Replicate Google Veo-3 — How to Run

This guide explains practical ways to run Veo‑3 style text-to-video generation using compatible APIs, how to structure prompts, and what to expect for speed, pricing, and limits.

Option A: Ulazai API (Veo 3 / Veo 3 Fast)

Generate 8-second videos with audio in 16:9 or 9:16. Current model availability, welcome credits and EUR pricing are shown in your UlazAI account.

Start with Veo 3 Fast

Option B: Third‑party Platforms

Some platforms expose Veo‑3–like models via hosted runtimes. Check pricing, regional access, and run‑time limits before committing.

Prompt Format

Use concise cinematic instructions: shot type, subject, movement, environment, lighting, and style. For API JSON, see our prompt guides.

Prompt Guide Examples JSON Format

Quick API Example

{
  "model": "veo3_fast",
  "prompt": "Aerial drone shot over misty mountains at golden hour, cinematic, smooth pan",
  "aspect_ratio": "16:9",
  "audio": true
}

Use Veo 3 Fast to iterate quickly, then switch to premium Veo 3 for final renders.

Cost & Limits

8s videos with audio (16:9 and 9:16)
Generation time depends on the model and current queue
Pricing in EUR — see details: Veo 3 Pricing

Step‑by‑Step (Studio)

Choose Veo 3 Fast for quick drafts
Pick aspect ratio and enable Audio
Paste a cinematic prompt (15–30 words)
Generate, review, and refine
Switch to premium Veo 3 for a final render

FAQ

Is Veo‑3 available on Replicate?

Availability changes over time. Check Replicate and your UlazAI account for the models, limits and EUR prices currently available to you.

Can I run long videos?

Use 8‑second clips and stitch scenes for longer outputs. See the Video Length Limits guide for best practices. Read more

How do I keep characters consistent?

Use reference images and our multi‑scene workflow. See the Consistent Characters guide for details. Learn more

Run Veo through a job flow you can observe and retry safely

Select the model, submit the prompt and media once, save the returned job ID and poll until completion. This is the practical way to run an asynchronous video model, whether the backend provider is Replicate or another route.

Do not submit again after a browser timeout until you have checked the first job. Compare duration and cost from the current endpoint rather than a third-party tutorial.