Use MiniMax H3 Image API through EvoLink for text-to-image and editing with up to 16 references. Test 1K, 2K or 4K output, estimate token costs, and integrate with one API key.
Test MiniMax H3 Image output quality, estimate the token cost of one request, and generate a sample before you formally integrate.
Create image task
Fixed to minimax-h3-imageImage generation and editing. Final charges follow actual token usage.
Reference images add input tokens to the request.
Target pixel area: 1K about 1.05 MP, 2K about 4.19 MP, 4K about 16.78 MP. 1K and 2K have the same image output charge; 4K uses about 4× the output tokens.
Results are kept for 24 hours — download anything you want to keep.
Compare text-to-image, reference editing and 4K request estimates. Image output, reference-image input and text input have separate token rates; the final charge follows actual usage.
First Test Cost
MiniMax H3 Image - 2K - 1:1 - 1 image - prompt optimization offOne image, prompt optimization off, no reference images
About 598 tests with $10 in credits.
To reproduce this estimate in the Playground, pick the Ratio size 1:1, choose 2K, turn prompt optimization off and generate 1 image from the prompt “Cat” with no reference images. Longer prompts, prompt optimization and reference images add input cost; the final charge follows actual usage.
MiniMax H3 Image testing budget guide
Pick an amount based on how many tests you expect.Good for a first validation.
Good for prompt iteration.
Good before production integration.
Each estimate covers one generated image at your current rates and leaves out the tokens of your own prompt text. The final charge follows actual token usage.
Rates are per 1K tokens. The calculator estimates a full request; the final charge follows actual usage.
minimax-h3-imageThere is no fixed price per image: the cost depends on the output size, prompt optimization, reference images and the number of images.
| Item | What it covers | Rate |
|---|---|---|
| Image output | Tokens of each generated image: about 4,200 at 1K or 2K and about 16,800 at 4K. Exact pixel sizes above 2K count width × height ÷ 1,024. | $0.0038/1K tokens0.2584 cr/1K tokens |
| Image input | Reference images in an edit: about 10,300 tokens per reference image, counted once for each generated image. | $0.0015/1K tokens0.102 cr/1K tokens |
| Text input | Your prompt, plus about 4,000 tokens for each generated image when prompt optimization is on. | $0.0023/1K tokens0.1564 cr/1K tokens |
This is the estimate held when you submit. The final charge follows actual token usage.
The model picks the aspect ratio from your prompt and reference images; resolution still sets the output area.
Prompt optimization stays on when size is auto or the request includes reference images.
MiniMax H3 Image is MiniMax's image generation and editing model. EvoLink exposes it through one API route for product scenes, campaign visuals and concept exploration, with up to 16 references and up to 20 megapixels per image. Review outputs against your own brief before production use.
One model ID, minimax-h3-image, covers both modes: send a prompt for text-to-image, or add image_urls to edit with reference images. Requests run as asynchronous tasks and are billed by tokens.
Modes: Text-to-image and image editing · Billing: By tokens: image output, reference-image input, and text input · Delivery: Asynchronous task; result links stay valid for 24 hours
minimax-h3-imageOne model ID for text-to-image and image editing on POST /v1/images/generations. Send a prompt to generate, or add image_urls with 1–16 reference images to edit. The response returns a task ID, not a finished image.
Send a prompt of up to 32,000 characters and get 1–10 images back. With size left on auto, the model picks the aspect ratio from the prompt.
Add 1–16 image URLs in image_urls and describe the change, referring to the images as "image 1", "image 2" in the prompt. The same model ID handles the edit, and each reference image adds image input tokens.
auto lets the model choose the aspect ratio. A ratio X:Y takes integers from 1 to 300 and must stay between 1:3 and 3:1. Exact pixels WxH take 655,360 to 20,000,000 total pixels and are never scaled or rounded.
resolution sets the output area for auto and ratio sizes and defaults to 2K; it is ignored for exact pixels. 1K and 2K have the same image output token charge; 4K uses about four times as many output tokens. Text and reference-image input also contribute to the total.
On by default: the model rewrites your prompt into a detailed one and returns the prompt it actually used with each image. It can be turned off only for text-to-image with a ratio or exact-pixel size; with auto size or reference images it stays on.
Set n from 1 to 10 to get several images from one request. Text input and reference-image input are charged once per generated image, so the cost scales with n.
