> ## Documentation Index > Fetch the complete documentation index at: https://evolink.ai/docs/llms.txt > Use this file to discover all available pages before exploring further. # GPT Image 2.6 Image Generation > - GPT Image 2.6 (gpt-image-2.6-sunburst / gpt-image-2.6-flare) supports text-to-image, image-to-image, image editing and markup-guided editing, with animated GIF and SVG vector output - Asynchronous processing: use the returned task ID to [query the task](/en/api-manual/task-management/get-task-detail) - Generated image links are valid for 24 hours. Save the results promptly ## OpenAPI ````yaml en/api-manual/image-series/gpt-image-2.6/gpt-image-2.6-image-generation.json POST /v1/images/generations openapi: 3.1.0 info: title: gpt-image-2.6 API description: >- Create image tasks using AI models with support for multiple models and parameter configurations license: name: MIT version: 1.0.0 servers: - url: https://api.evolink.ai description: Production security: - bearerAuth: [] tags: - name: Image Generation description: AI image generation related APIs paths: /v1/images/generations: post: tags: - Image Generation summary: gpt-image-2.6 API description: >- - GPT Image 2.6 (gpt-image-2.6-sunburst / gpt-image-2.6-flare) supports text-to-image, image-to-image, image editing and markup-guided editing, with animated GIF and SVG vector output - Asynchronous processing: use the returned task ID to [query the task](/en/api-manual/task-management/get-task-detail) - Generated image links are valid for 24 hours. Save the results promptly operationId: createImageGenerationGptImage26 requestBody: required: true content: application/json: schema: $ref: '#/components/schemas/ImageGenerationRequest' examples: text_to_image: summary: Text to Image (simple) value: model: gpt-image-2.6-sunburst prompt: A beautiful colorful sunset over the ocean text_to_image_hd: summary: Text to Image (HD 16:9) value: model: gpt-image-2.6-sunburst prompt: Cinematic wide shot of a futuristic city skyline at dusk size: '16:9' resolution: 4K quality: high 'n': 1 complexity: high text_to_image_pixel: summary: Text to Image (explicit pixels) value: model: gpt-image-2.6-sunburst prompt: Minimalist logo design size: 1024x1024 quality: medium image_edit: summary: Image-to-Image / Edit value: model: gpt-image-2.6-sunburst prompt: Add a cute cat next to her size: '1:1' resolution: 1K quality: medium image_urls: - https://example.com/input.png callback_url: https://your-domain.com/webhook/image-done image_inpaint: summary: Inpainting (with mask) value: model: gpt-image-2.6-sunburst prompt: Replace the sky with a starry night size: '1:1' resolution: 1K quality: medium image_urls: - https://example.com/input.png mask_url: https://example.com/mask.png markup_edit: summary: Markup-guided editing value: model: gpt-image-2.6-flare prompt: >- Change the vase inside the red circle to solid cobalt-blue ceramic, keep everything else unchanged, and remove the red circle quality: medium image_urls: - https://example.com/kitchen.png markup_url: https://example.com/markup.png batch_generation: summary: Batch generation value: model: gpt-image-2.6-sunburst prompt: A cute robot in pixel art style size: '1:1' resolution: 2K quality: high 'n': 4 gif_animation: summary: Animated GIF value: model: gpt-image-2.6-flare prompt: >- A lantern swaying gently in the night breeze, with a seamless loop output_format: gif size: auto quality: medium model_params: frame_count: 9 frame_duration_ms: 200 svg_vector: summary: SVG vector image value: model: gpt-image-2.6-sunburst prompt: >- A flat-style logo for a bakery, featuring wheat ears and a circular badge output_format: svg quality: medium model_params: svg_model: gpt-6.1-terra gif_animation_pixel: summary: Animated GIF (explicit frame size) value: model: gpt-image-2.6-flare prompt: A kitten blinking naturally, in a seamless loop output_format: gif size: 720x720 quality: low model_params: frame_count: 16 frame_duration_ms: 200 responses: '200': description: Image generation task created successfully content: application/json: schema: $ref: '#/components/schemas/ImageGenerationResponse' '400': description: Invalid request parameters content: application/json: schema: $ref: '#/components/schemas/ErrorResponse' example: error: code: invalid_request message: Invalid request parameters type: invalid_request_error '401': description: Unauthenticated, invalid or expired token content: application/json: schema: $ref: '#/components/schemas/ErrorResponse' example: error: code: unauthorized message: Invalid or expired token type: authentication_error '402': description: Insufficient quota, recharge required content: