01
Omni-modal context
H3 can interpret combinations of text, images, video, and audio rather than limiting the workflow to one prompt and one starting frame.
AI video model guide
MiniMax H3
A general-purpose multimodal video system from MiniMax with text, image, video, and audio context, native stereo sound, and separate base and 2K regeneration workflows.
Published model guide
Model input mapMiniMax
Text
First / last frame
Reference media
Audio
model contextgenerated result
Video output
This diagram explains supported context types. It is not presented as a generated sample from the model.
Capability overview
What defines MiniMax H3
The points below summarize the model family as documented by its developer. Product-specific limits are kept separate.
02
Native stereo sound
The model jointly produces video and 32 kHz stereo audio, supporting dialogue, ambience, and scene sound in the generated result.
03
Base and 2K workflows
H3-Base produces 768p output. The official full workflow uses H3-Regenerate-2K to regenerate that result at 2K with the original context.
Specifications
At a glance
Published model facts and current UnblurVideo status, last checked 2026-09-27.
- Developer
- MiniMax
- Output duration
- 4–15 seconds
- Common aspect ratios
- 21:9, 16:9, 4:3, 1:1, 3:4, and 9:16
- Base resolution
- 768 px on the shorter side2K requires the separate H3-Regenerate-2K workflow.
- Frame rate
- 24 FPS
- Audio
- 32 kHz stereo
- UnblurVideo availability
- PlannedNo fallback to another model is used.
- Facts last checked
- September 27, 2026
Best fit
Where MiniMax H3 is most relevant
H3 is designed for briefs that need richer context, flexible framing, or synchronized picture and sound.
01
First and last frame control
Plan a transition between a starting image and an ending image, or use either frame independently.
02
Reference-heavy scenes
Use multiple images, video clips, and supported audio references when consistency depends on several source materials.
03
Sound-led short sequences
Create short scenes in which dialogue, effects, ambience, and visual timing need to be considered together.
Workflow
Understand the H3 pipeline before generating
The official system separates context processing, base generation, and optional 2K regeneration.
Prepare multimodal context
H3-Context-IR interprets the relationship between prompts and reference media before the generation stage.
Generate the base result
H3-Base creates the video and stereo audio at a base output whose shorter side is 768 pixels by default.
Regenerate at 2K when needed
The official 2K workflow feeds the base result and original context into H3-Regenerate-2K; 2K should not be described as native H3-Base output.
Model availability and verified limits
This public guide is the canonical UnblurVideo page for MiniMax H3 and separates official model capabilities from features currently offered by the product.
- This model guide is publicly indexable and ready to be expanded when the production integration is enabled.
- MiniMax H3 is not in the production generation allowlist yet.
- The page does not silently substitute a Kling, Seedance, or another MiniMax model.
- The base model outputs 768p by default; 2K is a separate regeneration stage in the official full workflow.
- Provider limits and credit pricing will be rendered from the deployed configuration after integration.
FAQ
Questions about MiniMax H3
Answers distinguish the published model family from the version and controls that may later be deployed here.
Related models
Continue comparing model families
Use the same availability and specification checks before choosing the next model for your workflow.
Production-ready options
Open generatorGenerate with a model available today
Open the shared workspace to compare the current production models, supported settings, duration, and credit cost before submitting a job.
Source for this guide: MiniMax H3 technical announcement.