Three distinct creation modes
Start MiniMax H3 with text, first or last frames, or multimodal references. Buble changes uploads with the mode to prevent incompatible fields.
Use MiniMax H3 on Buble to create native 2K video from text, first or last frames, or mixed image, video, and audio references. Choose 4-15 seconds and keep its three input workflows separate so every request matches the provider contract.
These five original Buble generations were created through the live MiniMax H3 integration at native 2K, four seconds, and 16:9. They test weather, character staging, transparent materials, high-speed tracking, architectural scale, and animal motion. Every result is stored on Buble R2 with its own poster.
Prompt
A lone traveler in a yellow raincoat waits on a rural train platform at blue hour. Wind moves the coat and wet grass as an old train enters through mist. Smooth lateral tracking shot, believable rain, warm carriage light, quiet cinematic tension, natural station ambience, no text.
Verified Capabilities
The current MiniMax H3 configuration follows the APIMart request contract. These are usable controls, not inferred features from a launch announcement.
Start MiniMax H3 with text, first or last frames, or multimodal references. Buble changes uploads with the mode to prevent incompatible fields.
Every MiniMax H3 request uses the documented 2K resolution, without invented lower-resolution options.
Choose any whole-second duration from 4 through 15 for a complete short shot.
Text mode offers six ratios. Frame mode inherits the image shape; reference mode supports adaptive or explicit framing.
Reference mode accepts 9 images, 3 videos, and 3 audio clips. Audio requires a reference image or video.
Keep the documented AIGC watermark off for a clean render or enable it when publishing policy requires disclosure.
Input Mode Guide
Text explores a new shot, frames control its endpoints, and references carry visual or audiovisual context into a new composition.
| Decision | Text to Video | Frame Guided | Multimodal Reference |
|---|---|---|---|
| Starting material | Prompt only | First frame, last frame, or both | Images, videos, and optional audio |
| Creative purpose | Invent a scene from a written brief | Control where motion begins or ends | Carry subject, style, movement, or sound context |
| Aspect ratio | Choose one of six concrete ratios | Inherited from the input frame | Adaptive by default or explicitly selected |
| Media capacity | No upload required | One first frame and one last frame | Up to 9 images, 3 videos, and 3 audio files |
| Important rule | Prompt remains required | Cannot be combined with reference fields | Audio must accompany an image or video |
Production Workflow
Treat each generation as one directed shot and make the input decision before writing details.
Step 01
Use text for exploration, frames for planned endpoints, and references when MiniMax H3 should carry recognizable context.
Step 02
Crop images, trim reference clips to useful moments, and remove conflicting material. More files do not guarantee fidelity.
Step 03
Describe subject, one action, environment, camera, timing, light, sound, and what must remain stable.
Step 04
Check picture and sound, then save successful MiniMax H3 output instead of relying on a temporary provider URL.
Practical Use Cases
MiniMax H3 fits high-resolution short shots with a clear reason for using text, frames, or references.
Direct a reveal, orbit, macro detail, or environmental shot around clean product references.
Use consistent views to keep a character, wardrobe, or creature recognizable through motion.
Give MiniMax H3 planned opening and ending frames for reveals, transformations, or precise landing compositions.
Pair visual references with short audio when timing, ambience, or performance should influence the clip.
Prototype landscape, square, portrait, and ultrawide compositions separately instead of relying on a late crop.
Use MiniMax H3 to test camera, blocking, atmosphere, and pacing before expensive production.
Prompt Method
The adjacent clip is a rejected MiniMax H3 iteration: a wristwatch became a ring-like object. It shows why prompts must be judged against the render, not the intention.
Open with the visible subject and one action. Concrete verbs give MiniMax H3 a clearer timeline.
Define space, light, shot size, camera path, pacing, and the resolution point. One movement often reads better than several cuts.
State which image defines identity, which clip supplies motion, and what audio contributes. Never leave reference roles ambiguous.
List details that must not drift, compare them with the result, and reject an attractive take when it misses the brief.
Limits and Review
A valid request still needs creative, legal, and technical review before publication.
The request body is limited to 64 MB. Reference video and reference audio each have a 15-second aggregate limit.
Do not combine frame controls with multimodal reference fields; the provider rejects mixed MiniMax H3 modes.
Logos, measurements, hands, crowds, typography, and fast transformations can drift. Simplify or split precision-critical shots.
Confirm source rights, consent, trademarks, safety, and required synthetic-media disclosure before publishing.
FAQ
Practical answers based on the current Buble configuration and documented APIMart limits.
Create on Buble
Choose the right input mode, prepare useful references, write one coherent shot, and generate MiniMax H3 video on Buble.
Compare: Hailuo 2.3 · Kling 3.0 Turbo · PixVerse V6