[Experiment] Night Light Source Comparison — How Do Neon, Street Lamps, and Moonlight Change Skin Color?

Conclusion First

All 7 light source specifications (neon, street lamps, headlights, illuminations, moonlight, storefront lights) produced clear differences in composition, background, and light color across the same seeds. Particularly illuminations, neon, and street lamps changed the background objects themselves, and differences in skin color cast were confirmed.

• Conditions with large background object changes: Neon (3/3 images: neon signs appeared), Street lamps (3/3 images: lamp poles appeared), Illuminations (3/3 images: fairy lights/trees appeared), Headlights (3/3 images: vehicles appeared)
• Conditions with confirmed skin color cast: Neon (pink/blue reflection), Street lamps (orange warm cast), Illuminations (warm soft light from multiple point sources)
• Conditions with confirmed color temperature change: Moonlight (slightly blue-white tone), Headlights (strong white light)
• Combined light sources (2–3 specified simultaneously): 4 conditions × 3 seeds = all 12 images showed all specified light sources without any dropping out. Semantically similar light sources tend to merge into a single scene

What This Article Covers

• How specifying light source types in night scenes changes the image atmosphere
• How light color (warm, cool, multicolor) is reflected in skin depiction
• Whether shadow direction and intensity change by light source type
• Rendering stability of each light source (how many of 3 images match intent)
• Whether all light sources are reflected when multiple are specified simultaneously

Experimental Conditions

Base Prompt

Only {LIGHTING} is swapped per condition. Pose, outfit, and framing are all fixed.

Evaluation Criteria

• Light color: Is the color temperature change appropriate for the light source type?
• Shadow direction and intensity: Are shadows drawn consistent with light source position?
• Skin reflection: Is the light color reflected on skin (color cast)?
• Background consistency: Does the background atmosphere match the light source?

Results for Each Condition

B00: Control — No Light Source Specified

Baseline without specifying a light source for a night city scene.

seed=42	seed=123	seed=789

Observations

• Light color: All 3 images show urban nightscape ambient light at intermediate color temperature, not biased toward any particular source. seed=42 shows the subject in the middle of the road with store lighting from both sides. seed=123 is near an intersection, slightly darker overall. seed=789 is also near an intersection with car lights visible behind but no direct color cast on the subject
• Shadow direction: All 3 images show diffuse light from multiple ambient sources (stores, street lamps, signs) with no clear shadows
• Skin reflection: All 3 images show uniform color with no notable color cast — the default lighting state for an urban nightscape
• Background consistency: All 3 images show a Japanese urban commercial area at night with typical store lighting and signage. No specific light source objects are emphasized

B01: Neon Signs — neon sign lighting, colorful neon glow

seed=42	seed=123	seed=789

Observations

• Light color: All 3 images show numerous red, pink, blue, green, and purple neon signs in the background — a clearly different multicolor light environment from the control. seed=42 shows neon signs on buildings to both sides with pink/blue bokeh light reflecting on the ground. seed=789 shows the densest neon-covered street scene
• Shadow direction: Diffuse light similar to control with no directional shadows — neon multicolor prevents clear shadow direction
• Skin reflection: seed=42 shows faint pink/blue color cast from neon on arms and legs. seed=123 and 789 show neon light in the surroundings but with weak direct skin color reflection. Clear skin color cast confirmed in 1/3 images
• Background consistency: Neon signs are rendered in the background across all 3 images, with the light source description reflected as background objects. The background is completely different from the control

B02: Street Lamps — street lamp lighting, warm orange glow

seed=42	seed=123	seed=789

Observations

• Light color: All 3 images show a street lamp pole rendered close to the subject with orange-glowing lamp. Slightly warmer color temperature overall compared to control. seed=42 shows the orange light from the street lamp reflected on the subject’s hair and shoulders. seed=123 and 789 similarly have street lamps placed at close range
• Shadow direction: seed=42 and 789 show the street lamp positioned directly beside the subject, functioning as a strong upper light source. However, no clear change in shadow direction was confirmed
• Skin reflection: Orange color cast from hair and shoulder tops confirmed in 3/3 images. Upper warm light influence is reflected on skin compared to the uniform color of the control
• Background consistency: Street lamp poles are rendered near the subject across all 3 images, naturally integrated as light source objects. The composition itself changed from the control

