Lighting is the single biggest tell between a portrait that reads as photography and one that reads as AI. Generic “studio portrait” prompts give you flat, omni-directional light no real photographer would set up: the giveaway is forehead, cheekbones, and chin all sitting at the same exposure. These 12 templates lock direction, hardness, and color temperature explicitly, so you can swap the subject and reuse the same lighting recipe across an entire series. Pair them with the broader realistic portrait prompts library.
TL;DR
- Always name three things: direction (camera-left at 45 degrees), hardness (soft 1.5m softbox vs. hard bare bulb), and key-to-fill ratio (3:1 flattering, 8:1 dramatic). Skip any one and the model averages to flat light.
- The recipes below are model-agnostic, but they land best on Flux 2 (strongest physical-light simulation as of June 2026) and Midjourney V8.1 (default since June 10, 2026). On Midjourney, add
--style raw --s 50to stop the engine from “prettifying” your shadows. - Add a texture word (
natural pore detail,light freckles) every time, or skin goes plastic. - Pick one direction per series and stick to it; multi-source recipes look fake because photographers rarely fire four heads at once.
Best for
- Editorial and author headshots that need a consistent lighting language
- Speaker and founder portraits across a brand site
- Series shoots where you want unity without re-prompting from scratch
- Replacing stock photography with on-brand AI portraits
- Practicing lighting setups before a real studio session
Which model handles lighting best (June 2026)
The templates are written in plain photographic language, so they transfer across engines. Engine choice mainly changes how literally shadow direction and ratio are obeyed.
| Model | Lighting strength | Notes for these prompts |
|---|---|---|
| Flux 2 Pro | Best optical-light realism | Released Nov 25, 2025; ~$0.055/image via API, 4MP output. Honors ratio and falloff most literally. Struggles only on multi-refraction (neon/wet-street) scenes. |
| Midjourney V8.1 | Polished, slightly idealized | Default since June 10, 2026; plans $10 Basic to $120 Mega ($8/$96 annual). Add --style raw and keep --s at 50-100 so it does not soften your shadow lines. |
| Nano Banana 2 (Google) | Very photoreal, fast/cheap | Top-tier skin and lighting in same-prompt tests; best speed-to-cost. Good default if you want volume. |
| GPT Image 2.0 | Strong all-rounder | Comparable photorealism to Nano Banana 2; convenient inside ChatGPT for iteration. |
1. Soft Window Light
realistic portrait of [subject], soft window light from camera-left at 45 degrees, gentle falloff on right cheek, mid-tone skin with natural pore detail, neutral white balance, 85mm f/1.8, shallow depth of field, sharp focus on near eye, calm expression2. Golden Hour Outdoor
realistic portrait of [subject], golden hour 30 minutes before sunset, warm rim light from back-right, soft fill from sky, cinematic warm-magenta cast, natural skin with light freckles, 35mm f/2.8, shallow depth of field, slight haze3. Beauty Dish
realistic studio portrait of [subject], beauty dish from above and slightly forward at 30 degrees, soft fill from below for catchlight balance, glossy skin highlights, dark gray seamless backdrop, fashion magazine style, 100mm f/4.04. Rembrandt Lighting
realistic portrait of [subject], Rembrandt lighting from camera-left at 45 degrees and slightly above eye line, characteristic triangular cheek highlight on shadow side, dark background, painterly mood, 85mm f/2.0, ratio 4:1 key-to-fill5. Split Lighting (Dramatic Half-Face)
realistic portrait of [subject], split lighting from camera-left at 90 degrees, half face in deep shadow with sharp shadow line down nose center, black background, high-contrast monochrome feel, 85mm f/2.8, ratio 8:1 key-to-fill, serious expression6. Loop Lighting (Most Universal Flattering)
realistic portrait of [subject], loop lighting from camera-left at 30-45 degrees and slightly above eye line, small nose-shadow loop on cheek not touching mouth line, soft 1.5m softbox quality, medium gray backdrop, 85mm f/2.5, ratio 3:17. Butterfly / Paramount Lighting
realistic glamour portrait of [subject], butterfly lighting from directly in front and above the camera at 45 degrees, symmetric butterfly-shaped shadow under the nose, soft fill from below to open under-eye, white seamless backdrop, 100mm f/4.0, classic Hollywood style8. Clamshell Lighting (Beauty / Cosmetic)
realistic beauty portrait of [subject], clamshell lighting: large softbox above at 30 degrees + silver reflector below at 30 degrees, even shadowless face, twin catchlights stacked vertically in the eye, glossy skin, 100mm f/5.6, beauty magazine style9. Hard Midday Sun
