Lighting Techniques: 10 Methods Compared

0. Why lighting is the single biggest mood lever

If you had to choose between writing a detailed subject prompt and writing a detailed lighting prompt, pick lighting every time. On modern AI models, the lighting tags you use set the palette, the contrast, the skin tone, and the emotional register of the scene. Below are ten techniques — the ones you'll reach for 95% of the time. Each comes with a one-line recipe you can copy into Promgrammer.

sidelightingSide Lighting

+Add

backlightingBacklighting

+Add

uplightingUplighting

+Add

butterfly lightingButterfly Lighting

+Add

Shared prompt

masterpiece, best quality, amazing quality, very awa, 1girl, solo, mature female, adult woman, short black hair, brown eyes, white blouse, upper body, dark studio, black background, <tag slot>

Negative prompt

school uniform, pleated skirt, blazer, serafuku, child, loli, teenage, bad quality, worst quality, worst detail, sketch, censor, jpeg artifacts, watermark, signature, text, extra limbs, bad hands, bad anatomy

Model

rinFlanimeIllustrious_v30 · euler_ancestral/normal · steps 28 · CFG 5 · 896×1152 · seed 777002

This comparison was measured on a single Stable Diffusion family checkpoint. The same tags can behave differently on NovelAI, Midjourney or another checkpoint — read it as a reference point, not a verdict.

moonlightMoonlight

+Add

fluorescent lightingFluorescent Lighting

+Add

stark white lightStark White Light

+Add

neon glowNeon Glow

+Add

Shared prompt

masterpiece, best quality, amazing quality, very awa, 1girl, solo, mature female, adult woman, short black hair, brown eyes, white blouse, upper body, plain interior wall, <tag slot>

Negative prompt

school uniform, pleated skirt, blazer, serafuku, child, loli, teenage, bad quality, worst quality, worst detail, sketch, censor, jpeg artifacts, watermark, signature, text, extra limbs, bad hands, bad anatomy

Model

rinFlanimeIllustrious_v30 · euler_ancestral/normal · steps 28 · CFG 5 · 896×1152 · seed 777002

This comparison was measured on a single Stable Diffusion family checkpoint. The same tags can behave differently on NovelAI, Midjourney or another checkpoint — read it as a reference point, not a verdict.

1. Rim light (backlighting, from behind)

Separates subject from background with a bright halo along the silhouette. Good for dark or busy backgrounds.

backlighting, from behind, dark background

Caveat — On booru-trained Stable Diffusion checkpoints, backlighting is the token that carries this look. In our test on rinFlanimeIllustrious v3.0 (seed 777002, dark background) both rim light and edge lighting rendered a flat, front-lit portrait with no contour light, while backlighting with from behind produced the halo immediately — see the comparison above. Generators that read plain English, such as NovelAI or Midjourney, understand rim light directly, so keep it there. Once the effect fires, it can overwhelm mid-tones; pair it with a weaker key light to avoid a cutout look.

2. Rembrandt lighting (rembrandt lighting)

Classic portrait technique — a triangular highlight under the shadow-side eye. Dramatic but flattering.

rembrandt lighting, 45 degree key light, soft shadow, dark background, portrait

Caveat — Requires the subject facing slightly away. On anime checkpoints, the effect flattens — prefer SDXL realistic.

3. Golden hour (golden hour)

Warm orange-gold light from a low sun (just after sunrise or before sunset). Signature of romantic outdoor photography.

golden hour, warm sunlight, long shadows, lens flare, backlit hair

Caveat — Adds a warm cast to skin tones; may push pale subjects to orange. Balance with cool-toned wardrobe.

4. Blue hour (blue hour)

The 20 minutes after sunset when the sky is deep blue. Good for moody cityscapes and windows lit from inside.

blue hour, twilight, cool color temperature, city lights, soft ambient

Caveat — Low light — combine with practical sources (street lamps, windows) to avoid flat scenes.

5. Studio lighting (studio lighting)

Controlled multi-source setup — key + fill + rim. Default for product and fashion photography.

studio lighting, softbox key light, fill light, backlighting, white backdrop

Caveat — Can look too clean/commercial. Add subtle grain or shadow to feel photographic rather than ad-like.

6. Volumetric (god rays) (volumetric lighting)

Visible light beams cutting through atmosphere (dust, fog, smoke). The defining look of fantasy and sci-fi scenes.

volumetric lighting, god rays, dust particles, forest, sun piercing through trees

Caveat — Needs a medium (fog, mist) to be visible. Without it the model may produce painterly stripes instead.

7. Film noir / chiaroscuro (chiaroscuro)

High-contrast shadow play — deep blacks, narrow strips of light. Classic detective, horror, drama.

chiaroscuro, high contrast, venetian blind shadows, single light source, black and white

Caveat — Loses subject detail. Use shallow blind patterns to retain facial features.

