# Gemini Omni 1.2 Flash API: Google video generation & editing | KovAPI

> Gemini Omni 1.2 Flash is the video generation and editing model Google released in November 2026. One API covers text to video, first frame, first + last frame, reference images and video editing, with dialogue, sound effects and music generated together with the picture. On KovAPI it is `gemini-omni-1.2-flash`, billed per second.

Web page: https://kovapi.com/en/model/gemini-omni-1.2-flash · Updated: 2026-11-12
Prices, limits and timings are live values from when this file was generated.

## Fact sheet

- Model ID: `gemini-omni-1.2-flash`
- Endpoint: `POST https://kovapi.com/v1/videos/generations` → `GET https://kovapi.com/v1/videos/jobs/{id}`
- Length: 3-15 s, any whole second; up to 10 s with a first frame or first + last frame
- Resolution: 720P / 1080P / 4K
- Aspect ratio: 16:9 / 9:16
- References: up to 3 reference images; 1 first frame; 2 for first + last frame
- Video editing: one 1-15 s source video, plus up to 5 reference images; the output is as long as the source
- Prompt: ≤ 20,000 characters
- Per second: 720P $0.035/s · 1080P $0.05/s · 4K $0.10/s
- Results kept: 7 days

## Edit with one sentence, get the text right, pick any length

- **Change the look, the text or the objects with one sentence**: Put a 1-15 s video in `video_urls` and say what to change. Turn it into anime, change only the words on a sign, make something disappear; what you do not mention stays as close to the source as possible. Up to 5 reference images can show what to add. Billed by the source seconds.
- **People, products and packaging text follow the references**: Up to 3 reference images, one for each person, pet or product. The coffee ad used a barista and a coffee bag; in the clip the barista and the words on the bag match the references.
- **Fix the opening, or the opening and the ending**: One image as the first frame and the clip starts from it; two images as first + last frame and the model makes the move in between. Both go up to 10 s.
- **Signs, plates and packaging spelled right**: Put the words in quotes and Omni writes them into the picture as given. The street clip got the sign, the storefront and the number plate right in one go; an edit can change just one of them.
- **3-15 s to the second, up to 4K**: Pick any whole number of seconds and time the story with a timecode such as `[0-5s]`; 720P, 1080P or 4K, landscape 16:9 or portrait 9:16, with sound in every clip.

## Modes and API fields

Every mode shares `model` (`gemini-omni-1.2-flash`), `prompt`, `resolution` and `size`. Images go in `images`, and `image_role` says whether they are a first frame or references; a source video goes in `video_urls`. In the prompt, `@Image1`, `@Image2` name the images in upload order.

| Mode | Send | You get | Limits |
| --- | --- | --- | --- |
| Text to video | Only `prompt` and the output settings; `duration` in whole seconds. | Picture, camera and sound from the text; put dialogue, music and sound effects in the prompt. | 3-15 s; 10 s when `duration` is left out. |
| First frame | One image in `images`. A single image is the first frame by default; you can also set `image_role: first_frame`. | The clip opens on that image and goes on with the motion and camera the prompt describes. | Up to 10 s. |
| First + last frame | Two images in `images` with `image_role: first_frame`: image 1 is the first frame, image 2 the last. | A move from the first picture to the second; the motion and camera in between follow the prompt. | Up to 10 s. |
| Reference images | Images in `images`: two or three are references by default; a single one needs `image_role: reference`. Say in the prompt who or what each image is. | The people, pets, products or style in the images appear in a new scene, not as a frame of it. | Up to 3 images, 3-15 s. |
| Video editing | One source video (mp4 / mov) in `video_urls` and a prompt saying what to change; add reference images if you like and name them `@Image1`. | A new look, changed text, objects added or removed, as asked; what the prompt does not mention stays as close to the source as possible. | Source 1-15 s, up to 5 reference images; the output is as long as the source and billed by its seconds. |

## What you can make with Gemini Omni 1.2 Flash

- **Product and brand ads**: Product shots and the presenter as references keep the brand name and colours steady on camera; dialogue and sound go in the prompt.
- **Many versions of one clip**: Restyle a clip, relight it or change its text without generating it again, handy for testing ad variants.
- **Bring a photo to life**: Use a product shot, street photo or illustration as the first frame, describe the camera and the action, and get a clip that opens on it.
- **Transitions and establishing shots**: Give a start and an end picture for a move between seasons, day and night, or two scenes.
- **Vertical social clips**: 9:16 straight away, up to 4K, for openers and mood shots on short-video platforms.
- **Previs and short stories**: Split a clip into beats with a timecode, check the rhythm and the shots, then decide what to film or polish.
- **Pets and lifestyle**: Photos of a pet or an object as references put them into a new scene.
- **Signs, packaging and titles**: Quote the words that must appear, for shop fronts, events and product shots.

