MiniMax H3 Image Generator

Turn a prompt into a 16-megapixel picture, or edit and combine up to 16 of your own images, with MiniMax’s newest image model. Online in your browser, from 1 credit an image.

Make a MiniMax image online

Model

Prompt

0/5,000

Format

Aspect ratio

Size

2K · Auto · No watermark

A woman in a long coat on a seaside cliff at dusk under a neon sign reading MiniMax H3, made with the MiniMax image modelExample

What is MiniMax H3 Image?

It is MiniMax’s open-weights model for making and editing pictures, from the same H3 family as the Hailuo 03 video model. One model does both jobs: text to image from a prompt, and image edit from 1 to 16 pictures you upload. On Melies AI it runs in your browser, with no GPU, app or API key.

MiniMax builds it for detail and instruction following. Every picture comes out at about 16 megapixels in the largest size, from 4096 × 4096 for a square to 6272 × 2688 for a 21:9 banner, made at that size rather than upscaled. It also draws readable text, so a poster title, a label or a sign can be part of the prompt.

Pick 1:1, 16:9, 9:16, 4:3, 3:4, 3:2, 2:3, 5:4, 4:5, 21:9 or Auto, and 1K, 2K or 4K (about 1, 4 and 16 megapixels), or type an exact width and height up to 20 megapixels. Prompts can run up to 5,000 characters in English or Chinese; by default a short prompt is rewritten into a detailed description first, and you can switch that off to use your words as written, 10 to 20 seconds faster. Save as PNG or JPEG. There is no visible watermark.

Largest size
16 MP
Reference images
16
Aspect ratios + Auto
10
Sizes
3
Prompt characters
5,000
Credit · one 2K image
1

By MiniMax · run through a third-party API · not affiliated with MiniMax · Last updated

What the H3 image model does well

Five things it is built for, from posters with real text to edits that combine your own photos.

A fantasy city of stone towers and waterfalls on sea cliffs at sunset, with sailboats in the bay below
01

Native 16 MP, no upscaler

The 4K size makes about 16 megapixels directly, at every aspect ratio, so skin, fabric, fur and small print hold up when you crop in or print large. Draft in 1K or 2K, then render the one you like in 4K.

A detective board titled CONFIDENTIAL FILE MISSING with six suspect photos, quoted statements and a true-or-false table
02

Text that reads right

Put the words in quotes and they appear in the picture: a poster title, a product label, a shop sign or a menu. It works for English and Chinese, so a bilingual poster is one prompt.

An empty room and a grid of furniture and decor items, combined into one furnished bedroom with every item in place
03

Edits from up to 16 images

Upload a product, a person, a place or a style, and say what to make of them: put the product on a new set, dress the person in the outfit, or build one scene from several pictures. Name each one as @Image1, @Image2 and so on.

A skeleton watch with a brown leather strap resting on folds of linen, with fine detail on the gears
04

Real-world detail, not the “AI look”

MiniMax aims the model at natural light, real materials and believable imperfection rather than glossy, airbrushed renders. Describe the light, the lens and the materials, and it follows them closely.

A tall picture of a wooden spiral staircase rising through a circular library under a domed skylight
05

Wide banners and tall stories

Ten fixed shapes from 21:9 to 9:16, plus Auto, which picks the shape from your prompt (or follows @Image1 in an edit). Make a cinematic website banner and a vertical story from the same idea.

Text to image vs image edit

Both tabs run the same model. Text to image starts from words only; image edit starts from your pictures.

Text to imageImage edit
Starts fromA prompt1 to 16 images and a prompt
Name images as—@Image1, @Image2…
Auto ratioPicked from the promptFollows @Image1
Sizes1K, 2K, 4K1K, 2K, 4K
Price1 credit (1K, 2K), 2 (4K)The same, plus 1 per 8 images
Best forPosters, concept art, banners, stock-style photosProduct shots, try-ons, scenes built from your photos

How to use the MiniMax image generator

  1. 01

    Pick a tab

    Text to image to start from words, or Image edit to start from your own pictures (up to 16).

  2. 02

    Write the prompt

    Describe the subject, setting, light and style; put any text you want in the picture in quotes. In an edit, name your images @Image1, @Image2 and say what to keep.

  3. 03

    Choose shape and size

    10 ratios or Auto, and 1K, 2K or 4K. The price is on the button: 1 credit for 2K, 2 for 4K.

