Text to image · Reference editing

MiniMax H3 Image
Generator for Text to Image and Editing

Create an image from a prompt, or add 1 to 9 references to guide a new scene, outfit, pose or art style. References help guide appearance, but faces and fine details can vary. Choose from fifteen aspect ratios and 1k or 2k output, directly in your browser.

Text to ImageImage Edit

PromptText to image

Describe the subject, setting, framing, light, and style.

Aspect ratio

16:9

Resolution

Format

JPEG

Seed

-1

Create images from text or edit with up to nine reference images. Sign in to use your credits.

Example image preview

Text to image · Reference editing

Illustrative image for composition and lighting inspiration.

Image workflowMiniMax H3 Image ParametersView parameters

Check the required inputs, output settings, aspect ratios, resolutions, and formats before submitting an image task.

Text to image

promptYesDescribes the whole image: subject, setting, framing, lighting
aspect_ratioSet itChoose one of fifteen presets
resolutionNo1k (default) or 2k
output_formatNojpeg (default), png or webp
seedNoRandom seed. -1 picks a random seed

Reference editing

promptYesThe edit instruction. Refer to references as <Picture 1> … <Picture N>
imagesYes1 to 9 reference image URLs
aspect_ratioNoDefaults to the first reference image’s ratio
resolutionNo1k (default) or 2k
output_formatNojpeg (default), png or webp
seedNoRandom seed. -1 picks a random seed

Fifteen aspect ratios

1:1Square

Feed posts, avatars, thumbnails

4:5Portrait

Instagram portrait

3:4Portrait

Print-style verticals, product cards

2:3Portrait

Standard photo print proportions

9:16Vertical

Stories, Reels, Shorts, TikTok

1:2Tall

Banner rails, tall mobile hero images

1:3Very tall

Sidebar and skyscraper placements

9:21Ultra tall

Full-bleed mobile splash

16:9Landscape

Web headers, presentations, video thumbnails

4:3Landscape

Classic and archival looks

3:2Landscape

Standard camera proportions

5:4Landscape

Large-format print proportions

2:1Wide

Blog headers, email banners

3:1Very wide

Site-wide banners

21:9Ultrawide

Cinematic inserts, title beds

Two resolutions

1kDefault

Iteration for testing wording and composition

2kHigher resolution

Larger output for final review; inspect dimensions and detail before publishing

Three output formats

jpegDefault

General use; compact photographic output

pngOptional

Lossless edges and flat colour

webpOptional

Small files for web delivery

Native image output: 2K
Reference images: 9
Aspect ratios: 15
Output formats: 3

Prompt examples

MiniMax H3 Image Generator Examples

Explore instructions for outfit swaps, scene relocations, style transfers and character sheets.

Virtual try-on

The same man from <Picture 1>, now wearing the navy wool overcoat from <Picture 2>, buttoned, standing three-quarter to camera against a plain grey studio backdrop. Even soft light. Photorealistic still.

Scene relocation

The same woman from <Picture 1>, now standing on a rain-wet cobblestone street at dusk, shop signs glowing behind her, light falling from her right. Keep her hair length and glasses. Photorealistic still.

Style transfer

The subject in <Picture 1>, re-rendered as a hand-painted oil portrait with visible brushwork and warm earth tones. This is a painting, not a photograph.

Character sheet

The same character from <Picture 1>, now seated at a workshop bench under a single hanging lamp, holding a small brass tool, seen from the side. Keep the red scarf and the short beard. Same illustration style as the reference.

The model behind it

What Is MiniMax H3 Image?

This page offers two image workflows: text to image for a still composed from a prompt, and reference editing for a new scene, outfit, pose or art style guided by 1 to 9 images. Both offer 1k and 2k output settings.

Describe lighting, skin and fabric texture, camera framing, and depth of field in your prompt. These are visual directions to evaluate in the result, not a guarantee of a particular rendering architecture or level of detail.

Reference editing uses supplied faces, hairstyles, accessories and clothing as visual guidance. Address references as <Picture 1>, <Picture 2>, and so on to distinguish a person, outfit or prop. Compare the output with each reference, especially when identity or small details matter.

