One elf. One thousand styles.
Style is the loudest word in an image prompt — louder than composition, louder than lighting. To measure just how loud, we held everything constant and changed a single thing. The base subject never moved: aesthetic elf woman. Only the modifier after the comma changed, one thousand times.
The modifiers span thirteen families — anime eras and manga lineages, studio-auteur looks, traditional media from oil glazing to thangka, printmaking and craft, retro game hardware, 3D render passes, subculture fashion, color science and lighting setups, regional folk arts, film stocks and camera moves, surrealism, material studies, and weather moods. Every image below was generated at 832×1216 as webp (~quality 85) through the same images3 pipeline, so differences you see are the style term doing all the work.
"aesthetic elf woman" starts every prompt. No seed lock — each run samples fresh, which keeps comparisons about style, not lucky rolls.
Everything after the first comma is a single style modifier, from "1960s retro anime style" to "wearing stormcloud tulle, framed by aurora sky".
All 1000 went through create_and_upload_image at 832×1216 and came back as ~50KB webp files on the netwrckstatic CDN.
Browse the gallery
Filter by family or search for a technique. Click any tile to open it large, read the full prompt, copy it in one tap, and step through with arrow keys.
Showing 60 of 1000 — every image is the same subject (aesthetic elf woman) with only the style modifier changed.
Run your own grid
This page is itself a prompt experiment you can fork: pick a subject, write twenty modifiers, and let the bulk tool iterate while you judge the results side by side. The same trick works for lighting audits, wardrobe boards, and finding a signature look for a character before you animate them.
Open the bulk image tool