After its African American Nazis and glue on pizza, it's understandable that Google is a bit more cautious when launching artificial intelligence products. Yet the web quickly spotted Imagen 3, the new image generator from Mountain View. Available through the ImageFX platform at AI Test Kitchen, Imagen 3 uses a dynamic prompt system, offering suggestions and modifications with a couple of clicks.
Generating Images with Google's Imagen 3
AI-powered image generation has returned to the spotlight. On one hand, Grok's system seems to be out of control, and on the other, Google barely made a sound when launching its new Imagen 3 model. The Mountain View giant mentioned it during its I/O conference in May, and the official paper was published on August 13. Initially enabled for US users, a quick visit to the ImageFX platform on the official AI Test Kitchen portal confirmed that it's now available.
All the platform asks is to sign in with a Google account. Once done, you're taken directly to the interface, with a section to enter your prompt, several recommended tags at the bottom, and a settings option that lets you play with the seed number. If inspiration takes a break, the 'I'm Feeling Lucky' button will generate an automatic prompt.
Safety and Limitations
Imagen 3 has multiple layers of protection, and it immediately rejects certain elements (public figures, modern weaponry), though I did find some exceptions. For example, the model blocked all requests for an automatic pistol, but it had no problem generating a Roman gladius or a French musket in the context of an exhibition. I imagine it's only a matter of time before users find cracks in its armor.
Anatomy and Quality
Now, what about anatomy? Imagen 3 often makes classic mistakes like very long arms and extra phalanges on fingers, but I must admit its results are superior to Stable Diffusion, and on par with what FLUX.1 offers. Link below!
Official site: Click here
Access the paper: Click here
