Google Veo 3 API: Generate Video with Native Audio

Veo 3 is Google's flagship video model, and the headline feature is native audio — dialogue, sound effects and ambient noise generated in the same pass as the picture, not bolted on afterward. Normally that means a GCP project, Vertex AI billing, and quota requests before you send a single prompt. Through the Aigenway gateway you skip all of that: one API key, prepaid USD balance, the same task contract you use for every other model.

What you get

Veo 3 ships in a few variants trading render time, resolution and cost against each other, and Google updates them on its own schedule. Confirm the exact codes live on your account before hardcoding a model string:

curl "https://api.aigenway.com/v1/methods?provider=veo" \
  -H "Authorization: Bearer sk_live_..."

That call returns the current model and method values, plus per-second pricing. Expect a standard tier for everyday clips and a fast tier for iteration — the audio generation is part of the base model, not a separate add-on you toggle.

Text-to-video with audio

Send a prompt describing both the scene and the sound, and the response includes audio baked into the output clip — no separate voice or SFX pass to orchestrate:

curl -X POST https://api.aigenway.com/v1/generations \
  -H "Authorization: Bearer sk_live_..." \
  -H "Idempotency-Key: $(uuidgen)" \
  -H "Content-Type: application/json" \
  -d '{
    "provider": "veo",
    "model": "veo-3",
    "method": "text_to_video",
    "params": {
      "prompt": "a barista steams milk behind a coffee counter, hiss of the steam wand, faint chatter in the background, morning light",
      "duration": 8,
      "aspect_ratio": "16:9"
    }
  }'

Idempotency-Key makes retries safe on a slow or dropped connection — resend the same key and you get back the same task instead of paying for a second render.

Image-to-video

Animate a still you already control — a product shot, a character frame, a generated keyframe — and describe the motion and sound you want layered on top:

curl -X POST https://api.aigenway.com/v1/generations \
  -H "Authorization: Bearer sk_live_..." \
  -H "Idempotency-Key: $(uuidgen)" \
  -H "Content-Type: application/json" \
  -d '{
    "provider": "veo",
    "model": "veo-3",
    "method": "image_to_video",
    "params": {
      "prompt": "camera pushes in slowly, distant traffic hum, a bird calls once",
      "input_image": "https://example.com/street-scene.png",
      "duration": 8
    }
  }'

Same endpoint, same provider/model/method/params shape as text-to-video — only the method and the added input_image change, so switching between the two modes is a one-line diff in your integration.

Poll or webhook for the finished clip

Video is never synchronous, and audio generation adds to the render time — the POST above returns a task in queued or processing state, not a finished file. Two ways to get the result:

1. Poll. GET /v1/generations/{task_id} until status becomes succeeded (or failed), then read the output URL off the response.
2. Webhook. Set a callback_url on the request and the gateway POSTs the finished task to your endpoint as soon as it completes — no polling loop to write or rate-limit.

curl -X POST https://api.aigenway.com/v1/generations \
  -H "Authorization: Bearer sk_live_..." \
  -H "Idempotency-Key: $(uuidgen)" \
  -H "Content-Type: application/json" \
  -d '{
    "provider": "veo",
    "model": "veo-3",
    "method": "text_to_video",
    "params": { "prompt": "waves breaking on a rocky shore at dusk, wind and surf audio", "duration": 8 },
    "callback_url": "https://your-app.example.com/webhooks/aigenway"
  }'

The task shape — fields, status values, webhook payload and signature — is identical across every model on the gateway, so this same code also drives Kling and Seedance. Full details are in the task lifecycle and webhooks docs.

Why one gateway beats a GCP project

• No cloud project to stand up. No Vertex AI enablement, no service account, no quota request — sign up and call the endpoint.
• One integration, many models. The same request shape that calls Veo also calls Kling, Seedance and every image model — swap a string, not your SDK.
• Real USD billing. Prepaid balance, no GCP invoicing or committed spend to forecast; check remaining balance from the API at any time.
• Production plumbing included. Idempotency keys, queueing, task state and signed webhooks are already built in.

When Google ships a new Veo version, you point at the new model code from /v1/methods and keep shipping — no re-provisioning required.