Realistic female portraits from AI tend to drift in two directions: plastic AI skin, and the generic “Instagram-pretty” look. Both are fixable with the right structural prompt. This article gives you 10 copy-ready templates covering the use cases people actually ship, the exact words that move output from “AI render” to “looks like a real photo,” and current (June 2026) notes on which model to run each prompt in.
TL;DR
- The three highest-weight words in any realistic portrait prompt are a lens spec (
85mm f/1.4), a film stock (Kodak Portra 400), and a light direction (north-facing window light). They outweigh every “beautiful / stunning” adjective. - For pure photorealism in June 2026, Flux 1.1 Pro renders skin pores and fabric weight most believably; Google Nano Banana 2 wins on 4K micro-detail; Midjourney V7 / V8.1 wins on aesthetic lighting and composition.
- For a consistent face across a series, use Midjourney Omni Reference (
--oref+--ow), not the old--crefflag, which V7 no longer accepts.
What separates great realistic portraits from “AI-pretty”
Three things, every time:
- Specific facial structure:
sharp cheekbones, prominent brow ridge, narrow jawbeatsbeautiful, pretty, gorgeous - Concrete skin description:
natural skin texture with visible pores, slight freckles across the nose - Real lens specs:
85mm f/1.4,50mm f/2,Kodak Portra 400instead of “high quality”
A high-quality realistic portrait prompt always uses these six layers:
- Subject: age, ethnicity if relevant, facial structure, hair, expression
- Skin: texture, freckles, makeup state
- Wardrobe: material, fit, color
- Lighting: direction, quality, temperature
- Lens: focal length, aperture, film stock
- Composition: head-shot / half body / full, angle, background
Which model for which look (June 2026)
The same prompt lands differently across tools. Match the brief to the model before you tune wording:
| Model (current version) | Best for | Photoreal skin | Notes |
|---|---|---|---|
| Flux 1.1 Pro | Real-person realism, “could be a real photo” briefs | Strongest pores + translucency | Default for documentary / headshot realism |
| Google Nano Banana 2 | Maximum micro-detail, 4K output | Sharpest fabric/skin grain | Best when you need texture at print size |
| Midjourney V7 / V8.1 | Aesthetic lighting, editorial composition | Excellent, slightly stylized | V8.1 (shipped April 30, 2026) renders ~4-5x faster than V7 |
| Seedream | Stylized portraits, multilingual text in-frame | Good | Strong if your prompt mixes Chinese/English text |
The templates below are written in plain natural language plus Midjourney-style --ar flags. They work as-is in Midjourney; drop the --ar suffix and they work in Flux, Nano Banana 2, and Seedream.
10 copy-ready prompt templates
1. Fresh natural
A 24-year-old East Asian woman, clean skin with subtle freckles, soft black hair down, light makeup, gentle smile, natural skin texture with visible pores, north-facing window light, 50mm f/1.8, Kodak Portra 400, half body, --ar 4:5Best for personal brand, lifestyle, podcast covers.