The fields that shape output and cost on POST /v1/images/generations. Generation is asynchronous: poll the returned task ID or use a callback, and save results promptly because image links stay valid for 24 hours. Full schemas and code samples are in the API tab.
| Parameter | Type | What it does |
|---|---|---|
| model | string · required | Always 'minimax-h3-image'. The same ID covers text-to-image and image editing. |
| prompt | string · required | What to generate, or how to edit the reference images. Up to 32,000 characters; with prompt_extend off, only about the first 4,096 tokens take effect. |
| size | string · default auto | 'auto', a ratio X:Y (integers 1–300, between 1:3 and 3:1), or exact pixels WxH with 655,360–20,000,000 total pixels. Out-of-range values return 400; nothing is scaled or rounded. |
| resolution | string · 1K / 2K / 4K | Output area for auto and ratio sizes; default 2K. Ignored for exact pixels. 1K and 2K use the same number of image output tokens; 4K uses about four times as many. |
| prompt_extend | boolean · default true | Prompt optimization. false is accepted only for text-to-image with a ratio or exact-pixel size; with size auto or with image_urls it returns 400. |
| n | integer · 1–10 | Images per request; default 1. Text input and reference-image input are charged once per generated image. |
| image_urls | array · up to 16 | Reference images for editing: jpg, jpeg, png, webp, heic or heif; each side 128–8192 px; long side at most 4 times the short side; up to 64 MB each. Each one adds image input tokens. |
| output_format | string · jpeg / png / webp | File format of the result; default jpeg. Transparent backgrounds are not supported. |
| callback_url | string · optional | HTTPS webhook that receives the task result, as an alternative to polling. |
Use the EvoLink API for product integration and batch jobs, or call it from Codex, Claude, or Gemini for fast creative and development workflows. Both paths share the same EvoLink API key, balance, model routes, and task history.
Best for: product backends, batch jobs, automated workflows
Call EvoLink’s unified image API from your server and control the model ID, parameters, task queue, callbacks, and result storage.
Best for: creative and development tasks in Codex, Claude, and Gemini
Give the Agent your output goal, assets, and acceptance criteria. It can choose the route, assemble the request, track the task, and return the result without requiring you to hand-code every step.
This example shows the shortest runnable flow: create a task, save the returned task ID, then poll or wait for an HTTPS callback. Open the API tab for the complete parameter and response reference.
curl -X POST https://api.evolink.ai/v1/images/generations \
-H "Authorization: Bearer $EVOLINK_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "minimax-h3-image",
"prompt": "A colorful sunset over the ocean, painterly style",
"size": "16:9",
"resolution": "2K",
"n": 1,
"callback_url": "https://your-domain.com/webhooks/image-task-completed"
}'
# Save the returned task ID, then query:
curl https://api.evolink.ai/v1/tasks/{task_id} \
-H "Authorization: Bearer $EVOLINK_API_KEY"prompt_extend: false returns 400 when size is auto or when image_urls is present. Use a ratio or exact pixels for text-to-image, or leave prompt optimization on. The request is rejected before anything is charged.
Exact pixels need 655,360 to 20,000,000 total pixels, and every size must stay between 1:3 and 3:1, with ratio terms from 1 to 300. Out-of-range sizes return 400; nothing is scaled or rounded for you.
More than 16 URLs in image_urls returns 400. Each image must also be jpg, jpeg, png, webp, heic or heif, 128 to 8192 px per side, at most 4:1 long side to short side, and no larger than 64 MB.
Check every value against the parameter cards above: ranges, allowed options, asset counts, and file types must match the selected route.
Check the Authorization bearer token and confirm the available balance and task billing in the console.
Review likeness rights, brand assets, readable text, sensitive content, and input-asset compliance.
A MiniMax H3 Image task that reaches a final failed state is not charged. Keep the task ID, inspect the final state, and identify whether the cause is validation, review, or execution before retrying.
callback_url must be a public HTTPS URL, not localhost or a private network address. Always keep the task ID as a polling fallback.
| Feature | MiniMax H3 Image | GPT Image 2.6 Flare | Nano Banana 2.1 | Seedream 5.0 Pro | Grok Imagine Image 2.0 |
|---|---|---|---|---|---|
| Billing | By tokens: image output, image input, text input | By tokens | Per image by resolution; reference images and searches are separate items | Per image by output tier; each input image is billed | Per image by quality and resolution |
| Output resolution | 1K / 2K / 4K, or exact pixels up to 20 MP | 1K / 2K / 4K | 1K / 2K / 4K | 1K / 1.5K / 2K | 1K / 2K |
| Size options | auto, any ratio from 1:3 to 3:1, or exact pixels | auto, 15 ratios, or custom pixels | auto or 14 ratios | auto, ratio, or custom pixels | auto or 13 ratio presets |
| Reference images | Up to 16 | Up to 16 | Up to 14 | Up to 10 | Up to 3 |
Send POST /v1/images/generations with model set to minimax-h3-image. Text-to-image and editing use the same route; adding image_urls is what turns a request into an edit.
The create call returns a task ID. Poll GET /v1/tasks/{task_id} until the task is completed or failed, or pass an HTTPS callback_url. Result links are valid for 24 hours, so save the files you need.
Three rates apply: image output, reference-image input, and text input, which covers your prompt and prompt optimization. An estimate is held when the task starts and the final charge follows the actual usage. Failed tasks are refunded in full.
