Tested on NOLGIA

Veo 3.1 vs Seedance 2.5

On our scored street scene both passed all eighteen checks. The differences were the camera, which Veo 3.1 moved far more than a slow push in, the render time, where Veo 3.1 came back sooner, and what each takes: Veo 3.1 renders 4, 6 or 8 second clips with speech, Seedance 2.5 runs up to 30 seconds and edits your own clips.

Our own test, run on September 23, 2026. Every clip below is a real output from it. Model pages: Veo 3.1 and Seedance 2.5.

The results at a glance

Two runs each on 23 September 2026. Render times are one reading each; provider queues change through the day.
MeasureSame inputVeo 3.1Seedance 2.5
Nine checks, two runsOne text prompt18 of 1818 of 18
How much the picture grew on a slow push inSame prompt1.76 and 2.65 times1.22 and 1.10 times
Frame and length returned720p tier, 6 seconds asked1280 by 720, 24 fps, 6.0 s1280 by 720, 24 fps, 6.1 s
Video bitrateSame prompt9.0 and 12.4 Mb/s10.4 and 8.3 Mb/s
Request to finished fileStarted within five minutes of each other84 and 113 seconds179 and 313 seconds

Side by side, on the same inputs

The scored street scene, first run

Same input: One bakery storefront prompt with nine checkable details: a sign spelled BAKERY, a red bicycle left of the door, three pots right of it, a yellow umbrella, a ginger cat walking left to right, a slow push in, and no people. 720p, 6 seconds, two runs each.

Veo 3.1's first run: the bakery storefront with the BAKERY sign, a red bicycle, a yellow umbrella and a ginger cat
Veo 3.1. Run 1: 9 of 9. The camera travels much further than a slow push in.
Seedance 2.5's first run: the same storefront with the sign, the bicycle, three pots and the cat, the camera pushing in
Seedance 2.5. Run 1: 9 of 9. Its window lettering is invented and changes from frame to frame.

Both passed every check. Veo 3.1's picture grew 1.76 times between 10% and 90% of the clip, Seedance 2.5's 1.22 times.

Method and every run: One prompt, six video models, nine measured checks

The same prompt, second run

Same input: The same prompt and settings, run again.

Veo 3.1's second run: a large camera move that takes the sign out of the top of the frame
Veo 3.1. Run 2: 9 of 9. The biggest camera move of the twelve clips in the test.
Seedance 2.5's second run: the bakery storefront with a gentle push in
Seedance 2.5. Run 2: 9 of 9.

Both passed again. Veo 3.1's picture grew 2.65 times, so the sign left the top of the frame; Seedance 2.5's grew 1.10 times.

Method and every run: One prompt, six video models, nine measured checks

What each costs today

Read from the live price list when this page loads. The Generate button shows the exact price of your settings before anything runs.

Veo 3.1

per 5s clip · Pro and up

QualityCreditsUSD
720p112$7.84
1080p112$7.84

Seedance 2.5

per 5s clip · Pro and up

QualityCreditsUSD
480p26$1.82
720p56$3.92
1080p137$9.59

Dollar figures use the 100-credit top-up price, $0.07 per credit.

What each takes and returns

From the public model catalog on 24 and 30 September 2026. Each model page shows today's version.
Veo 3.1Seedance 2.5
Length4, 6 or 8 seconds4 to 30 seconds
Frame shapes16:9, 9:1616:9, 4:3, 1:1, 3:4, 9:16, 21:9
SoundAlways on, with speechOptional, with speech in the same pass
From a photoA start frame, or up to 3 reference images, not bothA start frame, an end frame, up to nine reference images
From a clip of yoursNoReference, edit or extend it

Which to pick

Your shotPickWhy
A detailed brief with things that must not appearEitherBoth passed 18 of 18 on the scored scene
A slow, controlled camera moveSeedance 2.5Veo 3.1 moved 1.76 and 2.65 times where a slow push in asked for less
A quicker first draftVeo 3.184 and 113 seconds against 179 and 313, one reading each
A take longer than 8 seconds, or a square or 21:9 frameSeedance 2.5Veo 3.1 renders 4, 6 or 8 seconds at 16:9 or 9:16
Edit or extend a clip you already haveSeedance 2.5Veo 3.1 takes no video in

What these tests do not tell you

  • It is one prompt and one kind of scene: a still street, one animal, one slow camera move. Hands, fast action, dialogue or several people can rank the models differently.
  • There are no people in it, so it says nothing about how either model draws a person, and nothing about speech.
  • Two runs each, on the 720p tier.

The tests behind this page

  • One prompt, six video models, nine measured checksSeptember 23, 2026One prompt with nine details a tool can check, sent unchanged to six models on the same morning, two runs each at 720p and 6 seconds, scored with an object detector, text recognition and feature tracking.

FAQ

Questions about Veo 3.1 vs Seedance 2.5.

Is Veo 3.1 better than Seedance 2.5?
On our scored scene they tied at 18 of 18. Veo 3.1 came back sooner and moved the camera much more; Seedance 2.5 kept the slow move the prompt asked for and runs longer takes.
Which one should I use for a spoken line?
This test has no dialogue, so it does not answer that. Our Veo 3.1 guide shows a spoken line from a still, and our pick for a person who talks is in the post on the best AI video model for each job.
Which one costs less?
It depends on the tier and the length. The prices on this page are read live from the price list, and the Generate button shows the exact price of your settings before you run.