01
Native 16-Megapixel Output
The model generates about 16 megapixels directly, without an upscaler. On Kie.ai the 4K tier gives this size, from 4096×4096 at 1:1 to 6272×2688 at 21:9; the 1K and 2K tiers give about 1 and 4 megapixels.
Input
Click to upload or drag and drop
Supported formats: JPEG, PNG, WEBP, BMP Maximum file size: 20MB; Maximum files: 16
Reference images, 1–16, each up to 20 MB and 64 MP. The order defines <Picture 1>, <Picture 2>, ... in the prompt.
Editing instruction, up to 5000 characters. Refer to reference images as <Picture 1>, <Picture 2>, ... in upload order.
0/5000Aspect ratio of the output image. auto: matches the aspect ratio of the first reference image (limited to 1:3–3:1).
Output size tier: 1K (about 1 MP), 2K (about 4 MP, default), 4K (about 16 MP). Billed per image by this tier. Hidden and not sent when both Width and Height are set.
Optional custom output width in pixels (1–8192). Width and Height must be set together: filling one makes the other required. With both set, Aspect Ratio and Resolution are hidden and not sent (the upstream does not accept them together with a custom size); the output is exactly Width × Height, billed by its total pixels (≤1.2 MP 1K, ≤4.5 MP 2K, otherwise 4K). Total pixels 655,360–20,000,000, width : height 1:3–3:1.
Optional custom output height in pixels (1–8192). Must be set together with Width: filling one makes the other required.
Rewrite the prompt into a detailed English description before generating. Turn off to use your prompt as-is (10–20 seconds faster).
Output image format.
Random seed, 0–2147483647. Leave empty for a random seed; the actual seed is returned with the result.
Output
output typeimage
no output
Input
Text prompt, up to 5000 characters. Chinese and English are both supported.
0/5000Aspect ratio of the output image. auto: the prompt rewriter picks a suitable ratio (falls back to 1:1 when prompt expansion is off).
Output size tier: 1K (about 1 MP), 2K (about 4 MP, default), 4K (about 16 MP). Billed per image by this tier. Hidden and not sent when both Width and Height are set.
Optional custom output width in pixels (1–8192). Width and Height must be set together: filling one makes the other required. With both set, Aspect Ratio and Resolution are hidden and not sent (the upstream does not accept them together with a custom size); the output is exactly Width × Height, billed by its total pixels (≤1.2 MP 1K, ≤4.5 MP 2K, otherwise 4K). Total pixels 655,360–20,000,000, width : height 1:3–3:1.
Optional custom output height in pixels (1–8192). Must be set together with Width: filling one makes the other required.
Rewrite the prompt into a detailed English description before generating. Turn off to use your prompt as-is (10–20 seconds faster).
Output image format.
Random seed, 0–2147483647. Leave empty for a random seed; the actual seed is returned with the result.
Output
output typeimage
README
Complete guide to using MiniMax H3 Image
MiniMax H3 Image API for 16-Megapixel Generation and Editing
MiniMax H3 Image generates about 16-megapixel images directly in the 4K tier, from 4096×4096 squares to 6272×2688 ultrawide frames, and edits images with up to four references on Kie.ai. It samples in six fixed steps and takes prompts in Chinese or English.
Key Capabilities of MiniMax H3 Image
MiniMax describes MiniMax H3 Image as a state-of-the-art open-source model for image generation and editing. One checkpoint covers both tasks.
02
Ten Aspect Ratios in Three Size Tiers
Choose 1:1, 5:4, 4:5, 4:3, 3:4, 3:2, 2:3, 16:9, 9:16 or 21:9, or auto to let the prompt rewriter pick, at 1K, 2K (the default) or 4K. For an exact size, pass width and height together: up to 8,192 pixels per side, 655,360 to 20,000,000 pixels in total, and a width-to-height ratio between 1:3 and 3:1.
03
Editing With Up to Four Reference Images
MiniMax H3 Image also edits existing images. On Kie.ai an edit takes one to four reference images, and the prompt refers to each one by number in upload order: the first image is Picture 1, the second is Picture 2.
04
Six-Step Sampling
MiniMax H3 Image is distilled to six sampling steps with no classifier-free guidance. Steps and guidance are fixed and negative prompts are not supported, so each request has fewer settings to tune.
05
Text Rendering With Reasoning
The model renders text inside images. The MiniMax model card shows Chinese and English logic puzzles solved on a board.
06
Prompts in Chinese or English
Write the prompt in Chinese or English, up to 5,000 characters. By default Kie.ai first rewrites it into a detailed English description of the whole image; set expand_prompt to false to use a detailed description of your own as-is, which is 10–20 seconds faster.
MiniMax H3 Image Parameters
Model specifications for MiniMax H3 Image.
| Parameter | MiniMax H3 Image |
|---|---|
| Developer | MiniMax |
| Tasks | Text to image and image editing; one PNG or JPEG image per task |
| Output size | 1K, 2K (default) or 4K (about 1, 4 or 16 megapixels); or width and height together: up to 8,192 pixels per side, 655,360 to 20,000,000 pixels in total, and a width-to-height ratio between 1:3 and 3:1 |
| Aspect ratios | auto, 1:1, 5:4, 4:5, 4:3, 3:4, 3:2, 2:3, 16:9, 9:16, 21:9 |
| Reference images | 1 to 4 per edit, numbered in upload order; each up to 20 MB and 64 MP |
| Sampling | 6 steps, fixed; no classifier-free guidance |
| Negative prompt | Not supported |
| Prompt | 1 to 5,000 characters, Chinese or English; rewritten into a detailed English description by default |
| Architecture | Single-stream diffusion transformer, 33.1B parameters, 50 layers |
| Text encoder | Qwen3-VL text encoder of MiniMax H3 |
| License | MiniMax-H3-Image Community License |
MiniMax H3 Image Output Sizes at 16 Megapixels
4K output sizes on Kie.ai (about 16 megapixels); 5:4 and 4:5 are 4576×3648 and 3648×4576.
