
Original document
GPT Image 2 · en
Scene:
A square, straight-on architectural presentation image of a pencil facade sketch on clean white tracing paper. The crop is centered and evenly lit like a scan on a design studio desk, with soft graphite marks, faint construction lines, a visible horizon guide, and no perspective drama. The image should feel like the precise source drawing for a later photoreal render.
Subject:
A compact four-story corner apartment building sketched in graphite: ground-floor recessed entry at lower left, two shopfront bays at street level, three upper residential floors, a shallow roof terrace, and a narrow side return visible on the right. The facade has a regular window grid, slender balconies on the second and third floors, a vertical stairwell slot, and simple masonry panel divisions.
Important details:
Preserve strict architectural logic: stable horizon line, consistent floor heights, aligned window centers, believable balcony depth, clear street curb, and proportional columns. Use hand-drawn line weight variation without readable notes or labels. The drawing must be detailed enough to guide later rendering, but still unmistakably pencil, with no color fills, no collage materials, and no invented surrounding buildings.
Use case:
Sketch to Photoreal Render series study 1/8, serving as the structural reference for layout-faithful photoreal fanout renders.
Constraints:
No readable text, no watermark, no logos, preserve exact 1:1 aspect ratio, no blueprints, no ruler marks with numbers, no decorative frame, no extra floors, no exaggerated fantasy architecture, no people blocking the facade.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.