## Gemini Omni 1.2 Flash pricing

| Resolution | Gemini Omni 1.2 Flash | Google list price |
| --- | --- | --- |
| 720P | $0.035/s | $0.07 |
| 1080P | $0.05/s | $0.104 |
| 4K | $0.10/s | $0.209 |

- Generation: the per-second rate of the chosen resolution x output seconds; reference images are free.
- Video editing: the same per-second rate x source seconds (whole seconds; a part under one second is cut). Pick "Edit a video" in the calculator.
- The cost is reserved at submission and settled at the same amount on success; a failed render is refunded in full.

## Examples: prompts and parameters

### Tabby cat: one continuous shot

- Made with KovAPI · Text to video · 10s · 720P · 16:9
- $0.35 at current prices
- Output: https://img.kovapi.com/i/mp-c12a02b55c1bc34ba139487d.mp4

The "single scene" example from Google's guide: it opens by fixing a continuous handheld single shot, then small motion, light and sound design, and ends with "No dialogue". Prompt from Google's [Gemini Omni Flash developer docs](https://aistudio.google.com/docs/omni); rendered by KovAPI with that prompt.

Original prompt:

```text
Continuous, unbroken handheld shot of a fluffy tabby cat sitting on a sunny windowsill, looking out into a leafy garden. The cat's tail twitches slowly, and its ears rotate slightly toward ambient noises. Sunbeams illuminate dust motes in the air. Sound design: Gentle breeze, distant bird chirps. No dialogue.
```

### A paper airplane through a library

- Made with KovAPI · Text to video · 8s · 1080P · 16:9
- $0.40 at current prices
- Output: https://img.kovapi.com/i/mp-179512a5c3b6c91fa7b10f56.mp4

A one-line example from Google's docs: one unbroken tracking shot, at 1080P. Prompt from Google's [Gemini Omni Flash developer docs](https://aistudio.google.com/docs/omni); rendered by KovAPI with that prompt.

Original prompt:

```text
A single unbroken tracking shot following a paper airplane gliding through a bustling library.
```

### Timecode: into the portal in 15 s

- Made with KovAPI · Text to video · 15s · 1080P · 16:9
- $0.75 at current prices
- Output: https://img.kovapi.com/i/mp-29f31c6396192e4adf2ae1e3.mp4

Three timecodes, `[0-5s]`, `[5-10s]` and `[10-15s]`, set the beats: walking, stopping to turn, running into the portal; the clip follows them. Prompt from Google's [Gemini Omni Flash developer docs](https://aistudio.google.com/docs/omni); rendered by KovAPI with that prompt.

Original prompt:

```text
[0-5s] A person is walking down a quiet street
[5-10s] They stop and turn around toward a glowing doorway
[10-15s] They start running into the portal
```

### Words on a sign, a storefront and a plate

- Made with KovAPI · Text to video · 8s · 720P · 16:9
- $0.28 at current prices
- Output: https://img.kovapi.com/i/mp-add537be3ae6b285aae5d047.mp4

The "text in videos" example from Google's guide: the sign, the storefront and the number plate each have their words in quotes, and all three come out right. Prompt from Google's [Gemini Omni Flash developer docs](https://aistudio.google.com/docs/omni); rendered by KovAPI with that prompt.

Original prompt:

```text
There is a street sign that says: "This is an AI generation by Omni", there is a storefront that says: "All you need AI", there's a car with the number plate: "OMNI1.2"
```

### 4K portrait: a cyberpunk city

- Made with KovAPI · Text to video · 6s · 4K · 9:16
- $0.60 at current prices
- Output: https://img.kovapi.com/i/mp-78147b0630d708927a07f2cd.mp4

The line Google's docs use to explain aspect ratios, rendered here in 4K portrait (2160×3840). Prompt from Google's [Gemini Omni Flash developer docs](https://aistudio.google.com/docs/omni); rendered by KovAPI with that prompt.

