Veo 3.1 Gemini API shutdown: what changed and how to migrate

What shut down, and when

Google's Veo page in the Gemini API docs now opens with a notice that the Veo 3.1 Preview models "in the Gemini API" were shut down on October 22, 2026, and tells developers to "migrate to Gemini Omni Flash (gemini-omni-1.2-flash) or the GA models available through the Gemini Enterprise Agent Platform …" (quoted in part; the full notice is on the page linked below). The Gemini API pricing page no longer lists Veo. Here is the full timeline as Google and fal publish it:

If you reached Veo through fal rather than Google directly, the fal Veo 3.1 page says: "This model will be removed on November 17, 2026. Move to a supported model before then." It does not name a replacement. Our fal alternative page compares the options.

Model ID map: from the preview IDs to what still works

The preview IDs have Google Cloud equivalents. Note that Lite is still a Preview model on Google Cloud, while Veo 3.1 and Veo 3.1 Fast are generally available there.

Option 1: move to Veo 3.1 on Google Cloud

This keeps the exact same models at Google's list price, which is lower per clip than ours, at the cost of setting up Google Cloud. At list price an 8-second Veo 3.1 Fast clip at 1080p with audio is about $0.96 ($0.12 per "count", which we read as one second). The main steps, from Google's documentation:

1. 1

   Set up a billed Google Cloud project

   Veo on Google Cloud is called through the Agent Platform (formerly Vertex AI) API at aiplatform.googleapis.com, inside a project with billing turned on. You typically authenticate with Google Cloud credentials — the gcloud CLI on your machine, or a service account on a server — instead of a Gemini API key.
2. 2

   Swap the model ID and region

   Use the -001 IDs from the table above. Google's Veo 3.1 model page lists a single region, us-central1.
3. 3

   Update the request flow

   Over REST you start a job with predictLongRunning and poll it with fetchPredictOperation. Google's code samples also use the Google Gen AI SDK's generate_videos call configured for Google Cloud, which is the smallest change if you used that SDK with the Gemini API.
4. 4

   Choose where videos land

   Without a Cloud Storage URI, video bytes come back in the response; with one (a gs:// bucket path), Google writes the files there.
5. 5

   Check how you will be billed

   Google's Veo 3.1 model page lists provisioned throughput and fixed quota as supported and marks pay-as-you-go as not supported, while the pricing page shows per-unit rates. Confirm quota and billing terms for your project before you ship. Our Veo API vs Google Cloud comparison goes into the trade-offs.

Option 2: switch to Gemini Omni Flash

Omni Flash is Google's recommended replacement inside the Gemini API, so you keep your API key and billing. It is a different model, though, not a renamed Veo, and the integration changes:

• Requests go through the Interactions API (POST /v1beta/interactions) rather than Veo's video operations, and short videos come back in the response by default.
• Clips run 3 to 15 seconds (default 10), compared with Veo's 4, 6 or 8 seconds.
• Output is 360p or 720p, with 1080p and 4K as upscaled options; 16:9 and 9:16 are both supported.
• It accepts up to six reference images, can use images as first and last frames, and can edit and extend videos (with some editing limits in the EEA, Switzerland and the UK).
• There is no negative-prompt field; Google suggests writing exclusions into the prompt. Audio inputs are not supported.
• If you are on gemini-omni-1.1-flash, requests route to 1.2 from December 3, 2026.

Expect to re-tune prompts: the same text will not produce the same look. Google prices Omni Flash per token on the Gemini API pricing page and gives approximate per-second figures only as a guide. For a model-by-model comparison, see Veo 3.1 vs Gemini Omni Flash. imaginary.app does not offer Omni Flash.

Option 3: use the imaginary.app API

Our REST API calls the Google Cloud Veo 3.1 models for you, so there is no Cloud project, service account or quota request. You sign in with Google, create a key starting with veo_ on the API page, and send requests to three endpoints: POST /api/generate to start a video, GET /api/status/:taskId to poll it, and GET /api/logs to list past generations.

curl -X POST https://api.veo3gen.app/api/generate \
  -H "Authorization: Bearer veo_your_api_key_here" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "veo3-fast",
    "prompt": "A lighthouse on a rocky coast at dusk, waves crashing, slow aerial push-in",
    "audio": true,
    "options": { "aspectRatio": "16:9", "resolution": "1080p", "duration": 8 }
  }'

# The response includes a taskId. Poll it until the video is ready:
curl -H "Authorization: Bearer veo_your_api_key_here" \
  https://api.veo3gen.app/api/status/TASK_ID

https://api.imaginary.app is an alias for the same API; the samples use https://api.veo3gen.app, which matches our API documentation.

options.duration accepts 4, 6 or 8 (default 8), aspectRatio is 16:9 or 9:16, and resolution is 720p, 1080p or 4K (4K on Fast and Quality only). The request format is our own, so moving from the Gemini API means rewriting the call, not just swapping a URL. Image-to-video and first-and-last-frame requests add an image (and lastFrame) object. The full field list is in the API documentation, and getting started with the API walks through a first integration.

The request above costs 10 credits (about $2.13–$2.50). Credits are taken when the job starts and returned automatically if it fails. Typical 8-second costs: