Original document
GPT Image 2 · en
[INPUT IMAGE: USER_PHOTO] Use the person in the input image as the ONLY subject. Preserve their identity and facial features clearly.
Create a hyper-realistic high-fashion editorial photo inside a surreal 3D geometric “color box” room (a hollow cube / tilted cube set). Each render MUST randomly choose:
1) a bold single-color box (monochrome environment, vivid and saturated),
2) a dynamic “cool” fashion pose (gravity-defying or extreme stretch / leap / sideways bracing against the walls),
3) a dramatic camera angle (wide-angle 24–35mm equivalent, tilted horizon, strong perspective).
The subject appears full-body and sharp, wearing an avant-garde fashion styling that feels modern and editorial (clean silhouette, stylish layering, premium fabric texture). Keep clothing tasteful and fashion-forward. The subject’s pose should feel athletic, stylish, and unusual—like a magazine campaign shot.
Lighting: studio quality, crisp and cinematic; strong key light with controlled soft shadows, subtle rim light; realistic reflections and bounce light from the colored walls. Ultra-detailed skin texture, natural pores, realistic fabric weave, clean edges, high dynamic range.
Composition: subject centered with plenty of negative space and strong geometric lines; the box perspective frames the subject.
Color: the box color is a SINGLE bold color and MUST be different each run (random vivid hue). The subject’s outfit contrasts well with the box color.
Output: hyper-real, photorealistic, 8k detail, editorial campaign quality, sharp focus on subject, no motion blur, no distortion of face, natural proportions.How to use
- Copy the complete original prompt, including any parameters.
- Replace the subject, text and composition with your own requirements.
- Use a compatible external tool; check its current access conditions and review the result.
Before you start
- Check whether a reference image is required. A compatible model label does not mean this example was generated with that model.
- Keep names, figures and visible text accurate; results can vary between model versions.