2. High fashion editorial
A 27-year-old female model, sharp angular features, sleek straight black hair, neutral cool expression, oversized blazer with sharp shoulders, single beam top light, smoke haze in air, plain dark grey backdrop, 85mm f/4, Hasselblad, --ar 4:53. Retro film aesthetic
A 28-year-old woman with soft wavy auburn hair, light freckles, vintage red lipstick, cream silk blouse, late 70s editorial vibe, warm tungsten lighting, slight film grain, 50mm f/2.0, Kodak Gold 200, --ar 4:54. Edgy fashion
A 25-year-old woman, sharp cheekbones, slicked-back ponytail, dramatic eyeliner, oversized leather trench coat, single hard beam light from above, deep shadows under jaw, magazine cover quality, 85mm f/2.8, --ar 4:55. Bridal editorial
A bride in a soft ivory silk gown with delicate lace details, gentle wind in her hair, holding a small bouquet, golden hour backlight in a Tuscan field, dreamy soft focus background, 85mm f/1.4, --ar 4:56. Eastern beauty / Chinese aesthetic
A 26-year-old Chinese woman in modern qipao with subtle floral embroidery, long black hair tied loose, calm intelligent eyes, ink-painting backdrop, soft morning light, 85mm f/2, Fuji 400H, half body, --ar 4:57. Cinematic film still
A 32-year-old woman looking off-frame at city lights, brooding expression, anamorphic 35mm cinematic look, dim practical streetlight from behind, magenta neon reflection on cheekbone, shallow depth of field, --ar 21:98. LinkedIn / business headshot
A 30-year-old professional woman, calm confident expression, natural skin texture, light makeup, tailored navy blazer over white shirt, soft Rembrandt key light from camera-left, plain neutral grey backdrop, 85mm portrait, f/2.8, sharp eyes, half body, --ar 4:59. Athletic / sportswear campaign
A 26-year-old athletic woman, toned shoulders, hair tied up, focused expression, sweat highlight on skin, black sports bra and bike shorts, dramatic side rim light, deep shadow opposite side, low angle, hands wrapped in boxing tape, --ar 4:510. Golden hour outdoor portrait
A 24-year-old woman with shoulder-length wavy hair backlit by golden hour, soft warm rim light, slight smile looking off-camera, cream linen dress, wheat field background, gentle warm bokeh, 85mm f/1.4, Kodak Portra 800, --ar 4:5Per-use-case tuning
- Personal brand / LinkedIn: 85mm f/2.8, plain grey backdrop, soft Rembrandt — keep it sober.
- Magazine cover: hard single beam light, dark background, sharp wardrobe geometry.
- Bridal: golden hour backlight, 85mm f/1.4, soft fabric in motion.
- Edgy fashion: top light, deep shadows, slicked hair, neutral cool expression.
- Retro / film: name the film stock (Portra 400, Gold 200, Ektar 100).
Common mistakes
- Using empty adjectives (
beautiful,gorgeous,pretty) — they don’t anchor anything - Forgetting skin description — model defaults to plastic skin
- No lens spec — model defaults to oversharpened “AI” look
- Specifying both
model-likeandnatural— contradictions confuse the model - Generic background
studio— sayplain neutral grey backdropinstead
How to keep one face consistent across a series
If you’re shooting a series for the same person:
- Lock the structural block. Copy the subject + lens + lighting lines verbatim into every prompt; only change pose, wardrobe, and scene.
- Use Midjourney Omni Reference, not
--cref. As of V7 the old--crefflag is gone; it errors or is ignored. Drop a strong front-facing portrait into the Omni Reference bin (web) or pass--oref [image-url](Discord), then set strength with--ow. Default is--ow 100; keep it below400or results get unpredictable. Around--ow 70keeps the face but lets you change the outfit. - In Flux / SDXL, train a LoRA on 15-30 photos of one face for the tightest identity lock.
- Keep the same color palette and film stock in every prompt so the set reads as one shoot.
See How to Keep AI Image Style Consistent for the full workflow.
FAQ
Q: Which model gives the most photorealistic portraits in 2026? A: For “could pass as a real photo,” Flux 1.1 Pro renders skin pores and fabric most convincingly; Google Nano Banana 2 is sharpest for 4K detail. Midjourney V7 (or the faster V8.1) wins on lighting and composition but skews slightly stylized. See Midjourney’s official docs for version flags.
Q: Why does the face still look “AI plastic”?
A: Add natural skin texture with visible pores, subtle freckles and use real lens specs. Delete every flawless skin / perfect skin word — they push the model back toward plastic.
Q: How do I avoid the “Instagram filter” look?
A: Name a film stock (Portra 400, Gold 200, Ektar 100) instead of “high quality,” and drop vibrant, saturated, glowing.
Q: Can I prompt a specific real person?
A: No. Current models block named-celebrity and real-person generation. Describe facial structure (sharp cheekbones, narrow jaw, monolid eyes) instead of a name.
Q: Best aspect ratio for portraits?
A: 4:5 for social and editorial, 2:3 for full-length, 21:9 for cinematic film stills.
Q: Why are the eyes always slightly off?
A: Add sharp eyes, eye contact with camera or looking directly at the lens. Models default to a drifting gaze unless you pin it.