NOV 6, 2026

Table of contents

1. Step 1: a 15-second video from one prompt
2. Step 2: extend it by 10 seconds
3. Step 3: edit all 15 seconds
4. 720p or 1080p: what a 1080p extend does
5. Also through the MCP server

Omni 1.2 Flash in Google Vids

November 6, 2026

Google moved Google Vids to Gemini Omni 1.2 Flash on November 5, 2026, and the Google Vids API supports its longer clips now:

• POST /videos makes videos of 3 to 15 seconds, at 720p or 1080p, landscape or portrait, from text, a start image, reference images or voiced avatars
• POST /videos/edit edits up to 15 seconds of a clip
• POST /videos/extend adds up to 10 seconds a call, as Omni 1.2 does in Google Flow, to clips of up to 24 seconds. Call it again to go longer, up to about 34 seconds

There is nothing to choose: Vids has one video model, and Google is moving accounts to Omni 1.2 over a few days from November 5. A video uses its seconds of the account’s monthly Vids allowance, which is separate from Google Flow credits.

Below, one scene in three calls: a 15-second video, a 10-second extension and an edit of the first 15 seconds. Every request is the real one, and every video is the real output.

Step 1: a 15-second video from one prompt

POST /videos with duration: 15. Google answered in about 60 seconds with a 1280×720 video, and both lines are spoken as written. It used 15 seconds of the allowance.

curl — POST /videos

curl --location 'https://api.useapi.net/v1/google-vids/videos' \
--header 'Authorization: Bearer user:12345-…' \
--header 'Content-Type: application/json' \
--data '{
  "prompt": "Cinematic single continuous shot, 15 seconds, 16:9. Golden hour on a Tokyo rooftop garden, city skyline glowing behind. A young chef in a crisp white apron finishes plating a steaming bowl of ramen, steam curling in the warm light. She looks up into the camera and says with a confident smile: \"Fifteen seconds. One prompt. No editing.\" She slides the bowl toward the camera as it slowly pushes in, wind chimes ringing softly and city traffic humming far below. She laughs and adds: \"Now that'\''s fast food.\"",
  "duration": 15,
  "aspectRatio": "landscape",
  "resolution": "720p",
  "async": true
}'

The finished job, from GET /jobs/jobid (shortened):

{
    "status": "completed",
    "result": {
        "mediaId": "user:[email protected]:eyJ1Ijoi…",
        "width": 1280,
        "height": 720,
        "duration": 15,
        "resolution": "720p",
        "aspectRatio": "landscape",
        "model": "/flix/generate_videos_omni_t2v_psq/v1",
        "quota": { "video": { "limit": 500, "left": 485, "resetAt": "2026-12-01T08:00:00.000Z" } }
    }
}

Step 2: extend it by 10 seconds

POST /videos/extend with the first video’s mediaId. The result is one clip, the 15 seconds plus 10 new ones, 25 seconds long, after about 100 seconds. A friend walks in, tries the ramen, and both new lines are spoken.

Also on YouTube: Gemini Omni 1.2 Flash in Google Vids: 15 seconds from one prompt, extended to 25.

We made this clip while testing with duration: 15, which the Vids editor’s slider also offers. Google added only 10 seconds and still counted 15, so the API now takes up to 10 a call and answers a longer request with 400. The same request with duration: 10 returns the same 25-second clip and uses 10 seconds.

curl — POST /videos/extend

curl --location 'https://api.useapi.net/v1/google-vids/videos/extend' \
--header 'Authorization: Bearer user:12345-…' \
--header 'Content-Type: application/json' \
--data '{
  "mediaId": "user:[email protected]:eyJ1Ijoi…",
  "prompt": "Same shot continues on the Tokyo rooftop at dusk, the city lights now switching on. A friend in a denim jacket walks into frame, sits at the table and takes a first slurp of the ramen, eyes widening. He says: \"Wait, you made this in fifteen seconds?\" The chef folds her arms and grins: \"Fifteen more, and I'\''ll make dessert.\" They both laugh as string lights flicker on above them and soft city ambience fills the air.",
  "duration": 10,
  "async": true
}'

