GPT Image Prompt
Same exact real person from the active LoRA identity reference images, not a similar Asian woman and not a newly invented face. Preserve her unmistakable facial identity first: same oval face proportions, same light skin tone, same dark brown eyes, same eye spacing and almond eye shape, same eyelid structure, same nose bridge and nostril shape, same lips, same jawline, same cheek structure, same chin, same forehead proportions, same natural facial softness, and same recognizable presence from the reference. Preserve her exact full-body identity from the reference: same shoulder width, shoulder-to-head relationship, torso length, torso-to-leg balance, waist thickness, ribcage-to-waist transition, hip width, hip-to-thigh transition, thigh volume, knee-to-ankle scale, leg length ratio, and body mass distribution. Do not replace her with a cuter, rounder, younger, more generic, fitness-influencer, or K-pop-like substitute.
Her hair is the same dark brown hair from the reference, styled into two natural braids under a beige baseball cap, without changing her facial identity. She wears the beige cap with front patch, ribbed light gray-beige sleeveless cropped athletic top with diagonal wrap-like seam, dark taupe-gray high-waist leggings, beige striped patterned socks, and small rings. The outfit must sit on the exact LoRA body geometry without changing her waist, hips, thighs, torso ratio, or leg length.
She stands full-body on a pink exercise mat in a simple bright indoor room, feet close together and naturally grounded, torso mostly front-facing with a slight turn. Both elbows are bent outward, both hands holding the lower sides of her cropped top near the waist with realistic grip pressure, natural finger spacing, visible small rings, believable wrist angles, and real fabric tension. Her expression is casual and alive: looking off to her left under the cap brim with relaxed eyes and a small natural open smile, mildly pleased, not blank, not stiff, not seductive, not commercial, not mannequin-like.
All expression, hairstyle, cap, outfit, pose, lighting, room, camera, and realism behavior remain subordinate to the active LoRA identity reference. The final subject must still be unmistakably the exact same woman from the reference images, preserving both face and body identity. Do not generate a similar-looking woman. Do not use the previous failed output as identity reference. Do not copy the failed generated face. Do not make her face rounder, eyes larger, cheeks fuller, nose smaller, lips cuter, or expression more influencer-like than the reference.
Casual vertical iPhone-style full-body photo from realistic smartphone distance, ordinary handheld framing with slight asymmetry, simple pale wall, visible baseboard, tiled floor, pink mat, soft even indoor light, mild cap shadow across upper face, realistic indoor exposure, slight smartphone smoothing, subtle compression softness, uneven sharpness, no cinematic polish, no DSLR blur, no beauty-filter skin, no over-clean HDR. Maintain believable foot grounding, contact shadows under socks and mat, realistic scale between subject, mat, wall, baseboard, and floor.
Avoid: wrong face, generic Asian face, rounder face, K-pop idol face, doll-like beauty, influencer fitness redesign, copied failed-output identity, facial beautification, face slimming, eye enlargement, nose reshaping, lip enhancement, waist slimming, hip narrowing, thigh reduction, leg elongation, different torso-to-leg balance, different body mass distribution, fashion-model proportions, seductive pose, blank stare, dead eyes, frozen mouth, commercial smile, cinematic lighting, luxury editorial look, artificial symmetry, plastic skin, CGI texture, fused fingers, duplicated fingers, broken wrists, unrealistic grip, floating feet, disconnected shadows, or any camera perspective that changes the LoRA reference identity.
How to use: Copy and paste this prompt into ChatGPT, Gemini, Nano Banana, or your favorite AI image tool. If it mentions LoRA, reference image, uploaded photo, or identity reference, upload your own photo first. No reference photo? Replace the character with a clear age, gender, and nationality, such as “25-year-old Japanese woman” or “30-year-old American man”.