Google is improving its image generation technology to keep pace with its competitors.
At the company’s Mountain View I/O Developer Conference on Tuesday, Google announced Imagen 3, the latest addition to the tech giant’s Imagen family of generative AI models.
Demis Hassabis, head of DeepMind, Google’s AI research division, said Imagen 3 more accurately understands the text prompts it translates into images compared to its predecessor, Imagen 2, and is more “creative and detailed » in his generations. Additionally, the model produces fewer “annoying artifacts” and errors, he said.
“It is (also) our best model yet for rendering text, which is a challenge for image generation models,” Hassabis added.
To allay concerns about the possibility of creating deepfakes, Google says Imagen 3 will use SynthID, an approach developed by DeepMind to apply invisible cryptographic watermarks to media.
Signups for Imagen 3 private preview are available in Google’s ImageFX tool, and Google says the model is “coming soon” to developers and businesses using Vertex AI, Google’s generative AI development platform. Google company.
Google usually doesn’t reveal much about the source of the data it uses to train its AI models – and this time is no exception. There is a reason for this. Much of the training data comes from public sites, repositories, and datasets on the web. And some of this training data, especially copyrighted data scraped without the content creators’ permission, is a source of intellectual property-related lawsuits.
Google’s Web Publisher Controls allow webmasters to prevent the company from scraping data, including photos and videos, from their websites. But Google doesn’t offer an “opt-out” tool, and unlike some of its competitors, the company hasn’t committed to compensating rights holders for their (in some cases unknowing) contributions to datasets. training.
The lack of transparency is not surprising. But it’s disappointing, especially from a company with resources like Google’s.
yahoo
