Google Veo for architecture videos: Veo 3.1, Gemini Omni and Flow explained for renderings

You heard Google Veo makes the most convincing AI video and want to try it on a rendering, but the name has become a moving target. In October 2026 the Gemini app switched to a new model, the developer preview of Veo 3.1 shut down, and Veo now lives mainly inside Google Flow and Google Cloud. This guide sorts out where you can still use Veo for architecture videos, how to turn a rendering into a clip with a first and last frame, what the native audio and SynthID watermark mean for a studio, and what a project's worth of clips costs in Flow credits.

Key takeaways

• As of October 2026, Google Flow offers Veo 3.1 Lite, Fast and Quality alongside Gemini Omni Flash 1.1, and all of them can start a clip from a first frame or a first and last frame.
• The Gemini app's video generation now runs on Gemini Omni 1.1 Flash, which Google says replaces Veo 3.1 there; it makes 10-second videos from up to five photo references.
• Veo 3.1 Preview in the Gemini API shut down on October 22, 2026; the generally available Veo 3.1 models on Google Cloud list a discontinuation date of "November 17, 2026 or later."
• Every video made in the Gemini app carries a SynthID watermark, which is invisible and does not replace a written disclosure in your caption.
• Veo and Omni generate sound with the picture, which for a project film usually means replacing it with a licensed track before you post.

Can you still use Google Veo for architecture videos?

Yes, but mainly through Google Flow. As of October 2026, Flow offers Veo 3.1 Lite, Fast and Quality, and each can animate a rendering from a first frame or from a first and last frame, in clips of up to 8 seconds. The Gemini app has moved to Gemini Omni 1.1 Flash, and the Veo 3.1 developer preview shut down on October 22, 2026, so older tutorials that send you to either place for "Veo" are out of date.

For a studio, that makes Flow the practical place to work: it shows both model families side by side, lets you pin the start and end of a camera move to your own images, and bills everything in one credit balance. The rest of this guide assumes Flow unless it says otherwise.

Where Veo runs, as of October 2026

Google now has three consumer and developer surfaces for video, and the model behind each one is different. The table is built from Google's Flow model guide, the Gemini video overview, the Gemini API Veo page and the Veo 3.1 model card on Google Cloud.

Google describes Gemini Omni 1.1 Flash as a "multimodal AI video generation and editing model that will now replace the previous Google Gemini Veo 3.1 model" in the Gemini app. On Google Cloud, Veo 3.1 lists output at 720p, 1080p or 4K, 9:16 and 16:9 aspect ratios, audio, and a discontinuation date of "November 17, 2026 or later," so plan for the name to keep changing. If you rely on any one of these for client work, keep your source images and prompts somewhere you control; the Sora shutdown checklist applies just as well to a Google model retirement.

Turning a rendering into a Veo clip in Flow

Flow calls image-led generation Frames to Video. For architecture it is the mode to use, because it ties the clip to your image instead of to a text description of a building.

1. Choose the model before the image. Veo 3.1 Lite and Fast are the cheap ones for testing camera moves; Quality costs far more and is worth it only once the move works.
2. Set the aspect ratio first. Flow's model guide lists both orientations for its text and frames modes; pick 9:16 for Reels and Shorts rather than cropping a 16:9 clip later.
3. Add a first frame, and a last frame if you have one. Two renderings of the same view a few steps apart give the model both ends of the move, so it invents less of the building.
4. Write one camera move and what must not change. "Slow push-in toward the entrance, eye level, no people, no added objects, materials unchanged" is a better prompt than a paragraph of mood.
5. Generate short, review, then go longer. Start with 4 or 6 seconds; a longer clip gives geometry more time to drift.

Flow also offers Ingredients, its reference-image mode, on Veo 3.1 Lite and Fast at 8 seconds and on Omni Flash 1.1 at all lengths. References help hold a material or a piece of furniture, but they do not lock a building's geometry the way a first frame does. AI video prompts for architecture ranks camera moves by how well they keep lines straight.

What Veo and Omni do to buildings

Google pitches Veo and Omni on realistic motion and light, and on a rendering the appeal is convincing reflections, moving foliage and soft shifts in daylight. The limits are the same as every image-to-video model, because the model sees pixels, not your plans.

The Gemini API page also warns that Veo "will sometimes block a video from generating because of safety filters or other processing issues with the audio," with no charge for blocked attempts. On a rendering with nothing controversial in it, that is rare but not impossible; just re-run it.

