Create product scenes, compose posters and edit existing images with MiniMax H3 Image. MiniMax’s open-weight image model supports 4K output and up to 16 reference images.
Describe the image or the change to make.
Upload images or paste public image URLs. Up to 16 references, in the order you want them used.
Default 1; up to 10 images.
Optional auto or WIDTHxHEIGHT. When using pixel dimensions, leave area and aspect ratio unset.
Leave unset for automatic framing, or enter X:Y between 1:3 and 3:1.
Leave unset for the 4K default. Area is a pixel-area target, not a fixed edge length.
For new generations only. Requires an explicit aspect ratio or pixel size. Reference editing always enhances prompts.
JPEG by default; PNG and JPG are also accepted.
0–100, default 95. Zero is sent as zero.
Off by default.
URL delivery through the asynchronous task API.
Keep off: this integration uses asynchronous tasks.
Try one of these prompts
Rates per million tokens. The final cost combines text input, reference-image input and image output.
Estimate from token usage across the whole request. The final charge uses actual text-input, image-input and image-output tokens.
At the token usage entered here. 1 credit = $0.001 USD.
The starting values are an example, not a fixed price per image. Enter totals for all images in a batch; count shared inputs once. Image dimensions alone do not determine the final token usage.
MiniMax H3 Image is MiniMax’s image generation and editing model, released on November 11, 2026. It turns text into images and combines visual references with editing instructions; MiniMax also publishes the standard model’s weights. Its documented strengths include photorealistic detail and following complex visual instructions, while the hosted service offers custom output sizes up to 20 million pixels.
可以用这个模型构建并上线的真实工作流与生产级用例。
Use MiniMax H3 Image to explore how a product could look under a different light or on a new surface. Start with a clear product reference, then describe a limestone pedestal and soft side lighting. Treat the result as a concept until its shape, label and material match the real item.
Explore capabilities
Give MiniMax H3 Image one image for the subject and another for the setting. Say which details to preserve and which to change: move the bottle onto the second image’s table while keeping its silhouette. This is reference-based editing; there is no mask control for locking every untouched pixel.
Read the editing FAQ
Ask MiniMax H3 Image for a poster with a defined subject, headline position and visual hierarchy. Specify the wording, its position and line breaks in the prompt, as MiniMax’s official repository recommends. Proofread every rendered letter and finish exact brand typography in your design tool.
MiniMax H3 Image covers text-to-image and reference editing. Build an initial scene from a written brief, then use images to communicate the product, composition or visual direction for a revision.
MiniMax H3 Image supports 4K output and custom dimensions. This gives a poster or product scene more pixels to work with; it does not remove the need to inspect small text, edges and fine details.
MiniMax publishes MiniMax H3 Image weights and inference code for the standard model. Local deployment is a separate technical route with its own hardware and license requirements; it is not a feature currently provided by reAPI.
Give MiniMax H3 Image a subject, setting and composition, then name the detail that will decide whether the image works. For example: an amber bottle on pale stone, side lighting, and open space at the left for a headline.
For a MiniMax H3 Image edit, put the main subject first and explain what later images contribute. Use only references that help the brief, within the 16-image limit; conflicting instructions make it harder to judge the result.
Choose an area level with an aspect ratio, or exact dimensions, in the MiniMax H3 Image playground. Generate the image and inspect it against your brief before making another revision. The scene posters are concept illustrations; the playground shows an actual model output.
MiniMax H3 Image generates images from a written scene. Include subject placement and light direction when those details determine whether a composition fits your project.
MiniMax H3 Image accepts one to sixteen images for editing. Assign roles explicitly: subject, setting or appearance. More references are not automatically better direction.
MiniMax H3 Image offers 1K, 2K and 4K area levels, or custom dimensions up to 20 million pixels. Choose a canvas that matches the intended layout.
MiniMax H3 Image supports up to ten output images in a request. Review each returned image separately; a larger batch does not guarantee every candidate meets the brief.
MiniMax H3 Image expands prompts by default. Text-to-image can disable enhancement with fixed framing; edits always use it. Review the revised prompt when diagnosing an unexpected result.
MiniMax H3 Image does not support masked edits, transparent backgrounds or preview images in the documented service. Use another tool when your deliverable depends on those features.
MiniMax H3 Image and MiniMax H3 Image Highspeed share their documented image controls. Choose between their delivery options and speed positioning, then judge actual images against your brief.
Sources: MiniMax’s H3 Image announcement, official image generation guide and MiniMax-H3-Image repository, checked November 12, 2026. This compares documented models, not live reAPI services. Speed is vendor positioning; no reAPI benchmark was performed.
关于此模型的常见疑问。
curl https://reapi.ai/api/v1/images/generations \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"minimax-h3-image","prompt":"A white ceramic teapot on pale linen","area":"1K","aspect_ratio":"1:1","n":1}'提交时同步拒绝并指明出错字段。检查枚举值、范围与 URL 字段——不产生任何扣费。
401 表示 Bearer 密钥缺失或无效；402 表示预扣额超出余额。密钥与积分在控制台管理。
参考素材与生成结果都会经过自动审核。被拒的任务返回明确错误并全额退款。
进入失败状态的任务不会扣费——预扣自动退还。保留任务 ID，随时可重试。