One prompt, ten panels: Reddit's recursive zoom image-generation experiment
NVDA808 · reddit · 2026-09-16
A Reddit user shared a single-prompt image generation experiment: the model produces one large image containing 10 sequential panels, where each panel must be an enlarged interpretation of a real region of the previous panel.
Key prompt design:
- Panel 1: generate a richly detailed original scene full of explorable visual information, with no pre-planned path.
- Panels 2–9: after each panel, the model inspects its own output, marks the most interesting genuine feature with a crop box, and treats only that region as the structural parent of the next frame — a strictly nested recursive zoom.
- No lazy zooms: each step must reveal a new level of organization, avoid repetitive microscopic textures, and never invent elements with no visual precursor.
The whole loop of generate-observe-select-zoom is executed autonomously by the model with zero user intervention, framed as a 'visual contact-sheet exploration.'
More from Multimodal
- Midjourney wallpaper recipe: sref + profile mix for eerie film-blur portraits — michaelrabone · 2026-11-02
- Cavallaero Teaser: A Painterly Rider Races Through a Golden Forest — CommunicationTop4244 · 2026-11-02
- Anima's snapref Offers a New Reference-Based Fix for Character Consistency — Intrepid_Swim_8669 · 2026-11-02
- Pencil-textured folding fan: surreal product concept render goes viral — Delahuntagram · 2026-11-02
- Free Midjourney SREF gallery adds 1,200+ handpicked style codes overnight — tisch_eins · 2026-11-02
- Adobe's EditAgent beats prior image-editing agents but needs ~10x the compute — emmanuelvivier · 2026-11-02