application/json: schema: $ref: '#/components/schemas/ErrorResponse' example: error: code: insufficient_quota message: Insufficient quota. Please top up your account. type: insufficient_quota '403': description: Access denied content: application/json: schema: $ref: '#/components/schemas/ErrorResponse' example: error: code: model_access_denied message: 'Token does not have access to model: gpt-image-2.6-sunburst' type: invalid_request_error '429': description: Rate limit exceeded content: application/json: schema: $ref: '#/components/schemas/ErrorResponse' example: error: code: rate_limit_exceeded message: Too many requests, please try again later type: rate_limit_error '500': description: Internal server error content: application/json: schema: $ref: '#/components/schemas/ErrorResponse' example: error: code: internal_error message: Internal server error type: api_error components: schemas: ImageGenerationRequest: type: object required: - model - prompt properties: model: type: string description: >- Image generation model name, with improved stability and control for commercial use | Model ID | Best suited for | |---|---| | `gpt-image-2.6-sunburst` | Greater detail for precise visual work | | `gpt-image-2.6-flare` | Faster generation for quick iteration | Both models accept exactly the same parameters. Actual token consumption at the same `quality` tier may differ slightly; use the returned `usage` as the source of truth. enum: - gpt-image-2.6-sunburst - gpt-image-2.6-flare default: gpt-image-2.6-sunburst example: gpt-image-2.6-sunburst prompt: type: string description: >- Prompt describing the image to generate or how to edit the input image **Length limits (both must be met):** - Up to `32,000` characters (counted as Unicode code points) - The `prompt` text itself must not exceed `60,000` bytes when encoded in UTF-8 Different characters use different numbers of UTF-8 bytes. Use the actual encoded byte count. If either limit is exceeded, shorten the prompt before resubmitting. **Recommendation:** If the prompt exceeds `8000` tokens, the generated image may not match expectations; shortening the prompt is recommended. example: A beautiful colorful sunset over the ocean maxLength: 32000 image_urls: type: array description: >- Reference images as public URLs, for image-to-image and image editing **Note:** - Number of input images per request: `1~16` - Size of a single image: not exceeding `50MB` - Pixels of a single image: width × height not exceeding `178,956,970` px - Side length of a single image: width / height each not exceeding `23170` px; exceeding this may cause errors - Supported file formats: `.jpeg`, `.jpg`, `.png`, `.webp` - Image URLs must be directly accessible by the server, or the image URL should directly download when accessed (typically these URLs end with image file extensions, such as `.png`, `.jpg`) - In image-to-image / image editing scenarios, the reference images themselves also incur additional image input token consumption - With `markup_url`, markup applies to the first image and the others serve as references; the first image must have width × height ≤ `16,777,216` px - Reference images are not supported when `output_format` is `svg` items: type: string format: uri example: - https://example.com/image1.png - https://example.com/image2.png mask_url: type: string format: uri description: >- Inpainting mask URL — marks the region of the reference image to regenerate. **Only valid in image edit mode** (must be combined with `image_urls`); the mask is silently ignored in pure text-to-image requests. **Format requirements:** - **Must be a PNG with an alpha channel**: transparent pixels (`alpha < 255`) = regions to regenerate; opaque pixels = preserved - **Mask dimensions must exactly match the reference image dimensions** (width × height in pixels) - Single mask per request **Note:** - At least one reference image is required in `image_urls`; a mask sent alone has no effect - Cannot be combined with `markup_url` - Common errors: - `Invalid mask image format - mask image missing alpha channel`: the uploaded image has no alpha channel (JPEG, opaque PNG, etc.). Re-export the mask as a PNG with transparent regions. - `Invalid mask image format - mask size does not match image size`: the mask dimensions don't match the reference image. Resize the mask to the same pixel dimensions as your reference image. example: https://example.com/mask.png markup_url: type: string format: uri description: >- Public URL of a markup layer. Use circles, arrows or freehand marks to indicate what to change in the reference image, and describe the changes in `prompt`. **Only valid in