The Playground and the pricing calculator estimate a request from its size, resolution, prompt optimization setting, reference images and n. That estimate is held when the task starts, and the final charge follows the usage the model reports.
If the actual usage is lower or higher than the estimate, the difference is refunded or collected, so each task record shows what the request really used.
minimax-h3-image handles text-to-image and reference-image editing on the same endpoint. Moving from a text-only prototype to an editing workflow means adding image_urls to the request, not integrating a second route or model ID.
The API key that calls minimax-h3-image also calls MiniMax's video models, other image models such as Nano Banana 2.1 and Seedream 5.0 Pro, and the text models on EvoLink, with one balance and one task history.
For image work this keeps a pipeline in one place: generate a still here, use it as the first frame for a video model, or try another image model on the same brief without opening a new account.
Every task reports pending, processing, completed or failed, and failed tasks are refunded in full. When only some of the requested images succeed, the task completes with those images and bills only for them.
Poll the task or use a callback, confirm the final state, and keep the task ID so each charge can be matched to a request.
MiniMax's video models on EvoLink, available with the same account and API key as MiniMax H3 Image.
MiniMax H3 Max Turbo
MiniMax video model for 5–15 second text-to-video and first-frame animation with an optional last frame, at 480p or 768p.
View API
MiniMax H3 Max
MiniMax video model with text-to-video, first-frame image-to-video and reference routes for image, video and audio inputs, at 480p or 768p for 5–15 seconds.
View API
MiniMax H3
MiniMax video model with text-to-video, image-to-video and reference-to-video routes, 768p or 2K output and 4–15 second clips.
View API
GPT Image 2.6 Flare
Token-billed image generation and editing with five quality tiers, up to 16 reference images and 1K, 2K or 4K output.
View API
Nano Banana 2.1
Google image model with 1K, 2K and 4K output, up to 14 reference images and optional web and image search grounding.
View API
Seedream 5.0 Pro
BytePlus image model with 1K, 1.5K and 2K output tiers and up to 10 reference images for generation and editing.
View API
Grok Imagine Image 2.0
xAI image model where the number of input images selects the mode, with 1K or 2K output and up to 10 images per request.
View APIThe model ID is minimax-h3-image, in lowercase with hyphens. Send it as model on POST /v1/images/generations for both text-to-image and image editing. It is a different model from minimax-h3-image-to-video, which is a MiniMax H3 video route.
By tokens, at three rates: image output, reference-image input, and text input, which covers your prompt and prompt optimization. An estimate is held when the task starts, the final charge follows the usage the model reports, and the difference is refunded or collected. Text input and reference-image input are counted once per generated image, so the cost scales with n. Current rates are in the pricing section of this page.
Both tiers use the same number of image output tokens per image, so the image output charge is the same. The total also includes prompt processing and reference-image input. 4K uses about four times as many image output tokens. For exact-pixel sizes above the 2K area, output tokens are counted as width × height ÷ 1024.
prompt_extend is on by default. You can set it to false only for text-to-image with a ratio or exact-pixel size. With size set to auto, or when image_urls is present, false is rejected with a 400 error before anything is charged, because the automatic ratio and edits both rely on the optimization step. With it off, only about the first 4,096 tokens of the prompt take effect.
Output can reach 20,000,000 pixels, about 20 MP. To set custom pixels, pass size as WxH with 655,360 to 20,000,000 total pixels and an aspect ratio between 1:3 and 3:1; there is no multiple-of-16 rule, and the size is never scaled or rounded. With auto or a ratio, resolution picks 1K, 2K or 4K: for example, 16:9 gives 2720×1536 at 2K and 5472×3072 at 4K.
Up to 16 per request in image_urls. Supported formats are jpg, jpeg, png, webp, heic and heif; each side must be 128 to 8192 px, the long side at most 4 times the short side, and each file up to 64 MB. Refer to them in the prompt as image 1, image 2 and so on. Edits are described in the prompt: mask-based inpainting and transparent backgrounds are not supported.
The task still completes with the images that succeeded, for example when one of several is blocked by the content safety check, and only those images are billed. The number of results can therefore be lower than n. If the whole task fails, it is refunded in full.
Generation is asynchronous. The create call returns a task ID; poll GET /v1/tasks/{task_id} until the status is completed, or set callback_url to receive the result on an HTTPS endpoint. Result links are valid for 24 hours, so download and store the images you want to keep.
4K is an output-area tier with a target of 4096 × 4096 pixels, not a fixed width. Actual dimensions depend on the aspect ratio: the documented 16:9 example is 5472 × 3072. EvoLink defaults to 2K for auto or ratio sizes; explicit pixel dimensions override the resolution tier.
No. The minimax-h3-image route generates still images and supports reference-image editing. The minimax-h3-image-to-video route generates video and has separate inputs and billing. MiniMax direct API model names and defaults can also differ; use the EvoLink model ID and request options documented on this page.