Veo 3
#7 v Text na videogoogle · v3 · od Mai 2025 (Google I/O, 20.–21. Mai 2025) · 15× · naposledy 30. 6. 2026
33
Momentum
Google Veo 3 is a text-to-video model by Google DeepMind, introduced at Google I/O in May 2025. It generates short video clips (4, 6, or 8 seconds per clip) with natively synchronized audio – including dialogue, sound effects, and ambient noise – from text or image prompts. The model is exclusively cloud-based, accessible via Google's APIs and consumer products; local execution is not supported. Since late 2025, Veo 3.1 is available as the newer successor version; the Veo 3.0 API endpoints are deprecated as of June 2026.
Vývoj momenta
04.04.03.07.
Vlastnosti
| Fine-Tuning | Not available / not publicly documented. Veo 3 is a closed-weights model with no publicly accessible fine-tuning option. |
| Generation Time | Per official Gemini API documentation: minimum 11 seconds, maximum 6 minutes (during peak hours). |
| License | Proprietary / closed-weights – no public model download; usage only via Google's APIs and products under Google's terms of service. |
| Max Resolution | 720p or 1080p native (16:9 or 9:16); 4K via upscaling (Vertex AI / Gemini API, premium tier). Official docs: 720p, 1080p, 4K. |
| Max Video Length | 8 seconds per clip (selectable: 4s, 6s, or 8s). Via the Extend feature, up to 20 extensions of 7s each can be chained, enabling a total length of up to 148 seconds (~2.5 min). |
| Platform | Cloud-only: Gemini API, Vertex AI (Google Cloud), Google AI Studio, Google Flow (filmmaking tool), Gemini app (consumer). No local download/operation possible. |
| Release Date | May 2025 (announced and released at Google I/O, May 20–21, 2025) |