GPT Image 2.5
GPT Image 2.5
New

Generate and edit images with consistent subjects, accurate text, and up to 4K detail.

Flare is the fast default for everyday generation and editing. Sunburst spends longer per image to hold fine detail and the tightest edit control. Both cost the same.

Text-to-Image generates from a prompt only. Edit transforms one or more uploaded images according to your prompt.

Describe the subject, composition, exact text content, color palette, and style. Multi-part prompts and quoted in-image text work best.

Output canvas. Larger canvases use more tokens. Auto lets the model pick from the prompt or reference images.

Higher quality adds detail, latency, and tokens. High is the default; Extra High and Maximum are for fine detail at large sizes.

Transparent returns a real alpha channel — pair it with PNG or WebP.

1

Choose 1–4 variants. Each image is billed on its own.

Sign in to spend tokens

Prompt is required

Your generated images will appear here