# MiniMax Image-01 Guide: Hailuo Image Generation with an API > Published 2026-04-05, updated 2026-09-06 ยท https://www.promptzone.com/arne_suzuki/hailuo-image-fast-ai-image-generator-1cod --- title: "MiniMax Image-01 Guide: Hailuo Image Generation with an API" published: true description: "Generate images with MiniMax Image-01 using its hosted API, supported reference inputs, verified request limits, and output-saving steps." tags: ai,imagegeneration,prompting --- MiniMax Image-01 is MiniMax's hosted text-to-image model. To generate an image, send a prompt with the model identifier `image-01` to MiniMax's image-generation API; a character reference is optional. Its launch announcement connects Image-01's development to MiniMax's Hailuo Video-01 work. [Announcement][launch] [Guide][guide] [API][api] No open weights are provided for Image-01 in these sources. Access it as a service, with an account, API credentials, and requests to MiniMax's infrastructure. ## What are the key facts about MiniMax Image-01? | Field | Verified detail | | --- | --- | | Developer | MiniMax. [Announcement][launch] | | Released | February 15, 2025 in MiniMax's release notes; the launch blog is dated February 28, 2025. [Release notes][releases] [Announcement][launch] | | Type | Hosted text-to-image generation with an image-reference workflow. [Guide][guide] | | Size or parameters | Not published in the cited model announcement or API reference. [Announcement][launch] | | License and access | Hosted API service; no open-weight model license or download is provided. [Guide][guide] | | Where it runs | MiniMax infrastructure, accessed through its image-generation endpoint. [API reference][api] | Use the API identifier `image-01` when implementing an integration. Keeping the product name and request identifier together helps avoid confusion when a project also uses other MiniMax services. [API reference][api] ## What can MiniMax Image-01 generate from prompts and references? MiniMax's launch examples emphasize scene composition, lighting, human subjects, and objects. Those examples establish the intended uses; they are not a substitute for evaluating the model on your own creative brief. [Announcement][launch] For a useful trial, choose a scene whose success you can describe concretely. A product on a plain surface is easier to assess than an open-ended request for something impressive. List the object's material, camera position, surroundings, and lighting direction. Then inspect whether the result communicates those choices. This turns a subjective browsing session into a focused production exercise. The official guide also documents generating a new image from a subject reference and a prompt. It describes preserving the subject's key characteristics across different contexts. [Reference workflow][guide] That makes character continuity a reasonable thing to test. Use the same reference while changing the background, then compare the visible identity details before accepting an image into a series. Do not judge character continuity only from thumbnails. Inspect facial features, clothing, and silhouette at the intended delivery size. The documented reference type is `character`; the guide does not establish a separate product-reference mode. [Reference workflow][guide] ## What are MiniMax Image-01 prompt, batch, and output limits? The API caps prompts at 1,500 characters and supports requests for one to nine images. Returned image URLs expire after 24 hours. [Request limits][api] Choose either an aspect-ratio preset or custom dimensions deliberately. The reference says an aspect ratio takes priority when both are provided, so an unnoticed preset can change the expected output shape. [API reference][api] The reference-image guide supports one reference image per request. It describes subject-guided generation; do not assume that it provides every control found in a dedicated image editor. [Guide][guide] Neither the launch post nor these instructions establish a universal consumer-GPU memory requirement for Image-01. Your computer sends requests to a hosted model, so local checkpoint sizing is not the setup task. A batch request also needs a result check. The response example includes `metadata.success_count`, `metadata.failed_count`, and `base_resp.status_code`; inspect those fields and the returned images before displaying a completed gallery. [API reference][api] ## How do you call the MiniMax Image-01 API? 1. Create a MiniMax platform account and obtain an API key using the account link in the reference. [API][api] 2. Set `MINIMAX_API_KEY` in the environment used by your client. 3. Start with a single image request and a clearly described subject. 4. Check the response status and save the returned image. 