From Realistic to Flat: gpt-image-2 Image-to-Image Style Spectrum Illustration

Starting from the same realistic reference image, use gpt-image-2's image-to-image editing capability to gradually move the image along the style spectrum of "realistic—semi-realistic—flat illustration." The reference image is generated by the pipeline and serves as the sole image input; the prompt only describes the direction of style transfer and the elements to preserve (composition, subject placement, color relationships), without altering the content itself. In base, quality is set to high and resolution to 2K, to observe how well details and color-block transitions are preserved at higher resolution; aspect_ratio is fixed at 4:5, close to the common layout of editorial illustration. It is suitable for design workflows that need to rewrite photographic material into a unified illustration language, such as exploring brand illustration guidelines, unifying the style of article illustrations, and cross-version visual iteration. Commercial images with typographic layouts, high-fidelity photorealistic rendering, and other use cases of this model family are equally applicable, but this entry focuses only on the style dimension comparison.

Style style-transfer · editorial-illustration · Flat design

Model GPT Image 2

Reference inputs

Reference images

Try it yourself

Prompt

Try this model

Perform image-to-image style transfer on a single reference photo, converting it into a flat editorial illustration. Preserve the subject silhouette, object positions, and the overall composition of the source image; gradually simplify photographic details into simplified geometric shapes, clean vector-like edges, and a limited palette of 5–6 desaturated colors with a single warm accent. Remove texture noise, gradients, and specular highlights; replace shadows with solid tonal shapes. 4:5 portrait composition, subject centered with ample negative space, suggesting soft, even lighting through flat fills rather than rendered light. Crisp edges, no outlines thicker than a hairline, no photorealistic grain, and no additional elements beyond the source scene.

Copy the full prompt and adapt it for your own generation.