Veo

Veo is a video-generation model from Google DeepMind, announced on December 16, 2024. It creates videos from text prompts, with output up to 4K resolution.

What Veo does

Veo is designed to handle complex scenes and cinematic language, including camera directions and motion. In MovieGenBench tests, 85% of users preferred it over OpenAI’s Sora. It can generate videos up to 2 minutes long.

Key capabilities include:

• Up to 4K video generation (4096×2160)
• Long prompts (up to 100 words)
• Camera control (e.g., low angle, close-up)
• Physics simulation (gravity, collisions)
• 12 visual styles (e.g., realism, animation)
• 70% fewer artifacts vs. Veo 1
• Text-to-video and image-to-video generation

How to use Veo (Veo 3)

Veo is available through the VideoFX web app on labs.google (sign-ups have been open since December 2024).

• Sign up on labs.google
• Open VideoFX
• Enter a prompt
• Choose settings (angle, style)
• Generate and download the video

Notes and limitations

• SynthID watermark on every frame
• Max length: 128 seconds
• VideoFX initially limited to 720p
• English-only
• Free for early users; a paid version is planned for 2026 (pricing not announced)
• Planned integrations in 2025: YouTube Shorts (up to 15 seconds), Vertex AI