Veo 3.1 Fast: the eight-second ceiling, the native audio, and when to use it

Veo 3.1 Fast is the engine on our shelf most writeups misunderstand on sight. The duration options are four, six and eight seconds, nothing else. That isn't an initial limitation that'll relax next version; the provider's duration enum is three values. There's no plan, no coupon and no prompt trick that unlocks nine. Every ninety-second Veo video on the internet is twelve clips joined.

Which sounds like a reason not to use it, until you look at what that constraint actually buys you. Native audio. The cheapest voiced second on our shelf. One and a half times silent (so turning audio off cuts the cost by a third). Three images of reference. 720p and 1080p at the same price. For the right shot, the eight-second ceiling is a feature, not a wall.

Here's how we actually use it, with the credit math.

What Veo 3.1 Fast is

Google's second-generation Veo model, in its Fast tier. On our shelf it serves 4, 6 or 8 seconds; it renders at 720p or 1080p and bills the same credits for either (which is unusual — on Seedance 2.5 the resolution more than doubles the price); it takes up to three reference images; and it bills 30 credits a voiced second and 20 credits a silent one, at either resolution. A reference generation is pinned to 8 seconds, whatever the slider says, because that is the length the provider actually bills.

The headline number: 30 credits a second voiced, which is the same as Kling at 720p, and 20 credits a second silent, which is the cheapest silent second on our shelf. One 8-second silent Veo clip is 160 credits, about $1.60 on Lite and $1.20 on Creator. The cheapest finished second you can buy today.

The four / six / eight constraint, in one table

Not a slider. An enum of three values. Everything in the SERP that says otherwise is talking about Veo's extension feature, which is a different workflow — clips extended seven seconds at a time, up to 148 seconds total, each extension billed as a fresh render.

what you want	what Veo 3.1 Fast actually renders	credit cost at 720p voiced
4 seconds	4 seconds, in one generation	120
6 seconds	6 seconds, in one generation	180
8 seconds	8 seconds, in one generation	240
10 seconds	not available on Veo	—
12 seconds	two 8-second clips, trimmed to 12	about 480
30 seconds	four 8-second clips, trimmed to 30	about 960
90 seconds	twelve 8-second clips, trimmed to 90	about 2,880

Two things to notice. First, the per-second price is low enough that the trim overhead rarely matters: a 30-second finished scene on Veo lands at about 960 credits — around $9.60 on Lite — against 2,700 on Seedance 2.5 for the same length at the same resolution. Second, if your clip is a locked eight-second talking head, this is the cheapest voiced second on our shelf; Kling matches it, and nothing else gets close.

What the native audio actually solves

Veo renders audio inside the model, not through a lip-sync pass after the fact. That's a different problem than Hedra or Pixverse are solving, and it has one specific consequence: a reaction, a laugh, a sigh, a cough, a glance that comes with a tiny vocal beat, lands on the right frame. The audio tracks the mouth because the mouth and the audio were generated together.

For a locked eight-second shot where somebody says one line and reacts to a thing, there is no better engine on the shelf. For a long monologue where lip sync drifts across a whole scene, Veo is the wrong tool and a dedicated lip-sync pass over a silent render on Seedance or Kling is the right one. The eight-second ceiling is what keeps the drift inside what the model is good at.

Reference generations, and the eight-second pin

Three images, up to. A character ingredient, a prop, a background: Google's "ingredients to video" shape. The constraint hiding in the docs: a reference render is always eight seconds, whatever the slider says. We quote that length rather than the slider in the engine picker, because quoting six for a reference render on Veo is quoting a render the provider will bill at eight.

In practice this means Veo is where you spend an eight-second reference every time you spend one, and you plan the shot for that length. A six-second beat with one reference image on Veo is a six-second trim out of an eight-second render; the engine returns eight, we deliver six, you pay for eight. On Kling or Seedance the reference does not change the length.

Where Veo 3.1 Fast is the right engine

The opener. A locked eight-second establishing shot with one spoken line, 720p, voiced. 240 credits. The cheapest voiced opener you can ship. Where other engines ask for a reference or a long prompt to hold a scene, Veo takes a text description and a reference image and delivers eight seconds of audio-locked video.

Silent b-roll. 20 credits a silent second at 720p. A silent eight-second b-roll clip is 160 credits — about $1.60 on Lite. Nothing matches it. On Seedance the silent discount doesn't exist; on Kling silent is 24 credits a second.

A short talking head. One character, locked camera, native audio, 8 seconds. The lip sync lands. For a creator shipping daily shorts where every clip is one line delivered to camera, Veo is the engine that keeps the format consistent at the lowest cost.

1080p where other engines charge the premium. 720p and 1080p are the same credit rate on Veo 3.1 Fast. On Seedance 2.5 1080p is more than double 720p. For a client deliverable that insists on 1080p but doesn't need Seedance's 30-second take, Veo is the cheaper 1080p on the shelf.

Where Veo 3.1 Fast is the wrong engine

Anything longer than eight seconds in one continuous take. Full stop. The engine does not render it. A 15-second shot is two Veo clips joined; a 30-second shot is four. If the continuity matters, Kling (15s one gen) or Seedance 2.5 (30s one gen) is the engine.

Scenes with four or more named characters. Veo takes three references and does not address them by name in the prompt — the engine uses whatever addressing you type, which is less reliable than Kling's @Element1..5 or Seedance 2.5's @Image1..30. The refusal routes you to a roomier engine.

Camera moves across a locked character over a long beat. The eight-second ceiling is what helps the audio stay locked; it is also what cuts off a slow dolly before the shot has landed. Kling's element binding over a 15-second shot is the better fit.

The credit math on a 90-second finished video

A 90-second vertical story with one locked character across six scenes. Three real paths through the shelf.

path	engines	finished credits
all Veo 3.1 Fast	12 × 8s voiced clips at 240 ea, trimmed	about 2,880
mix: Veo opener + Kling middle + Seedance 2.5 signature	1 × 8s Veo + 4 × 15s Kling + 1 × 15s Seedance 2.5	about 3,600
all Kling	6 × 15s voiced at 450 ea	about 2,700

Veo-only is not the cheapest finished video on our shelf, because the trim overhead on a 15-second scene is a whole extra render. The mix usually wins — Veo for the one place its advantages compound (an eight-second opener with native audio, 720p), Kling for everywhere a character has to hold through a camera move, Seedance 2.5 for the one shot the clip hangs on.

If you want to run the arithmetic on your own clip, the engine picker inside the studio quotes every length on every engine before you press render; the number you see is the number you'll pay. A free account comes with 300 credits, which is one voiced Veo clip, one silent Kling test, and a character sheet before you spend a dollar. Open the shelf and look at the numbers against your idea.

The honest part

Veo 3.1 Fast is the engine most new creators underestimate because the eight-second number looks like a limitation. On the shelf it is the engine with the sharpest audio, the cheapest silent second, the only one where 1080p costs the same as 720p, and the one with native audio that lands on the right frame. Our own 9.7 million views in 30 days on @cevrico came from reaching for the right engine per shot; Veo earned its share. Own results, not a projection; what works for your channel will depend on the format you ship. The eight-second ceiling is a feature when you choose shots that fit it.