Since Jul 2023
Sound and music
Voice, sound effects and score, then sound in the same pass as the picture.
For two years AI films were silent until someone scored them. Creators cut to stock music, then generated whole songs with Suno and Udio (spring 2024) and sound effects with ElevenLabs (May 2024), layering them under silent clips. In May 2025 Veo 3 made dialogue, effects and ambience in the same pass as the picture, and by 2026 models took music and sound as references. Many music videos and trailers still start with the song.
Tutorials3
Veo 3 and Flow: native audio inside a director app
Theoretically Media · YouTube · May 2025
How it works
- Pick or make the music first when the film is cut to it, and map its beats before generating anything.
- Let the model make the sound when it can: native dialogue, effects and ambience (Veo 3, Kling 3.0, Seedance), with separate stems where offered.
- Layer what it misses: effects prompted per shot in ElevenLabs, voices from a fixed cloned voice per character.
- Score with a music model (Suno, Udio, ElevenLabs Music) and keep the stems so the mix stays editable.
- Mix under the voice in the editor; sound often covers more cuts than a picture fix.
Behind the scenes1
How films were actually made, from the people who made them.
Related on the timeline3