Image gallery

MiniMax H3 Image Examples

What it changes

What You Can Change with MiniMax H3 Image

Two endpoints, one model. Start from a text description and get a new still, or start from images you already have and change what is around the subject.

Supported image workflows
What you want	Supported	What you provide
Change the scene	Yes	Subject reference + a description of the new location
Change the outfit	Yes	Subject reference, optionally a garment reference
Change the pose	Yes	Subject reference + the new pose in words
Change the art style	Yes	Subject reference + an explicit style instruction
Combine several subjects or props	Yes	Up to 9 references, addressed as <Picture 1>…<Picture 9>
Generate from text alone	Yes	A prompt describing the whole image, with no reference needed

From a prompt alone

MiniMax H3 Text to Image

MiniMax H3 Text to Image composes a single photorealistic still from a written prompt, with no reference image involved, at 1k or 2k across fifteen aspect ratios.

Direct Light and Texture

Specify the light source, surface texture and depth of field you want. Inspect faces, fabric and fine detail in the downloaded image before using it as a final asset.

Choose 1K or 2K Output

Use 1k for initial iterations and 2k when you need a larger output. Check the downloaded dimensions and detail at your intended display size.

Choose From Fifteen Aspect Ratios

Square, tall, wide and cinematic presets, listed in the parameter guide beside the generator.

Generate One Image per Request

Submit your prompt and settings, then wait for the returned image. Completion time varies with uploads, queue load and processing; review the displayed credit cost before submitting.

Reference-guided editing

MiniMax H3 Image Edit Features

Keep the subject, change the setting. Use references for identity, outfits, style and composition.

Guide Appearance With References

Use clear references and name the facial features, hairstyle, accessories and clothing you want to retain. Identity and fine details can vary between outputs, so compare each result before adding it to a series.

Change the Scene, Outfit and Pose

State the new location, what the subject wears and how they stand, and the model re-renders them into it with matching light and perspective. Supply a garment as an extra reference when you want it matched.

Restyle a Photo as Painting or Animation

For oil painting, anime or illustration, the model re-renders the same subject in a different medium.

Combine Up to Nine References

Address each one positionally in the prompt as <Picture 1> through <Picture 9>: a person from one, an outfit from another and a prop from a third.

Three steps

How to Use MiniMax H3 Image

Choose your starting point, describe the image you want, then set the output and submit.

01
Start From Text, or Upload References
For text-to-image, skip straight to the prompt. For editing, upload one clear image of the subject, then add more for outfits, props or additional people, up to nine.
02
Describe the Image You Want
Working from text, describe the whole frame: subject, setting, framing, lighting. Working from references, describe what changes and point at them: “The same woman from <Picture 1>, now wearing a red leather jacket, standing in a neon-lit Tokyo alley at night. Photorealistic still.”
03
Set Resolution and Ratio, Then Submit
1k for iteration, 2k for final assets. Set aspect_ratio when working from text, or when you need a shape other than the first reference’s. The image comes back as a URL in your chosen format.

Real project workflows

MiniMax H3 Image Use Cases

Four kinds of work this fits: key art, concept exploration, virtual try-on and character sheets. The first two start from text; the last two start from images you already have.

Image use cases
Use case	Endpoint	What it produces
Key art and thumbnails	Text to image	Cinematic stills that match the look of your video content
Concept exploration	Text to image	Scenes, costumes and moods to settle before you generate video
Virtual try-on and outfit swaps	Reference editing	Explore different clothes using an authorized subject reference
Character sheets	Reference editing	Develop variations, then check identity across settings, poses and framing

Prompt craft

How to Write MiniMax H3 Image Prompts

Five habits that help you describe the image you want more clearly.

Describe the Whole Frame When Working From Text

Name the subject, the setting, the framing and the light. These details give the model a complete composition to work with.

Describe the New Scene in Full When Editing

Say where the person from <Picture 1> is, what they wear, how they stand and how the scene is lit.

Name the Medium When You Want a Style

For style transfer, say it outright: “This is a painting, not a photograph,” rather than naming an artist or a genre alone.

Keep the Features That Matter in the Prompt

Hair, glasses, beard, a specific accessory: write them down when they are part of the identity you need held.

Fix the Seed to Iterate on Wording

Keep the seed and other settings fixed while you adjust the instruction, so you can compare the effect of your wording more consistently.