B03: Car Headlights — car headlights illumination

seed=42	seed=123	seed=789

Observations

• Light color: All 3 images show vehicles rendered near the subject with headlights on. seed=42 has a white car beside a subject in navy tank top, with white headlight reflected on the road surface. seed=123 has a black Mercedes in the subject’s left rear with strong headlights. seed=789 has multiple vehicles flanking the subject with headlights from both sides
• Shadow direction: seed=123 shows headlights illuminating from the subject’s left rear, slightly brightening the front of the subject. seed=789 shows the subject sandwiched by light from vehicles on both sides with high contrast
• Skin reflection: White light highlight on legs and arms from headlights confirmed in 2/3 images (seed=123, 789). Stronger white light contrast compared to control
• Background consistency: Vehicles are rendered in the background across all 3 images, the light source description reflected as objects. Natural composition with vehicles on the road

B04: Illuminations — fairy lights, christmas illumination

seed=42	seed=123	seed=789

Observations

• Light color: All 3 images show fairy lights, Christmas trees, and street tree illuminations in the background — completely different from the control. seed=42 has curtain-style fairy lights hanging in front of the subject with a Christmas tree visible behind. seed=123 similarly has fairy light curtains and Christmas tree. seed=789 shows a tree-lined avenue with white LED fairy lights wrapped around the trees. Overall warm-to-white soft light from multiple point sources
• Shadow direction: Diffuse light from numerous point sources with no clear shadows — similar to control in terms of shadow contrast
• Skin reflection: Soft warm color cast from illumination light confirmed on skin across all 3 images. Particularly in seed=42 and 123, where fairy lights are close, the warm tones on skin are prominent
• Background consistency: Illuminations and Christmas decorations are naturally rendered in the background across all 3 images — the light source description is changing the scene itself. The largest background change of all 7 conditions

B05: Moonlight — moonlit, soft moonlight

seed=42	seed=123	seed=789

Observations

• Light color: A round moon is rendered in the sky across all 3 images. seed=42 shows a moon between buildings with slightly blue-tinted overall color temperature. seed=123 also has a moon above buildings with slightly cooler tones. seed=789 has a large moon rendered high above the intersection. However, urban street lamps and store lighting remain the dominant light sources — it does not become moonlight-only lighting
• Shadow direction: No shadows from moonlight from above confirmed. Urban ambient light is dominant with the moon’s contribution limited to subtle color temperature adjustment
• Skin reflection: Slightly blue-white skin tone compared to control confirmed in 2/3 images (seed=42, 789). However, the change is subtle and cannot be called a prominent blue color cast
• Background consistency: Moon is rendered in the sky across all 3 images, confirming its reflection as a light source object. However, the background scene composition (urban nightscape) doesn’t change much from the control

B06: Storefront Light — storefront window light

seed=42	seed=123	seed=789

Observations

• Light color: All 3 images position the subject directly in front of a shop window, with interior lighting through the glass hitting the subject. seed=42 has a large shop window on the left with a visible mannequin. The warm-tone interior lighting illuminates the left side of the subject. seed=123 similarly stands in front of a shop window with a visible mannequin. seed=789 stands before a glass facade with interior lighting transmitting through from behind
• Shadow direction: Faint left-right light asymmetry from side lighting of the shop window confirmed in seed=42 and 123. However, no dramatic shadow contrast
• Skin reflection: Warm light color cast on the left half of the skin from the shop window confirmed in 2/3 images (seed=42, 123). Light source directionality is somewhat clearer compared to control
• Background consistency: Shop windows are rendered close and large near the subject across all 3 images. Stores were also rendered in the control, but the storefront condition has the shop window closer and larger

Cross-Condition Comparison (Same Seed)

Comparing all 7 conditions side-by-side at seed=42. Focus on light color and skin reflection differences.

