Cast Two
Two faces is a scene. Three is a crowd scene, and the model starts blending features.
An AI flirting video generator built for tension: two people, one room, one look held a beat too long.
SET THE SCENE









Prompt Guide
Model
Reference Image
Three choices that separate chemistry from a stock-photo smile.
Two people at 40 centimetres read as a couple. Three metres read as colleagues. The gap is the scene.
A look that lasts a beat too long is the shot. Cut away early and the tension goes with it.
Let one person step out of the light. Interruption costs nothing and works better than a kiss in six seconds.
Rooftop bars, first dates, late-night texts.
Two people stand apart on an empty platform at night, one steps closer, the other looks at the departure board instead, cold fluorescent light with warm vending machine glow, static wide shot, held beat, 16:9.
Try This StyleA man delivers an obviously bad pickup line at a party, the woman blinks, then laughs and shakes her head, warm string lights behind them, handheld medium shot, slight crowd blur, 16:9.
Try This StyleA woman in an empire dress reads a folded letter beside a piano in a candlelit room, her expression changes, she folds it and holds it against her chest, static medium shot, warm candle light, painterly period look, 16:9.
Try This StyleA phone screen lights a face in a dark bedroom, the person reads a message, smiles, types and deletes, types again, cool screen light on skin, static close shot, shallow focus, cinematic framing.
Try This StyleTwo people under one umbrella at a crossing, one pauses mid step, rain on the fabric, medium shot, night street.
Try This PromptA hand sliding a folded napkin across a cafe table, other hand hovering, steam between cups, tight top shot.
Try This PromptA figure at an upper balcony window looking down once, lace curtain moving, cobbled street below, static wide.
Try This PromptTwo people in a lit doorway, one steps back instead of leaning in, hallway light behind, medium shot, no cut.
Try This PromptTwo people leaning over one menu, shoulders touching, restaurant candlelight, over the shoulder framing.
Try This PromptThumbs typing a message then deleting it, screen glow on a face, dark bedroom, tight vertical framing.
Try This PromptDistance, hands and light are where these clips go stiff. Fix them in this order.
Two faces is a scene. Three is a crowd scene, and the model starts blending features.
Hands on a table, on a glass, on a sleeve. Hidden hands warp into the other person.
Bar light, street lamps, a phone screen. A practical source justifies the close-up.
Blinks at normal speed read as glitches. Hold the eyes and let the blink happen once.
Six seconds cannot carry a line. A look plus a lip movement reads better than subtitles.
No real faces without consent, and nothing explicit. Likeness and platform rules apply to both people.
Notes from creators blocking two-person scenes.
I write short romance serials for a reading app and needed teaser clips. The doorway version where he steps back instead of leaning in is the one that got people commenting.
Comedy was easier than sincerity here. A bad pickup line survives a small motion error that would ruin a serious scene, and the laugh covers the rest.
I make character backstory videos for a tabletop campaign. Two invented faces, one room, one line, and my table immediately started speculating about how they met.
A near-miss works; contact often merges features. A lean that stops short reads better on camera anyway.
Too much movement in frame. Lock the camera, move one shoulder and hold.
Only if the people have to be specific. Invented faces avoid consent and likeness problems.
One free render on Grok Imagine Video 1.5, no subscription.
Get written consent first. Non-consensual likeness breaks platform rules and is illegal in several states.