Veo 3.1 and Runway Gen-4 get compared on taste more often than on paper. Both vendors publish paper, and the two sheets answer different questions: what a shot can contain, and what a generation costs.
Every figure below is printed on the vendor’s own page. Where one side has no row to print, the row says so.
The short answer: eight seconds with sound, or ten without
Veo 3.1 is documented at eight seconds per clip, in 1080p and 4K, with dialogue, sound effects and ambient noise generated with the picture. Runway Gen-4 is documented at five or ten seconds, in six roughly one-megapixel sizes, from a required input image, with no audio row in its spec sheet.
If your brief is a person talking on camera, the audio row decides it. If it is a long move out of a locked still frame, the duration row does.
What each vendor’s own pages publish
| Published item | Veo 3.1 · Google | Runway Gen-4 · Runway |
|---|---|---|
| Where it is printed | DeepMind’s Veo model page | The Gen-4 help article, plus Runway’s API docs |
| Clip length | “Veo videos are 8 seconds long.” | 5 or 10 seconds |
| Resolution | 1080p and 4K | Six sizes, 1280x720 at 16:9 among them |
| Frame rate | Not published | 24fps |
| Audio with the video | “generating all audio natively” | Absent from the Gen-4 spec sheet |
| Text-to-video | Documented | Input image required on both Gen-4 models |
| Character reference | Reference images, first and last frame | “a single reference image” |
| Published rate | Nothing on the model page | 12 credits per second, plans from $0 |
| Marking | SynthID marking for generated video | “No watermarks” on the Standard plan |
Two rows are empty on purpose. Google prints no frame rate, and Runway’s Gen-4 sheet prints no audio.
Duration and resolution: two sheets, two shapes
Google prints one length, “Veo videos are 8 seconds long”, then documents the way around it: with Extend you “can create longer videos, even lasting for a minute or more”, generated from the last second of the clip before.
Runway prints two options — “Gen-4 creates videos in 5 or 10 second durations based on an input image and text prompt you provide” — and its help article explains the choice. Simple actions suit five seconds; several movements benefit from ten.
Resolution diverges harder. Google prints “Generate outputs in 1080p and 4K”. Runway’s Gen-4 sheet prints six sizes, from 1280x720 at 16:9 to 1584x672 at 21:9, all around one megapixel, at 24fps. You will see Gen-4 called a 4K model. On Runway’s own pages, 4K is a plan’s upscaling feature, not a Gen-4 output size.
Audio: the row that decides most briefs
Google is blunt: Veo adds sound effects, ambient noise and dialogue, “generating all audio natively”. The Gen-4 sheet has no audio field, and Runway’s developer docs file audio under separate models with their own rates. Runway can do sound. It is a second purchase and a second step there, and part of the same generation here.
The rate card Runway prints for Veo 3.1
Almost no comparison mentions this. Runway’s API docs list Veo 3.1 as a model you can call through Runway, with a published per-second rate.
| Model on Runway’s API | With audio | Without audio |
|---|---|---|
| veo3.1 | 40 credits per second | 20 credits per second |
| veo3.1_fast | 15 credits per second | 10 credits per second |
Runway states that credits cost one cent each, so its own rate works out at $0.40 per second for Veo 3.1 with audio. That is Runway’s price for reselling the model, not Google’s price for making it.
What a clip costs, and whose number it is
Google’s Veo page publishes no price, no tier, no credit balance and no free allowance. Runway publishes all four: a free tier with 125 one-time credits, then Standard at $12 a month billed annually, Pro at $28 and Max at $76.
For per-second arithmetic use the API rate card, not the plan page, because the plan page quotes credits in seconds of video and its own lines disagree about Gen-4 Turbo. The card gives one number per second — Gen-4 at 12 credits, Gen-4 Turbo at 5. Google publishes nothing comparable, which is why the price and free access guide refuses to convert a Veo credit balance into a clip count.
Worked example: the same beat on both models
One person, one line of speech, a locked camera, eight seconds. Written for Veo 3.1 first, because the audio row makes it a one-step job there.
Case one: a talking shot on Veo 3.1
Framing and motion: medium shot, static camera at eye level
Style: documentary naturalism, 35mm, shallow depth of field
Lighting: soft window light from camera left, no fill
Character: a woman in her late thirties, dark curly hair, green work apron
Location: a small bakery counter, trays of bread behind her
Action: she wipes the counter, looks up and speaks, then returns to work
Dialogue: "We sold out by nine. That has never happened."
Audio: quiet kitchen room tone, a tray set down on metal, no musicParameters: one generation, 1080p or 4K, native audio. Two things to check. Whether the line lands whole, because Google states that natural spoken audio for shorter segments remains an area of active development. And whether the tray sound lands on the tray, which the prompt guide covers as matching audio to visuals.
Case two: the same beat on Gen-4
Gen-4 works from a still, so one beat becomes two decisions. The input image comes first: the same woman behind the same counter, lit from the left, framed 16:9. Runway’s article says the image “acts as the first frame of your output video” and that the prompt “should be almost entirely focused on describing the desired motion”.
She wipes the counter in two slow passes with her right hand, sets the cloth down
and straightens up. She looks into the lens, holds the look for a beat, then turns
back toward the bread trays. Camera stays locked. No other people enter frame.Parameters: 5 seconds, 16:9 at 1280x720, 24fps, 60 credits on the published rate. The dialogue moves to post, where Runway sells sound separately. For the speech half in more detail, the audio and dialogue guide goes further.
We have not run the two side by side, and this page is not a quality verdict.
Where the third-party numbers come from
Search this comparison and three corrections keep returning. Gen-4 gets listed at 1080p or native 4K, while Runway’s own sheet stops around one megapixel. Veo 3.1 gets a duration menu, while Google prints eight seconds and documents Extend. And both models get per-second prices with no stated owner, some of them a hosting platform’s own rates. The Veo 3.1 vs Kling 3.0 comparison runs the same separation on a vendor that does print a full card.
What neither vendor has published
Google’s Veo 3.1 page publishes no price, no tier, no credit conversion, no free allowance, no dialogue language list and no visible watermark policy, though it states that output is marked with SynthID. Runway’s Gen-4 pages publish no audio support for the model, and no professional delivery formats for Gen-4 — those are documented for Gen-4.5 and Aleph 2.0.
Neither vendor publishes a joint benchmark. Each shows internal comparisons on its own page, which is a vendor testing itself.
Frequently asked questions
Which model makes longer clips? Gen-4, on the published figures: 5 or 10 seconds against a fixed 8. Google documents Extend for longer work, up to a minute or more.
Does Gen-4 generate audio? Its spec sheet has no audio row. Runway sells audio through separate models, billed separately.
Which is cheaper? There is no single answer. Google publishes no price for Veo 3.1, and Runway publishes both a plan ladder and a per-second rate for the same model on its own API.
Can I run both from one place? Runway’s API docs list Gen-4.5, Gen-4 Turbo and Veo 3.1 as available models, so the choice is not always either-or. The API and Vertex AI pricing guide covers the developer route into Veo.
Two spec sheets, two different jobs. Pick by the rows that decide your brief — audio and duration — and price the rest from the vendor’s own rate card rather than from a comparison table.