Important points
- Veo produces high-quality, consistent videos with a focus on cinematic style and natural language input for creators.
- Imagen 3 improves image generation by rendering text more accurately, interpreting long prompts, and generating images in a wider range of styles.
- Both Veo and Imagen 3 are available in private preview to select creators and showcase Google's advances in AI models for video and image generation.
Today is a big day for AI at Google I/O, and in addition to all the talk about Gemini 1.5 Pro, Google's DeepMind Lab announced several new AI models for video and image generation. The new image generation model is Imagen 3, which has some major improvements over the previous model, while the equivalent model for video is called Veo.
Project Astra is Google's answer to GPT-4o, powered by Gemini, and coming to your Google Pixel.
Google takes OpenAI seriously.
Veo produces high-quality, consistent video
Veo is the better of the two models because the video generation is newer and improved faster. Google's Veo is up against his OpenAI's Sora, which is also very impressive and promises to deliver 1080p high-quality video with an emphasis on consistent results. Google says it can generate videos in a “wide range of cinematic and visual styles” and understands terms like “time-lapse,” allowing creators to create all kinds of shots using natural language.
Google also highlights how Veo learns from years of generative models to understand video content and simulate real-world physics to produce more realistic results. To demonstrate Veo's potential, Google teamed up with Donald Glover to create a project with a new model featuring all kinds of shots that could be mistaken for real footage.
This tool is currently available in private preview on VideoFX for some creators.
Imagen 3 raises the bar for image generation
On the still image front, Google introduced Imagen 3. This is the latest version of the image generation model that can produce realistic images with more detail and less articulation than before. One of the big improvements in IMagen 3 is that it renders text much better. This has been one of the telltale signs of AI-generated images so far. This should actually result in more consistent and readable text.
Imagen 3 better interprets long prompts and also incorporates the finer details found in those prompts. Additional details can be used to describe foreground and background elements, and imagen 3 can generate output that meets all of the prompt's criteria. Moreover, thanks to advanced features, you can generate images in a wider range of styles. As an example, the image above uses the following prompt:
A weathered wooden mechabot covered in flowering vines stands quietly in a field of tall wildflowers, a small blue bird perched in its outstretched hand. A digital manga featuring warm colors and soft lines. A large cliff with a waterfall looms behind it.
Imagen 3 is currently available to some creators at ImageFX and will soon be available at Vertex AI.
