Veo 3.1: what Google's video model can do with sound

What Veo 3.1 is

Veo 3.1 is a Google DeepMind video model that creates a clip with sound in one pass. Google introduced it on 15 October 2025 as an update to Veo 3. The model follows the prompt more closely, holds a character more accurately across scenes and works with richer audio, from dialogue to effects that match the picture.

The family has three models. Veo 3.1 is the highest quality and goes to 4K. Veo 3.1 Fast is faster and cheaper. Veo 3.1 Lite is the most affordable and came out on 31 March 2026. In January 2026 Google added the vertical 9:16 format for reference mode and upscale to 1080p and 4K.

Veo 3.1 is not Google’s latest video model. In May 2026 Gemini Omni came out, and in the Gemini API docs Google recommends it as the default. Veo 3.1 remains the choice for extending scenes, last-frame mode and 4K.

What Veo 3.1 can do

Sound with the video. Veo 3.1 generates speech, effects and background noise together with the picture. It does not lay them on afterwards. A line in quotes becomes the character’s speech.

Photo animation. Upload an image as the first frame and describe the motion. The model keeps the face, clothing and light of the source.

References. Up to three images of a character, object or scene. The model keeps their look for the whole clip. That is how you make a series with one character or one product.

First and last frame. You set two pictures, and the model builds a transition between them with sound. Useful for transformations: an empty room into a furnished one, a sketch into a finished product.

Extension. In the Gemini API you can extend a clip by 7 seconds up to 20 times, 148 seconds in total. The model continues the action from the last second of the previous clip.

What Veo 3.1 is good for

• Ads and product shots. Close-ups of a product with sound: coffee pouring, a package opening, a bottle turning.
• Short clips for social media. Vertical 9:16 with no crop, speech and music in one file.
• Animating photos. A portrait, a product or an interior becomes an 8-second clip.
• Storyboards and previz. A scene with camera move, light and sound before a real shoot.
• Atmosphere shots. Nature, weather and city views with ambient sound.

Veo 3.1 and similar models

	Veo 3.1	Gemini Omni 1.1 Flash	Seedance 2.5
Developer	Google DeepMind	Google DeepMind	ByteDance Seed
Released	15 October 2025	27 August 2026	2026
Length per generation	4, 6, 8 s	3–10 s	up to 30 s
Extension	up to 148 s (API, 720p)	up to 40 s	2 extensions
Max resolution	4K	4K (upscale)	n/a
Sound	yes	yes	yes
References	up to 3 images	images and up to 3 videos	text, images, video, audio
In AI2Face	yes	yes	yes

Veo 3.1 wins on frame quality and sound in a short clip, and on last-frame mode. Gemini Omni Flash is more flexible with inputs, and in the API it costs the same as Veo 3.1 Fast, which is less than full Veo 3.1. Seedance 2.5 makes the longest clips in one generation, and it is available in AI2Face too.

How to write prompts for Veo 3.1

Google suggests a five-part formula: camera + subject + action + setting + style and mood. Describe sound separately:

• Dialogue — in quotes: The woman says: "It's time for us to leave."
• Effects — after the label SFX:: SFX: thunder rumbles in the distance.
• Background — Ambient noise: or “background sound”: quiet café hum.
• Camera move — operator terms: dolly in, tracking shot, crane, drone flyover, POV.
• Several shots in one clip — timecodes: [00:00–00:02] medium shot…, [00:02–00:04] reverse angle….
• Exclusions — by description, not by negation: “a deserted landscape with no houses or roads” instead of “no buildings”.

A drone flies toward a white lighthouse on a cliff in a storm at dusk, huge waves
break on the rocks and throw up spray, the lighthouse beam cuts through heavy rain.
SFX: roaring waves, howling wind, distant rolls of thunder. Cinematic, dark blue tones

You can write the prompt in your own language, but camera terms in English give the model less ambiguity. Google does not publish the list of languages Veo 3.1 can speak. Test a spoken line in your language on a short clip.

Veo 3.1, Fast and Lite: how they differ

	Veo 3.1	Veo 3.1 Fast	Veo 3.1 Lite
Best for	final quality	fast iterations	drafts and bulk clips
Resolution	720p, 1080p, 4K	720p, 1080p, 4K	720p, 1080p
References	up to 3	up to 3	none
First and last frame	yes	yes	yes
API extension	yes	yes	no
Price	base	lower	Google says under 50% of Fast

In the AI2Face generator, 4K is available only on full Veo 3.1. Fast and Lite go up to 1080p. Start with Lite or Fast to find a prompt that works, then make the final on full Veo 3.1.

How to use Veo 3.1 in AI2Face

1. Open the generator, choose video, and pick the model: Veo 3.1, Fast or Lite.
2. Choose a mode: from text, from a photo, from references, or from a first and last frame.
3. Set the aspect ratio (16:9 or 9:16), the duration (4, 6 or 8 seconds) and the resolution.
4. Turn sound on if you need it, and start. The credit cost is shown before you start.

A finished image from the generator can become the opening frame of a clip right away. A portrait from Nano Banana 2, a model in the Nano Banana family, can turn into video. Veo 3.1 is available in the AI2Face generator, and the cost is shown before you generate.

Limits

• 1080p and 4K are only for 8-second clips. Shorter clips are generated at 720p.
• References work only at 8 seconds.
• Clip extension is available in AI2Face. In the Gemini API, Google extends a clip only at 720p.
• People in the frame. In the EU, the UK, Switzerland and MENA countries, Google allows only generation of adults. The model does not create recognizable people, celebrities or brands.
• Two-frame transitions are not always smooth. The final frame matches, but details in the middle can drift.
• API storage is 2 days. Clips made through the Gemini API are deleted from Google’s server after two days. In AI2Face the clip stays in your history.
• Labeling. Every clip carries an invisible SynthID watermark.

Conclusions

Veo 3.1 is the strongest choice when you need a short clip with solid sound: ads, product shots, atmosphere, photo animation. Speech and effects from the prompt skip a separate voiceover step. For long clips in one generation, look at Seedance 2.5. For flexible editing and mixed inputs, look at Gemini Omni Flash. A practical way to save credits: tune the prompt on Lite or Fast at 720p, then make the final on Veo 3.1 at 1080p or 4K. All three versions are available in the AI2Face generator, and the cost is shown before you generate. Finished images from the generator become the first frame of a clip.