realistic portrait of [subject], hard direct midday sun from overhead-right, sharp-edged shadow under nose and chin, contrasty skin tones, dust particles visible, urban concrete background, 50mm f/4.0, documentary feel, no fill, embrace the contrast10. Low-Key Single Source
realistic portrait of [subject], single hard source from camera-left at 60 degrees, ratio 16:1 key-to-fill, deep shadows occupying 70 percent of frame, only the cheek and one eye fully lit, charcoal background, 85mm f/2.0, chiaroscuro mood11. Neon / Mixed Color Source
realistic portrait of [subject], magenta neon from camera-right and cyan neon from camera-left, color contrast across face, slight haze for color bleed, wet street reflection in background, 50mm f/1.8, cyberpunk editorial mood, mixed color temperature embraced not corrected12. Candlelight / Low Lux
realistic portrait of [subject], single candle flame as only light source held at chest level frame-left, warm 1800K color temperature, soft falloff to deep shadow on right side, visible grain consistent with high ISO, 50mm f/1.4 wide open, intimate quiet moodHow to refine
Always specify the lens focal length (85mm for compression and skin, 35mm for environmental context) and the explicit ratio (3:1 is flattering, 8:1 is dramatic); diffusion models default to flat ratios. Pick one direction (left, right, above) and stick with it across the series. On Midjourney V8.1, append --style raw --s 50; on Flux 2, the prompt is usually obeyed as written, so add less, not more. The AI realistic portrait workflow covers the full production loop including consistency across a series.
Common mistakes
- Plastic skin from no texture words; always add
natural pore detail,light freckles, orsubtle skin imperfections. - No lens specified, so the model picks a wide default that warps facial proportions.
- Conflicting light directions in the same prompt (
light from leftandrim from leftandkey from right). - Background brighter than the subject’s face, so the eye is dragged away from the catchlights.
- Forgetting the ratio; without
ratio 3:1the model averages to a flat omnidirectional look that screams AI.
FAQ
Which AI model is best for realistic portrait lighting in June 2026?
For literal control over shadow direction and key-to-fill ratio, Flux 2 Pro has the strongest physical-light simulation (released Nov 25, 2025, around $0.055 per image via API). Midjourney V8.1 (default since June 10, 2026) gives a more polished look but idealizes shadows, so pair it with --style raw. Nano Banana 2 and GPT Image 2.0 are excellent fast, low-cost defaults when you need volume.
What is a key-to-fill ratio and why does it matter?
It is the brightness difference between your main light (key) and the fill that opens the shadows. 3:1 (the key is three stops brighter region than the shadow side) reads as natural and flattering; 8:1 is dramatic; 16:1 is near-silhouette low-key. Diffusion models default to roughly 1:1 flat light, which is exactly what makes a portrait look AI-generated, so naming the ratio is the highest-leverage edit you can make.
Why does my AI portrait still look flat after I describe the light?
Three usual causes: you named a direction but no ratio, you used soft language (“nice lighting”) instead of a setup, or the model softened the shadow. Fix it by adding an explicit ratio, naming the modifier (soft 1.5m softbox vs. hard bare bulb), and on Midjourney lowering --s to 50 with --style raw.
What Midjourney settings give the most photorealistic lighting?
Keep stylize low. --s 50 to --s 100 with --style raw keeps skin reading as real rather than illustrated, and lets your stated shadow line and ratio survive. Pushing --s above 200 reintroduces the smooth, idealized Midjourney “look” that flattens lighting.
Can I use these AI portrait images commercially?
It depends on the tool’s license and tier, not on the prompt. Midjourney grants commercial rights on paid plans; Flux and Google’s image models have their own usage terms; always confirm faces and likeness rights before publishing. Check the vendor’s current terms, since image-AI licensing changed several times through 2025-2026.
Related
- Realistic portrait prompts — broader portrait library
- Headshot portrait prompts — corporate variant of the same recipes
- Environmental portrait prompts — extend the lighting recipes outdoors
- AI realistic portrait workflow — production tutorial
- Midjourney V8.1 version notes — official parameter and version reference
Tags: #Portrait #Realistic