5. Add reference inputs only after the basic path works. The following original prompt uses MiniMax's documented JSON request format. It prints the response so you can inspect the image URL and status fields. [API][api] ```bash curl --fail-with-body https://api.minimax.io/v1/image_generation \ -H "Authorization: Bearer $MINIMAX_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "model": "image-01", "prompt": "A ceramic teapot on a pale wooden table, soft window light, uncluttered background", "aspect_ratio": "1:1", "response_format": "url", "n": 1 }' ``` Keep a record of the prompt and settings alongside the downloaded file. When reviewing a variation, annotate the specific failure rather than using a general label such as low quality. For example, distinguish an incorrect object shape from an unwanted background. Those observations suggest different prompt revisions and make later comparisons more useful. The official guide offers a base64 response example if you prefer to decode image data directly. Follow that example when your application needs to write an output file immediately. [Saving outputs][guide] For the reference workflow, use the documented `subject_reference` structure with a character image. Keep the reference constant during an evaluation so the changing scene is the main variable. [Guide][guide] A practical prompt log can contain the scene description, reference filename, selected settings, and acceptance notes. Build that habit before introducing a large batch or a more elaborate interface. Find starting structures in [PromptZone's prompt library](/ai-prompts), then rewrite them around the object and composition you can actually evaluate. ## How does MiniMax Image-01 compare with Ideogram and local FLUX? Ideogram 3.0 is another hosted image model with documented style controls. Compare it with Image-01 when the task depends on visual treatment, and use MiniMax's subject-reference path when testing character continuity. [Ideogram API][ideogram] FLUX.1-schnell offers downloadable weights and a documented local workflow. That is a different deployment choice from sending requests to MiniMax. [FLUX model card][flux] | Decision | What to evaluate | | --- | --- | | Hosted subject-reference images | MiniMax Image-01's documented reference workflow | | Hosted image style controls | Ideogram 3.0's documented generation inputs | | Running a model on your own hardware | FLUX.1-schnell's published weights and runtime | These options do not establish a quality winner. Choose a shared brief and judge the resulting assets using the same acceptance rules. The [ComfyUI pillar](/tomas_novak/comfyui-2026-the-complete-guide-to-power-user-ai-image-generation-1g17) explains the local workflow context if deployment control is part of your decision. ## What else should you know about MiniMax Image-01? ### Is Hailuo Image a downloadable model? MiniMax documents Image-01 as a hosted API service and provides no open weights in the cited release materials. Use the `image-01` API identifier for the image-generation workflow. [Guide][guide] [Announcement][launch] ### Can Image-01 work from a reference image? MiniMax Image-01 accepts one subject-reference image per request. The guide demonstrates the `character` reference type for generating the same subject in different settings. [Guide][guide] ### Can I keep the returned image URL permanently? MiniMax Image-01 image URLs expire after 24 hours. Download the file for storage, or use the documented base64 response option. [API][api] ### Is batch generation the same as predictable throughput? MiniMax Image-01's batch-size allowance does not specify a fixed completion time. Measure the time and accepted output count for your own workload before planning a production schedule. [API][api] ## Sources - [MiniMax Image-01 launch announcement][launch] - [MiniMax model release notes][releases] - [MiniMax image-generation guide][guide] - [MiniMax text-to-image API reference][api] - [Ideogram 3.0 generation reference][ideogram] - [FLUX.1-schnell official model card][flux] [launch]: https://www.minimax.io/news/image-01 [releases]: https://platform.minimax.io/docs/release-notes/models [guide]: https://platform.minimax.io/docs/guides/image-generation [api]: https://platform.minimax.io/docs/api-reference/image-generation-t2i [ideogram]: https://developer.ideogram.ai/api-reference/generate-images/generate-v3 [flux]: https://huggingface.co/black-forest-labs/FLUX.1-schnell ## Related guides on PromptZone - [Best SDXL Models in 2026](/tara_suzuki/best-sdxl-models-in-2026-realistic-anime-and-all-purpose-checkpoints-116) - [ComfyUI 2026: The Complete Guide](/tomas_novak/comfyui-2026-the-complete-guide-to-power-user-ai-image-generation-1g17) - [AI Model Releases Timeline](/ai-model-releases)