Veo 3.1 Deep-Dive
Google's Veo 3.1 delivers photorealism, cinematic polish and native spoken dialog. Learn its structured prompt formula and when to pick it over Sora 2.
If Sora 2 is a physics engine with a camera, Veo 3.1 is a cinematographer with a sound department. Google DeepMind's model is the one people point at when they say I genuinely could not tell that was AI — skin texture, lens behavior, natural light. And its party trick changed the game: characters that speak, with voices and lip movement generated together.
Where Veo 3.1 Shines
- Photorealism: faces, skin, fabric and natural lighting that hold up under scrutiny
- Native audio and dialog: spoken lines with synced lips, plus ambience and effects, generated with the video
- Cinematic quality: strong grasp of shot types, lens language and film lighting terms
- Prompt adherence: a structured prompt gets followed with unusual discipline
- Image-to-video: give it a first frame (or reference images) and it respects them faithfully
The spec: 8-second clips at up to 1080p, 24fps, with native audio. Shorter than Sora 2 — but 8 disciplined seconds that nail your brief usually beat 12 improvised ones.
The Veo Prompt Formula
Where Sora 2 loves prose, Veo 3.1 rewards structure. Build your prompt as ordered components: shot type, subject, action, location, lighting, camera movement, and audio cues. Dialog goes in quotes with the speaker named and a tone: The barista says warmly: followed by the line. This maps straight onto the prompt anatomy you already know — Veo just takes each slot more literally than any other model.
Structured Veo 3.1 prompt with dialog
Medium close-up. A food truck owner in her 50s with flour-dusted apron leans out of the service window at golden hour. She smiles and says warmly: "Last one's on the house." Handheld camera with subtle movement, warm backlight, sizzling grill ambience in the background. No subtitles, no on-screen text.
Every slot filled in order: shot type, subject, action, dialog with tone, camera, lighting, audio, and the anti-subtitle spell at the end.
Sora 2 or Veo 3.1? The Producer's Cheat Sheet
| Your shot needs | Pick |
|---|---|
| A character speaking on camera | Veo 3.1 — native dialog with synced lips |
| Physics-heavy action, stunts, liquids | Sora 2 — the physics engine |
| Maximum photorealism for ads and brand work | Veo 3.1 |
| More runway: an action needing 10+ seconds | Sora 2 |
| Animating an approved still with high fidelity | Veo 3.1 — first-frame discipline |
Both deep-dives done — you now speak two dialects of video prompting. Time to think bigger than one shot: storyboarding multi-scene stories is where clips become content.
Write one structured prompt with a spoken line and hear your character talk. Generate with Veo 3.1