What Veo 3.1 is
Veo is Google DeepMind’s video generation model line. Veo 3.1 is the current version and the one GATA renders with — if you have been searching for “Veo 3”, 3.1 is its successor and what you get here. It generates short clips with strong prompt adherence and native audio.
In GATA, Veo 3.1 is one of five per-shot renderer choices, alongside Kling o3, Kling v3, ByteDance Seedance 2.0, and Grok Imagine 1.5. It bills at 45 credits per second of finished video, the top of the range. Locked cast and locked look feed in automatically, so the model receives a structured brief rather than a paragraph somebody has to keep rewriting.
Where Veo 3.1 shines
- Hero shots where prompt adherence and fidelity justify the highest per-second rate
- Dialogue beats, thanks to native audio
- Subtle camera moves with consistent geography
- Shots that need a 4K master — Veo is one of three 4K-capable models here
The duration rule that catches teams out
Veo 3.1 only renders 4, 6, or 8 seconds. A requested duration is rounded up to the next supported value, and anything longer than 8 seconds is clamped to 8. Ask for a 5-second shot and you get — and are billed for — 6 seconds. GATA shows a confirmation before generating whenever the model would adjust your shot’s duration, so the change is never silent.
Two consequences worth planning around:
- Budget from the rounded duration, not the one you typed. At 45 credits per second, a 5-second shot billed as 6 costs 270 credits, not 225.
- Veo cannot cover a long take. For anything past 8 seconds the shot has to be split, or rendered on a model without the bucket restriction.
The other limitation: no locked-cast references
Veo 3.1 does not accept locked-cast composites as references. Only Kling o3 and Kling v3 do. Veo will render an excellent shot of a person, but not reliably your person, shot after shot.
For character-driven sequences this is the deciding factor. The usual pattern is to render recurring-character shots on Kling o3 at 20 credits per second and reserve Veo 3.1 for the one or two hero beats where fidelity earns its cost. Both read from the same project records, so the cut still holds together — see character consistency for why identity has to be bound structurally rather than described in a prompt.
What it costs
| Rate | |
|---|---|
| Base | 45 credits / second |
| 4K output | +30 credits / second |
| Lip-sync | +15 credits / second |
| Background noise | +5 credits / second |
A 6-second hero shot in 4K with dialogue is (45 + 30 + 15) × 6 = 540 credits. The same 6 seconds with no uplifts is 270. Plan inclusions and worked project math are on the pricing page; the Veo 3 pricing breakdown works through what those credits mean in money.
Where to swap models
For fast action with hard transitions, or any shot longer than 8 seconds, GATA’s per-shot picker switches to Kling o3 / v3 or Seedance 2.0 without re-doing cast, look, or script. The shot inherits the same project context; only the renderer changes. See shot inheritance for the mechanism and the full model comparison for the side-by-side.
Why the model is not the product
Veo 3.1 is available directly from Google. A subscription gets you a prompt box and clips. What it does not get you is the thing that makes twelve clips cut together: a locked cast, an approved look, locations that persist between shots, and a script the shot list reads from.
That structure is what GATA is, and it is why the renderer is a swappable detail inside it rather than the product itself. See script to video, in parallel.