The announcement
On 20 May 2025, Google introduced Flow as a filmmaking tool combining Veo, Imagen, and Gemini. In a separate announcement the same day, Google presented Veo 3 as a video model able to generate sound alongside moving images, including environmental audio and dialogue. These were vendor announcements, not independent evidence of production reliability.
The shift is still useful to study. Earlier AI-video experiments often paired a silent generated clip with separately recorded or selected sound. A generation that proposes picture and sound together can suggest a scene's mood quickly. It can also make an editor's job harder when the sound implies a performance, location, or timing that the picture cannot sustain across several shots.
Sound is a story choice
For each proposed scene, write a sound brief before generation: which sounds are diegetic, which are off-screen, where silence matters, and who owns a voice. Judge generated dialogue for intelligibility and dramatic intent, but also ask whether its voice and performance were authorized. Do not assume a plausible voice is a performer you can use.
Cut the scene once with the generated track and once with an independent sound sketch. If the sound is doing all the narrative work, the images may need clearer action. If the imagery is strong but the sound is inconsistent, a recorded performance or conventional post-production track may offer better control.
A bounded test
Choose a 15-second beat with one action and one off-screen cue. Save the script, storyboards, source images, model name, settings, outputs, and edit notes. Score the result for story clarity, camera continuity, sound perspective, and ability to revise. That test compares a project's needs with the tool's actual behaviour; it is not a published benchmark.
Flow's 2025 feature and plan descriptions should remain attached to their date. Product access, generation limits, and supported models change. A later feature page cannot be used to backfill what was available on launch day.
Sources & context
- Google · Introducing Flow (20 May 2025)
- Google · Veo 3 and generative media announcement (20 May 2025)
Sources checked 16 September 2026. The proposed sound review is Screen Method's editorial method, not a report of a completed test.