Overview
veo-3-fast is Google’s Veo 3 fast tier — 720p output, no audio, fastest gen in the Veo family. Best default for budget cinematic clips, social shorts, and rapid iteration. Async long-running operation (LRO) pattern: submit returns immediately with a job_id, generation runs in the background, poll for completion.
Flagship tier with native audio is veo-3.
Specs
| Field | Value |
|---|---|
| Model ID | veo-3-fast |
| Creator | |
| Backend | Vertex AI (veo-3.0-fast-generate-001, us-central1) |
| Best for | Budget cinematic clips, social shorts, rapid iteration |
| Resolution | 720p |
| Audio | No |
| Aspect ratios | 16:9 (default), 9:16 |
| Duration | 4, 6, or 8 seconds (Vertex enum) |
| First-frame I2V | Yes — pass image_url |
| Pricing mode | Per second × duration |
| Default latency | ~30–60s end-to-end (LRO submit → poll done → blob mirror) |
| Output | Blob-hosted MP4 (Vercel CDN, durable URL) |
Pricing
Per second of generated video. List = provider cost × 1.35.| Variant | Provider $/s | Kyma list $/s | 8s clip |
|---|---|---|---|
veo-3-fast | $0.10 | $0.135 | $1.08 |
GET https://kymaapi.com/v1/pricing.
Compared to other video models on Kyma
| Strength | veo-3-fast | veo-3 | kling-3-pro | seedance-2-fast | hailuo-02-768p |
|---|---|---|---|---|---|
| Cheapest Veo | ★★★★★ | ★★ | n/a | n/a | n/a |
| Native audio | — | ★★★★★ | — (use -audio) | ★★★★★ | — |
| Cost $/8s | $1.08 | $4.32 | $0.90 | $1.94 (audio) | $0.42 (flat) |
| Resolution | 720p | 1080p | configurable | 720p | 768p |
| Photoreal humans | ★★★★ | ★★★★★ | ★★★★ | ★★★★ | ★★★ |
Use this when
- You need Veo’s specific style on a budget.
- Audio not required — pure visuals are enough.
- 720p is acceptable (social, drafts, thumbnails).
Pick something else when
- You need native audio (dialogue + ambient + lip-sync) →
veo-3. - Cost is the only constraint →
hailuo-02-768pat $0.42 flat for 6–10s. - You want multi-shot action with audio →
seedance-2-pro.
Example — text-to-video
Example — image-to-video (first frame)
202 with job_id; poll GET /v1/jobs/{id} until status is succeeded (~30–60s for 4s clips). Output URL is a durable Vercel blob (Kyma mirrors the Vertex response before exposing the URL — no expiry concerns).