Recently Google published Imagen 3, their in-house artificial intelligence (AI) model for image generation. The tech giant did not mention the rollout; instead, it delivered it secretly to users.
A research paper outlining the operation of the image creation model was also published in an online journal. Currently, the text-to-image creation technique is only available in the United States, and there is no news on when it will be made available to users worldwide.
The web giant’s AI Test Kitchen now allows users to sign up and make images using the AI model. The third version of the Imagen model is expected to have improved texture production and word recognition skills, as well as stricter prompt adherence.
The AI model is only available in the United States. A Reddit user reported that he was able to generate images in a range of styles, including Nikon DSLR style, GoPro style, wide angle lens, and more. However, the model is believed to be having difficulty producing close-up photographs with multiple people as well as underlit images, which its predecessor could do.
Another area where Imagen 3 struggles is limbs. The user claimed that the model was producing erroneous results when using prompts such as “a guy holding a cup of coffee”. The AI would generate extra limbs, create a random limb holding the object, or fuse the object and the limb. The image generation model is also said to have very strict censorship in prompts.
Also Read:
- Chrome OS Is Missing or Damaged: Tips to Fix This Error
- Gogoanime Facts and 10 Best Working Alternatives
Notably, the free Gemini chatbot can make images, but it does so using Gemini’s capabilities. Imagen 3 is built on a different architecture, and because its dataset consists primarily of pictures, it is better trained to generate AI images.