GPT-Image parameter and API differences

Understand GPT-Image parameter differences and usage for image generation and editing.

Adjust generation, editing and output parameters for your chosen model. See Create image and Edit image for complete request and response structures.

Applicable models

Model	Supported APIs
gpt-image-2.6-sunburstNew gpt-image-2.6-flareNew gpt-image-2.5-sunburst gpt-image-2.5-flare gpt-image-2 gpt-image-1.5	OpenAI Images: Create image, Edit image

Generation parameter differences

Field	Difference
`quality`	2.6 and 2.5 support `auto/low/medium/high/xhigh/max`; 2 and 1.5 support `auto/low/medium/high`. All default to `auto`
`size`	1.5 supports `auto`, `1024x1024`, `1024x1536` and `1536x1024`; 2 and 2.5 support custom dimensions, with edges divisible by 16 and no larger than 3840, an aspect ratio of 1:3–3:1 and 655360–8294400 total pixels. Sizes above 3686400 pixels are experimental
`size` (2.6)	Supports `auto` or custom dimensions divisible by 16, with an aspect ratio of 1:3–3:1. GIFs also have frame-count and total pixel-budget limits; do not reuse static-image size settings unchanged
`complexity`	2.6 supports `low/medium/high`, defaulting to `high`. Controls planning before generation; `quality` controls rendering detail. Set them independently
`background`	For transparency, explicitly set `transparent` and use `png` or `webp` for raster output. JPEG does not preserve alpha. Transparent backgrounds on 2 are an official preview feature

Editing parameter differences

Field	Difference
`input_fidelity`	1.5 supports `low/high`, defaulting to `low`; choose `high` to preserve input details. 2 always uses high fidelity, so omit this field. Also omit it for 2.6; do not copy 1.5 settings directly to newer models
`markup`	2.5 and 2.6 support markup-guided edits. Supply a PNG overlay with transparent areas, matching the first input image’s dimensions after orientation, and describe the changes in the prompt
`mask`, `markup`	Choose one; with multiple inputs, it applies to the first image. A mask identifies the edit area through transparency; markup uses circles, arrows or other annotations to indicate where to edit
`output_format`	Use `png/jpeg/webp` for edits; use the image generation API for SVG or GIF output

Output format differences

Field	Difference
`output_format`	All six models support `png/jpeg/webp`, defaulting to `png`; 2.6 image generation also supports `svg/gif`
`output_compression`	Only for `jpeg/webp`, in the range `0～100`; omit for PNG, SVG and GIF
SVG	For 2.6 image generation, set `output_format: "svg"` and `n: 1`; do not use streaming or partial images. Decode `data[0].b64_json` and save it as `.svg`; its contents are UTF-8 XML
`svg`	Selects the SVG authoring model, separate from the top-level image `model`. Set `svg.model` only when SVG model selection is enabled; otherwise omit the entire object. `svg.reasoning` and image `complexity` are independent controls
GIF	For 2.6 image generation, set `output_format: "gif"`; do not use streaming, partial images or custom compression. Decode `data[].b64_json` and save it as `.gif`
`frame_count`	GIF only, `2～16` frames, default `9`. Frame count, frame dimensions and `n` share a total pixel budget; reduce one if the request exceeds it
`frame_duration_ms`	GIF only, `20～2000` milliseconds, default `200`, rounded to 10 ms

Request examples

Replace MODEL_ID with the full model ID provided by the platform. The generation, SVG and GIF examples below use 2.6; the markup edit uses 2.5 or 2.6.

curl 'https://api.agtcloud.ai/v1/images/generations' \
  -H 'Authorization: Bearer your-api-key' \
  -H 'Content-Type: application/json' \
  --data '{
    "model": "MODEL_ID",
    "prompt": "A simple cafe opening poster titled OPEN TODAY, with clear typography",
    "quality": "medium",
    "complexity": "high",
    "size": "1024x1024",
    "output_format": "png"
  }'