image editing mode** (requires `image_urls`). **Format requirements:** - **Must be a PNG with transparent pixels**: draw only the annotations and leave the rest transparent. Do not draw directly on the original image - **Dimensions must exactly match the first reference image** (width × height in pixels) - The first reference image and the markup layer must each have width × height ≤ `16,777,216` px - File size: no more than `50MB` **Note:** - Markup applies to the first image in `image_urls`; the remaining images serve as references - Cannot be combined with `mask_url` - Supports standard image and GIF editing, but not SVG - Can also be supplied as `model_params.markup_url`; the top-level value takes precedence when both are present - In `prompt`, also specify what to preserve, for example, "keep everything else unchanged and remove the red circle" example: https://example.com/markup.png size: type: string description: >- Generated image dimensions. Defaults to `auto`. Standard images and SVG reference images share the following rules; GIF uses the separate rules in section ④. **① Standard images / SVG reference images: ratio format (15 options)** - `1:1`: Square - `1:2` / `2:1`: Extreme portrait / landscape - `1:3` / `3:1`: Ultra portrait / landscape (3:1 limit) - `2:3` / `3:2`: Standard portrait / landscape - `3:4` / `4:3`: Classic portrait / landscape - `4:5` / `5:4`: Common social media - `9:16` / `16:9`: Mobile / desktop widescreen - `9:21` / `21:9`: Ultra-wide **② Standard images / SVG reference images: explicit pixel format**: `WxH` (or `W×H`), e.g. `1024x1024`, `1536x1024`, `3840×2160` - Both width and height must be multiples of `16` - Each edge range: `[16, 3840]` - Pixel budget: `655,360 ≤ width × height ≤ 8,294,400` (about 0.65 MP ~ 8.29 MP) - Aspect ratio: `≤ 3:1` **③ Standard images / SVG reference images: `auto`**: The model decides the size automatically (`resolution` does not apply in this mode) **④ GIF output**: `size` specifies the dimensions of each frame. Only `auto` and explicit pixel dimensions are supported, not ratios. - With `auto`, each frame is `512×512` - Explicit frame width and height must be positive multiples of `16` - The `655,360` minimum pixel count for standard images does not apply to individual GIF frames; for example, `512x512` is valid - The image combining all frames must satisfy the edge, total pixel and aspect ratio limits; see `model_params.frame_count` for grid layouts and per-frame limits - `resolution` does not apply **Out-of-range handling:** - For standard images / SVG reference images, a ratio + `resolution` combination that exceeds the budget is automatically scaled down proportionally (e.g. 4K 2:1 → 3840×1920) - Explicit GIF frame dimensions that exceed the limits return HTTP 400 and are not automatically scaled down default: auto example: auto resolution: type: string description: >- Resolution tier shortcut, only effective when `size` is a ratio; ignored in explicit pixel mode and for GIF output **Pixel budget rules** (dimensions are derived from the target pixel count and the `size` ratio, aligned to multiples of 16): - `1K`: ~1 MP (1024² = 1,048,576 pixels) - `2K`: ~4 MP (2048² = 4,194,304 pixels) - `4K`: ~8.29 MP (3840×2160 = 8,294,400 pixels, the maximum) **Landscape / square output dimensions** (portrait dimensions are the landscape width/height swapped, e.g. `2:3` = `3:2` reversed): | Ratio | 1K | 2K | 4K | |---|---|---|---| | `1:1` | 1024×1024 | 2048×2048 | 2880×2880 | | `2:1` | 1456×720 | 2896×1456 | 3840×1920 \* | | `3:1` | 1776×592 | 3552×1184 | 3840×1280 \* | | `3:2` | 1248×832 | 2512×1680 | 3520×2352 | | `4:3` | 1184×880 | 2368×1776 | 3312×2480 \* | | `5:4` | 1152×912 | 2288×1824 | 3216×2576 | | `16:9` | 1360×768 | 2736×1536 | 3840×2160 (UHD) | | `21:9` | 1568×672 | 3136×1344 | 3840×1648 \* | \* Marks combinations that are auto-downscaled to fit the pixel budget. Values are case-insensitive. enum: - 1K - 2K - 4K default: 1K example: 1K quality: type: string description: >- Rendering quality, which directly affects output token count and cost. Defaults to `medium` | Value | Output tokens (1024²) | Relative image output cost (1024²) | |---|---|---| | `low` | 824 | ~0.57× | | `medium` | 1,449 | 1.0× | | `high` | 1,756 | ~1.21× | | `xhigh` | 3,122 | ~2.15× | | `max` | 7,024 | ~4.85× | **Note:** - This table compares image output costs only, excluding input costs and SVG author model costs - Calculated for `1024×1024`; output tokens vary with total pixels and aspect ratio. Use the returned `usage` for actual consumption - For GIF output, tokens are calculated for the combined image containing all frames; for SVG output, quality