AI video with sound: rain on the roof, a door, one line that lands

• Soundscape: rain, traffic, a fridge hum, wind in a field. It's written per shot, so a cut to outside sounds like outside.
• Effects: a door, a phone, a glass on a table, the doorbell.
• Music: only when the scene wants it, and named by mood.
• Dialogue: short lines in English, each speaker with one voice kept across the short.
• Silence: sometimes the strongest choice, especially in horror.

Each category has a line budget built in: about one line per 15 seconds in horror, one per 8 seconds in romance and one per 6 seconds in comedy. Fewer lines mean fewer chances for a mouth to slip.

Some of the best formats avoid on-screen speech altogether: AI ASMR runs on texture sounds, nature documentary parodies use an off-screen narrator, VHS home videos put the voice behind the camera, and doorbell camera clips get by on a chime and footsteps.

• Stiff lip sync on longer lines. Our reviewer flagged it as one of the most common faults.
• Voices that sound a little synthetic in emotional moments. Underplay it: a small line beats a big one.
• No voice cloning and no real voices. Characters are invented, and so are their voices.

• A father and his grown daughter sit in a parked car outside her new flat. Rain on the roof. He says: “You'll call?” She nods, doesn't move.

  Make this

• A late-night radio host takes a call from a listener who describes, exactly, what the host is wearing right now. Only the host's voice and the caller's on the line.

  Make this