MiniMax H3 Video API

Overview

MiniMax H3 is a general-purpose multimodal video model that unifies generation, reference, editing, motion transfer, audio-video creation, and multi-shot work instead of treating them as separate tasks.

Its core idea is Contextual Omni Representation: text, images, video, and audio form one context, while natural language explains how each reference should affect the target video.

H3 supports V2V Motion Transfer, native stereo, multi-shot modeling, stronger text and brand rendering, and 2K generation through In-context Regeneration. On iCreat, output is 768P or 2K for 4–15 seconds.

minimax/h3-video

Key Features

Unified Multimodal Generation. H3 combines generation, reference, editing, First/Last Frame control, and motion transfer in one model, reducing the need to switch between separate task-specific video workflows.

Contextual Omni Representation. H3 understands how references relate to each other and the target video, so one source can guide motion while another defines character, sound, or appearance.

Generalized Reference and Editing. Natural-language instructions can describe what to preserve, change, or borrow from image, video, and audio references instead of relying only on a fixed list of editing operations.

V2V Motion Transfer. H3 can transfer performance, body movement, timing, or camera behavior from a reference video while generating a different subject, environment, product, character, or visual style.

Native Stereo Audio. H3 jointly models video and audio, including voice, sound effects, and music, so sound can be generated as part of the video rather than added afterward.

Native Multi-Shot Modeling. H3 models shot changes inside the video-generation process, helping visual direction, movement, characters, and audio remain connected when a sequence moves between multiple shots.

Text and Brand Rendering. Stronger instruction following and text rendering make H3 useful for ads, e-commerce, UI concepts, title sequences, product visuals, and branded scenes containing words or logos.

In-context Regeneration for 2K. H3 regenerates higher-resolution output with access to the original multimodal context, helping restore fine visual details, small text, and structure beyond conventional standalone upscaling.

Model Comparison

MiniMax H3 vs Hailuo 02 Pro vs Hailuo 02 I2V Pro

MiniMax H3 vs Seedance 2.0

Inputs

A Text Prompt is required. H3 can also use images, video, and audio as multimodal references.

Omni Reference supports up to 12 mixed reference files. In Text-to-Video mode, ratio is required; reference-based generation can use adaptive.

Parameters

Pricing

Total Cost = Unit Price × Output Video Duration

Quick Start

This example uses First Frame and Last Frame to control the start and end of one shot.

curl --fail-with-body --connect-timeout 10 --max-time 60 \
  -X POST https://api.icreat.ai/v1/task/submit/minimax/h3-video \
  -H "Authorization: Bearer ${ICREAT_API_KEY}" \
  -H "Content-Type: application/json" \
  -d '{
    "content": [
      {
        "type": "text",
        "text": "Keep the astronaut and suit from the first frame. Pull the camera back as warning lights pulse, then transition smoothly into the final wide shot of the approaching planet."
      },
      {
        "type": "image_url",
        "url": "https://cdn.example.com/astronaut-close-up.png",
        "role": "first_frame"
      },
      {
        "type": "image_url",
        "url": "https://cdn.example.com/station-wide-shot.png",
        "role": "last_frame"
      }
    ],
    "resolution": "2K",
    "duration": 8,
    "ratio": "adaptive",
    "callback_url": "https://example.com/webhooks/icreat-video",
    "aigc_watermark": false
  }'

A successful request returns:

{ "task_id": "task-xxx" }

Output

submit → task_id → query-status / callback → get-result

Submit the request, track the task with query-status or callback_url, then use the same task_id with get-result after the task reaches SUCCEEDED.

Use Cases

Film Opening Titles. H3 combines multi-shot modeling, text rendering, and native audio for title sequences where typography, shot changes, camera movement, pacing, and sound must follow one creative direction.

Product Websites and UI Motion. Product renders, interface visuals, text, and brand elements can guide hero videos or UI concepts where motion, layout, and branded details must preserve a designed identity.

Animated Posters. A finished poster can remain the visual anchor while H3 adds movement, camera behavior, typography changes, depth, and sound without rebuilding the original composition from scratch.

Advertising and E-commerce. Product references, brand assets, motion examples, written direction, and audio can be combined when a commercial video needs several visual, motion, text, and sound constraints to work together.

Motion-Transfer Creative. Use V2V Motion Transfer when the movement is already right but the character, product, environment, camera context, or art direction needs to change without rebuilding the performance.

Limitations

Fine visual detail can still vary in demanding scenes, especially when several small elements must remain precise.

Generalization across very different reference types is still an active improvement area for H3.

Complex reference and editing instructions work better when each source has one clear role.

In-context Regeneration improves 2K detail, but it does not guarantee perfect text or logo preservation.

FAQ

When should I choose MiniMax H3 instead of Seedance 2.0?

Choose H3 for one generalized model that combines mixed references, editing, motion transfer, native stereo, and 2K. Choose Seedance when broader Resolution or workflow-specific routes matter more.

Should I use 768P or 2K on iCreat?

Use 768P for cheaper iteration and motion tests. Use 2K when the concept is selected and the extra visual detail is worth the higher per-second price.

When should I use First/Last Frame instead of general references?

Use First/Last Frame when both the opening and ending composition matter. Use general references when appearance, motion, sound, or style should guide the result more broadly.

Do I need text, images, video, and audio in every H3 request?

No. A Text Prompt is required on iCreat, while image, video, and audio references are optional. Add only the references that provide useful control for the intended result.

Is H3 a good choice when I need audio in the final video?

Yes. H3 jointly models audio and video and supports native stereo, making it a strong option when voice, sound effects, music, and visuals should be generated together.