Veo 3 API: Google Veo 3.1 with native audio

Veo models on this Veo 3 API

Two tiers of Google’s Veo 3.1 are available. Lite is the budget option and the lowest-priced way to get a clip with sound from any model we host. Fast is the higher-fidelity tier, with synchronized dialogue and effects. Both take the same parameters, so switching tiers means changing one string.

Veo 3 API price per clip

A clip costs its per-second rate times its length, rounded up to a whole credit, and the server bills exactly the figures below. On the Starter pack one credit costs about 0.90 cents.

Sound adds 12 credits per second on Lite at 720p. Fast charges one rate at every resolution, so on that tier 1080p costs nothing extra. Other models and the credit packs are on the video API pricing page.

What native audio means

Most video models hand back a silent file and leave music, effects and voice to another tool. Veo 3.1 produces the soundtrack in the same pass as the frames, so a door slams on the frame where it shuts and a character can speak a line you wrote in the prompt. Picture and sound come from one generation, so nothing drifts out of sync.

Sound is opt-in on the Veo 3 API. Add "audio": true to the request body to get it; leave the field out or send false and you receive a silent MP4 at the lower rate. You steer the soundtrack with words: the closing phrase of the latte prompt, gentle cafe chatter and clinking cups, describes only what the viewer hears.

Text to video and image to video with the Veo 3 API

Text to video needs only model and prompt. Anything you omit falls back to the model default: 6 seconds on Lite, 8 on Fast, 720p and 16:9 framing on both. Choose 9:16 for Shorts, Reels and other vertical feeds.

Image to video starts with an upload. Send the raw bytes of a JPG, PNG or WebP file, 10 MB at most, to POST /v1/uploads and you get an upload id back. Pass it as image_id in POST /v1/videos and Veo animates from your picture. Here aspect_ratio also accepts auto, the default, which keeps the shape of the image. Uploads are deleted after 30 days.

# 1. Upload the starting frame
curl https://sivideoapi.com/v1/uploads \
  -H "Authorization: Bearer $SIVIDEOAPI_KEY" \
  -H "Content-Type: image/jpeg" \
  --data-binary @frame.jpg
# {"object": "upload", "id": "…"}

# 2. Animate it
curl https://sivideoapi.com/v1/videos \
  -H "Authorization: Bearer $SIVIDEOAPI_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "veo-3.1-lite", "image_id": "UPLOAD_ID", "prompt": "Steam curls up from the cup as the camera slowly pushes in", "duration": 6, "audio": true}'

The create call returns at once with status pending and the job’s cost_credits. Check GET /v1/videos/{id} every few seconds; once it says succeeded, download video_url, an MP4. A failed job carries an error message, and its credits are already back.

Writing prompts for the Veo 3 API

Prompts may run to 2,000 characters, yet two or three focused sentences usually serve Veo 3.1 better than a pile of adjectives.

• Open with subject, action and place. The fox prompt below covers all three before its first comma.
• Name one camera move. Close-up, slow tracking shot or push-in: Veo understands film terms, and one move per clip keeps motion clean.
• Set the light and lens. Golden hour, neon or shallow depth of field shift the mood more than extra adjectives.
• Give sound its own clause. List the ambience and effects, and put spoken lines in quotation marks so the model treats them as dialogue.
• Keep to one moment. Four to eight seconds is a single shot; for a sequence, render several clips and edit them together.
• From an image, describe motion. The frame already shows who and where, so spend the words on what moves.
• Draft cheap, finish rich. A 4-second silent Lite draft costs 72 credits; render the keeper at full length with sound.

Veo 3 API or Seedance and Kling on the same key

One balance pays for every model, so picking Veo is a per-request choice, not a vendor contract:

Lite stays the sensible default for short clips with sound. Seedance 2.0 earns its higher rate when a shot needs ten seconds, a square frame or several cuts in one clip, and Kling 3.0 Turbo suits fast action. Only the allowed values change between models: GET /v1/models lists them with prices and needs no key, and an out-of-range value gets HTTP 400 before any credits are held.

Veo 3 API FAQ

Is this the official Google Veo 3 API?

No. SiVideoAPI is an independent service and is not affiliated with Google. We run Veo 3.1 through our inference provider and resell it by the second, so you need no Google Cloud project, Vertex AI quota or cloud billing account, only a key from our dashboard.

Which Veo version is behind the Veo 3 API?

Veo 3.1, in two tiers with the model ids veo-3.1-lite and veo-3.1-fast. People often search for Veo 3; 3.1 is the release of that family we serve.

How long does a Veo render take?

It depends on the tier, the length and the provider queue, so poll GET /v1/videos/{id} every few seconds instead of sleeping a fixed time. A job that never completes is cancelled after 45 minutes and refunded.

What does a refused prompt cost?

Nothing. Prompts pass a safety check before any credits are held, and if Veo itself declines the request later, the job ends as failed and the full cost returns to your balance on its own.

How many Veo videos can run in parallel?

Each account can have ten renders in progress and send twenty create requests per minute. Going over either limit returns HTTP 429, which you can retry after a short wait.

Can I test the Veo 3 API for free?

Yes. A new account created with Google receives 80 credits, and a silent Lite draft of 4 seconds at 720p uses 72 of them.

Read the API documentation for every endpoint, create a key on the API key dashboard, compare all models on the SiVideoAPI home page, and see the refund policy before you buy credits.