gemini-omni

Quick example

curl https://reapi.ai/api/v1/videos/generations \
  -H "Authorization: Bearer rk_live_xxx" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gemini-omni",
    "prompt": "A kitten playing piano, slow camera push-in",
    "duration": 6,
    "resolution": "1080p",
    "aspect_ratio": "16:9"
  }'

import requests

resp = requests.post(
    "https://reapi.ai/api/v1/videos/generations",
    headers={
        "Authorization": "Bearer rk_live_xxx",
        "Content-Type": "application/json",
    },
    json={
        "model": "gemini-omni",
        "prompt": "A kitten playing piano, slow camera push-in",
        "duration": 6,
        "resolution": "1080p",
        "aspect_ratio": "16:9",
    },
    timeout=30,
)
print(resp.json())

const r = await fetch("https://reapi.ai/api/v1/videos/generations", {
  method: "POST",
  headers: {
    Authorization: "Bearer rk_live_xxx",
    "Content-Type": "application/json",
  },
  body: JSON.stringify({
    model: "gemini-omni",
    prompt: "A kitten playing piano, slow camera push-in",
    duration: 6,
    resolution: "1080p",
    aspect_ratio: "16:9",
  }),
});
console.log(await r.json());

package main

import (
    "bytes"
    "encoding/json"
    "fmt"
    "io"
    "net/http"
)

func main() {
    body, _ := json.Marshal(map[string]any{
        "model":        "gemini-omni",
        "prompt":       "A kitten playing piano, slow camera push-in",
        "duration":     6,
        "resolution":   "1080p",
        "aspect_ratio": "16:9",
    })
    req, _ := http.NewRequest("POST",
        "https://reapi.ai/api/v1/videos/generations", bytes.NewReader(body))
    req.Header.Set("Authorization", "Bearer rk_live_xxx")
    req.Header.Set("Content-Type", "application/json")

    resp, _ := http.DefaultClient.Do(req)
    defer resp.Body.Close()
    out, _ := io.ReadAll(resp.Body)
    fmt.Println(string(out))
}

Submit response

{
  "id": "task_018f5a3a1b6e7d9f8c2b4d6e8f0a2c4e",
  "model": "gemini-omni",
  "status": "processing",
  "created_at": 1735000000
}

Poll GET /api/v1/tasks/{id} (see the Tasks reference) until status === "completed". The completed payload's output.video_urls holds the generated MP4 URL, valid for 7 days.

Authentication

Every call needs a Bearer token. Generate keys at reapi.ai/settings/apikeys.

Authorization: Bearer YOUR_API_KEY

Keys carry the active workspace's billing scope — there is no separate project header.

Endpoint

POST /api/v1/videos/generations
GET  /api/v1/tasks/{id}

Submission is async. The POST returns immediately with a task_id; the task endpoint returns the same envelope until completion. Polling does not consume credits.

Mode routing

gemini-omni picks its mode from the counts of image_urls and video_urls you send — there is no mode parameter:

Request body

model — required

string. Must be "gemini-omni".

prompt — string, required

Up to 2,000 characters. Required in every mode (text-to-video, image-to-video, three-image fusion, reference-to-video). Empty / whitespace-only prompts are treated as missing.

duration — integer, default 6

One of 4, 6, 8, 10 seconds. Other values are rejected with 400 duration must be 4, 6, 8, or 10 seconds, got N.

resolution — string, default "720p"

720p / 1080p / 4k. Lowercase is canonical; uppercase forms ("4K") are accepted and normalized. Drives the per-generation rate — 720p and 1080p share the same price; only 4K is uplifted.

aspect_ratio — string, default "16:9"

Output framing. One of:

Unknown ratios are rejected with 400 invalid aspect_ratio.

size — string, alias for aspect_ratio

The same value the supplier doc lists as a separate field. If both are sent, they must match; otherwise 400 aspect_ratio and size disagree. The reapi playground does not surface size; the JSON body still accepts it for parity.

image_urls — string[]

Array of public HTTP(S) URLs. Allowed counts: 0, 1, or 3.

• 0 entries — text-to-video.
• 1 entry — image-to-video; the image is treated as the starting frame.
• 3 entries — three-image fusion. The model combines all three references into one motion shot.

video_urls — string[]

Array of public HTTP(S) URLs. Allowed counts: 0 or 1.

• 0 entries — non-reference modes (text / image / fusion).
• 1 entry — reference-to-video. The source clip drives the output; it must be ≤ 30 seconds (longer is rejected with 400), and the first ≤ 10 seconds are used as the reference. Billed at a flat per-generation rate (not per second). duration MUST be omitted in this mode.

Response envelope

Submit and poll share the same shape — only status and output fill in over time.

