Google's DeepMind team has rolled out the new Imagen 3, an advanced AI text-to-image generator.
The latest Imagen 3 is said to be better than the current crop of AI generators such as Midjourney, DALL-E 3, and X's Grok-2.
With default settings, Imagen 3 is capable of generating photo-realistic images at full HD (1024 × 1024) resolution, and this can be upscaled by 2X, 4X, or 8X times.
"Our model is trained on a large dataset comprising images, text and associated annotations. To ensure quality and safety standards, we employ a multi-stage filtering process," the Google DeepMind team said.
It should be noted Google, earlier this year, in February disabled image generation on Gemini AI bots on phones over inaccurate depictions of people.
For instance, when users asked Gemini to produce Viking soldiers' photos, it churned out irrelevant images of people of mixed ethnicities in Viking gear.
It paused image generation on the Gemini AI bot. Now, with the Imagen 3, Google has improved the tool to offer accurate pictures.
Google Imagen 3's sample photos.
The tool -- if it senses the text query contains harmful, and encompasses areas such as child sexual abuse and exploitation, hate speech, harassment, sexually explicit content, and violence and gore -- will decline to entertain such a request.
Google Imagen 3's sample photos with 4X upscaling.
To avoid copyright issues and misinformation lawsuits, Google Imagen 3 will also avoid generating pictures of company brand logos and high-profile celebrities.
For now, Imagen 3 is available for public testers only in the US. It is expected to expand to other global regions including India in the coming months.
Get the latest news on new launches, gadget reviews, apps, cybersecurity, and more on personal technology only on DH Tech.
Published 19 August 2024, 09:39 IST