Overview
MiniMax H3 is the July 31, 2026 video model release for Hailuo AI and MiniMax's video generation API. It moves the product line beyond Hailuo 2.3's standard text-to-video and image-to-video workflows by accepting a wider multimodal context: text, images, short video clips, and audio references can all guide the generated result.
The release is most relevant for creators and developer teams that need more control over subject consistency, motion, style, and editing rhythm without switching between separate video, image, and audio tools. H3 outputs 2K video clips up to 15 seconds and uses MiniMax's asynchronous video generation workflow for API integration.
What's New
Unified Multimodal Video Generation
H3 is described by MiniMax as an open, general-purpose multimodal video model. Instead of treating text-to-video, image-to-video, and reference-based creation as separate product families, the H3 API accepts a content array where text, image URLs, video URLs, and audio URLs can all be passed into the same generation request.
For creators, the practical change is that a prompt can now be grounded by richer evidence: a first frame, a last frame, reference images, short reference videos, or audio cues. This makes H3 better suited to brand clips, character-driven assets, product shots, and social videos where the source material carries more intent than a prompt alone.
2K Output and Up to 15-Second Clips
Hailuo 2.3 topped out at combinations such as 768p for 6-10 seconds or 1080p for 6 seconds, depending on workflow. H3 raises the listed output resolution to 2K and supports integer clip durations from 5 to 15 seconds.
The duration range matters for commercial and social production because many short-form clips need more than a six-second motion test but still fit inside a single generated asset. H3 also supports common aspect ratios, with adaptive framing available for workflows where the source image or reference material should drive the final frame shape.
First and Last Frame Control
H3 carries forward first- and last-frame control previously available in MiniMax-Hailuo-02, but integrates it into the unified v2 multimodal content API. A first frame can lock the opening composition, while a last frame can define the desired destination state for the motion path.
This gives teams a cleaner workflow for controlled transitions, product reveals, before-and-after shots, and storyboard-driven animation. It also addresses one of the practical limitations of earlier video models: getting a clip to start and end in a predictable way without repeated regeneration.
Reference-Based Creation Across Media Types
Reference generation is the largest workflow expansion in H3. The API can accept up to 9 reference images, up to 3 short reference videos, and up to 3 audio clips, with a mixed-input cap of 12 files in total.
MiniMax positions these references as ways to preserve character, motion, camera behavior, style, voice, or editing rhythm. Audio cannot be sent alone and must be paired with image or video input, which keeps the workflow anchored to visual creation rather than standalone sound generation.
Availability & Access
H3 is available through MiniMax's Video Generation API under the model name MiniMax-H3. The workflow is asynchronous: create a generation task, poll the task status with the returned task_id, then retrieve the generated video URL when the task succeeds.
Key launch limits:
| Area | MiniMax H3 limit |
|---|---|
| Output resolution | 2K |
| Output duration | 5-15 seconds, integer values only |
| Prompt length | Up to 7,000 characters |
| Reference images | Up to 9 images |
| Reference videos | Up to 3 clips; each 2-15 seconds; total video duration up to 15 seconds |
| Reference audio | Up to 3 clips; each 2-15 seconds; total audio duration up to 15 seconds |
| Mixed files | Up to 12 files total |
| File size | Video up to 50 MB, image up to 30 MB, audio up to 15 MB |
Supported formats include H.264/AVC and H.265/HEVC video, AAC or MP3 in-video audio, JPG/JPEG/PNG/WEBP/HEIC/HEIF images, and WAV or MP3 audio.
Pricing & Plans
MiniMax has published API documentation and pay-as-you-go pricing for H3, but the public Video Packages pricing page says H3 is not supported by the existing prepaid video point packages yet. H3 is excluded from those prepaid Video Packages; instead, MiniMax publishes separate pay-as-you-go pricing for H3.
MiniMax lists H3 pay-as-you-go output costs at $0.13 per second for 2K output and $0.09 per second for 768P output. Audio references are free; the first five input images are free, additional images cost $0.04 each, and reference video is billed by duration at the selected output resolution rate. MiniMax's current H3 guide documents 2K output, so confirm 768P availability in the API console before building workflows around that lower price.
The Hailuo web product remains the consumer entry point for MiniMax video generation, while developers integrating H3 should use the MiniMax Open Platform API, purchased Credits, or standard account balance billing rather than the older Hailuo video packages.
Best For
- Content teams producing social clips up to 15 seconds where a prompt, product image, and reference material need to drive one coherent output.
- Agencies creating brand videos that need subject, style, or motion consistency across batches of short generated clips.
- Product marketers building controlled reveals, before-and-after visuals, or storyboarded transitions from first and last frames.
- Developer teams integrating AI video generation into apps through a task-based API rather than a manual web workflow.
- Creators who need richer reference control than Hailuo 2.3, but do not need long-form video generation in a single model call.
FAQ
What is MiniMax H3?
MiniMax H3 is MiniMax's July 31, 2026 multimodal video generation model. It supports text-to-video, first/last-frame image-to-video, and reference-based generation using images, short videos, and audio clips.
How is H3 different from Hailuo 2.3?
Hailuo 2.3 focused on improved character movement, physics, stylization, and a Fast variant for lower-cost image-to-video. H3 adds 2K output, clips up to 15 seconds, and multimodal image, video, and audio references, while carrying earlier first/last-frame control into its unified v2 API.
What model name should developers use?
The API model name is MiniMax-H3. MiniMax's example requests submit jobs to the video generation endpoint and poll the returned task_id until the task succeeds or reaches a terminal failure state.
Does H3 support audio-to-video generation?
H3 can use audio as a reference, but audio cannot be submitted alone. MiniMax requires audio references to be accompanied by image or video input, so the workflow remains video generation with multimodal guidance.
How much does MiniMax H3 cost?
MiniMax lists H3 pay-as-you-go pricing at $0.13 per second for 2K output and $0.09 per second for 768P output. The Video Packages page separately says MiniMax H3 is not supported by prepaid video packages yet, so H3 uses public pay-as-you-go pricing or purchased Credits.