{
  "id": "task_018f5a3a1b6e7d9f8c2b4d6e8f0a2c4e",
  "model": "gemini-omni",
  "status": "completed",
  "created_at": 1735000000,
  "output": {
    "video_urls": ["https://cdn.reapi.ai/media/tasks/018f5a3a1b6e7d9f8c2b4d6e8f0a2c4e/0.mp4"]
  },
  "error": null
}

output.video_urls URLs are valid for 7 days. Re-host to your own storage if you need them longer.

Validation errors

All cases below return HTTP 400 with code 20003 unless noted. Pattern-match on code, not message — message strings carry request-specific context (field names, observed values, etc.) and are not a stable contract.

The full envelope is { "error": { "code", "message", "request_id" } } — see Errors catalog for the wire format and request_id correlation tips.

Recipes

Text-to-video — minimum request

{
  "model": "gemini-omni",
  "prompt": "A little girl walking down a sunset coastal road"
}

Text-to-video — full parameters

{
  "model": "gemini-omni",
  "prompt": "A kitten playing piano, slow camera push-in, cinematic warm tones",
  "duration": 8,
  "resolution": "1080p",
  "aspect_ratio": "16:9"
}

Image-to-video — animate a single frame

{
  "model": "gemini-omni",
  "prompt": "Bring the scene to life with a gentle camera dolly forward",
  "image_urls": ["https://your-cdn.com/first_frame.jpg"],
  "duration": 6,
  "resolution": "1080p"
}

Three-image fusion

{
  "model": "gemini-omni",
  "prompt": "Compose a 10-second product spot mixing scene, character, and product",
  "image_urls": [
    "https://your-cdn.com/scene.jpg",
    "https://your-cdn.com/character.jpg",
    "https://your-cdn.com/product.jpg"
  ],
  "duration": 10,
  "resolution": "1080p",
  "aspect_ratio": "9:16"
}

4K hero shot

{
  "model": "gemini-omni",
  "prompt": "A neon city street in the rain, slow camera pan, reflections on the asphalt",
  "duration": 4,
  "resolution": "4k",
  "aspect_ratio": "16:9"
}

Reference-to-video — match a source clip

{
  "model": "gemini-omni",
  "prompt": "the same scene but at night with neon lights",
  "resolution": "720p",
  "aspect_ratio": "16:9",
  "video_urls": ["https://your-cdn.com/source.mp4"]
}

The output tracks the source clip (which must be ≤ 30s; the first ≤ 10s are used as the reference); omit duration. Billing is a flat per-generation rate (see Pricing below).

Choosing a mode

Polling pattern

The task endpoint behaves identically to other video tasks — the only difference is the completed output shape (video_urls instead of image_urls). A pragmatic schedule:

0–5 minutes:    poll every 5s
5 min – 1 h:    back off gradually toward 1 min
≥ 1 h:          cap at 3 min between polls

A typical task completes in a few minutes. The worker's wall-clock cap is 48 hours, comfortably above any realistic queue.

Pricing

Two billing modes — picked by request shape.

Per generation — text / image / fusion

Charged once per submitted job. 720p and 1080p share the same rate at every duration; only 4K is uplifted. Duration tiers are 4s / 6s / 8s / 10s. See current per-tier rates on the Gemini Omni model page.

Flat per generation — reference-to-video

Triggered when video_urls is set. Charged once per job at a flat rate by resolution — independent of clip length. 720p and 1080p share the same rate; 4K is uplifted. The source clip must be ≤ 30 seconds: the gateway probes its decoded length server-side and rejects longer clips with 400; the first ≤ 10 seconds are used as the reference. See current rates on the Gemini Omni model page.

Bill formula

1 credit = $0.001. Integer credits = ceil(usd × 1000). Failed jobs refund automatically.

Tips

• Prompt motion, not just scene. "Slow push-in, warm tones, shallow depth of field" outperforms a pure noun-list of what's on screen.
• Pick 720p first if you're iterating. It's the same per-generation price as 1080p, but renders faster and lets you change your mind on the final tier without re-doing the bill math.
• Three-image fusion needs cohesive references. Pick three images that share lighting and composition cues — the model fuses them more cleanly than three random shots.
• Pick 4K only when shipping. A 4K render is roughly 2× the cost of 720p / 1080p; reserve it for the final keeper.
• Keep reference clips ≤ 30s. Reference-to-video requires a source clip of ≤ 30 seconds (the server probes its length via ffmpeg and rejects anything longer with 400); only the first ≤ 10 seconds are used as the reference. Billing is a flat per-generation rate regardless of length. Omit duration — the server rejects requests that pass both.

Related

• Errors catalog
• Authentication
• Quickstart