Create 250 AI-generated images

This project requires a highly standardized dataset of 250 distinct, fictional adult women (750 PNG images total) for private ML research. Each identity must be captured in three perfectly consistent views—a frontal headshot (1536 x 1536), a frontal full-body shot (1536 x 2304), and a strict 90-degree side profile (1536 x 2304). Every subject must maintain exact anatomical and stylistic consistency across all views using a reliable identity-locking or 3D workflow. The capture protocol mandates a non-sexualized, photorealistic studio setting with a plain warm-gray background, diffuse lighting, and a neutral, closed-mouth expression. Subjects must maintain a rigid, straight posture with arms slightly away from the torso, wearing identical matte dark-charcoal fitted athletic tops and leggings. The dataset must deliberately avoid uniform "model" bodies, requiring a balanced distribution instead across five distinct structural variations for frame, upper torso, waist-to-hip, and lower-body contours. Deliverables include organized per-identity folders with standardized JSON metadata, a master manifest, SHA-256 hashes, and full reproducibility documentation alongside complete commercial ML research rights. An additional, separately priced tier of 90 nuisance-robustness variants (minor lighting, background, or seed shifts) is required for 30 approved identities. The project will proceed through paid, iterative milestones (3 identities, then 10 identities) before full batch production.