# Gemini Omni Flash (video)

> Gemini Omni Flash 1.2 makes video with synchronized audio at 360p to 4K, 3 to 15 seconds long, from text, a first frame (with an optional last frame), or a reference set: up to ten images bound with `<IMAGE_REF_n>` tags plus up to three clips of at most 15 seconds bound with `<VIDEO_REF_n>` tags. It is best at coherent physics, on-screen text and literal camera direction.

Status: roasOS API v1.1. Catalog `model-catalog@2026-10-06.v5`, pricing `customer-pricing@2026-11-05.v15`. The endpoint `POST /v1/video/gemini-omni-flash` is live. `POST /v1/generations` stays for one minor version as a deprecated alias (see "Today").

## Endpoint
POST https://api.roasos.com/v1/video/gemini-omni-flash
Auth: Authorization: Bearer roasos_sk_… (scope api:write). Idempotency-Key required. quote_only / quote_id / webhook_url as in https://docs.roasos.com/llms.txt.

## Good at
The model is grounded in Gemini world knowledge, so plain cause-and-effect action renders coherently. Audio is always on: the model lip-syncs a script and adds sound; there is no audio switch and no audio-off discount. You cannot send an audio file: no endpoint takes a voice or a soundtrack. Reference images bind to roles with inline `<IMAGE_REF_0>`, `<IMAGE_REF_1>` tags, so one prompt can point at a person, a product and a location. Camera words are taken literally ("one continuous shot", "locked off", "push in"), and on-screen text is a strength. The `resolution` field sets the output tier and the price: 360p, 720p, 1080p or 4k, billed per output second. In image-to-video, `end_image_url` pins the last frame; with a last frame the clip is at most 10 seconds. Reference clips (`video_urls`, up to three, each at most 15 seconds) carry motion and choreography: "the dog performs the dance from `<VIDEO_REF_0>`". A lone clip with no references is a video edit instead. It is not the fidelity leader for image-to-video talking heads; use Kling 4.0 or Seedance for those. Video edit from a source video is a separate endpoint (`/v1/video-edit/gemini-omni-flash`).

## Pricing
Unit: clip. Pricing key: `gemini-omni-flash`.

| request | credits |
|---|---|
| `3s_720p` | 3 |
| `9s_720p` | 9 |
| `15s_720p` | 15 |

Credits come from the roasOS pricing table (`MODEL_COSTS`) through the catalog. No other field changes the price.

## Modes
- `text-to-video`
- `image-to-video`
- `reference-to-video`

The worker derives the mode from the media fields of the request.
Provider endpoints: `google/gemini-omni-flash/v1.2/image-to-video` · `google/gemini-omni-flash/v1.2/reference-to-video` · `google/gemini-omni-flash/v1.2/text-to-video`.

## Request fields
| field | type | required | default | rule |
|---|---|---|---|---|
| `aspect_ratio` | enum | no | `16:9` | `9:16`, `16:9` |
| `avatar_id` | string | no | — | UUID; only in `reference-to-video` |
| `avatar_outfit` | string | no | — | ≤ 500 characters |
| `duration_seconds` | integer enum | no | `5` | `3`, `4`, `5`, `6`, `7`, `8`, `9`, `10`, `11`, `12`, `13`, `14`, `15` |
| `end_image_url` | string | no | — | https URL; ≤ 4,096 characters; only in `image-to-video` |
| `extend_of_generation_id` | string | no | — | UUID |
| `image_url` | string | no | — | https URL; ≤ 4,096 characters; only in `image-to-video` |
| `mode` | enum | no | — | `text-to-video`, `image-to-video`, `reference-to-video` |
| `prompt` | string | yes | — | ≥ 1 character; ≤ 10,000 characters |
| `quote_id` | string | no | — | UUID |
| `quote_only` | boolean | no | `false` | — |
| `references` | string[] | no | — | 1–10 items; only in `reference-to-video` |
| `resolution` | enum | no | `720p` | `360p`, `720p`, `1080p`, `4k` |
| `scene_id` | string | no | — | UUID |
| `video_urls` | string[] | no | — | 1–3 items; only in `reference-to-video` |
| `wait_timeout_seconds` | integer | no | `0` | 0–60 |
| `webhook_url` | string | no | — | https URL; ≤ 4,096 characters |

The schema is strict: a field that is not in this table returns `400 unsupported_option`. Field names are the roasOS role names, one name per role across every family.

## Media
- `avatar_id` — identity; the saved character reference sheet, with a derived portrait only where the model requires a single face or source frame
- `references` — style and appearance references
- `image_url` — first frame (or the single source image)
- `end_image_url` — last frame
- `scene_id` — first frame — a scene this workspace owns
- `video_urls` — reference videos

| role | items | max size | content types | duration |
|---|---|---|---|---|
| `end_image_url` | ≤ 1 | 10 MB | `image/jpeg`, `image/png`, `image/webp` | — |
| `image_url` | ≤ 1 | 10 MB | `image/jpeg`, `image/png`, `image/webp` | — |
| `references` | ≤ 10 | 10 MB | `image/jpeg`, `image/png`, `image/webp` | — |
| `video_urls` | ≤ 3 | 100 MB | `video/mp4` | 0–15 s |

- Refused: `elements`, `product_id`, `product_images`, `reference_images`, `scene_image`, `video_url`, `audio_urls`, `audio_url`, `voice_id`. A refused role returns `400 unsupported_option`.
- Every URL must be an immutable object on `https://cdn.roasos.com/…` that your workspace owns; any other URL is `400 reference_media_not_workspace_owned`. Upload with `POST /v1/uploads` (direct POST of the bytes to `put_url`, or a `source_url` import), then use the returned CDN URL.

