NanoBananas offers three main families of AI video models: Veo from Google DeepMind, Seedance from ByteDance and Wan from Alibaba's Tongyi Lab. They overlap, but each has a clear center of gravity. Veo makes polished short clips with sound included. Seedance makes the longest single takes and handles references and cinematic formats best. Wan makes 10 to 15 second clips at the lowest cost per second and supports the most feed formats.

Everything here reflects the NanoBananas video studio on November 9, 2026: the models it offers, the lengths, formats and modes it supports for each, and its credit prices. Release dates and background come from the developers' own announcements, linked in the sources section. Upstream APIs sometimes offer more than the studio exposes, and those extras are not listed here.

The three families at a glance

Capabilities as offered in the NanoBananas video studio.
Veo 3.1Seedance 2.0 / 2.5Wan 3.0
DeveloperGoogle DeepMindByteDance SeedAlibaba Tongyi Lab
ReleasedOct 2025; Lite Mar 20262.0 Feb 2026; 2.5 Jul 2026Aug 2026
Longest clip in the studio8 s15 s (2.0), 30 s (2.5)15 s
Top resolution4K (upscaled on Fast and Lite)4K (2.0), 1080P (2.5)1080P
Aspect ratios16:9, 9:1616:9, 9:16, 1:1, 4:3, 3:4, 21:916:9, 9:16, 1:1, 4:3, 3:4
Start and end framesYesYesYes
Image referencesFast and Lite: up to 3 (8 s clips)Up to 7 (2.0 Fast: 9)Up to 10
Video referencesNoYesYes, 1 to 15 s
SoundAlways onOptionalOptional

Every model in the studio

Each family has several versions at different price points. This is the full line-up you can pick from.

"All four" means text to video, first frame, start and end frames, and reference mode. Wan 2.6, 2.6 Flash and 2.2 Turbo are 16:9 only.
ModelClip lengthsResolutionsModesCredits per second (720P)
Veo 3.1 Quality4, 6, 8 s720P, 1080P, 4KText, first frame, start and end22
Veo 3.1 Fast4, 6, 8 s720P, 1080P, 4KAll four, incl. image references6
Veo 3.1 Lite4, 6, 8 s720P, 1080P, 4KAll four, incl. image references3
Seedance 2.54 to 30 s480P, 720P, 1080PAll four22
Seedance 2.04 to 15 s480P, 720P, 1080P, 4KAll four15
Seedance 2.0 Fast4 to 15 s480P, 720PAll four6
Seedance 2.0 Mini4 to 15 s480P, 720PAll four3
Wan 3.05, 10, 12, 15 s480P, 720P, 1080PAll four6
Wan 3.0 Prime5, 10, 12, 15 s480P, 720P, 1080PAll four9
Wan 2.75, 10 s720P, 1080PAll four8
Wan 2.65, 10, 15 s720P, 1080PText, first frame, video reference5
Wan 2.6 Flash5, 10, 15 s720P, 1080PFirst frame, video reference2 (4 with sound)
Wan 2.2 Turbo5 s480P, 720PFirst frame6

What a clip actually costs

Prices are per second of output, so the fair comparison is a clip of the same length. Veo stops at 8 seconds, so the table uses 8 seconds where every model can make it, and adds a 10-second column for the rest.

Credits in the NanoBananas studio for text to video. The studio shows the exact quote before you generate.
Model8 s at 720P10 s at 720P8 s at 1080P
Veo 3.1 Lite24Not available40
Veo 3.1 Fast48Not available64
Veo 3.1 Quality176Not available200
Seedance 2.0 Mini2430Not available
Seedance 2.0 Fast4860Not available
Seedance 2.0120150288
Seedance 2.5176220440
Wan 3.0Not available60Not available (10 s: 120)
Wan 2.7Not available80Not available (10 s: 120)
  • Sound does not change the price on Veo, Seedance or Wan 3.0. Veo always includes it; on Seedance and Wan 3.0 it is a free toggle. Wan 2.6 Flash is the exception: sound doubles its rate.
  • Cheapest with sound: Veo 3.1 Lite and Seedance 2.0 Mini both cost 24 credits for 8 seconds at 720P.
  • Cheapest long clip: Wan 3.0 makes 15 seconds at 720P for 90 credits; Seedance 2.0 Mini makes 15 seconds at 720P for 45.
  • 1080P gets expensive fast on Seedance 2.5, at 55 credits per second. Draft at 480P or 720P first.

Veo 3.1: polished short clips with sound

Veo is the simplest family to use well. Every clip comes with dialogue, effects or ambience generated together with the picture, and the three tiers let you trade cost for detail without changing your prompt.

  • Strengths: reliable motion and lighting, sound always included, 4K output, strong prompt following for short scenes.
  • Limits: 8 seconds per clip, only 16:9 and 9:16, no video references, image references only on Fast and Lite.
  • Status: Google shut down the Gemini API preview endpoints on October 22, 2026 and lists Gemini Omni 1.2 Flash as the successor. Veo 3.1 remains available in the studio.

For a full breakdown of the three tiers, read Veo 3.1 Quality vs Fast vs Lite.

Seedance: long takes, references and cinematic framing

Seedance is ByteDance's video family and the most flexible in the studio. It accepts the widest range of lengths, the most aspect ratios including cinematic 21:9, and both image and video references.

  • Seedance 2.5: single takes of up to 30 seconds, the longest in the studio. Use it for continuous shots that would otherwise need stitching.
  • Seedance 2.0: up to 15 seconds and the only Seedance model with 4K output.
  • Seedance 2.0 Fast and Mini: the same modes at 480P and 720P for a fraction of the price. Mini is one of the cheapest ways to test motion.
  • Limits: sound is optional rather than built in, and 1080P on 2.5 is costly. Draft low, render high.

