Google has developed artificial intelligence that's capable of creating images based on simple text input. Imagen, as it's known, is said to easily outperform the competition to create realistic artwork based on the user's words.
We've seen tech like this before, where Generative Adversarial Networks and AI like VQ-GAN+CLIP, Latent Diffusion Models and DALL-E 2 can be used to create interesting images.
Indeed we've used AI before to create intriguing portraits of people. But Imagen promises an "unprecedented degree of photorealism and a deep level of language understanding".
Imagen comes from Google Research's Brain Team. That team says Imagen works so well due to a breakthrough in the training model whereby scaling the language model helped improve performance. It's still early days, but Imagen could represent an interesting view of the future where AI could be used to create images in place of Photoshop.
Though it's not available to the public yet, Imagen has been used to conjure up some pretty impressive images.
No comments:
Post a Comment