Google DeepMind has just opened up access to Project Genie. This new AI tool lets users create interactive game worlds using text prompts or images. As of Thursday, subscribers to Google AI Ultra in the U.S. can try out this experimental prototype.
Project Genie is powered by several technologies, including Genie 3, an advanced world model, and the image generator Nano Banana Pro. This release aims to collect user feedback, crucial for refining these AI systems. DeepMind believes that world models like Genie are key to achieving artificial general intelligence (AGI). In the short term, they plan to use this technology in video games and simulations for robot training.
The competition in this space is heating up. Other companies, such as Fei-Fei Li’s World Labs and former Meta scientist Yann LeCun’s AMI Labs, are also developing their own world models. As Shlomi Fruchter, a DeepMind research director, told TechCrunch, “It’s exciting to have more users access it and provide feedback.”
While using Project Genie, you start by creating a “world sketch.” You provide details about the environment and a main character. The image generation model then creates a base image. You can modify this image to guide the world-building process. However, the results can be inconsistent. For instance, users might ask for green hair and end up with purple. Using real photos can also lead to hit-or-miss outcomes.
Once you’re happy with your image, Genie generates an explorable world in seconds. Currently, users can navigate for only 60 seconds due to budget and computing limits. Each session relies on dedicated compute power. Fruchter explained that this limitation helps them manage resources while gathering valuable insights from user interactions.
Despite its limitations, Project Genie shows promise. The model shines when creating whimsical, stylized worlds but struggles with photorealistic settings. A user creating a castle made of marshmallows was delighted by the results, which delivered a vibrant, imaginative space. On the other hand, attempts at realistic environments often fell short, resembling gaming graphics instead of lifelike settings.
Safety features are also in place to prevent copyright issues. For instance, users can’t create content related to Disney characters after a cease-and-desist from the company last year. This emphasizes the importance of balancing creativity with legal boundaries.
User reactions highlight the tool’s potential and current gaps. Many enjoy the whimsical worlds but find navigation tricky. Feedback indicates that non-gamers struggle with the controls, making for a frustrating experience at times. DeepMind is aware of these issues and is dedicated to improving user interaction in the future.
Overall, Project Genie offers a glimpse into the future of interactive AI. While still in the experimental stage, it holds promise for creating unique experiences that can’t be replicated elsewhere. As it evolves, the hope is that it becomes easier and more enjoyable to use.
Source link
Google,Exclusive,gemini,google deepmind,world models,genie 3,nano banana pro,project genie