8. Backlight (backlight)

Light from behind the subject — silhouette or strong hair rim. Dramatic, emotional.

backlight, silhouette, bright rim on hair, hazy atmosphere

Caveat — Face becomes dark. For portraits, add a fill light or use reflective surfaces (water, glass) to bounce light.

9. Soft box / diffused (soft lighting)

Low-contrast, even lighting with feathered shadows. Good for beauty, food, slice-of-life.

soft lighting, diffused light, gentle shadows, pastel palette, even exposure

Caveat — Can feel flat. Add a subtle key direction to keep depth.

10. Practical / ambient (practical lighting)

Light motivated by in-frame sources — lamps, neon signs, fire, screens. Feels lived-in and cinematic.

practical lighting, warm lamp glow, phone screen illumination, window light, cozy interior

Caveat — Source placement matters — position the subject where the light naturally falls.

11. Neon (cyberpunk) (neon lighting)

Saturated colored lights (cyan, magenta, pink) characteristic of cyberpunk and 80s aesthetics.

neon lighting, cyan and magenta, wet streets, reflections, night

Caveat — Easily over-saturates skin. Keep one dominant color and a secondary accent rather than three equal colors.

11. Stacking and combinations

Lighting tags compose well when they don't fight each other. Compatible pairs:

• • golden hour + backlit — outdoor portrait with hair glow
• • blue hour + practical lighting — city dusk with window/lamp motivation
• • volumetric lighting + backlighting — fantasy hero shot
• • studio lighting + backlighting — commercial portrait
• • neon lighting + rainy streets — cyberpunk default

Incompatible — avoid stacking:

• • soft lighting + chiaroscuro (contradictory contrast)
• • studio lighting + practical lighting (conflicting source logic)
• • golden hour + blue hour (can't be both)

chiaroscuroChiaroscuro

+Add

rembrandt lightingRembrandt Lighting

+Add

silhouette lightingSilhouette Lighting

+Add

saturated red lightSaturated Red Light

+Add

Shared prompt

masterpiece, best quality, amazing quality, very awa, 1girl, solo, mature female, adult woman, short black hair, brown eyes, white blouse, upper body, plain interior wall, <tag slot>

Negative prompt

school uniform, pleated skirt, blazer, serafuku, child, loli, teenage, bad quality, worst quality, worst detail, sketch, censor, jpeg artifacts, watermark, signature, text, extra limbs, bad hands, bad anatomy

Model

rinFlanimeIllustrious_v30 · euler_ancestral/normal · steps 28 · CFG 5 · 896×1152 · seed 777002

This comparison was measured on a single Stable Diffusion family checkpoint. The same tags can behave differently on NovelAI, Midjourney or another checkpoint — read it as a reference point, not a verdict.

god raysGod Rays

+Add

volumetric lightingVolumetric Lighting

+Add

bokehBokeh

+Add

dreamy lightingDreamy Lighting

+Add

Shared prompt

masterpiece, best quality, amazing quality, very awa, 1girl, solo, mature female, adult woman, short black hair, brown eyes, white blouse, upper body, plain interior wall, <tag slot>

Negative prompt

school uniform, pleated skirt, blazer, serafuku, child, loli, teenage, bad quality, worst quality, worst detail, sketch, censor, jpeg artifacts, watermark, signature, text, extra limbs, bad hands, bad anatomy

Model

rinFlanimeIllustrious_v30 · euler_ancestral/normal · steps 28 · CFG 5 · 896×1152 · seed 777002

This comparison was measured on a single Stable Diffusion family checkpoint. The same tags can behave differently on NovelAI, Midjourney or another checkpoint — read it as a reference point, not a verdict.

12. Model notes

• SDXL / RealVis: Responds best to photographic lighting vocabulary (softbox, rim, golden hour).
• Animagine / NovelAI: Uses Booru-style tokens (dramatic lighting, backlight, god rays). Rembrandt and chiaroscuro are weaker.
• Flux.1: Natural language. Write sentences like "soft window light from the left, long shadow across the face".
• Midjourney: Leans cinematic by default; lighting tags have strong effect plus you can layer on --style raw for realism.

rim lightRim Light

+Add

edge lightingEdge Lighting

+Add

backlighting, from behindBacklighting + from behind

hard lightingHard Lighting

+Add

Shared prompt

masterpiece, best quality, amazing quality, very awa, 1girl, solo, mature female, adult woman, short black hair, brown eyes, white blouse, upper body, dark studio, black background, <tag slot>

Negative prompt

school uniform, pleated skirt, blazer, serafuku, child, loli, teenage, bad quality, worst quality, worst detail, sketch, censor, jpeg artifacts, watermark, signature, text, extra limbs, bad hands, bad anatomy

Model

rinFlanimeIllustrious_v30 · euler_ancestral/normal · steps 28 · CFG 5 · 896×1152 · seed 777002

This comparison was measured on a single Stable Diffusion family checkpoint. The same tags can behave differently on NovelAI, Midjourney or another checkpoint — read it as a reference point, not a verdict.