Original prompt:

```text
A futuristic city with neon lights and flying cars, cyberpunk style
```

### First frame: a street photo comes to life

- Made with KovAPI · First frame · 8s · 720P · 16:9 · image_role=first_frame
- $0.28 at current prices
- Output: https://img.kovapi.com/i/mp-7d449482e0fc8a7d8cc45039.mp4
- Image 1: https://img.kovapi.com/i/mp-b7c2b45a774cc22b88c191f2.png

The first frame is a street photo; the prompt describes a concrete camera move and actions, as Google recommends: a push-in, a baker setting out croissants, a pigeon landing on the basket and flying off. Input images made by KovAPI with Nano Banana Pro; prompt written by KovAPI the way Google's guide recommends.

Original prompt:

```text
The camera slowly pushes in toward the bakery window. Inside, a baker in a white apron places a fresh tray of golden croissants in the window display. A pigeon lands on the bicycle basket, looks around and flutters away. In a single continuous shot. Sound design: quiet morning street ambience, a distant church bell, pigeon wings flapping. No dialogue. No music.
```

### First + last frame: summer morning to snowy night

- Made with KovAPI · First + last frame · 8s · 1080P · 16:9 · image_role=first_frame
- $0.40 at current prices
- Output: https://img.kovapi.com/i/mp-05622ab6ca908a072b78254b.mp4
- Image 1: https://img.kovapi.com/i/mp-d76700384255b66384f59138.jpg
- Image 2: https://img.kovapi.com/i/mp-601efc3d56bcd883c7565db0.jpg

The prompt from the first + last frame section of Google's docs; the two frames share one camera position, the second made from the first. Prompt from Google's docs; input images made by KovAPI with Nano Banana Pro; rendered by KovAPI.

Original prompt:

```text
A smooth cinematic transition from a lush green forest at sunrise to a snowy forest under a starry night sky.
```

### Reference images: a cat and a ball of yarn

- Made with KovAPI · Multi-image reference · 8s · 720P · 16:9 · image_role=reference
- $0.28 at current prices
- Output: https://img.kovapi.com/i/mp-82df85e079207c922446172d.mp4
- Image 1: https://img.kovapi.com/i/mp-8a3ed962b9a5b10874c8645d.png
- Image 2: https://img.kovapi.com/i/mp-f62963a2437f9b0fd21e4342.jpg

The subject-reference example from Google's docs: one image of a cat, one of yarn, a one-line prompt. The cat and the yarn in the clip follow the references. Prompt from Google's docs; input images made by KovAPI with Nano Banana Pro; rendered by KovAPI.

Original prompt:

```text
A cat playfully batting at a ball of yarn.
```

### Product ad: the person and the packaging from references

- Made with KovAPI · Multi-image reference · 10s · 1080P · 16:9 · image_role=reference
- $0.50 at current prices
- Output: https://img.kovapi.com/i/mp-e1dd3e4ef52fd3d2cc17da18.mp4
- Image 1: https://img.kovapi.com/i/mp-1d600f5a04da572b6a4c2bc2.jpg
- Image 2: https://img.kovapi.com/i/mp-804b9708cd179c2ec07a7543.jpg

Two references: a barista and a coffee bag. The prompt gives the action, a close-up, one spoken line and the sound, and asks for a sharp label; the person and the words on the bag match the references, and the line is spoken as written. Input images made by KovAPI with Nano Banana Pro; prompt written by KovAPI the way Google's guide recommends.

Original prompt:

```text
The barista from the first image stands behind a wooden cafe counter in the morning. She picks up the coffee bag from the second image, turns it so the label faces the camera, opens it and pours whole beans into a grinder. Close-up of the beans, then a medium shot as she smiles at the camera and says: "Freshly roasted, every Monday." Keep the label text sharp and readable. Warm morning light. Sound design: cafe ambience, beans rattling into the grinder, the grinder starting. No music.
```

### Video edit: anime with one sentence

- Made with KovAPI · Video editing · 10s · 720P · 16:9
- $0.35 at current prices
- Output: https://img.kovapi.com/i/mp-905cc7d194d76bd05fe02ed9.mp4
- Source video (10s): https://img.kovapi.com/i/mp-c12a02b55c1bc34ba139487d.mp4

The source is the "Tabby cat" clip, and the edit prompt is only Google's line "Make this video anime": the whole look changes, the camera and the motion stay. Billed by the source seconds. Prompt from Google's [Gemini Omni Flash developer docs](https://aistudio.google.com/docs/omni); rendered by KovAPI with that prompt.

