GPT Image 2.5 Sunburst: FAQ

1. What is Sunburst?

Sunburst is the higher-fidelity member of the GPT Image 2.5 family. It shares the exact same capability surface as its faster sibling, Flare, but prioritizes image quality over speed. Use it for text-to-image and reference-guided editing when final fidelity matters, especially multi-reference composition and face-accurate stills.

2. What resolutions does Sunburst support?

• 1K (drafts and exploration)

• 2K (recommended default)

• 4K (highest detail, more expensive)

3. What aspect ratios are available?

10 ratios, from ultra-wide to ultra-tall:

• 3:1, 21:9, 16:9, 3:2, 4:3, 1:1, 3:4, 2:3, 9:16, 1:3

The default is 16:9. Output snaps to the nearest supported ratio with no deviation cap, so it is safe for any source shape. Extreme banner ratios (like 4:1 or 1:8) are not covered here.

4. Does Sunburst support transparency?

Yes, natively. Set background to transparent for an RGBA cutout for compositing.

• Background options: auto (model decides), transparent (RGBA cutout), opaque (solid).

• Pair transparency with png or webp output. Transparent plus jpeg is a hard error.

5. Can you use reference images?

Yes, up to 16 reference images for style or content guidance.

• Create: up to 16 references.

• Modify: 1 source plus up to 15 references (16 total).

When using two or more, label each image's role in the prompt (style, character, pose, or composition reference). Sunburst is strong at literal face copy-paste on unfamiliar faces (exact identity with no drift) and at blending brand, character, and product elements from labeled references.

6. What quality and format options are there?

• Quality: low, medium (default), high, xhigh, max, in increasing quality and generation time.

• Format: png (lossless, supports transparency), jpeg (smaller file), webp (modern compression, supports transparency).

7. Create vs Modify?

• Create: generates a new image from a prompt, optionally reference-guided, with full aspect-ratio and resolution control.

• Modify: edits an existing source image using up to 16 total input images. Good for targeted edits, face swaps, and multi-reference composition on an existing image.

At a glance