A prompting method that treats directing an AI model like directing a real actor: shoot multiple takes of the same beat, adding incremental notes pass by pass — breath, emotion, physicality — rather than trying to specify a perfect performance in one prompt. Hesitation Beat for Human-Feeling AI Performance is one example of a note added this way. Feeds into Take Compositing for Best Performance, where the best take of each performer across these passes is selected for the final composite.
A sample of the directorial, actor-note style of prompting this method uses for a single beat:
"He takes the glasses off like he's the face of a luxury eyewear brand. The flares streaks across the frame and the eyes come up half closed."
The prompt describes intention and feeling (a brand-ad flourish, a specific half-closed-eyes beat) rather than only mechanical camera/action instructions — the same register a director would use talking to an actor between takes.
Each successive generation pass adds exactly one more specific directive on top of the prior broad prompt, mirroring how a director gives one note at a time across multiple takes rather than one long list of instructions upfront. A pass might start from a broad emotional/action prompt, then the next pass adds a breath, the next a specific body-physics detail, the next a finer emotional beat — narrowing in on the performance rather than re-describing the whole shot each time. Framed explicitly as directing a performance, not debugging a technical output: the notes given between passes are the same kind a director gives a human actor between takes.
Из тем: Unsorted, Directing Performance, Dialogue, and Sound