curl https://api.tokport.io/v1/images/generations \
-H "Authorization: Bearer $TOKPORT_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"gpt-image-2","prompt":"Create one portrait diptych in a selected 2:3 output size. Use ONLY the photographic upper panel of the supplied reference as the content source. Ignore the existing geometric lower panel and design a new geometric interpretation rather than copying it.\n\nDivide the final canvas by a straight horizontal line into exactly two equal-height panels, with no border. TOP: show the same recognizable bridge, arch, cables, river, reflections and sunset. Preserve their key spatial relationships and natural photographic appearance. Crop or extend the background to fit; do not stretch or warp the bridge. Do not add geometric shapes or print texture to this top panel.\n\nBOTTOM: reinterpret that same bridge scene as an original flat geometric screen print. Preserve the arch, cable rhythm, horizontal deck, sun and reflected shapes as recognizable correspondences. Derive circles, arcs, rectangles and parallel lines from the actual scene, not unrelated decoration. Use 3 to 5 source-derived colors: warm ivory, deep blue, muted teal and a restrained sunset-orange accent. Use asymmetrical balance, about 30 percent clean negative space, crisp flat color and subtle paper grain and slight ink misregistration. Avoid gradients, 3D rendering and excessive distress.\n\nThe lower panel must be a new design, not a mirror, filter or duplicate of the photo. No title, lettering, signature, watermark, logo or invented objects. English adaptation inspired by LLL (@allen79005518); the reference is the publicly shared comparison example, not the unpublished original source photo.","size":"1024x1536","n":1,"response_format":"url"}'