ChatGPT Image 2.5 Flare and Sunburst generate images from text prompts and edit images you upload. Adding one or more images switches either tool into edit mode automatically. Describe the change you want and specify which parts of the image should stay the same.
In generate mode, Image Size sets the output dimensions. In edit mode, the tool uses the supplied images to determine the output size, so the Image Size setting does not apply. You can supply up to 16 images to edit.
Quality ranges from Low to Max, with High selected by default. Higher settings allow more detail but take longer and use more tokens. Extra High and Max cost more per result than High.
Set Background to Transparent when you want a cut-out subject on an alpha channel rather than a filled backdrop.
Same prompt: Flare and Sunburst
Left: Flare. Right: Sunburst. Same prompt, High quality, 1536 × 1024, opaque background, no reference images.
View Flare at full size · View Sunburst at full size
The prompt used for both images
Create a photorealistic architectural visualization of a contemporary courtyard house. View from the shaded interior living room through fully open floor-to-ceiling sliding glass doors toward a planted courtyard. Warm oak ceiling and cabinetry, pale travertine floor continuing outside, a low cream linen sofa at left, one sculptural dark timber lounge chair at right. In the courtyard, a single mature olive tree rises from a rectangular bed of native grasses beside a shallow reflecting pool. Beyond it stands a two-storey facade of pale limestone and vertical terracotta fins. Soft morning daylight, natural material colors, realistic shadows and construction details, eye-level 24mm architectural photograph, straight verticals. Clear foreground, middle ground and background. No text, logos, labels or watermarks.
Pick a model and start generating
These models are tuned for detail — open one and put this guide to work.