applies to the reference image - Use `low` for quick drafts. For final assets, compare tiers to balance detail, latency and cost enum: - low - medium - high - xhigh - max default: medium example: medium complexity: type: string description: >- Planning depth before image generation, independent of `quality`. Defaults to `high` - `low`: Suitable for local changes such as replacing a background or changing sign text - `medium`: Balanced - `high`: Suitable for text-heavy images, complex layouts or complex relationships between objects **Note:** - Higher levels take longer and may introduce unwanted changes; compare results for your task - Changing `complexity` does not incur an additional charge - For SVG output, this applies to the reference image - Can also be supplied as `model_params.complexity`; the top-level value takes precedence when both are present enum: - low - medium - high default: high example: high background: type: string description: |- Alpha channel of the output image. Defaults to `opaque` - `opaque`: Flat background, no alpha channel - `transparent`: Keeps the alpha channel **Note:** - `transparent` is a Preview feature and results may be unstable - When using `transparent`, `output_format` must be `png` or `webp` - When `output_format` is `gif` or `svg`, only `opaque` is supported enum: - opaque - transparent default: opaque example: opaque output_format: type: string description: >- Output file format. Defaults to `png` - `png`: Lossless, supports transparent backgrounds - `jpeg`: Lossy compression with smaller files - `webp`: Balances file size and quality; also supports transparency - `gif`: Animated GIF; see `model_params` for frame count and frame duration - `svg`: SVG vector image, suitable for logos, icons and other graphics that need lossless scaling **Note:** - When `background` is `transparent`, only `png` or `webp` is allowed - `gif` supports both text-to-image and image editing - For `gif` and `svg`, `n` must be `1` - `svg` supports text-to-image only - SVG output is created by first generating a reference image, then using an author model (`model_params.svg_model`) to convert it to vector graphics. Author model tokens are billed separately at that model's rates enum: - png - jpeg - webp - gif - svg default: png example: png 'n': type: integer description: >- Number of images to generate, each billed independently **Note:** - Text input tokens scale linearly with `n` - Support for `n > 1` is currently unreliable: even with a value greater than 1, only one image is often returned. The actual count is whatever `results` holds in the response, and billing follows the returned `usage`. If you need several images, send multiple single-image requests instead - When `output_format` is `gif` or `svg`, this must be `1` minimum: 1 maximum: 10 default: 1 example: 1 model_params: type: object description: >- Model-specific parameters. GIF frame parameters and SVG author parameters must be supplied in this object. - `frame_count` and `frame_duration_ms` are only allowed with `output_format=gif` - `svg_model` and `svg_reasoning_effort` are only allowed with `output_format=svg` - Parameters incompatible with the output format return HTTP 400; they are not silently ignored properties: frame_count: type: integer description: >- Number of GIF frames. Only available when `output_format` is `gif`. Defaults to `9`. All frames are generated together as one combined image. Explicit frame width and height must be positive multiples of `16`; individual frames are not subject to the minimum pixel count for standard images. Edge lengths below are in pixels. | Frames | Per-frame limits | Example size | |---|---|---| | `2`~`8` | ≤ 1,036,800 pixels; one edge ≤ 960, the other ≤ 1920 | `960x1072` | | `9` | ≤ 921,600 pixels; both edges ≤ 1280 | `1280x720` | | `10`~`12` | ≤ 691,200 pixels; one edge ≤ 1280, the other ≤ 960 | `960x720` | | `13`~`16` | ≤ 518,400 pixels; both edges ≤ 960 | `720x720` | **Grid layout (columns × rows):** - `2`~`8` frames: `4×2` or `2×4` - `9` frames: `3×3` - `10`~`12` frames: `3×4` or `4×3`, with `4×4` tried if necessary - `13`~`16` frames: `4×4` **All of the following must also hold:** - At least one layout must keep each edge of the combined image ≤ `3840` px, total pixels ≤ `8,294,400` and aspect ratio ≤ `3:1` - When using a `4×4` layout, each frame must satisfy the area and edge limits listed for `13`~`16` frames - If no available layout fits, the request returns HTTP 400; the area and edge limits in the table are not the only constraints With `size=auto`, each frame is `512×512`, which works for every supported frame count. Output tokens are calculated