Skip to main content

MiniMax-H3-Image / MiniMax-H3-Image-Highspeed

Overview​

MiniMax-H3-Image is MiniMax's open-weight H3-family model for image generation and editing, released on November 11, 2026. It combines detailed instruction following, multi-reference composition, and native 4K generation for creative and product design workflows.

For generation and editing requests, reference-image uploads, streaming events, and error handling, see the MiniMax Image Generation API.

Model Variants​

ModelDescription
MiniMax-H3-ImageStandard variant with lower image-input and image-output token prices.
MiniMax-H3-Image-HighspeedFaster generation with the same supported inputs, output sizes, and reference-image limits.

Key Features​

  • Generation and Editing: Creates images from text or transforms existing images through reference-based instructions.
  • Multi-Reference Composition: Combines up to 16 references to guide subjects, visual styles, layouts, and product arrangements.
  • Native 4K Detail: Generates 4096 x 4096 square images directly, with detailed textures and lighting; custom output sizes extend to 20 megapixels.
  • Creative Control: Follows detailed briefs for scene composition, object placement, viewpoints, and visual relationships.
  • Prompt Enhancement: Expands short prompts into more detailed descriptions, with optional disabling for text-to-image generation.

Best Use Cases​

  • Game and Concept Art: Explore environments, character sheets, creatures, and cinematic key art from written briefs and visual references.
  • Product and E-commerce Visuals: Combine product references into lifestyle scenes and compare arrangements or backgrounds.
  • Campaign and Packaging Concepts: Develop promotional artwork and packaging directions with coordinated colors, graphics, and short labels.
  • Creative Iteration: Compare visual styles, refine viewpoints, and revise existing compositions; Highspeed suits workflows that prioritize faster turnaround.

Capabilities and Limitations​

The following specifications apply to both variants.

CapabilityDescription
Input and OutputText input for generation; text plus reference images for editing; image output.
Reference ImagesEditing accepts 1-16 images, each up to 64 MiB, with 128-8192 pixels per side and a maximum 4:1 aspect ratio.
Output Size1K, 2K, and 4K area presets; custom dimensions from 655,360 to 20,000,000 total pixels.
Aspect RatiosOutput ratios from 1:3 to 3:1; presets describe target image area rather than a fixed edge length.
Images per Request1-10 images, with 1 as the default.
Output FormatsJPEG, PNG, and WebP.
Prompt LengthUp to 32,000 characters; with prompt enhancement disabled, only approximately the first 4,096 tokens take effect.
Prompt EnhancementEnabled by default for text-to-image and always enabled for image editing.

Known Limitations​

  • Masked editing, transparent backgrounds, and preview images are not supported.
  • Long paragraphs and dense text are less reliable than titles, labels, and short annotations.
  • The total request body is limited to 64 MiB; use image URLs or uploaded-file references when combined uploads would exceed it.
  • Generated image URLs have no guaranteed retention period; save the returned images for continued use.

Standard Pricing​

Prices are shown in USD per 1 million tokens. Both models are billed by actual text input, reference-image input, and image output token usage, rather than a fixed price per image.

ModelText InputImage InputImage Output
MiniMax-H3-Image$2.30$1.50$3.80
MiniMax-H3-Image-Highspeed$2.30$3.00$7.60

Credits settlement: B.AI converts charges at 1 USD = 1,000,000 Credits and deducts Credits from the account balance.

Pricing note

Model prices in the documentation are based on the model providers' official list prices and serve as B.AI standard reference prices. B.AI may offer discounts or limited-time promotions on this basis; offer details are shown in Promotions and Pricing Updates and the website's discount groups. Applicable prices, offer terms, and final charges are subject to the platform display and actual billing records.