gpt-image-2.6-sunburst ★

Specifications and API

Creation Methods

Text generation; reference image editing

Output Formats

PNG, JPEG, WebP, GIF

Quality Options

auto, low, medium, high, xhigh, max

Choose a Model

Flare focuses on fast generation and editing, while Sunburst focuses on precise composition and editing. Explicitly specify gpt-image-2.6-sunburst; do not use gpt-image-2.6, which omits the variant name.

Text-to-Image Generation

Submit a prompt through the image generation API, describing the subject, scene, lighting, color palette, and aspect ratio. You can first generate one image to check the composition, then adjust the quality and size as needed.

Edit Reference Images

Submit image and prompt through the image editing API. JSON accepts image URLs, while multipart accepts local files; provide up to 16 reference images. Explain the role of each image and what needs to be preserved or modified.

Billing and Results

Billing is based on the number of successfully generated images, with a base price of 0.22 Credits/image. Returned Token statistics do not change the per-image billing method. By default, read the image links in data; when b64_json is selected, only one image is returned. For long-running tasks, pass async or callback_url, and use the original task_id to query the completed result.

Usage Limits

Use auto or WIDTHxHEIGHT for size. Width and height must be multiples of 16, the longest side must not exceed 3840, total pixels must be 655,360–8,294,400, and the aspect ratio must not exceed 3:1. n is 1–10, and b64_json requires n=1. Static PNG/JPEG/WebP supports 4K pixel presets. For native GIFs, 512x512, 9 frames, and 200ms per frame are recommended; the actual size may be adjusted by the model. GIFs cannot directly use static 4K specifications. SVG is not currently supported. Version 2.6 does not configure cross-model substitution.

Frequently Asked Questions

Advanced Parameters

Supports complexity=low/medium/high and static image markup annotation editing. GIFs support 2–16 frames, with each frame 20–655350ms (rounded to 10ms increments), with defaults of 9 frames and 200ms. The annotation image must be a PNG of the same dimensions as the first reference image and contain transparent pixels; it cannot be used together with mask or GIF.