Sunburst Style Spectrum: Four-Stage Illustration from Realistic to Flat

Anchored on the same subject, gpt-image-2.5 (Sunburst variant, precision-oriented) is used to demonstrate a continuous transition of illustration style from realistic to flat. The pipeline first generates a realistic baseline image with text-to-image, then uses that image as the reference image for three edits, advancing respectively to thick-paint painting, screen printing, and minimalist flat stages. In the final submission, the four images are combined into a style spectrum at the same aspect ratio. Sunburst's improvements in reference-subject consistency and repeated-edit stability keep the silhouette and pose of the same subject in a recognizable correspondence across the four generations, which is exactly the key to this kind of style-comparison work. It is suitable for illustration style exploration, brand visual style setting, art direction communication drafts, and proposal scenarios where a team needs to be shown "how the same subject appears at different levels of abstraction." Reference images are passed in via image_urls, and the aspect ratio and quality tier are explicitly fixed in base, ensuring the four images can be placed side by side for direct comparison.