage products, protect logos, and rewrite backgrounds or VO while keeping the hero object stable.
Run multimodal boards for cast, camera, and voice when clients expect editable AV drafts.
Prototype title sequences, action previs, and instruction-led edits before a full shoot.
Turn still boards plus motion and sound refs into short proofs stakeholders can hear and see.
Explore stylized motion, UI demos, and world concepts with controllable multimodal cues.
Map frames and multimodal refs, write camera and stereo notes, then generate or instruction-edit the beat that matters.