Condition	seed=42
B00: Control
B01: Neon Signs
B02: Street Lamps
B03: Headlights
B04: Illuminations
B05: Moonlight
B06: Storefront Light

In the seed=42 comparison: compared to control (general urban nightscape), neon adds colorful signs to the background, street lamps add an orange lamp pole beside the subject, headlights place a vehicle, illuminations render fairy lights and trees in the foreground, moonlight adds a moon to the sky, and storefront light enlarges the shop window at close range. Even with the same seed, composition and background change significantly by light source condition.

Deep-Dive Test: Combined Light Source Reproducibility

Since all 6 conditions in Phase 1 showed clear differences from the control, Phase 2 tests whether specifying 2+ light sources simultaneously reflects both characteristics.

Experimental Conditions

D00: Neon + Moonlight

seed=42	seed=123	seed=789

Observations

• Neon sign appearance: Background neon signs rendered in 3/3 images. seed=42 has colorful neon signs on the left, seed=789 also has neon signs on both sides of the street
• Moon appearance: Moon in the sky rendered in 3/3 images. seed=42 has a large moon in the sky straight ahead, seed=789 also has a moon visible between buildings
• Skin color cast: seed=42 shows a slightly blue-white tone from moonlight rather than strong neon color reflection. Overall skin color cast is subtle across all 3 images
• Comparison with standalone conditions: Neon color cast is slightly weaker than B01 (neon alone), and the moon is more clearly rendered than B05 (moonlight alone). Both objects appeared stably without either one dominating

D01: Street Lamps + Neon

seed=42	seed=123	seed=789

Observations

• Street lamp pole appearance: Street lamp poles rendered right beside the subject in 3/3 images. seed=42 and 789 have classically decorated lamp poles prominently in the foreground
• Neon sign appearance: Background neon signs confirmed in 3/3 images. However, the street lamp poles occupy the central area, making neon signs more subdued background elements
• Skin color cast: Orange color cast on hair and shoulders from the street lamp’s upper orange light source confirmed in 3/3 images — same tendency as B02 (street lamps alone)
• Comparison with standalone conditions: Street lamp pole rendering dominates the composition similarly to B02, with neon more subdued than B01 standalone. Street lamps tend to dominate the scene composition in 3/3 images

D02: Illuminations + Storefront Lighting

seed=42	seed=123	seed=789

Observations

• Fairy light appearance: Fairy light curtains hanging on the shop window glass surface in 3/3 images. Unlike B04 (illuminations alone) with Christmas trees and street tree illuminations, fairy lights here appear as shop window decorations
• Shop window appearance: Large glass shop windows rendered behind the subject in 3/3 images, similarly close as B06 (storefront lighting alone)
• Skin color cast: Warm fairy light through the glass illuminating the subject — warm soft color cast confirmed across the skin in 3/3 images
• Comparison with standalone conditions: B04’s fairy lights and B06’s shop window fused into an integrated scene of “shop with fairy light decorations in the window.” Both characteristics are incorporated naturally into one scene without contradiction

D03: 3 Light Sources (Headlights + Neon + Moonlight)

Testing the limit of specifying 3 light sources simultaneously.

seed=42	seed=123	seed=789

Observations

• Vehicle appearance: Vehicles near the subject with headlights on in 3/3 images. seed=42 has a white car at left front, seed=123 has a black car at left rear, seed=789 has a silver car at left front
• Neon sign appearance: Background neon signs confirmed in 3/3 images. However, density is lower than B01 (neon alone), appearing as scattered background elements
• Moon appearance: Moon rendered in the sky in 3/3 images. Visible between buildings
• Skin color cast: seed=42 shows white headlight reflected on the road surface with highlight on the subject’s legs. seed=789 also shows white headlight from vehicle hitting the subject. Neon color cast is weak across all 3 images
• 3 light source coexistence: All 3 light sources (vehicle, neon signs, moon) rendered in 3/3 images, successfully appearing simultaneously. However, each light source’s rendering intensity is somewhat more subdued than their respective standalone conditions