Audio: a feature you will usually replace

Veo generates sound natively, and the Gemini app lets you "ask Gemini to generate audio with your video." For a film or ad, that is a headline feature. For a building, generated footsteps, wind or murmured speech rarely help, and a 6-second clip with its own soundscape is hard to cut into a longer edit. Most studios mute the generated track and lay one licensed piece of music across the whole edit, so the film plays the same on Instagram, LinkedIn and your website.

If you keep any generated audio, listen for speech. Invented dialogue in a project film reads as a person saying something about your building that nobody said.

SynthID, disclosure and where the file goes next

Google states that "all videos generated in the Gemini app are embedded with SynthID, our imperceptible watermark for identifying Google AI-generated content," and that you can upload a file and ask Gemini whether it was made with Google AI (Gemini video overview). The Gemini API page says the same for Veo: "Videos created by Veo are watermarked using SynthID." As of October 2026, those are the two statements Google makes on the pages read for this guide.

SynthID is invisible. Your viewers will not see it, and a social platform may or may not detect it. Write the disclosure in your caption either way: "AI-generated concept video from design renderings. Not a built project." Content credentials and AI labels explains which platforms read which signals, and AI rendering disclosure for architects has wording for built and unbuilt work.

What a project costs in Flow credits

Flow bills each generation in credits, and Google AI plans add a monthly allowance on top of a small daily one. As of October 2026, from Google's Flow credits page:

Allowances on the same page: 50 Flow credits a day without a subscription, plus 200 a month on Google AI Plus, 1,000 on Google AI Pro, 10,000 on Google AI Ultra at USD 100 and 25,000 on Ultra at USD 200. Unused monthly credits "do not roll over." Upscaling to 1080p costs 0 credits on Plus, Pro and Ultra, and 4K costs 50 on Ultra only. Non-subscribers may be unable to generate during peak hours, roughly 2 PM to 5 PM UTC. Google's plan page did not display Plus and Pro prices when read, so those could not be verified as of October 2026.

Remember the re-rolls. Six clips at Veo 3.1 Fast are 120 credits; if half need a second attempt, budget 180, then add your own editing and music time.

Veo clips or a finished six-image film

Veo in Flow is the right tool when you want a few striking clips, need a landscape format, or enjoy directing each shot. It is a lot of work when the deliverable is a 30-second vertical film of one project, because you generate, review and assemble six or more clips and score them yourself.

Arch2Video is built for that second job: you upload exactly six images from one project, its AI director plans six connected five-second scenes, and you get a 1080 × 1920 MP4 of about 30 seconds with an original soundtrack, for USD 15 per video as a one-time payment. It offers no prompt box and no per-shot control, every film includes a small fictional cast in at least one scene, and generation usually takes an hour or more. The comparison of AI video generators for architects puts Veo beside Runway, Kling and Luma.

Frequently asked questions

Is Google Veo 3.1 being shut down?

Partly. Veo 3.1 Preview in the Gemini API shut down on October 22, 2026, and Google recommends Gemini Omni Flash or the generally available models on Google Cloud instead. The Gemini app has moved to Gemini Omni 1.1 Flash. Veo 3.1 Lite, Fast and Quality remain in Google Flow as of October 2026, and the Google Cloud model card lists a discontinuation date of "November 17, 2026 or later."

Can the Gemini app make a vertical video from my rendering?

Yes, with limits. Google's help page says you "choose the aspect ratio of the video before it's created" and can upload up to five images; the app makes 10-second videos, requires a Google AI plan for personal accounts, and is not available to users under 18, as of October 2026. For more control over a building, use Frames to Video in Flow, where your rendering becomes the clip's first frame.

Does a Veo video carry a watermark?

Every video generated in the Gemini app is embedded with SynthID, and Google's API page says the same of Veo. SynthID is invisible, so it does not tell your viewers anything; whether a visible mark also appears depends on the surface and plan, so check your exported file. Either way, add a written disclosure in the caption for AI-generated work.

Can Veo make a walkthrough of my building from the model?

No. Veo and Omni work from images, video and text, not from a BIM or 3D model, and they generate clips of up to 8 or 10 seconds. A sequence of clips made from renderings is a series of views, not a measured route through the plan. For a true walkthrough, export the camera path from your renderer; use Veo for short atmospheric shots around it.