What’s new
- Consistent references (Ingredients to Video): feed it up to 3 reference images and a character’s look or a style can hold across several shots; the January 2026 update added native 9:16 vertical output for this mode plus better dialogue and character/background consistency.
- Scenes over a minute long (Scene Extension): you’re not capped at 8 seconds — Google says you can keep extending a clip and stay coherent past a minute, adding 7 seconds per extension for up to 20 extensions, though the footage you extend from has to be 720p.
- First and last frame, AI fills the middle (Frames to Video): give it a start and end frame and it generates the camera movement and transition between them; unlike before 3.1, this mode now comes with synchronized audio too.
- Up to native 4K, vertical included: beyond the default 720p, the 8-second option can render at 1080p or 4K, and paired with Ingredients to Video it can output native 9:16 vertical without cropping.
What it’s good for
- A piece where a character or product has to look the same across multiple shots: hand references to Ingredients to Video.
- A story that needs more than 8 seconds with a real beginning, middle and end: generate one clip, then keep going with Scene Extension.
- You already have a start and end frame and want the camera move and transition between them handled automatically: use Frames to Video.
Prompting tips
- Build the prompt in Google’s own order: camera move, subject, action, scene, style/mood — that structure works better than a pile of adjectives.
- Keep camera move and composition separate: dolly, tracking shot, crane, aerial, slow pan describe movement; wide shot, close-up, extreme close-up, low angle describe framing — don’t blend the two.
- Put dialogue in quotes: write it as
she says, "we should go now."so the model treats it as spoken line rather than a scene description. - Tag sound effects and ambience separately: something like “SFX: distant thunder” and “ambience: low hum of a starship cockpit” works better than a vague “nice sound”.
- Write exclusions as positive descriptions: instead of “no man-made buildings”, write “bare ground, no houses, no roads” — a concrete description beats a negation.
- Break long shots into timestamps: a single prompt can use segments like [00:00–00:03], [00:03–00:06] to control multiple camera moves within one shot.
16:9, 8 seconds. Low-angle tracking shot of an old sea captain in a worn leather jacket standing on a fishing boat deck, looking out at the horizon, sea wind lifting his white hair.
A foggy harbor at dawn, the rising tide slapping against the hull. Realistic, nostalgic look, film grain, cold morning light.
[00:00–00:03] The captain turns toward camera. [00:03–00:06] He raises a hand and points toward a distant lighthouse. [00:06–00:08] The camera slowly pulls back to reveal the whole harbor.
He says, "Time to set sail." SFX: waves against the hull, rope creaking. Ambience: distant gulls, a foghorn.
An empty deck, no other crew, no modern equipment.
Known limits
- Google says natural, coherent dialogue — especially short lines — is still being refined, and lines can come out disjointed sometimes.
- 1080p, 4K, and reference images (Ingredients to Video) all require the 8-second duration; the 4- and 6-second options only come with default 720p generation.
- Scene Extension only works on Veo 3.1 and Veo 3.1 Fast, not the Lite version, and the footage you extend from must itself be 720p.
Using it in Nomi
On the generation canvas, add a video card, pick “Veo 3.1” in the model picker and choose a mode (Text to video, First frame to video, First and last frame). Add your references, write the prompt and generate. In the storyboard you can pick it per shot. The first time, connect any one of the providers above under Settings → Models.
Sources
- Introducing Veo 3.1 and new creative capabilities in the Gemini API (Google Developers Blog, 2025-10-15)
- Veo 3.1 model page (Google DeepMind)
- Veo 3.1 technical docs (Google AI for Developers, Gemini API)
- Veo 3.1 Ingredients to Video update (Google official blog, 2026-01-13)
- Ultimate prompting guide for Veo 3.1 (Google Cloud Blog, 2025-10-15)
- Veo 3.1 video generation API reference (Kie.ai)
