Промт для восстановления портретных фото без искажения лиц

This new prompt is specifically designed to address the user’s complaint that **"the faces were changed and do not look like the original."** 

The primary failure in AI restoration is often "hallucination," where the AI replaces blurry features with generic high-definition faces. This prompt adds strict "Anti-AI-Hallucination" safeguards and shifts the focus from *recreating* features to *upscaling* the original pixels.

***

### Improved Restoration Prompt:

**GOAL:** Perform a grain-preserving, high-definition upscale of this sepia photograph. The absolute priority is **Facial Identity Fidelity.** 

**CRITICAL: NO FACE SWAPPING OR RE-GENERATION**
The user noted that previous attempts changed the subjects' identities. You must NOT use generative "face restoration" models that invent new features (different eyes, noses, or teeth). 
- **Preserve the exact geometry:** The distance between eyes, the specific curve of the man’s smile, and the unique lip shapes of all five subjects must remain 100% identical to the source.
- **Micro-Detail Enhancement:** Instead of replacing faces with generic ones, sharpen the existing pixels. Use "De-blur" and "Luminance Noise Reduction" rather than "Generative Fill."
- **Gaze and Expression:** The man’s squint and the children's specific glances must be maintained exactly. Do not "fix" lazy eyes or "improve" expressions.

**UNIFORM IMAGE SHARPENING (ANTI-BIAS)**
- **Background Integrity:** The AI must apply the same sharpening intensity to the stone wall background as it does to the faces. Restore the mortar lines and rough stone textures to eliminate the hazy, washed-out look.
- **Clothing Micro-Textures:** Focus heavily on the weave of the man's jacket, the lapels of the woman's coat, and the wool texture of the children's headwear. 
- **Equal Focus Rule:** Every inch of the frame—from the girl’s legs on the left to the baby’s hood on the right—must have uniform clarity. No depth-of-field blur.

**DAMAGE REMOVAL & TONAL REPAIR**
- **Manual Spot Removal:** Remove the white vertical scratches across the man’s arm and the baby’s forehead. Remove the mold/dust spots on the stone wall and the white chemical artifacts on the clothing.
- **Contrast Balance:** Deepen the shadows to recover the lost definition in the woman’s hair and the man’s clothing. Ensure the highlights on the skin do not become "blown out" or plasticky.
- **Strict Sepia Preservation:** Maintain the original monochrome sepia palette. Do not introduce any blue tints, skin-tone colorization, or grayscale shifts.

**MANDATORY CHECKLIST BEFORE FINAL RENDER:**
1. **Identity Check:** Does the man still look like the exact same individual, or has the AI "beautified" him into a stranger? (Failure = changed features).
2. **Texture Check:** Is the stone wall as sharp as the faces? (Failure = blurry background).
3. **Artifact Check:** Are all those white specks and vertical scratches gone? (Failure = remaining damage).
4. **Expression Check:** Is the man's smile identical in shape to the original? (Failure = generic AI smile).

**DO NOT REIMAGINE. DO NOT REGENERATE. ONLY RESTORE AND UPSCALE.**

ANTI-BIAS ALERT: You naturally focus restoration/colorization on FACES while leaving background, clothing, and foreground under-processed. This is the #1 quality failure. Consciously counteract this.

MANDATORY UNIFORM QUALITY CHECK:
- Background sharpness = Face sharpness (NOT less)
- Clothing detail = Face detail (fabric weave, buttons fully visible)
- Foreground objects = Faces (flowers, grass, objects NOT blurry)
- Edges/corners = Center of image (NOT softer or less restored)
Any area less sharp or colorized than faces is a FAILURE — redo that area.

ABSOLUTE RULES — NEVER VIOLATE:
- YOU ARE RESTORING/COLORIZING AN EXISTING PHOTO — not generating a new one. Every pixel must correspond to the original image.
- DO NOT recreate, reimagine, or regenerate any part of the scene — work only with what is already in the photo
- DO NOT add, invent, or hallucinate any part of the image that is not in the original
- DO NOT change the framing or composition — keep it exactly the same
- DO NOT add any new damage, folds, creases, tears, dark borders, vignettes, or artifacts
- DO NOT change, replace, or alter any person's face, body, or identity
- DO NOT change facial expression (мимика лица) — mouth shape, lip curve, and overall expression must be IDENTICAL to the original
- DO NOT change eye gaze direction — eyes must look in the exact same direction as in the original photo; do NOT redirect a gaze or correct an asymmetry that exists in the original
- DO NOT add or remove any person or object
- DO NOT modernize clothing, hairstyles, or any visual elements
- Keep the exact same framing, crop, and aspect ratio as the original
- SKIN COLOR: Eastern European skin = warm-neutral to cool-neutral, medium tone. NEVER red or orange. NEVER ghostly pale or washed-out. NEVER airbrushed or plastic. DO NOT add acne, spots or marks not in the original
- COLOR SATURATION: Keep overall saturation MODERATE — natural analog film look. NOT vivid digital. NOT pale or desaturated`
- CRITICAL: Preserve original tonal character EXACTLY — black and white photos MUST stay black and white. Sepia photos MUST stay sepia. DO NOT add ANY color whatsoever.
- OUTPUT MUST BE GRAYSCALE if the input is grayscale — no warm tones, no blue tints, no color cast of any kind
- Apply uniform sharpness to the ENTIRE image — background and foreground equally. NO background blur
- DO NOT smooth, airbrush, or beautify faces — preserve natural skin texture exactly as in the original