Use this when an AI portrait looks waxy, over-lit, perfectly centered, or detached from its environment. The main copy box is intentionally compact: it is the generation version. If you already have a permitted reference image, use the separate edit version below so the model is not asked to generate and preserve at the same time.
The two panels came from the same filled version of the final compact prompt, executed twice. Both results keep the bookstore legible, give the subject a real task, and render linen, wood, and worn books with useful texture. They do not return the same fictional person, pose, outfit, shop, or lighting balance. For generation, the template controls direction; it is not a seed, identity lock, or deterministic scene specification.
Open the exact filled prompt used for both panels
Create one candid, photorealistic smartphone snapshot of one original fictional 47-year-old Middle Eastern woman restocking paperbacks in a neighborhood used bookstore under mixed late-afternoon window light and ordinary overhead fluorescent light. Compose for vertical 2:3 and emphasize a rumpled linen shirt, worn book covers, and wooden shelving. Use eye-level standard-wide framing from a plausible distance, with no extreme wide-angle facial distortion. Keep the composition slightly imperfect and meaningful surroundings legible; do not use synthetic portrait blur. Use ordinary automatic exposure, scene-appropriate white balance, restrained phone processing, and physically consistent shadows and reflections. Retain natural facial characteristics with subtle skin microtexture and local tonal variation, natural lip texture, irregular hair edges, believable material weight, seams, folds, and object contact. The pose, gaze, hands, and clothing tension must agree with the action. Do not beautify, smooth skin, glamorize, over-sharpen, add cinematic grading, exaggerate pores, or invent blemishes, scars, illness, injury, fatigue, asymmetry, or age. No influencer pose, studio light, text, logo, trademark, watermark, fake camera interface, RAW claim, or metadata claim. Return one coherent review candidate, not a collage.What this prompt controls
The prompt distributes realism across five systems: capture, face, hair, materials, and behavior. That matters because “add pores” cannot repair cinematic blur, flattering studio light, stiff hands, clean showroom backgrounds, or a presentation pose. The prompt also avoids claiming a literal iPhone, RAW file, DNG, or EXIF history. It asks for visible appearance, not invented provenance.
| Layer | Primary control | Common failure |
|---|---|---|
| Capture | Plausible distance, auto exposure, legible context | Beauty light, fake bokeh, facial distortion |
| Face | Retain natural characteristics and restrained microtexture | Porcelain skin or deliberately manufactured defects |
| Hair | Irregular edges and individual strands | Helmet-like silhouette or noisy flyaways everywhere |
| Materials | Weight, seams, folds, contact, reflection | Clothing painted onto the body |
| Behavior | One small action with matching gaze and hands | Default influencer pose |
Reference-edit version
Attach only an image you are permitted to edit. Replace the bracketed treatment note, then keep the preserve list short and visible. This version does not ask the model to invent a new subject.
Image 1 is the sole edit target. Make only its capture treatment feel like an everyday phone photo without retouching; do not turn it into a beauty portrait.
Change: [ONE OBSERVED TREATMENT PROBLEM]. Use ordinary auto exposure, scene-appropriate white balance, restrained skin microtexture, natural lip texture, irregular hair edges, believable fabric and object contact, and meaningful environmental focus.
Preserve exactly: the same adult identity, existing age cues and natural facial characteristics, expression, gaze, hair, body proportions, pose, hands, wardrobe, crop, camera angle, scene geometry, objects, and lighting direction.
Never add a facial mark, health signal, injury, tiredness cue, asymmetry, makeup change, age change, prop, or background feature that is absent from the source. No skin smoothing, artificial subject blur, glamour light, cinematic grade, extreme wide-angle distortion, text, logo, or watermark. If a protected detail moves, reject the result and restart from Image 1.Add only the module your scene needs
| Situation | Optional line |
|---|---|
| Low or mixed indoor light | Use ordinary mixed window and ceiling light with believable color crossover; do not neutralize every surface into studio white. |
| Curly or textured hair | Preserve the curl pattern and irregular outer edge; show individual curl groups without turning the hair into uniform strands. |
| Rain, glass, or metal | Make wet reflections and hard-surface highlights agree with the same scene light; avoid decorative shine. |
| Recurring drama character | Use the accepted character reference and trait inventory; this prompt controls one frame, not continuity across episodes. |
Use a short test loop
- Generate one baseline. Name the visible failure: smooth skin, artificial blur, staged behavior, wrong material response, or lighting inconsistency.
- Change one layer. Do not add every optional module at once.
- Review the full frame. Check hands, object contact, text-like artifacts, reflections, depth, and protected details—not only the face.
- Restart when an invariant moves. Return to the accepted source instead of stacking edits on a drifting candidate.
The retained evidence includes all eight call records, seven distinct exact prompt texts, hashes, dimensions, rubric decisions, and one disclosed failed edit. Open the portrait realism test record. For the complete reasoning, comparison board, diversity cases, and twelve-question review method, read How to Make AI Portraits Look Real Without Plastic Skin.