Original prompt:

```text
Make this video anime
```

### Video edit: change only the words on the sign

- Made with KovAPI · Video editing · 8s · 720P · 16:9
- $0.28 at current prices
- Output: https://img.kovapi.com/i/mp-ee06321042c206161cb6ee34.mp4
- Source video (8s): https://img.kovapi.com/i/mp-add537be3ae6b285aae5d047.mp4

The source is the "Words on a sign" clip, and the prompt is Google's example line: the sign now says "Omni Flash", while the storefront, the plate and the camera are unchanged. Prompt from Google's [Gemini Omni Flash developer docs](https://aistudio.google.com/docs/omni); rendered by KovAPI with that prompt.

Original prompt:

```text
Change the text on the sign to say "Omni Flash"
```

### Video edit: make the violin disappear

- Made with KovAPI · Video editing · 8s · 720P · 16:9
- $0.28 at current prices
- Output: https://img.kovapi.com/i/mp-ff452614135cc0c0c6899ffe.mp4
- Source video (8s): https://img.kovapi.com/i/mp-5e0b9ff56c3378f85dd34815.mp4

Google's multi-turn editing example: generate "A woman playing violin outdoors.", then send "Make the violin invisible.". Here the first clip is the source video; the violin is gone and the bowing stays. Prompt from Google's [Gemini Omni Flash developer docs](https://aistudio.google.com/docs/omni); rendered by KovAPI with that prompt.

Original prompt:

```text
Make the violin invisible.
```

### Edit + reference image: put on that hat

- Made with KovAPI · Video editing · 8s · 720P · 16:9
- $0.28 at current prices
- Output: https://img.kovapi.com/i/mp-19f905c62d058768f7273e73.mp4
- Image 1: https://img.kovapi.com/i/mp-f336123368a69708fd348e71.jpg
- Source video (8s): https://img.kovapi.com/i/mp-5e0b9ff56c3378f85dd34815.mp4

Google's example line is "Put a fashionable hat on this person"; here a reference image picks the hat and the prompt names it `@Image1`. The colour, the brim and the black band follow the reference. Reference image made by KovAPI with Nano Banana Pro.

Original prompt:

```text
Put the hat from @Image1 on this person. Keep everything else the same.
```

## Prompting Gemini Omni 1.2 Flash: Google's own guide

Based on Google's [Gemini Omni Flash prompt guide](https://aistudio.google.com/docs/omni#prompt-guide): fix the shot and the subject first, then motion, light and sound; when editing, write only the one thing to change. The example lines below come from that guide and are the prompts this page's examples actually sent.

- **Ask for one shot if you want one**: By default Omni cuts between a few shots and builds a little story. For a single continuous shot, write "In a single continuous shot" or "No scene cuts". — Tabby cat original: "Continuous, unbroken handheld shot of a fluffy tabby cat…"
- **Describe the sound**: Leave sound out and the model picks a track for you. Give music a genre and a mood, put ambience and dialogue in too, and quote every spoken line. — Tabby cat original: "Sound design: Gentle breeze, distant bird chirps."
- **Write exclusions as short sentences**: `negative_prompt` is not available; put exclusions in the prompt itself, such as "No dialogue", "No embellishments", "No extra sound effects". — Last line of the tabby cat original: "No dialogue."
- **Edit one thing at a time**: Short edit prompts work best; a long description can change things you meant to keep. When changing one detail, add "Keep everything else the same". — Sign original: "Change the text on the sign to say "Omni Flash""
- **Time the events**: Natural language such as "After 3 seconds…" or "At 5s…" works, and so does a timecode like `[0-5s]`; use it for your own cuts and rhythm. — Portal original: "[10-15s] They start running into the portal"
- **Quote on-screen text**: Omni writes text correctly and legibly. For signs or plates that would appear in the scene anyway, put the words you want in quotes. — Street original: "there's a car with the number plate: "OMNI1.2""
- **Detail the people and places**: Google's "meta prompting": ask for micro-detail, expression and timing, describe characters like a costume designer, and be specific about background elements. — Google's guide: "Be extremely detailed in your descriptions of characters and environments."
- **Describe motion for image to video**: Use a sharp image and describe the camera move, the subject's action and the setting changing; that beats "make it move". If the image is only a sketch or a reference, say so, or it may be taken as the first frame. — Bakery (written to the guide): "The camera slowly pushes in toward the bakery window."

