OpenAI says GPT-Image-2.6 Sunburst and Flare add a complexity setting, markup-guided edits and SVG and animated GIF output in the Image API. This guide is our own method, written from OpenAI’s documentation. We have not benchmarked the models, so compare results on your own images.
Pick the model and complexity
OpenAI describes Sunburst as for detailed visual work and Flare as for fast iteration. A simple way to work:
- Draft with Flare to explore ideas quickly.
- Finish with Sunburst once the idea is settled and the layout matters.
- Match complexity to the job. OpenAI says lower settings suit focused edits like changing a background, and higher settings help with dense text, complex layouts and relationships between objects. It also notes higher settings can add latency and unwanted changes, so compare.
Prompts for high complexity
A one-page infographic about a five-step prompt workflow. Five numbered cards in a row, each with a short title and one line of readable text. Soft blue and white palette, plenty of margin, no logos.A cafe menu board with three sections (Coffee, Tea, Snacks), four items each, aligned prices on the right, chalk-style lettering on dark green.Prompts for low complexity (focused edits)
Keep the product and its shadow exactly as they are. Change only the background to a plain warm grey studio sweep.Markup-guided edits
OpenAI says you send a transparent PNG layer (markup) that shows where the change should happen. The layer must match the first input image’s width and height, and the prompt describes the change. Markup applies only to the first image; extra images are references. You cannot combine it with mask or prompt_rewriter.
Replace the object inside the circled area with a small potted plant. Leave everything outside the markup unchanged.Change the text inside the boxed area to "Open Late". Keep the font style and colour.SVG and animated GIF
OpenAI says both models support SVG and animated GIF output but the pages we read do not explain how to request them. Check the image generation guide first. Then try simple, flat subjects.
A flat icon of a lighthouse in three colours, simple shapes, no gradients, suitable as a scalable vector.Habits that help
- Say what must not change in every edit.
- Quote any text you want rendered.
- Change one thing at a time so you can see what the complexity setting did.
More in how to write better image prompts, GPT Image and Midjourney vs GPT Image.