for the combined image, not by adding up individual frames. minimum: 2 maximum: 16 default: 9 example: 9 frame_duration_ms: type: integer description: >- Duration of each GIF frame in milliseconds. Only available when `output_format` is `gif`. Defaults to `200`, rounded to 10-millisecond increments minimum: 20 maximum: 2000 default: 200 example: 200 svg_model: type: string description: >- SVG author model that converts the reference image into SVG. Only available when `output_format` is `svg`. Defaults to `gpt-6.1-terra` **Note:** - Author model tokens are billed separately at the selected model's rates - The task result's `usage` includes `svg_model`, `svg_input_tokens` and `svg_output_tokens` - Author model input and output tokens are listed separately and are not included in `usage.total_tokens` enum: - gpt-6.1-terra - gpt-6.1-sol - gpt-6-astra - gpt-6-sol - gpt-6-luna default: gpt-6.1-terra example: gpt-6.1-terra svg_reasoning_effort: type: string description: >- Reasoning effort for the SVG author model. Only available when `output_format` is `svg`. If omitted, the author model's default level, `medium`, is used | Author model | Allowed values | |---|---| | `gpt-6.1-terra`, `gpt-6-sol`, `gpt-6-luna` | `none`, `low`, `medium`, `high`, `xhigh`, `max` | | `gpt-6.1-sol`, `gpt-6-astra` | `low`, `medium`, `high`, `xhigh`, `max` | **Note:** - Reasoning tokens are included in the author model's output tokens enum: - none - low - medium - high - xhigh - max example: medium callback_url: type: string description: >- HTTPS callback address after task completion **Callback Timing:** - Triggered when task is completed, failed, or cancelled - Sent after billing confirmation is completed **Security Restrictions:** - Only HTTPS protocol is supported - Callback to internal IP addresses is prohibited (127.0.0.1, 10.x.x.x, 172.16-31.x.x, 192.168.x.x, etc.) - URL length must not exceed `2048` characters **Callback Mechanism:** - Timeout: `10` seconds - Maximum `3` retries on failure (retries after `1` second/`2` seconds/`4` seconds) - Callback response body format is consistent with the task query API response format - Callback address returning 2xx status code is considered successful, other status codes will trigger retry format: uri example: https://your-domain.com/webhooks/image-task-completed ImageGenerationResponse: type: object properties: created: type: integer description: Task creation timestamp example: 1757156493 id: type: string description: Task ID example: task-unified-1757156493-imcg5zqt model: type: string description: Actual model name used example: gpt-image-2.6-sunburst object: type: string enum: - image.generation.task description: Specific task type progress: type: integer description: Task progress percentage (0-100) minimum: 0 maximum: 100 example: 1 status: type: string description: >- Task status. Queued and running tasks both return `processing`; the final status is `completed` or `failed`. enum: - processing - completed - failed example: processing task_info: $ref: '#/components/schemas/TaskInfo' description: Asynchronous task information type: type: string enum: - text - image - audio - video description: Task output type example: image usage: $ref: '#/components/schemas/Usage' description: Usage and billing information ErrorResponse: type: object properties: error: type: object properties: code: type: string description: Error code identifier message: type: string description: Error description type: type: string description: Error type TaskInfo: type: object properties: can_cancel: type: boolean description: Whether the task can be cancelled example: true estimated_time: type: integer description: Estimated completion time (seconds) minimum: 0 example: 100 Usage: type: object description: >- Reserved credits at task creation. Query the task after completion for actual consumption and token details. properties: billing_rule: type: string description: >- Billing rule. GPT Image 2.6 is billed per 1,000 tokens. Image usage, text input and SVG author model usage are charged separately at their respective rates. enum: - per_1k_tokens example: per_1k_tokens credits_reserved: type: number description: Estimated credits consumed minimum: 0 example: 2.5 user_group: type: string description: User group category example: default securitySchemes: bearerAuth: type: http scheme: bearer description: >- All APIs require Bearer Token authentication. **Get API Key:** Visit [API Key Management Page](https://evolink.ai/dashboard/keys) to get your API Key **Add to request header:** ``` Authorization: Bearer YOUR_API_KEY ``` ```` This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.