Google announced on Wednesday (13) Imagen 2, the new (and second) generation of its AI model for image creation. This artificial intelligence is only available to customers of Vortex AI, a service available from Google Cloud. In the previews shown by big tech, Imagen 2 delivers ultra-realistic results.
Google also claims that Imagen 2 has had its understanding of prompts improved, which should lead the user to achieve the desired results (or as close to it as possible). Unfortunately, because it’s a paid service within a paid service, we can’t test it and there’s not even a testing space made available by the company — and the case of Gemini leaves the foot behind.
In the official announcement of Imagen 2, made on the website of DeepMind, a subsidiary of Google, the company reveals that to improve the understanding of the prompts, it has worked on expanding the description of the images used in the training of the AI. According to Google, this allowed Imagen 2 to improve its understanding of the context and nuances of commands.
The AI model used in the tool also received “specialized training” in aesthetics based on human preferences. Google has created an aesthetic scale (the popular Instagrammable) for each image, taking into account factors such as lighting, exposure time, framing, and sharpness. The company claims that this allows Imagen 2 to generate even higher quality images.
This ability, added to the example given by Google, raises a question: how soon will AIs be present in social media feeds through beautiful instagrammable images? Images of people are becoming more and more realistic, but why would anyone bother to post a photo of someone else or someone similar? Images of beautiful landscapes are still more attractive for likes.
Imagen has a tool to recognize your images
A differential of Imagen’s AI is that it has a watermark that recognizes images made by itself. This Google technology creates an identifier in the pixels of the photos and artwork created by artificial intelligence and, according to the company, cannot be removed in any way. So, a user can upload an image to Imagen and it will identify if it was made by AI or not.
The downside to Imagen is that it’s still restricted to Google customers. Bard isn’t able to generate images yet (although it says it does, it just searches for them on Google). Google’s generative AI utilizes the Gemini LLM, but Google’s plans to bring image creation to Bard are unknown. However, Imagen already gives a preview of what we may see in the future (distant or near).