| Part | Purpose | Example |
| --- | --- | --- |
| Shot | Fix the shot first: continuous, handheld, one take. | Original: "Continuous, unbroken handheld shot" |
| Subject and setting | Who the subject is, where, looking at what. | Original: "of a fluffy tabby cat sitting on a sunny windowsill, looking out into a leafy garden." |
| Small motion | Small movements bring the frame to life; the subject need not do much. | Original: "The cat's tail twitches slowly, and its ears rotate slightly toward ambient noises." |
| Light | One sentence for light and mood. | Original: "Sunbeams illuminate dust motes in the air." |
| Sound design | A line of its own for ambience; quote any dialogue. | Original: "Sound design: Gentle breeze, distant bird chirps." |
| Exclusions | What you do not want, last. | Original: "No dialogue." |

```text
In a single continuous shot, [shot size and camera move] of [subject] in [setting].
[0-4s] [first action]
[4-8s] [second action]
[8-10s] [closing image]
Lighting: [light and colour].
Sound design: [ambience, sound effects]. [Character] says: "[line]"
No subtitles. [No music / No dialogue]
```

## How to generate video with Gemini Omni 1.2 Flash on KovAPI

1. **Sign up and create an API key** Log in to KovAPI and create a key in the console; one key works for every model on the site.
2. **Try a prompt above** Pick length, resolution and aspect ratio, upload images or a source video if you need them, and see the live quote before you submit.
3. **Submit an async job** Send `gemini-omni-1.2-flash` to `POST /v1/videos/generations`; keep the job id from the HTTP 202. The docs have complete request examples.
4. **Poll or receive a webhook** Poll the job id, or set `callback_url` when submitting. On success, download the MP4 from `data[0].url`; results are kept 7 days, so save them in time.

### Minimal request

```bash
curl https://kovapi.com/v1/videos/generations \
  -H "Authorization: Bearer $KOVAPI_KEY" -H "Content-Type: application/json" \
  -d '{"model": "gemini-omni-1.2-flash", "prompt": "...", "duration": 3, "resolution": "720P", "size": "16:9"}'
# -> HTTP 202 {"id": "..."}; poll:
curl https://kovapi.com/v1/videos/jobs/<id> -H "Authorization: Bearer $KOVAPI_KEY"
```

## How Gemini Omni evolved

- 2026-05-19 · Gemini Omni Flash: **Gemini Omni Flash · launch** Announced at Google I/O as the first model of the Omni family: video from text, images and video, edited through conversation, first in the Gemini app and YouTube Shorts.
- 2026-06-30 · gemini-omni-flash-preview: **Public preview in the Gemini API** Developers could call the preview in the Gemini API: 3-10 s at 720P, text to video or animating a still, then editing the result.
- 2026-08-27 · Gemini Omni 1.1 Flash: **Omni 1.1 · generally available** The GA release added first + last frame and resolutions up to 4K (1080P and 4K are upscaled). [Gemini Omni Flash](https://kovapi.com/en/model/gemini-omni-flash) on KovAPI is this generation: 6 / 8 / 10 s, 720P / 1080P.
- 2026-11-05 · Gemini Omni 1.2 Flash: **Omni 1.2 · 15 seconds, steadier references** Any length from 3 to 15 s; people, outfits and packaging text from reference images hold better, with more natural dialogue and lip-sync; Google's list price is 30% below 1.1. Google says requests to 1.1 move to 1.2 automatically from 2026-12-03.