  4. 04

    Generate and download

    Press Generate; the picture is usually ready in under a minute. It stays under “My images” for 14 days.

H3 image pricing per picture

One price per picture, set by its size. Image edit adds 1 credit for every 8 images you upload. Packs from $9.90, no subscription, failed images refunded.

Model1K2K4K
H3 Image · text to image1 credit1 credit2 credits
H3 Image · edit with 1–8 images2 credits2 credits3 credits
H3 Image · edit with 9–16 images3 credits3 credits4 credits

MiniMax image FAQ

Something else? Email support@meliesai.com.

You can start for free. New accounts that sign up with Google get 4 free credits, enough for 4 pictures at 2K. After that each image costs 1 credit at 1K or 2K and 2 at 4K, from one-time packs starting at $9.90, with no subscription.

1 credit per picture at 1K or 2K and 2 credits at 4K (about 16 MP). Image edit adds 1 credit for every 8 uploaded images, so an edit from 9 to 16 images costs 2 more. The $9.90 pack has 100 credits, enough for 100 images at 2K.

Open the Image edit tab, add 1 to 16 pictures and say what to make of them, naming each as @Image1, @Image2 in the order you added them, for example “Put the bottle from @Image1 on a marble counter; keep the label unchanged.” Pick a ratio (Auto follows @Image1) and a size, then press Generate.

Yes. MiniMax publishes the weights on Hugging Face with ComfyUI templates for text to image and editing, plus support in diffusers, SGLang and vLLM-Omni. It is heavy: the image transformer is about 66 GB in BF16 and the text encoder another 67 GB, so it needs data-center GPUs. Here it runs online with no setup.

No, it makes still pictures. To animate one, use the MiniMax H3 video model, which MiniMax also calls Hailuo 03: open its page, pick Image to video and upload your picture as the first frame. It makes 4 to 15 second clips with stereo sound, in the same generator family on this site.

The weights are open: MiniMax releases them on Hugging Face and ModelScope under MiniMax’s own community license for the H3 image model, which sets the terms for use. It is not an OSI open-source license, so read the license before shipping it in a product. Using it here does not need the weights at all.

Three sizes: 1K (about 1 MP), 2K (about 4 MP) and 4K (about 16 MP). Ten shapes: 1:1, 16:9, 9:16, 4:3, 3:4, 3:2, 2:3, 5:4, 4:5, 21:9, plus Auto, which picks the shape from your prompt in text to image and follows your first image in an edit. A 4K square is 4096 × 4096; a 4K 21:9 banner is 6272 × 2688.

Yes. Put the exact words in quotes in your prompt, such as a poster title, a label or a sign, and say where they go and in what style. It handles English and Chinese. Keep each piece of text short and check spelling in the result; for long copy, add it in your design tool.

No. MiniMax distilled the model to six sampling steps with no guidance setting, and negative prompts are not supported. Say what you want instead, in positive terms: “a clean white background” rather than “no clutter”. With Rewrite prompt on, your prompt is expanded into a detailed description first, which also helps.

Up to 16 in Image edit, as JPG, PNG or WebP. They are numbered in the order you add them, and you call them @Image1, @Image2 and so on in the prompt. More images make an edit slower; 8 images or fewer add 1 credit to the price, 9 to 16 add 2.

They are siblings. MiniMax built the image model on the H3 architecture that powers Hailuo 03 video, without the audio layers, and it shares the same text encoder. The image model makes still pictures; MiniMax H3 (Hailuo 03) makes video. Both are on this site, each with its own generator.

No visible watermark or logo is added to pictures made on Melies AI. Pick PNG or JPEG under Format; 4K starts on JPEG, about 6 MB against 30 MB for a 4K PNG. Each picture stays under “My images” for 14 days, so download the ones you want to keep.

Yes. Pictures you generate are yours to use, including in ads, client work and products, as long as your prompt and uploaded images do not infringe anyone else’s rights and follow our content standards. Edits of real people need their consent.

No. Melies AI is an independent platform that gives you access to MiniMax’s image model through a third-party API provider. It is not affiliated with, endorsed by or sponsored by MiniMax. MiniMax’s own app is Hailuo AI (hailuoai.video).

Make your first 16 MP picture

Write a prompt or upload your photos, pick a shape and a size, and download the result.

Create an image