Step 3: edit all 15 seconds

POST /videos/edit on the first video with duration: 15. With Omni 1.1 an edit returned at most 10 seconds. Now the whole 15-second clip came back edited, every frame of it, after about 140 seconds. In this run it used 10 seconds of the allowance.

curl — POST /videos/edit

curl --location 'https://api.useapi.net/v1/google-vids/videos/edit' \
--header 'Authorization: Bearer user:12345-…' \
--header 'Content-Type: application/json' \
--data '{
  "video": "user:[email protected]:eyJ1Ijoi…",
  "prompt": "Make it a snowy winter evening: snow falling gently over the rooftop and the city, the chef wearing a red knit scarf. Keep everything else the same.",
  "duration": 15,
  "async": true
}'

720p or 1080p: what a 1080p extend does

We recommend making and extending Vids clips at 720p and going to 1080p once, at the end. This test shows why. A 1080p video from Google is a 720p draft that Google upscales (Gemini API docs: “1080p output (upscaled)”). An extend at 1080p processes the whole clip again, the seconds you already had included, and each pass sharpens them further.

One detailed shot (a fisherman’s face, rain on his skin, a knitted cap and a painted sign), made at 1080p and then extended twice at 1080p, against the same shot made and extended at 720p with 1080p set on the last extend only. Each panel plays the same first 10 seconds, cropped to the face:

1. The 1080p shot as Google made it.
2. The same seconds after one 1080p extend: skin and wrinkles come back harder and crunchier.
3. After two 1080p extends: the face looks plastic and over-drawn.
4. Made and extended at 720p, with resolution: "1080p" on the last extend only: one 1080p pass, and the face stays natural.

A still from 5 seconds in, in the same order:

How much of the first 10 seconds each extend changed, measured against the shot as Google made it (SSIM, where 1.00 means identical):

Clip	First 10 seconds compared with the original
One extend at 1080p	0.92, re-rendered and sharper
Two extends at 1080p	0.83, re-rendered again
One extend at 720p	0.99, left as it was

So give a clip one 1080p pass, at the end:

• Make and extend your clips at 720p.
• Finish a clip of up to 20 seconds with one upscale to 1080p.
• For a clip of 21 to 34 seconds, which upscale doesn’t take, set resolution: "1080p" on the last extend only.
• A single clip you won’t extend or edit can be made at 1080p from the start: it uses the same seconds as 720p.

The prompts, in order: the shot (duration: 10), then two extends of 10 seconds each.

Static camera, extreme close-up portrait of an elderly fisherman on a harbor at dawn: deep wrinkles, grey stubble, rain droplets on his skin, a hand-knitted navy wool cap with clearly visible yarn stitches. Behind him a weathered wooden sign with small painted letters reading HARBOR MARKET OPEN 6AM. He looks into the camera and says calmly: "The sea gives, and the sea takes." Gulls cry softly, waves lap against the pier.

Same static shot continues. He lifts a wet fishing net close to the camera, every rope knot sharp and dripping, and says: "Tomorrow, we go out again." Waves lap softly.

Same static shot continues. He lowers the net, smiles, and the first sunlight catches the droplets on his cap. He nods once. Gulls cry in the distance.

Also through the MCP server

The Google Vids MCP server has the same limits: vids_create_video and vids_edit_video take up to 15 seconds, vids_extend_video up to 10 a call on clips of up to 24 seconds. Omni 1.2 arrived in Google Flow on the same day, with 15-second videos and extend with voices: Google Flow: Omni 1.2 Flash, 15-Second Videos, and Extend with Voices.

The Gemini Omni 1.2 Flash API guide compares Omni 1.2 in Google Flow and Google Vids, with the price per video on every plan.