prompt
Generate the image described by this JSON specification. Obey the
input_images role mapping exactly: image 1 controls camera/geometry
only; the remaining images are the appearance ground truth.
{
"task": "Clean studio \"silo\" product photograph for an e-commerce catalogue.",
"subject": "Sofa — Two-Tone Reclining Sofa",
"camera": "azimuth 305 deg around the subject, camera 7 deg above, looking slightly down. Matches image 1 exactly - keep the same viewpoint, framing, zoom and crop; do not re-frame.",
"input_images": {
"image_1": "Rough 3D turntable render - CAMERA & GEOMETRY AUTHORITY ONLY. Reproduce its exact camera angle, framing, perspective and aspect ratio. IGNORE its colours, materials, lighting, textures and low-poly / render artifacts.",
"images_2_to_4": "Real photographs of the actual product - APPEARANCE & STRUCTURE GROUND TRUTH. The output product must look IDENTICAL."
},
"style": "Ultra-high-end commercial studio product photography, sharp focus.",
"background": "100% PURE FLAT WHITE #FFFFFF AND ONLY WHITE. Only the subject + a single soft contact shadow beneath it.",
"output": "Output ONLY the silo photograph."
}