## Gemini Omni 1.2 Flash vs other video models

| Model | Length (s) | Resolution | Refs image / video / audio | Price | Good for |
| --- | --- | --- | --- | --- | --- |
| Gemini Omni 1.2 Flash (`gemini-omni-1.2-flash`) | 3–15 | 720P / 1080P / 4K | 3/0/0 | $0.035/s | Editing an existing video with one sentence, or short clips that need accurate on-screen text, 4K, or a length to the second. |
| Gemini Omni Flash (`gemini-omni-flash`) | 6 / 8 / 10 | 720P / 1080P | 3/0/0 | $0.035/s | Projects already working on the previous generation: 6 / 8 / 10 s, 720P / 1080P, same price as 1.2. |
| Veo 3.1 (`veo-3.1`) | 8 | 720P / 1080P | 3/0/0 | $0.20/clip | Fixed 8-second clips billed per clip. |
| Seedance 2.5 (`seedance-2.5`) | 4–30 | 480P / 720P / 1080P | 30/10/10 | $0.10/s | Lots of references in one job (images, videos and audio together), or story clips up to 30 s. |
| Wan 3.0 (`wan3.0-video`) | 4–30 | 480P / 720P / 1080P | 10/5/5 | $0.04/s | Reference video or audio, or video from a document or web page, up to 30 s per clip. |
| MiniMax H3 (`minimax-h3`) | 4–15 | 480P / 768P / 1080P | 9/3/3 | $0.015/s | Batches of short clips: draft at 480P / 768P, then render 1080P; first + last frame, and reference mode takes reference audio and video. |
| Grok Imagine Video 1.5 (`grok-imagine-video-1.5`) | 4–15 | 480P / 720P | 7/0/0 | $0.30/clip | 480P / 720P clips billed per clip, with up to 7 reference images. |

If another model already gives you stable results, run the same inputs and criteria through both and compare picture, motion, sound, render time and real cost before switching.

## How the inputs combine

- How images are read: one image is a first frame by default, two or three are references. For one image as a reference, set `image_role: reference`; for a first + last frame, send two images with `image_role: first_frame`. With a source video, every image is a reference.
- When editing, `duration` does not apply: the output is as long as the source, and a part past the last whole second is cut (a 4.5 s video is edited and billed as 4 s). Without `size`, the edit keeps the source's orientation. An edit cannot take first or last frames.
- `@Image1`, `@Image2` follow upload order. When generating, the tag is passed to the model as words such as "reference image 1"; when editing, it is bound to that exact image, and images the prompt does not name are passed along too.
- Every clip has sound, and `audio: false` is refused. Write what you do not want into the prompt, such as "No dialogue" or "No music"; `negative_prompt` and `system_prompt` are refused without charge.

## Why use Gemini Omni 1.2 Flash on KovAPI

- **Below Google's list price**: 720P is 50% below Google's list price, 1080P 51% and 4K 52%; video editing costs the same per second. The playground shows the price of each job before you submit it.
- **No charge for failures, automatic retries**: A failed render refunds the reserved cost in full; when the upstream is busy or errors, the system retries for you, so you write no retry logic.
- **No concurrency cap**: Accounts have no concurrency limit, so submit batches as they are; at peak times jobs queue, the API returns `queue_position` and `eta_seconds`, and a webhook can tell you when they finish.
- **One API for every video model**: Submitting, polling and webhooks work the same for every video model on the site; change `model` to send the same prompt to Seedance, Wan, Veo and others for a comparison.
- **Easy sign-up and payment**: Sign up with email, Google or GitHub; top up by WeChat Pay, card, PayPal or USDT; pay as you go with no monthly fee. The same key also calls the chat and image models on the site.

## Gemini Omni 1.2 Flash API FAQ

### What is Gemini Omni 1.2 Flash?

Google's video generation and editing model released in November 2026, the new generation of Gemini Omni Flash. On KovAPI it is `gemini-omni-1.2-flash`: text to video, first frame, first + last frame, reference images and video editing, with sound in every clip.

### How long can a video be?

Text to video and reference images: 3-15 s in whole seconds, 10 s when `duration` is left out. First frame and first + last frame: up to 10 s. An edit is as long as its source video.

### Which resolutions and aspect ratios?

720P / 1080P / 4K, 720P by default; aspect ratio 16:9 / 9:16; 24 fps. Google states that 1080P and 4K are upscaled. Other values are refused at submission, with no charge.

