Google pushed out Veo 3.1 shortly after Sora 2 landed, and now I've got two serious audio-capable video models competing for the same slot in my pipeline. I've run the same handful of prompt types through both over the past week rather than trusting either company's showcase reel, and the results split in ways that actually matter for picking a default tool.
Prompt adherence versus visual polish
Veo 3.1 follows detailed camera and motion instructions more consistently in my testing, which matters a lot for the kind of structured, directed shots I need for brand work. Sora 2's output often looks slightly more cinematic out of the box, better default color and lighting choices when the prompt is looser, but it takes more liberties with specifics I actually specified. For work where I'm matching a brand's exact visual language, that consistency gap makes Veo the safer default right now.
Audio quality, a closer call than I expected
Both handle ambient sound and basic sync convincingly. Sora 2 has a slight edge on dialogue-adjacent audio, mouth movement and voice timing feel a touch more natural in my tests. Neither is where I'd source final dialogue for anything client-facing yet, but for background chatter, room tone, and non-verbal sound, both are usable straight out of generation.
- Structured, brand-specific shots: Veo 3.1's prompt adherence wins
- Loose, moody, exploratory shots: Sora 2's default look often wins
- Audio: close, slight edge to Sora 2 on dialogue-adjacent sound
My actual working setup now
I've landed on using both rather than picking a single winner, which sounds like a cop-out answer but reflects how differently they behave. Veo goes first when I have a shot list with specific camera moves that need to be respected. Sora goes first when I'm exploring concepts and want a few options to react to before locking a direction. Running the same prompt through both and picking the better result adds time, but for anything client-facing, that extra pass has paid for itself more than once already.