The latest version of Google’s image-generation diffusion network is now available for everyone in the US to try.
When people talk about the AI race, most folks’ thoughts immediately go toward large language models powering functions like ChatGPT or Google Assistant. There’s a lot more going on in the AI space than just language models, though. Before convincingly-human language models became commonplace, everyone was obsessed with AI image generators. Development hasn’t stopped on those, though, and the latest iterations are probably way better than you expect.
Case in point: Google’s just unveiled the latest version of its own image diffusion model known as Imagen 3. There’s an academic paper to read if you’re interested, but it’s mostly just Google presenting results where people prefer Imagen 3’s output over that of other contemporary models and then several pages of Google’s Deepmind researchers insisting that they have tried very hard to keep people from misusing the image generator. The researchers apparently compared Imagen 3 against DALL-E 3, Stable Diffusion 3 Large, and Midjourney v6 and generally their model came out ahead, although Midjourney apparently creates more « visually pleasing » images.
You can go try out Imagen 3 right now on Google’s « AI Test Kitchen » website, in the ImageFX section. The UI is very simple and easy to use, and in fact features one of the best UIs we’ve seen for a web-based image generation service.