| Aspect ratio | Size (width × height) |
|---|---|
| 1:1 | 4096×4096 |
| 4:3 | 4736×3552 |
| 3:4 | 3552×4736 |
| 3:2 | 5024×3360 |
| 2:3 | 3360×5024 |
| 16:9 | 5472×3072 |
| 9:16 | 3072×5472 |
| 21:9 | 6272×2688 |
How to Prompt MiniMax H3 Image
A four-step workflow for text-to-image generation and multi-reference editing.
- 1
Step 1: Describe the Whole Image
Describe the full frame: subject, setting, lighting, style and any text that should appear in the image. By default the prompt is rewritten into a detailed English description. - 2
Step 2: Pick a Size for Your Aspect Ratio
The default tier is 2K. Set resolution to 4K for about 16 megapixels, for example 4096×4096 at 1:1 or 5472×3072 at 16:9, or pass width and height together for an exact size within the limits in the specifications table above. - 3
Step 3: Add Reference Images for Edits
For an edit, use minimax-h3/image-edit with one to four reference images and refer to each one by number in upload order, for example Picture 1 for the product and Picture 2 for the style. - 4
Step 4: Iterate on the Prompt
Sampling steps and guidance are fixed and there is no negative prompt, so changes come from rewriting the description or swapping the references. To reproduce a result, turn expand_prompt off and reuse the seed.
What You Can Build With MiniMax H3 Image
Typical uses for 16-megapixel generation and multi-reference editing.
Posters and Large-Format Visuals
Generate poster, packaging and banner visuals at about 16 megapixels without a separate upscaling pass.
Wide and Tall Campaign Assets
Produce 21:9 hero banners and 9:16 vertical creatives at the same pixel budget as a square image.
Multi-Reference Product Edits
Combine product shots, characters and style frames in one edit, with up to four reference images numbered in the prompt.
Images With Readable Text
Create posters, diagrams and puzzle-style visuals where Chinese or English text is rendered inside the image.
MiniMax H3 Image on Kie.ai
MiniMax H3 Image API Quick Start
Send a POST request to https://api.kie.ai/api/v1/jobs/createTask with an Authorization: Bearer YOUR_API_KEY header, model set to minimax-h3/text-to-image and an input object with prompt, resolution and aspect_ratio. For edits, set model to minimax-h3/image-edit and add image_urls with one to four images. Each image is billed by output size tier, and edits add a charge for each reference image; the current prices are in the Pricing line at the top of this page.
Image Models on Kie.ai Today
Browse the text-to-image models and the image editing models available on Kie.ai today.
One Key Across the Kie.ai Market
All models in the Kie.ai market use the same authentication method: an Authorization: Bearer YOUR_API_KEY header with the key from your Kie.ai API Key Management page.
Nothing to Host
Kie.ai provides its models as an API; there are no GPUs or model weights to manage.
Frequently Asked Questions
What is MiniMax H3 Image?
MiniMax H3 Image is an open-source model for image generation and image editing, part of the MiniMax H3 family. One checkpoint does both text to image and editing with one or more reference images.
What resolution does MiniMax H3 Image generate?
On Kie.ai you choose a tier: 1K (about 1 megapixel), 2K (about 4, the default) or 4K (about 16), for example 4096×4096 at 1:1 or 5472×3072 at 16:9 in 4K. The model generates the 16-megapixel size directly, without an upscaler. For an exact size, pass width and height together: up to 8,192 pixels per side, 655,360 to 20,000,000 pixels in total, and a width-to-height ratio between 1:3 and 3:1.
Which aspect ratios does MiniMax H3 Image support?
On Kie.ai: 1:1, 5:4, 4:5, 4:3, 3:4, 3:2, 2:3, 16:9, 9:16 and 21:9, plus auto, which lets the prompt rewriter choose. With expand_prompt off, auto falls back to 1:1 for text to image; for edits, auto follows the first reference image.
Can MiniMax H3 Image edit existing images?
Yes. Use minimax-h3/image-edit with one to four reference images. You refer to each reference by number in the prompt, in the order the images were uploaded.
Does MiniMax H3 Image support negative prompts or custom step counts?
No. The model is distilled to six sampling steps with no classifier-free guidance; steps and guidance are fixed and negative prompts are not supported.
How should I write prompts for MiniMax H3 Image?
Describe the whole image in Chinese or English, including any text that should appear in it, in up to 5,000 characters. Kie.ai rewrites the prompt into a detailed English description by default; turn expand_prompt off if you already write detailed English descriptions.
How is MiniMax H3 Image different from MiniMax H3?
MiniMax H3 on Kie.ai is the video model of the family, covering text to video, image to video and reference to video. MiniMax H3 Image is the image model: it follows the H3 base design with the audio branch removed and generates and edits still images.
How do I call the MiniMax H3 Image API on Kie.ai?
Create a key on the Kie.ai API Key Management page and send it in an Authorization: Bearer YOUR_API_KEY header. Send a POST request to /api/v1/jobs/createTask with model set to minimax-h3/text-to-image or minimax-h3/image-edit and the prompt and other settings in the input object; the response returns a taskId. Get the result from the task details endpoint, or pass a callBackUrl to receive it when the task finishes. Result links expire after about 14 days, so store the images yourself. See the API documentation.