## Prompt
Limit 10,000 characters including the roasOS route-facts block (`ROUTE-VERIFIED FEATURE FACTS:` with lines for the first frame, the references, the character and the outfit) and the appended `<IMAGE_REF_n>` tags. fal accepts 20,000; 10,000 is the roasOS cap.
- Describe cause-and-effect motion plainly; the model renders physics coherently.
- Place `<IMAGE_REF_0>` inline to bind a reference to a role ("the jar in <IMAGE_REF_1>"); unbound references are appended.
- Camera words are literal: "one continuous shot", "locked off", "natural smartphone zoom".
- Use negative instructions to control additions: "No music, just realistic sound", "Don't add text".
- When you give dialogue or audio direction, keep it explicit; when you give none, the model still adds sound.
- Only 16:9 and 9:16, 3–15 s (10 s with a last frame): keep the action tight enough to resolve inside the clip.

## Not exposed
- `duration_seconds`: values not exposed — `3`, `4`, `6`, `7`, `9`.
- `prompt`: pinned to `characters 10,001–20,000`.
- `generate_audio`: pinned to `always on; no switch and no audio-off price`.
- `seed`: not accepted by this public endpoint.
- `negative_prompt`: not accepted by this public endpoint.
- `cfg_scale`: not accepted by this public endpoint.
- `shot_type`: not accepted by this public endpoint.
- `multi_prompt`: not accepted by this public endpoint.
- `video_url`: not accepted by this public endpoint.

## Result
Result kind `video`. Always present: `url`.
```json
{ "data": { "id": "gen_…", "kind": "video", "model": "gemini-omni-flash", "status": "completed", "result": {"url":"https://cdn.roasos.com/workspaces/ws_01/result"} } }
```
Files live on `https://cdn.roasos.com` and are immutable; a result URL can be a media input to the next call. Poll `GET /v1/jobs/{id}/status` → `{ id, status, updated_at }`, then `GET /v1/jobs/{id}/result` (`409 run_not_completed` until the job is `completed`, `failed` or `canceled`). Statuses: `queued → processing → completed | failed | canceled`.

## Example
```bash
# 1. quote — nothing is spent
curl -X POST https://api.roasos.com/v1/video/gemini-omni-flash \
  -H "Authorization: Bearer roasos_sk_live_…" \
  -H "Idempotency-Key: 0f1c9e2a-gemini-omni-flash-0001" \
  -H "Content-Type: application/json" \
  -d '{"quote_only":true,"prompt":"A woman holds the bottle at eye level in a bright kitchen, morning light, static camera."}'
# → 200 { "data": { "kind": "quote", "quote_id": "q_7Hk…", "credits": 3, "expires_at": "…", "confirmation": { "model": "gemini-omni-flash", "example": "3s_720p" } } }

# 2. run — the same body and the same Idempotency-Key, quote_id instead of quote_only
curl -X POST https://api.roasos.com/v1/video/gemini-omni-flash \
  -H "Authorization: Bearer roasos_sk_live_…" \
  -H "Idempotency-Key: 0f1c9e2a-gemini-omni-flash-0001" \
  -H "Content-Type: application/json" \
  -d '{"quote_id":"q_7Hk…","prompt":"A woman holds the bottle at eye level in a bright kitchen, morning light, static camera."}'
# → 201 { "data": { "id": "gen_…", "kind": "video", "model": "gemini-omni-flash", "status": "queued", "credits_reserved": 3, "links": { "status": "/v1/jobs/gen_…/status", "result": "/v1/jobs/gen_…/result" } } }

# 3. poll
curl https://api.roasos.com/v1/jobs/gen_…/status -H "Authorization: Bearer roasos_sk_live_…"
curl https://api.roasos.com/v1/jobs/gen_…/result -H "Authorization: Bearer roasos_sk_live_…"
```

## Today
`POST /v1/generations { "type": "video", "model": "gemini-omni-flash", "options": { prompt, image, reference_images (≤ 9), duration (3–15), aspect_ratio, resolution (360p|720p|1080p|4k) } }`. `avatar_id`, `product_id` and `avatar_outfit` are app-only today.

## See also
https://docs.roasos.com/models/video/gemini-omni-flash · https://api.roasos.com/v1/catalog/video/gemini-omni-flash · google/gemini-omni-flash/v1.2/image-to-video · google/gemini-omni-flash/v1.2/reference-to-video · google/gemini-omni-flash/v1.2/text-to-video