For Developers/Models/MiniMax H3-Image
Open SourceMiniMaxReleased 2026-11

MiniMax H3-Image

MiniMax's open-weight image model for 4K generation and editing in under 10 seconds

Context window

Not published

Input / 1M tokens

$1.50

Output / 1M tokens

$3.80

Provider

MiniMax

MiniMax API, per 1M tokens (Standard mode): text input $2.30, image input $1.50, image output $3.80. Highspeed mode: text input $2.30, image input $3.00, image output $7.60. MiniMax estimates about $0.022 for a 1K text-to-image and $0.069 for a 4K image in Standard mode. Images of 16.6 MP or more are 50% off until November 17, 2026 at 16:00 UTC. The listed $1.50/$3.80 is image input / image output. · Data verified 2026-11-11

MiniMax H3-Image (November 11, 2026) is an open-weight image model from Chinese lab MiniMax, built on its H3 multimodal model, that handles text-to-image generation, reference-based generation and editing in one 33.1B-parameter diffusion transformer. MiniMax says it renders a 4096 x 4096 (16 MP) image in under 10 seconds and accepts up to 16 reference images. It is on the MiniMax API in Standard and Highspeed modes, in MiniMax Code and in Hailuo AI. The weights are on Hugging Face under the MiniMax-H3-Image Community License, which excludes the EU, UK, South Korea and the US and requires MiniMax's written permission above $1M in annual revenue or for hosted services.

How to access it

MiniMax API (model IDs MiniMax-H3-Image and MiniMax-H3-Image-Highspeed) through an OpenAI-compatible images endpoint, MiniMax Code and Hailuo AI. Weights: MiniMaxAI/MiniMax-H3-Image on Hugging Face, runnable with SGLang, vLLM-Omni or diffusers, under a license that excludes the EU, UK, South Korea and the US.

Strengths

  • ✓4K (16 MP) images in under 10 seconds, per MiniMax
  • ✓Generation and editing in one model, with up to 16 reference images
  • ✓Open weights on Hugging Face under a restricted community license
  • ✓About $0.02 per 1K image and $0.07 per 4K image in Standard mode

Best for developers who...

High-resolution marketing and product imagesReference-based editing with many input imagesSelf-hosted image generation where the license allows it

When to choose it (and when not to)

Reach for MiniMax H3-Image when...

  • →You want 4K images quickly at a low per-image price
  • →You need many reference images in one edit
  • →You want to self-host an image model in a region the license covers

Look elsewhere if...

  • ✕You want to use the weights in the EU, UK, South Korea or US (excluded by the license)
  • ✕You need published benchmark comparisons (MiniMax published none)
  • ✕Your company has over $1M in yearly revenue and cannot get MiniMax's written permission for the weights

How to use it

  • ›Highspeed mode is faster but costs about twice as much per image as Standard
  • ›When editing with several references, name them in the prompt by number (image 1, image 2)

Quickstart

Python
import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.minimax.io/v1",
    api_key=os.environ.get("MINIMAX_API_KEY"),
    timeout=600,
    max_retries=0,
)
result = client.images.generate(
    model="MiniMax-H3-Image",
    prompt="An orange cat lying on a windowsill basking in the sun, warm afternoon light, film texture",
    size="2720x1536",
    response_format="url",
)
for item in result.data:
    print(item.url)

From MiniMax's image generation guide (OpenAI SDK against the MiniMax endpoint). Retries are off because a generation request is not idempotent.

API model id: MiniMax-H3-Image

Compare MiniMax H3-Image

Compare MiniMax H3-Image with any other model

Build a comparison →
All model comparisons →

Learn the concepts