Tasks and inputs
MiniMax H3 Image is listed for Image-text-to-image, Image-to-image, Text-to-image. It accepts text, image and produces image. Use the publisher’s input format and preprocessing for your chosen deployment.
MiniMax H3 Image is an open-weight 33.1B image generation and editing model producing native ~16MP output in six steps, with up to 16 reference images.
The specification score summarizes recorded features and limits. It is not a benchmark result.
Pricing conditions: Indicative USD per 1M tokens (standard tier): $2.30 text input, $1.50 image input, $3.80 image output; about $0.022 per 1K text-to-image request. Launch promo 50% off outputs >=16.6MP until 2026-11-17 16:00 UTC. Confirm on your own account.
Recorded modality support shows what this model can accept and produce. An undocumented modality is shown as unknown rather than assumed to be unsupported.
Accepts
Produces
Most image APIs give you a 1–2 megapixel picture and an upscaler. This one renders around 16 megapixels natively, at every aspect ratio, without an upscaling pass. That single fact decides whether it belongs in your pipeline: if you print, crop hard or ship hero banners, it does.
MiniMax released it on 10 November 2026 as an open-weight model built on the MiniMax H3 family. One checkpoint handles text-to-image, reference-guided generation and editing. (The launch blog is dated a day later, 11 November; the release notes say the 10th.)
guidance_scale to 1 or omit it; other values return HTTP 400. Negative prompts are not supported.Expect roughly 69 GB for the image repository alone. Diffusers, SGLang, vLLM-Omni, LightX2V, DiffSynth-Studio and ComfyUI are supported.
| Ratio | Example size |
|---|---|
| 1:1 | 4096 × 4096 |
| 4:3 / 3:4 | 4736 × 3552 |
| 3:2 / 2:3 | 5024 × 3360 |
| 16:9 / 9:16 | 5472 × 3072 |
| 21:9 | 6272 × 2688 |
The effective range runs from about 4MP to 20MP. MiniMax says a 4K image takes under ten seconds on its service.
Editing accepts up to 16 reference images. Upload order maps to <Picture 1>, <Picture 2> and so on inside the prompt, which makes multi-product composites practical.
Standard tier, indicative USD per 1M tokens: $2.30 text input, $1.50 image input, $3.80 image output. In practice MiniMax's own examples put a 1K text-to-image request at about $0.022 and a 2K one at about $0.023. A 4K render lists at about $0.069, or half that during the launch promotion: 50% off any request whose output reaches 16.6 million pixels, ending 17 November 2026 at 16:00 UTC. After that, assume list price. Prompt-expansion tokens add to the bill, so confirm costs on your own account.
There is a faster, pricier hosted variant; see the MiniMax provider page for H3 Image Highspeed.
Weights use the MiniMax-H3-Image Community License Agreement, which also covers the prompt expander. A section of that licence requires separate written authorisation for certain uses, requested through MiniMax's form. Read it before commercial self-hosting.
Hosted access is through the MiniMax API, MiniMax Code and Hailuo AI.
No independent quality rankings existed at launch. So build your own small benchmark:
Compare against MiniMax image-01 if you used it, and against whatever you ship today in the text-to-image category.
Compare the model on the work you need to complete.
MiniMax H3 Image is listed for Image-text-to-image, Image-to-image, Text-to-image. It accepts text, image and produces image. Use the publisher’s input format and preprocessing for your chosen deployment.
The listed route has starting rates of $1.50 input and $3.80 output per million tokens. Choose the output size and quality your design needs. Include drafts and revisions when budgeting an accepted image.
The weights are downloadable under MiniMax-H3-Image Community License Agreement. Check runtime support and size the complete model and working context for your expected concurrency.
These records share a directory category. Check the source card, output type and deployment requirements before treating a model as a replacement.
recraft
Recraft V4.1 Flash V2 is Recraft's fastest, cheapest image model: 1K raster output in about 1.1 s for $0.007, quality close to Recraft V4.1.
MiniMax
MiniMax H3 Image Highspeed is the faster hosted tier of MiniMax's H3 Image generator and editor, at double the image-token price of the standard tier.
OpenAI
GPT-Image-2.6 Flare is OpenAI's fast image generation and editing model for interactive iteration, with SVG, animated GIF and markup-guided edits.
OpenAI
GPT-Image-2.6 Sunburst is OpenAI's detail-first image generation and editing model, adding SVG, animated GIF output and markup-guided edits.
krea
Krea 3 is Krea's in-house image model for text-to-image and instruction-based edits at 1K or native 2K, priced from $0.03 per image.
Alibaba
Qwen-Image-2.1-Turbo creates and edits images, including reference-guided and transparency workflows; hosted billing is separate from local deployment.