Combined Light Source Summary

• All light sources appeared stably even with combined specifications: All 4 conditions × 3 seeds = 12 images rendered objects corresponding to the specified light sources. No drop-outs (specified but not appearing) confirmed
• Semantically similar light sources tend to fuse: D02 (illuminations + storefront lighting) generated an integrated scene with fairy lights decorating the shop window. Semantically distant light sources (D00: neon + moonlight) tend to appear as independent objects
• Differences in compositional dominance: In D01 (street lamps + neon), the lamp pole occupied the area near center, pushing neon to the background. There are differences in the tendency to be placed large and close
• Even 3 light sources don’t break down: In D03, all 3 light sources appeared and the scene held together naturally. However, each light source’s rendering intensity was more subdued compared to standalone conditions

Summary

This experiment confirmed clear differences from the control across all 6 conditions when adding 7 types of light source descriptions to prompts. The deep-dive testing further confirmed that specifying 2–3 light sources simultaneously maintains stable appearance of all light sources. Light source type changes background objects, composition, and color temperature, with skin color cast also confirmed in some conditions.

Confirmed Results

• How much light color reflects on skin: Orange warm cast on skin confirmed in street lamps (3/3), warm soft light in illuminations (3/3), pink/blue cast from neon (1/3). Moonlight (2/3) produced subtle changes. Headlights (2/3) reflected as contrast changes
• Whether shadow direction matches light source position: No clear shadow direction change confirmed in any condition. Diffuse ambient light from multiple city night sources is dominant, making directional shadows from single sources difficult to generate
• How much difference from control: Background object changes confirmed in all 6 conditions. Particularly illuminations, neon, street lamps, and headlights changed the composition significantly. Storefront lighting moved in a similar direction to the control but with the shop window rendered closer and larger. Moonlight showed the smallest background change, limited to moon rendering and subtle color temperature adjustment
• Whether light source objects appear in background: All conditions showed stable rendering of corresponding objects — neon signs (3/3), lamp poles (3/3), vehicles + headlights (3/3), fairy lights + trees (3/3), moon (3/3), shop windows (3/3)
• Whether all light sources appear stably with combined specifications: All 4 conditions × 3 seeds = 12 images rendered objects for all specified light sources without drop-outs. Semantically close light sources (illuminations + storefront) fused; distant ones (neon + moonlight) appeared independently

Strength of Effect by Condition

Practical Points

1. Light source specifications largely change background scene composition: Specifying a light source type causes corresponding objects (neon signs, lamp poles, vehicles, etc.) to appear in the background. The effect is more of a scene setting change than a lighting change
2. Skin color cast varies by condition: Warm-tone light sources like street lamps and illuminations consistently reflect color on skin. Neon’s multicolor appears in the background but direct skin color is unstable
3. Moonlight has weak effect in urban nightscapes: city street ambient light is strong, limiting the moon’s contribution. To make moonlight the main light source, combine with locations with less ambient light (parks, rural roads, etc.)
4. Light source specification alone is insufficient for shadow direction control: No directional shadows confirmed in any condition. To control shadow direction, more direct description is likely needed (e.g., lit from the left by a single spotlight)
5. Combined light sources work stably up to 2–3: Specified light sources appeared without drop-outs. Semantically close light sources fuse into natural scenes; distant ones appear independently. However, as the number of light sources increases, each one’s rendering intensity becomes somewhat more subdued

Note: All 7 light sources produced confirmed background changes in this test. Illuminations are effective for a winter-portrait mood, and the orange cast from street lamps also reads well. Combined light sources held up to 2–3 without drop-outs, so they’re worth combining when a richer scene is desired. Shadow direction control remains an open challenge.

Related Articles

• Lighting-related experiments: After-school library emotional composition (golden hour lighting experiment)
  PR RunPod クラウドGPUでAI画像生成 RunPodを始める →