GPT Image 2.6 Sunburst API

This page uses gpt-image-2.6-sunburst. Choose image size, quality, references, and masks for this version.

Start with the subject and scene, then specify composition and lighting. For image text, include exact wording and placement. Keep the requested changes separate from details to preserve.

model and prompt are required. Optional fields are size (auto, 15 aspect ratios or exact pixels), resolution (1k, 2k or 4k, default 1k), quality (low, medium, high, xhigh, max or auto), n (1–4), output_format (png / jpeg / webp / gif / svg), output_compression, background, moderation, image_urls (up to 16) and mask_url. complexity / markup_url / markup / frame_count / frame_duration_ms / svg.model / svg.reasoning.effort / safety_identifier. Up to 32,000 characters. SVG: n=1.

Yes. Add up to 16 public image URLs in image_urls to edit or combine images. For local inpainting, add mask_url: a PNG with an alpha channel, the same size as the first input image, where fully transparent pixels mark the area to change. Keep masks under 4 MB.

Use auto or one of 15 aspect ratios at 1k, 2k or 4k, for example 1:1 at 2880×2880 or 16:9 at 3840×2160 in 4k. Exact sizes must be multiples of 16, at most 3840 px per side, no wider than 3:1 and between 655,360 and 8,294,400 pixels. Sizes above 2560×1440 are experimental.

Yes. Set background to transparent with output_format png or webp. JPEG has no alpha channel, so transparent JPEG requests are not supported.

Choose PNG, JPEG, or WebP for still images, GIF for animation, and SVG for vectors. Explicit transparent backgrounds require PNG or WebP; SVG has a transparent background by default. GIF and SVG accept text prompts only and have additional restrictions.

Both views describe the same draft. Valid supported JSON values update the form; invalid JSON keeps your edited text and shows an error so you can correct it.

The common APIMart pattern submits POST /v1/images/generations, reads data[0].task_id, then queries GET /v1/tasks/{task_id}. A completed image is available at data.result.images[0].url[0].

In the common task response, a failed status is accompanied by data.error.message. Check the error before deciding whether to revise the input or retry.

Check the current price for the selected model and settings before submitting. The final charge is shown in the task result.

Use Get API key to open APIMart’s key-management page after signing in. Keep your APIMart key on the server or in an environment variable and out of public client code.

GPT Image 2.6 is not yet live. Availability and prices will follow the service rollout.

Choose low, medium, or high independently of quality. High is the default. Complexity adds no separate charge or image tokens.

Use markup_url with an HTTPS or data URL, or markup as a URL or {"image_url":"..."}; file_id is unsupported. A reference image is required. The transparent PNG overlay must match the first image. Each image may have at most 16,777,216 pixels and the overlay at most 50 MB. Markup and mask_url are mutually exclusive. The overlay adds no image tokens; a small amount of text usage may still apply.

GIF is text-to-image only: no image_urls, mask_url, markup, or output_compression. frame_count is 2–16, default 9; frame_duration_ms defaults to 200, is at least 20, and uses 10 ms increments. size is per-frame pixels in multiples of 16, or auto for 512×512; ratio names are unsupported. The combined sheet must fit within 3840 pixels per side, 8,294,400 total pixels, and 3:1. Smaller frames or fewer frames may be needed. n is 1–4, billed per animation using the whole sheet.

SVG is text-to-image only with n=1, no references, masks, markup, or output_compression. svg.model is required and must have platform text-token pricing. svg.reasoning.effort may be none, minimal, low, medium, high, xhigh, or max where supported; omit it for the model default. Quality and complexity control the generated reference image. The static SVG has a transparent background by default and contains no scripts, external links, or embedded bitmaps. Charges combine reference-image and writing costs. Accounts on fixed per-image contracts do not support SVG.

Auto reserves funds at max quality; observed tasks typically settle at low, with the difference returned after actual usage is known. Compared with 2.5, low uses 4.2× and medium 3.3× the output tokens; high, xhigh, and max are unchanged. Flare and Sunburst share token rates and consumption. Final charges use current prices and actual usage. SVG costs include the reference image and the writing model’s actual text usage. GIF costs use the full frame sheet, with each animation charged separately.

Optional stable user identifier, up to 128 characters; use a hash of a username or email. It is not forwarded by default.

You can reach us via the live chat in the bottom-right corner, email us at [email protected], or join our Discord community. Our team will get back to you as soon as possible.