Common failures and the fix

Image prompt troubleshooting
What you get	Why	Fix
The output looks like the reference, barely changed	The prompt named the reference but did not describe a new image	Write the location, outfit, pose and lighting in full
The style did not change	A style name alone reads as a description, not an instruction	Say the medium outright: “This is a painting, not a photograph”
The face drifted between runs	Reference guidance does not guarantee identical facial details	Use clear references, describe key features, and compare each result; rerun or edit if needed
The composition changes every time you reword	The seed is random	Fix the seed, then change only the wording
The wrong reference was used for the outfit	References were not addressed positionally	Point to them explicitly as <Picture 1>, <Picture 2>

Same account, different job

Images or Video: Which One You Need

The difference is what you hand over and what you get back. Choose a still for the look, or a clip for motion and sound.

Image editing and video generation compared
Workflow	Image editing	Video generation
What you supply	1–9 reference images plus an instruction	A prompt, a frame, or a reference pack
What comes back	One still image	A 5–15 second clip with audio
Output	1k or 2k	480p, 768p or 2K
Charged	Flat rate per image	Metered per second of output
Best for	Stills, whether composed or carried across a series	Motion, camera work and sound

A common order of work is to settle the look first, including outfit, styling and location, then take the approved still into the video generator as a first frame.

Open the video generator See credits and pricing

Frequently asked questions

MiniMax H3 Image FAQ

Answers about text-only generation, reference images, resolutions, aspect ratios, output formats and prompt wording.

What is MiniMax H3 Image?

MiniMax H3 Image is an AI image model with two endpoints: text to image, which composes a still from a prompt alone, and reference editing, which takes 1 to 9 images and re-renders your subject into a new scene, outfit, pose or style.

Can MiniMax H3 Image generate an image from text alone?

Yes. MiniMax H3 Text to Image is a separate endpoint that takes a prompt and no images at all, rendering one photorealistic still at 1k or 2k. Set aspect_ratio explicitly in this mode.

How many reference images can MiniMax H3 Image use?

MiniMax H3 Image accepts 1 to 9 reference images per request. Address each one positionally in the prompt as <Picture 1> through <Picture 9> so the model knows which reference does which job.

What resolution does MiniMax H3 Image output?

Choose 1k or 2k, with 1k as the default. Use 1k while you iterate and 2k for a larger output. Check the dimensions and visible detail of the downloaded image before preparing a final asset.

What aspect ratios does MiniMax H3 Image support?

H3 Image supports fifteen aspect ratios: 1:1, 1:2, 2:1, 1:3, 3:1, 2:3, 3:2, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9, 9:21 and 21:9. In reference editing, the output defaults to the first reference image’s ratio.

Does MiniMax H3 Image keep the same face across images?

References can guide appearance, but an identical face is not guaranteed. Use clear images, describe important features and compare every output with the original, especially for character sheets or client work.

Can MiniMax H3 Image change clothes on a person?

Yes. H3 Image handles outfit swaps: describe the new garment, or supply it as an extra reference image. Write out the clothing details you want to preserve.

Can MiniMax H3 Image turn a photo into a painting?

Yes. MiniMax H3 Image performs style transfer to painting, anime or illustration. State the medium outright. “This is a painting, not a photograph” works better than naming a style or artist alone.

What image formats does MiniMax H3 Image return?

H3 Image returns jpeg by default, and png or webp on request via the output_format parameter. Choose png for lossless edges and flat colour, and webp when file size matters at the destination.

How do I get the same MiniMax H3 Image twice?

Fix the seed and keep the prompt and settings the same. MiniMax H3 Image uses a random seed unless you set one. A fixed seed gives you a more consistent starting point for repeated runs.

Why does MiniMax H3 Image reproduce my reference instead of changing it?

An underspecified prompt can contribute to this. Describe the new location, outfit, pose and lighting rather than only pointing to “the person from <Picture 1>”. Try changing one instruction at a time and compare the results.

Get started

Edit Your First MiniMax H3 Image

Start with a reference and describe the new scene. Explore the image workflow and find the credit pack that fits your next project.

Create an Image See pricing

MiniMax H3 ImageGenerator for Text to Image and Editing