# MiniMax H3 Image

> MiniMax H3 Image (official ID MiniMax-H3-Image, RouteMux slug minimax/minimax-h3-image) is MiniMax's image generation and editing model. It supports text-to-image generation and editing with 1 to 16 reference images, 1K, 2K, and 4K resolution tiers (up to roughly 16 megapixels) or custom pixel dimensions, 1 to 10 output images per request, and token-based billing. Its sibling model minimax/minimax-h3-image-highspeed shares identical endpoints, parameters, and limits: the standard model is the family default with lower per-token rates for everyday and batch workloads, while Highspeed generates images faster with lower latency.

Canonical: https://routemux.com/minimax/minimax-h3-image
Provider: MiniMax

## Key facts

| Field | Value |
| --- | --- |
| Model ID | minimax/minimax-h3-image |
| Official model ID | MiniMax-H3-Image |
| Input modalities | text, image |
| Output modalities | image |
| Capabilities | Image Generation, Image Edit |
| Upstream release | 2026-11-10 |
| Availability | available |

## Pricing

- RouteMux: $1.9 per 1M tokens
- Official: $3.8 per 1M tokens
- Saving vs official: 50%
- Official price source: https://platform.minimax.io/docs/pricing/overview, verified 2026-11-11

Billing is prepaid wallet, charged per successful request only.

## API access

Base URL: https://api.routemux.com

Supported protocols:

- openai_images

## What is MiniMax H3 Image

MiniMax H3 Image (official ID MiniMax-H3-Image, RouteMux slug minimax/minimax-h3-image) is MiniMax's image generation and editing model. It supports text-to-image generation and editing with 1 to 16 reference images, 1K, 2K, and 4K resolution tiers (up to roughly 16 megapixels) or custom pixel dimensions, 1 to 10 output images per request, and token-based billing. Its sibling model minimax/minimax-h3-image-highspeed shares identical endpoints, parameters, and limits: the standard model is the family default with lower per-token rates for everyday and batch workloads, while Highspeed generates images faster with lower latency.

On RouteMux, the model speaks the OpenAI Images protocol: send text-to-image requests to POST https://api.routemux.com/v1/images/generations and image editing requests to POST https://api.routemux.com/v1/images/edits. The prompt accepts up to 32,000 Unicode codepoints, and each request can generate n=1 to 10 images. Image editing accepts 1 to 16 reference images via the JSON images array (each item containing an image_url string) or multipart form uploads under image or image[]. Mask inpainting (mask) and streaming (stream must be omitted or false) are not supported.

You can control dimensions in two ways: pair area ("1K", "2K", or "4K", default "4K") with aspect_ratio ("auto" or X:Y where X and Y are integers from 1 to 300 and X/Y is between 1/3 and 3, such as 1:1, 16:9, 9:16, 21:9, 3:1, or 1:3), or pass an explicit pixel size (WxH with total area between 655,360 and 20,000,000 pixels and aspect ratio between 1/3 and 3; when size is WxH, both aspect_ratio and area must be omitted; size also accepts "auto"). Prompt expansion is enabled by default for text-to-image; if you set disable_prompt_enhanced to true on text-to-image, you must explicitly specify either a WxH size or an X:Y aspect_ratio (image editing always uses prompt expansion and rejects disable_prompt_enhanced=true). Supported output_format values are "png", "jpeg", "jpg", and "webp", and response_format accepts "url" or "b64_json". output_compression takes an integer from 0 to 100 (default 95) and applies to jpeg and webp only. Reference image_url values must be public https URLs or data URLs; mm_file:// references are not supported.

Because image generation is non-idempotent (non_idempotent), upstream continues generating and billing tokens even if the client disconnects or times out, so re-sending a request causes duplicate generations and double billing. The RouteMux gateway retries this model only when upstream returns HTTP 429, and never retries on 5xx responses, timeouts, or network errors. On the client side, set max_retries=0 in the Python OpenAI SDK (or maxRetries: 0 in TypeScript) and omit --retry in curl.

## Key capabilities

- **1K/2K/4K & flexible aspect ratios** — Choose area tiers 1K, 2K, or 4K (up to ~16 megapixels) or pass exact WxH pixel dimensions from 655,360 to 20,000,000 pixels across 1:3 to 3:1 ratios.
- **1–16 reference image editing** — Pass 1 to 16 reference images to /v1/images/edits for image editing, with n=1 to 10 outputs per generation or edit call.
- **Non-idempotent retry protection** — Gateway retries only on upstream 429 — never on 5xx, timeouts, or network errors. Set max_retries=0 (or maxRetries: 0) in your SDK.
- **50% below official pricing** — Billed per text input, image input, and image output token at 50% off MiniMax official rates.

## Use cases

- **High-resolution & multi-aspect asset generation** — Use area 1K/2K/4K (~16 megapixels) or custom WxH pixel sizes across 1:3 to 3:1 aspect ratios with prompts up to 32,000 characters to generate visual assets.
- **Multi-reference subject-consistent editing** — Supply 1 to 16 reference images in /v1/images/edits together with n=1 to 10 outputs to edit and iterate consistently from a shared set of visual references.

## Other variants in this family

- MiniMax H3 Image Highspeed (`minimax/minimax-h3-image-highspeed`)

## Sources

- [MiniMax Image Generation Guide](https://platform.minimax.io/docs/guides/image-generation)
- [MiniMax Pricing Overview](https://platform.minimax.io/docs/pricing/overview)

_Content reviewed by the RouteMux team on 2026-11-11._