Veo 3 was the Google DeepMind model that made video generation audible: announced at Google I/O on May 20, 2025, it generated dialogue, sound effects and ambience together with the picture. Veo 3.1, released on October 15, 2025, keeps that and improves on it. Google says it brings "richer audio, more narrative control, and enhanced realism", "stronger prompt adherence", and better picture and sound when turning images into video.
Is it worth the upgrade? For most people the question has already been answered: Google now lists Veo 3 as deprecated in the Gemini API and offers Veo 3.1 in its place, in three tiers: Veo 3.1, Veo 3.1 Fast and Veo 3.1 Lite. What remains useful to know is what actually changed, so you know what to expect when you move your prompts across.
The short answer
- Sound: both generate audio with the video; Google describes Veo 3.1's as richer.
- Prompt following: Veo 3.1 sticks more closely to what you write, from the order of actions to the camera move.
- Image to video: Google singled out better picture and sound when you start from a photo.
- New controls: first and last frames, and up to three reference images, arrived with Veo 3.1.
- Resolution: Veo 3.1 added a 4K option in January 2026; Veo 3 tops out at 1080p.
- Your prompts: Veo 3.1 reads the same prompt structure, so you can paste your Veo 3 prompts in. Expect a new take, not the exact clip you got before.
Veo 3 vs Veo 3.1 at a glance
| Veo 3 | Veo 3.1 | |
|---|---|---|
| Announced | May 20, 2025 (Google I/O) | October 15, 2025 |
| Status in Google's API | Deprecated | Current, in three tiers (Veo 3.1, Fast, Lite) |
| Sound generated with the video | Yes | Yes, richer |
| Follows the prompt | Yes | More closely than Veo 3 |
| Photo to video | Yes | Yes, with better picture and sound |
| First and last frames | — | Yes |
| Reference images | — | Up to 3 |
| Clip length in Google's API | 8 seconds | 4, 6 or 8 seconds (8 s for 1080p and 4K) |
| Resolution | 720p, 1080p | 720p, 1080p, 4K (upscaled) |
| Aspect ratios | 16:9; Google added 9:16 in September 2025 | 16:9, 9:16 |
| Extension in Google's tools | — | Yes (+7 s per step in the API; a minute or more in Flow) |
Sound: the feature Veo 3 introduced
Before Veo 3, Veo clips were silent and any voice or sound had to be added in an edit. Veo 3 changed that: characters could speak lines written in the prompt, and effects and background noise arrived with the picture.
Veo 3.1 keeps native audio on every tier and, in Google's words, brings "richer audio". The way you ask for sound has not changed:
- put spoken lines in quotes and say who speaks;
- name the sound effects you want ("a door slams", "rain on the window");
- describe the ambience ("café chatter", "distant traffic").
If you leave sound out, the model still builds a soundtrack for the scene. For a detailed walk-through, see our Veo 3.1 prompt guide.
Closer prompt following
Google's headline claim for Veo 3.1 is "stronger prompt adherence". In practice this matters most for prompts that describe several things in order: an action, then a reaction; a specific camera move; a named style. The more precisely you write the shot, the more of it you get back, which makes a precise prompt worth the effort.
This is also why old Veo 3 prompts transfer well. Veo 3.1 reads the same kind of prompt, so you can paste it unchanged and then tighten the parts that Veo 3 used to ignore.
Better image to video
Google specifically called out "improved audiovisual quality when turning images into videos" in the Veo 3.1 release. When you start from a photo, that photo is the first frame, and the prompt only needs to describe the motion, the camera and the sound. This is the mode to use for product shots, portraits and any still you need to keep.
New controls in Veo 3.1
First and last frames
Give Veo 3.1 a start image and an end image and it builds the motion between them, with sound. Use it when a shot has to land on a known composition: a reveal, a product turning to face the camera, a transition from one scene to the next. Veo 3 had no equivalent.
Reference images
Up to three images of a person, character or product tell the model what they look like; the model then places them in a new scene. Google calls this Ingredients to Video, and in January 2026 it extended it to native vertical 9:16 video. It is the most direct way to keep a character or product consistent across several clips.
Extension
In Google's own tools, Veo 3.1 clips can be extended: the Gemini API adds 7 seconds per step, up to 20 times, at 720p, and Flow's Extend feature builds continuous videos of a minute or more.
Resolution: 4K arrives in January 2026
Veo 3 generates at 720p or 1080p; Google added 1080p and vertical 9:16 to Veo 3 in September 2025. On January 13, 2026, Google added a 4K option to Veo 3.1, which it describes as "state-of-the-art upscaling". Treat 4K as a sharper master for large screens rather than extra generated detail.
Is it worth moving to Veo 3.1?
In most cases the choice is already made: Veo 3 is deprecated in Google's API, so new work belongs on Veo 3.1. The useful question is how to make the move smoothly:
- Reuse your prompts. The structure (scene, camera, dialogue in quotes, sound) is the same, so paste your Veo 3 prompts in as a starting point.
- Tighten what Veo 3 ignored. Because Veo 3.1 follows prompts more closely, details you gave up on, like the exact order of actions or a specific camera move, are worth writing back in.
- Try the new controls. If a shot has to end on a particular frame, add an end frame. If a character has to look the same across clips, use reference images.
- Pick the tier by job. Draft on Veo 3.1 Lite, use Veo 3.1 Fast for everyday clips and reference images, and use the full Veo 3.1 tier for the take you publish.
Using Veo 3.1 on Veo Video
Every Veo video you make on Veo Video runs on Veo 3.1, including the generator on our Veo 3 page. There is no separate Veo 3.0 model here. Each generation is an 8-second clip with sound, in 16:9 or 9:16, at 720p, 1080p or 4K. You can start from text or a photo and add an end frame on any tier. Reference images run on Veo 3.1 Fast, with up to three images per clip.
New accounts get 30 free credits, enough for one Veo 3.1 Lite clip at 720p, a quick way to check how one of your old Veo 3 prompts plays on Veo 3.1. When it works, render the keeper on Veo 3.1 at 1080p or 4K.