Keeping the same character recognizable across a series of stills and short clips is mostly a prompt-discipline problem. The structure that works best for me is a fixed "identity block" followed by a variable "shot block". The identity block describes the character in the same words every time: age, face shape, hair, one or two signature wardrobe items, and a color palette. The shot block then changes only framing, action, lighting and camera movement. I usually lock the look first with an AI image generator, then reuse that reference image plus the identical identity block when moving to video, so the model has both text and visual anchors.
Example prompt:
IDENTITY: Mara, late 20s, oval face, short copper bob with blunt bangs, freckles, olive rain jacket, silver hoop earrings, muted teal and orange palette.
SHOT: medium close-up, Mara looks up from a cafe window as rain starts, soft overcast light, shallow depth of field, slow push-in, 35mm film look.
Tip: never paraphrase the identity block between shots; even small wording changes ("red hair" vs "copper bob") cause drift.
Top comments (0)