Gemini Advanced users can now create and share videos using the new video model, Veo 2, which translates text prompts into dynamic videos. This feature is part of Google Labs' Whisk, allowing users to animate images created from both text and image prompts.
Video Creation with Gemini
Veo 2 enhances video generation by producing high-resolution, detailed videos with cinematic realism. It understands real-world physics and human motion, resulting in fluid character movements and lifelike scenes. Users can generate an eight-second video clip at 720p resolution in MP4 format by selecting Veo 2 from the model dropdown in Gemini. There is a monthly limit on video creation, with notifications provided as users approach this limit.
Creating videos involves describing the desired scene, with more detailed descriptions offering greater control over the final output. This allows for creative exploration across various visual styles, from realism to fantasy. Sharing videos is straightforward, enabling uploads to platforms like TikTok and YouTube Shorts.
Example Prompts
- Glacial Cavern: A wide shot of a glacial cavern illuminated by pale cyan light, featuring frozen candy figures and figures in white exosuits.
- Cozy Forest Den: An animated shot of a mouse with oversized glasses reading by a glowing mushroom.
- Pacific Coastline: An aerial view of a cliff and beach at sunrise or sunset, capturing the serene beauty of the coastline.
- Melting Ice Cream: A time-lapse of voxel-style ice cream melting under a blue sky, focusing on texture and form changes.
Whisk Animate
Whisk Animate, part of Google Labs, allows users to animate images into eight-second videos using Veo 2. This feature is also available to Google One AI Premium subscribers.
Safety Measures
To ensure a safe video generation experience, Google has implemented extensive evaluations to prevent policy violations. All videos generated with Veo 2 include SynthID, a digital watermark indicating they are AI-generated. While Gemini's outputs depend on user prompts, there may be instances of objectionable content. Feedback mechanisms are in place to facilitate ongoing improvements.