Google Releases Imagen 3 AI Image Generation Model with Enhanced Capabilities

Improved capabilities have been added to Google's Imagen 3 AI image generation model, which is now available to all US users.

On Thursday, Google made Imagen 3, its in-house AI model for creating images, available to the public. The tech giant quietly distributed the model to users rather than making any public announcements about it. Moreover, an exploration paper specifying the operations of the picture age model was likewise distributed in a web-based diary. The text-to-image generation model is currently only available to US users, and it is unknown when it might be made available to users in other regions.

Google's AI Test Kitchen has released the Imagen 3 AI model, allowing users to sign up for the platform and use the AI model to create images. Improved texture generation and word recognition capabilities, in addition to stricter prompt adherence, are said to be included in the third generation of its Imagen model.

Gadgets 360 was unable to test the platform because the AI model is only available in the United States. However, a Reddit user claimed to be able to produce images in a variety of styles, including GoPro-style, wide-angle lens, and Nikon DSLR quality. However, unlike its predecessor, the model is said to have difficulty producing close-up images of multiple people and images with low lighting.

Legs are another area where Imagen 3 struggles. The client guaranteed that the model was creating mistaken results while utilizing prompts, for example, "a person holding some espresso". The AI would ultimately produce additional limbs, a random limb that would hold the object, or a fusion of the limb and the object. Additionally, prompt censorship is said to be extremely stringent under the image generation model.

A research paper was also published by Google in the online pre-print journal arXiv. There, the company made it clear that it used a latent diffusion model, which is a version of the Stable Diffusion diffusion model that was popularized. Additionally, the company stated that new approaches have been employed to minimize the Imagen 3 model's potential harm.

Notably, the free version of the Gemini chatbot uses Gemini's capabilities to generate images. Imagen 3 is better trained to generate AI images because it is built on a different architecture and its dataset primarily consists of images.

Source - https://www.gadgets360.com/ai/news/google-imagen-3-ai-image-generation-model-released-improved-capabilities-6351128