Open Midjourney (or your image model) and ask for one mid-1990s 35mm film still of your subject. Name the city, season, outfit fabrics, jewelry, and the candid mid-stride glance.
Keep it photographic: pores, flyaways, fine grain, lifted blacks. This still is the face and wardrobe lock for everything after.
Do not start in Seedance.
Pass that still into GPT Image (or Nano Banana Pro) and ask for a CHARACTER REFERENCE SHEET:
The sheet is what Seedance uses as the identity ref, so identity holds when the camera moves and the accent flips.
Generate empty or near-empty location stills next: NYC diner booth, Village sidewalk, night street. Period props only, no phones.
Soft cloudy daylight or warm pendants, Kodak 35mm grade. Keep one clean diner still as the location ref.
Your remake used a teal vinyl booth, Formica, chrome napkin holder, neon OPEN in the window.
In Seedance 2.5, wire the character sheet + location still as reference images. Write a REFERENCE MAP (which image is face vs place), then a GLOBAL SETUP (year, light, handheld friend-cam, length), then MOMENT blocks with clocks: lean-in line, accent shift, tiny giggle, shrug.
Optional accent audio ref. One continuous take, live sync dialogue, no music, no cuts.
Use one master prompt and swap only your phrase. The ready-made master prompt is in the members pack.
Pull a 1 to 2s grade pass if needed (warm lift, fine grain), keep the friend-cam shake, and post a still from the method under the clip so people see the craft. Proof from the remake: a 7s diner friend-cam where she says a pizza line in thick old Brooklyn, giggles once, then shrugs "it's all in the prompt." That is the video people will make: short cinematic 90s NYC micro-films with dialogue and natural acting.
This guide is the full process. The ready-made version lives inside AI Creative OS on Skool, next to every other prompt pack I post.