Wan: low cost per second and the most feed formats

Wan is Alibaba's family. Its newest model, Wan 3.0, is the value pick for 10 to 15 second clips, and the Wan line covers the most specialised options in the studio.

  • Wan 3.0: 5 to 15 seconds, five aspect ratios including square and 4:3, up to 10 image references and video references of 1 to 15 seconds, at 6 credits per second at 720P.
  • Wan 3.0 Prime: the same model on a faster, higher-priced queue.
  • Wan 2.7: 5 or 10 seconds at 720P or 1080P with a strong reference mode; no sound.
  • Wan 2.6 Flash and 2.2 Turbo: very cheap first-frame animation in 16:9 for quick tests.
  • Limits: 1080P maximum, and older versions are 16:9 only.

Which family for which job

JobPickWhy
Short social clip with dialogue or ambienceVeo 3.1 FastSound is built in and 8 s is enough
Product hero shot in 4KVeo 3.1 Quality or Seedance 2.0Both render 4K; Veo Quality is Google's native-4K tier
One continuous 20 to 30 second shotSeedance 2.5The only model that goes past 15 s
Character or product held steady from referencesSeedance 2.0 or Wan 3.0Up to 7 to 10 image references plus video references
Cinematic widescreenSeedanceThe only family with 21:9
Square or 4:3 feed ads on a budgetWan 3.0Five formats at 6 credits per second
Cheapest motion test of a stillSeedance 2.0 Mini or Wan 2.6 Flash2 to 3 credits per second

How each family uses references

Reference mode lets you guide a clip with images of a character, product or style, and on some models with short video clips for motion. Capacity differs a lot between families.

ModelImage referencesVideo references
Veo 3.1 Fast and LiteUp to 3, 8 s clips onlyNo
Seedance 2.0, 2.5, MiniUp to 7Yes
Seedance 2.0 FastUp to 9Yes
Wan 3.0 and PrimeUp to 101 to 15 s
Wan 2.7Up to 5Up to 15 s

When you add a reference video, the studio bills the output seconds plus the reference seconds at 70% of the normal rate. A 10-second Wan 3.0 clip at 720P guided by a 5-second reference video costs 63 credits: 6 × 15 seconds × 0.7. Describe in the prompt what each reference is for, such as the dancer's movement from the reference video, the outfit from the first image.

Three prompts to compare the families yourself

Run each prompt at 720P on one model from each family, at the closest length each supports, and judge the checkpoints rather than your overall impression.

1. Dialogue and sound

A street food vendor in a busy night market flips noodles in a smoking wok and calls out to the camera: "Two more minutes!" Handheld medium shot, warm string lights, sizzling sound and crowd chatter.

  • Does the spoken line match the lip movement?
  • Do the wok, flames and noodles move believably?
  • Is the ambient sound layered and natural?

2. A long continuous take

One continuous shot following a cyclist from a quiet residential street onto a busy bridge at sunset, the camera tracking alongside, then rising to reveal the river and city skyline. No cuts.

  • Does the shot stay continuous without jump cuts?
  • Do the cyclist and bike keep the same look throughout?
  • Does the camera rise smoothly at the end?

3. Product from a reference

The sneaker from the reference image rotates slowly on a floating white platform in a soft gray studio, light catching the mesh texture and the logo on the side. Calm, premium product video.

  • Do the logo, colors and sole match the reference?
  • Does the mesh texture stay stable as it rotates?
  • Is the rotation smooth and evenly paced?

Mixing families in one project

You do not have to commit to one family. A typical project drafts motion on a cheap model (Seedance 2.0 Mini or Veo 3.1 Lite), renders long shots on Seedance or Wan, and uses Veo for short hero clips with sound. All models share one credit balance in the video studio.

To keep characters and products consistent across models, start every shot from a still made with Nano Banana. The image to video guide shows how. If you are considering Google's newer model, see Gemini Omni 1.2 Flash vs 1.1.

Sources

Frequently asked questions

Which AI video model is cheapest on NanoBananas?

For 8 seconds at 720P, Veo 3.1 Lite (with sound) and Seedance 2.0 Mini both cost 24 credits. Wan 2.6 Flash is cheaper per second at 2 credits, but it only offers first-frame and video-reference modes in 16:9.

Which model makes the longest videos?

Seedance 2.5, with single takes of up to 30 seconds. Seedance 2.0, Wan 3.0 and Wan 2.6 go up to 15 seconds, and Veo 3.1 up to 8 seconds.

Which models can use reference videos?

Seedance and most Wan models support reference clips in the studio. Veo 3.1 does not. Gemini Omni 1.1 Flash also accepts one reference video.

Do these models generate sound?

Veo 3.1 always includes sound. Seedance and Wan 3.0 offer it as a free option. Wan 2.7, 2.6 and 2.2 Turbo do not generate sound, and Wan 2.6 Flash doubles its price with sound on.

Which model is best for 4K video?

Veo 3.1 Quality and Seedance 2.0. Veo 3.1 Fast and Lite also offer 4K, upscaled from the generated clip. Seedance 2.5 and Wan top out at 1080P.

Is Seedance better than Veo?

Neither is better overall. Veo is simpler and always has sound; Seedance is longer, more flexible with references and supports more formats. Run the same brief on both.

Which model supports square video?

Seedance and Wan 3.0, 3.0 Prime and 2.7 support 1:1, 4:3 and 3:4. Veo 3.1 supports only 16:9 and 9:16.