When generating AI portraits, backgrounds tend to default to “glamorous” locations like studios, beaches, or city nightscapes. But what happens when you deliberately specify unglamorous, everyday spaces? This experiment tests how well prompts can control backgrounds featuring convenience store fluorescent lights, laundromat washing machines, and back alley graffiti walls.
Experiment Setup
The base subject prompt was kept constant, with only the location description changed across conditions.
- Seeds: 3001, 3100, 3200 (fixed)
- Variable changed: Location/background description (single variable)
- Base subject:
1girl, 32yo japanese actress, standing indoors, casual clothes, t-shirt and shorts, looking at camera
4 conditions with 3 images each, 12 images total.
Control (neutral background)
First, the baseline with no specific location — just neutral background.
| seed 3001 | seed 3100 | seed 3200 |
|---|---|---|
 |  |  |
All 3 images produced gray-to-beige plain walls as the background. Lighting is flat with minimal shadows, resembling a studio or outdoor wall setting. The t-shirt and shorts instruction was reflected in all 3 outputs.
Experiment 1: Convenience Store
Adding inside a convenience store, cluttered shelves behind her, fluorescent ceiling lights.
| seed 3001 | seed 3100 | seed 3200 |
|---|---|---|
 |  |  |
All 3 images clearly depicted a convenience store interior. Key observations:
- Product shelves: 3 out of 3 images generated an aisle composition with products on both sides
- Fluorescent lights: Ceiling fluorescent lights confirmed in all 3 images. The
fluorescent ceiling lightsdescriptor is clearly effective - Product details: PET bottles, detergents, and condiments arranged in a Japanese convenience store style across all 3 images
- Floor: White-to-gray tile flooring consistent with convenience store aesthetics
Compared to the control, the lighting shifted to the flat, white quality typical of fluorescent tubes, also affecting the subject’s skin tone.
Lab Director’s comment: The convenience store reproduction is insane. The cluttered shelf vibe is totally there —
cluttered shelvesis doing the heavy lifting. The washed-out fluorescent look just adds to the everyday feel.
Experiment 2: Laundromat
Adding inside a laundromat, washing machines in background, harsh fluorescent lighting.
| seed 3001 | seed 3100 | seed 3200 |
|---|---|---|
 |  |  |
The laundromat was also reproduced with high fidelity across all 3 images.
- Washing machines: Front-loading drum machines visible in the background of all 3 images, with circular door windows clearly rendered
- Fluorescent lights: Overhead tube lighting confirmed in all 3 images.
harsh fluorescent lightingproduced a slightly harder light impression compared to the convenience store condition - Control panels/signage: Pricing displays and operation panels were visible in 2 out of 3 images
- Floor: Tile flooring confirmed in all 3 images, lending the clean utilitarian feel typical of laundromats
The direct descriptor washing machines in background reliably placed the core background element.
Experiment 3: Back Alley
Changing to standing in a narrow back alley and adding graffiti walls, wet pavement, dim streetlight. This is an outdoor condition.
| seed 3001 | seed 3100 | seed 3200 |
|---|---|---|
 |  |  |
The outdoor back alley was also consistently reproduced.
- Narrow alley: All 3 images depicted a narrow passage with walls on both sides, matching the
narrow back alleydescriptor - Graffiti: Spray-painted graffiti confirmed on walls in all 3 images, with varied colors and styles
- Wet pavement: Wet ground visible in all 3 images. Puddle reflections were observed in 2 out of 3
- Streetlight: A dim light source was visible in the seed 3200 background. However, the “dim” effect from
dim streetlightwas limited — overall brightness remained at a daytime-to-evening level
An interesting observation: 2 out of 3 back alley images added a shoulder bag to the subject, despite no bag being specified in the prompt. The location descriptor appears to influence subject accessories as well.
Lab Director’s comment: The bag thing is interesting, right? No one prompted a bag, but “person in an alley” apparently comes with accessories. Location prompts have more side effects than you’d think.
Cross-Condition Comparison (seed 3100)
Comparing all 4 conditions side by side at seed 3100.
| Control | Convenience Store | Laundromat | Back Alley |
|---|---|---|---|
| | | |
With the same seed, the background changes dramatically based on location descriptors. The subject’s pose and body type are broadly maintained, but clothing color and design vary between conditions. Location prompts appear to influence clothing color choices as well.
Summary
Key findings from this mundane location experiment:
- Location name + specific objects is an effective combination. Adding
cluttered shelvestoconvenience storeincreases background density - Lighting descriptors affect the location’s atmosphere.
fluorescent ceiling lightsvs.harsh fluorescent lightingshowed subtle differences in light hardness in 2 out of 3 images - Outdoor conditions showed location prompts spilling over into subject accessories like bags (observed in 2 out of 3 images)
wet pavementwas reflected in all 3 images — surface texture descriptors have high reproduction ratesdim streetlighthad limited dimming effect. Achieving clearly dark images likely requires descriptors likenightorlow key lighting
Even “unglamorous” everyday spaces can be reproduced with high fidelity when you describe specific objects and lighting conditions. This approach is effective when you want to move beyond studio-style shoots and create portraits with narrative context.
Lab Director’s comment: Everyday-life photos are actually harder to nail than fancy ones. The fact that you can get different fluorescent vibes between a convenience store and a laundromat? Quietly impressive.
![[Verified] Image Generation Prompt Best Practices](/tips/prompt-best-practices/cover_0_0000_4517457392071889496.webp)
