A Fully Reusable Video Prompt: Turning Four Reference Photos into a 15-Second Surrealist Short
umesh_ai · x · 2026-10-17
Prompt sharer umeshai posted a fully structured video-generation prompt:
- Inputs: four stylistically unified monochrome long-exposure reference photos (monumental staircase with ghostly commuters, rainy bus stop under translucent umbrellas, highway light streaks, a banyan sheltering a lone tea seller) — the model must treat them as one physically connected city, not separate scenes
- Narrative constraint: a man stops to protect a tiny bird while everyone rushes past; that decision folds the city around him and leads to unexpected kindness — told entirely through imagery and gestures
- Specs: exactly 15 seconds, 16:9, photoreal, 4K, 24 fps; no dialogue, captions, logos or narration
- Style control: preserve the fine-art monochrome long-exposure look — black silhouettes, silver wet pavement, mist, analog grain and halation
The prompt demonstrates a reusable pattern for video models: multi-reference world-building + single-event narrative + strict spec constraints, applicable to Veo/Sora-class tools.
More from Multimodal
- Seedance 2.5 keeps Midjourney reference style remarkably well in video gen test — CurieuxExplorer · 2026-10-17
- Creative AI videos made with Gemini Omni & Flow, prompts shared — michaelrabone · 2026-10-17
- Midjourney style code --sref 1602964516 yields cinematic giant-monster imagery — tisch_eins · 2026-10-17
- CEB SAM 3D Body Loader for Blender goes free: full AI mocap & animation studio — ssh4net · 2026-10-17
- Runway video packs cinematic destruction into 15 seconds, sparks realism debate — umesh_ai · 2026-10-17
- MOSS Image Sampler Tests: dpmpp_sde + radiance_shift Follows Prompts Best at cfg 3 — mikemend · 2026-10-17