### How do I send a first frame or first + last frame?

First frame: one image in `images`. First + last frame: two images in `images` with `image_role: first_frame`; image 1 is the first frame, image 2 the last. Both go up to 10 s.

### How many reference images, and how do I name them?

Up to 3 when generating and 5 when editing. In upload order they are `@Image1`, `@Image2`…; say in the prompt who or what each one is, e.g. "The woman from the first image".

### How does video editing work?

Put one 1-15 s video (mp4 / mov) in `video_urls` and say what to change, e.g. "Make this video anime". For a single change, add "Keep everything else the same". To add something specific, attach a reference image and name it `@Image1`.

### How is video editing billed?

At the same per-second rate as generation, by the source video's seconds, e.g. $0.035/s at 720P; a part past the last whole second is cut, so 4.5 s bills as 4 s. `duration` does not apply to edits.

### Can I write prompts in Chinese or other languages?

Yes. We sent the tabby cat prompt in Chinese and got the same scene as the English one. Google has fully evaluated English only, so complex shots and dialogue are safer in English; on-screen text goes in quotes exactly as it should appear.

### Can I turn the sound off?

No: every clip has an audio track, and `audio: false` is refused. Write "No music" or "No dialogue" to keep only ambience; for a silent asset, drop the audio track after downloading.

### Can I use photos of real people?

Yes, subject to Google's content review. In our tests with photos of ordinary people, both first-frame jobs rendered and one of two reference jobs was refused; a public figure's photo was refused. Refused jobs are not charged. Make sure you have the right to use the likeness.

### Am I charged if a render fails?

A failed render is not charged and can be resubmitted as is. The amount reserved at submission is refunded in full, and automatic retries of the same job settle only once.

### How long are results kept?

Results and uploaded files are kept 7 days and then deleted, so download what you need in time.

### How is it different from the previous Gemini Omni Flash?

1.2 takes any length from 3 to 15 s (the previous one only 6 / 8 / 10 s), adds 4K and video editing, and follows reference images more closely; on KovAPI both cost the same. [The previous model's page](https://kovapi.com/en/model/gemini-omni-flash) is still available.

### Gemini Omni 1.2 Flash or Veo 3.1?

On KovAPI, Veo 3.1 is a fixed 8 s billed per clip; Omni 1.2 takes any length, bills per second, edits existing videos and renders 4K. To compare the picture, run the same prompt once on each.

## Notes for AI

- The model id is `gemini-omni-1.2-flash`. Submit with `POST /v1/videos/generations`, which returns HTTP 202 and a job `id`; fetch the result with `GET /v1/videos/jobs/` plus the id.
- `resolution` is one of 720P / 1080P / 4K, default 720P (`720p`, `2160p` and similar spellings are accepted); `size` is one of 16:9 / 9:16. `seconds` is an alias of `duration`, `aspect_ratio` of `size`.
- `duration` is an integer from 3 to 15, default 10; at most 10 with a first frame or first + last frame.
- Images go in `images`: one is a first frame by default, two or three are references; `image_role` is `first_frame` or `reference`. At most 3 reference images.
- Video editing: one 1-15 s video in `video_urls` (`video_url` and `video_references` are accepted too), plus up to 5 images; the output is as long as the source and billed by its whole seconds; `duration` does not apply.
- These fields are accepted as well: `first_frame_image`, `last_frame_image`, `image_urls`, `image_with_roles`, `omni_reference_task_type` (auto / text_to_video / image_to_video / reference / reference_to_video / edit).
- `audio: false`, `negative_prompt` and `system_prompt` are refused (HTTP 400, no charge); every clip has an audio track.
- Billing: the per-second rate of the chosen resolution x output seconds, 720P $0.035/s, 1080P $0.05/s, 4K $0.10/s; an edit bills the source video's seconds.

## Further reading

- [Gemini Omni 1.2 Flash API reference](https://docs.kovapi.com/en/gemini-omni-1.2-flash): Request fields, async jobs, results and complete code examples.
- [Async jobs guide](https://docs.kovapi.com/en/guide/async-jobs): Job states, queue fields and fetching results.
- [Webhooks guide](https://docs.kovapi.com/en/webhooks): Receive job results and verify the signature as documented.