MiniMax H3 Highspeed | Reference to Video.

MiniMax H3 Highspeed | Reference to Video Overview

MiniMax H3 Highspeed | Reference to Video generates short video clips from a written prompt and reference media. Created by MiniMax within the H3 family, it accepts image, video or audio guidance rather than requiring a designated first frame. The Highspeed integration on each::labs exposes 768P output, keeping the available resolution setting straightforward.

Every request needs a prompt and at least one reference file. You can combine supported reference types, describe their intended role in the scene, and request a clip between 4 and 15 seconds with a concrete output aspect ratio.

On each::labs, this MiniMax video generation workflow gives creative teams a way to develop shot ideas around existing assets. The references guide generation; they do not turn the request into a conventional editing timeline or guarantee exact reproduction of the supplied material.

Capabilities

• Generate short video from a written prompt combined with at least one supported reference file.
• Accept up to nine image references to provide visual context for the requested shot.
• Accept up to three video references and up to three audio references within provider media limits.
• Combine supported reference types in one request while keeping each media list within its own cap.
• Request output durations from 4 to 15 seconds and choose 768P resolution.
• Select six concrete output aspect ratios, including landscape, square and portrait compositions.
• Return completed generated footage as a video URL for inspection, downloading and further editing.

Use Cases for MiniMax H3 Highspeed | Reference to Video

MiniMax H3 Highspeed | Reference to Video can support reference-led development across several creative roles. A creator can supply visual assets and request “A quiet morning scene with the subject beside a window,” then review how well the generated shot fits a short social edit.

A marketer can explore a campaign mood around existing images without treating the result as an exact reproduction of a product photograph.

A designer can reference a short video when developing a shot concept, describing the desired scene and camera movement rather than assembling a frame-accurate timeline.

A developer can offer separate image, video and audio reference fields in a creative application. Each scenario benefits from combining a written brief with supporting media, while retaining a review step before footage is selected for publication.

Tips and Tricks

Write a scene description that explains how the references relate to the desired result. For example: “Create a quiet interior scene guided by the supplied room images; a curtain moves gently while the camera slowly approaches the window.” This gives the visual references a clear role.

When supplying a video reference, try: “Create a new outdoor shot with a similarly gentle camera movement to the reference clip; show trees beside a lake at dawn.” Treat this as a direction to explore, not a promise of exact motion transfer.

Begin with the smallest useful reference set and change one file or prompt instruction at a time. Keep video and audio excerpts focused and within their separate duration limits. Match the output ratio to your intended layout before comparing takes, and inspect the result before adding further references.

Technical Specifications

The MiniMax H3 Highspeed | Reference to Video API exposes these reference-generation inputs:

• Prompt: required, nonempty text, with a maximum length of 7000 characters.
• Images: up to nine reference_image_urls; videos: up to three reference_video_urls; audio: up to three reference_audio_urls.
• Reference requirement: at least one file across the three media lists.
• Duration: integer output seconds from 4 to 15, default 5.
• Resolution: 768P; default 768P.
• Aspect ratio: 21:9, 16:9, 4:3, 1:1, 3:4 or 9:16; default 16:9.
• Output: video URL after successful asynchronous completion, with no exposed codec selector.

Things to Be Aware Of

Each reference video or audio clip must last 2–15 seconds. Videos together must not exceed 15 seconds; audio files have their own separate 15-second combined limit. These input totals do not change the requested output duration.

Image files may be up to 30 MB, videos up to 50 MB, and audio up to 15 MB each. Check file formats and dimensions before submitting.

An empty reference set is rejected even with a valid prompt. Keep file URLs reachable, and inspect generated details rather than assuming every reference instruction will be reproduced.

Key Considerations

Select references that support one coherent shot, then explain what the generated clip should show. Adding files without a clear purpose can make the brief harder to interpret. Keep the reference counts and duration limits separate from the requested output duration.

The catalog price is $0.06 per requested output second at 768P. A five-second clip costs $0.30; a ten-second clip costs $0.60.

Submit media you are permitted to provide, and check every URL before generation. Review the resulting scene for continuity and suitability; reference guidance is not a guarantee of an exact copy or a finished edit.

Limitations

This mode does not expose first-frame or last-frame anchoring fields. It is not a frame-accurate editor or a guarantee of exact subject identity, motion copying or audio reproduction. Output clips are limited to 4–15 seconds, and 2K is unavailable.

No seed, negative-prompt, output frame-rate or codec control is exposed. Reference inputs must satisfy both the count caps and the provider’s file, dimension and duration requirements.