After 15 months of complete silence, Google has lastly launched an replace to its AI picture generator, Imagen, and boy, the outcomes look so good.
Imagen 2.0 was introduced quietly throughout Google’s I/O convention in May 2023. Today, it’s lastly right here, however solely accessible to Google Cloud prospects utilizing Vertex AI.
What is Google Imagen?
Google Imagen is a text-to-image AI mannequin that may create photorealistic pictures from a textual content description. Like different AI picture mills corresponding to Dall-E 3 or Midjourney, Imagen relies on a diffusion mannequin, which is a sort of neural community that may step by step refine pictures to match a given textual content immediate.
Imagen is educated on a large dataset of textual content and pictures, which permits it to generate pictures which might be each correct and detailed.
A conditional diffusion mannequin maps the textual content embedding right into a 64×64 picture. Imagen additional makes use of text-conditional super-resolution diffusion fashions to upsample the picture from 64×64 to 256×256 and 256×256 to 1024×1024.
If you need to dig deeper into how Imagen works, try the whitepaper right here.
What’s new in Google Imagen 2?
These are the important thing enhancements in Imagen 2.0:
- Improved image-caption understanding
- More practical picture technology
- Fluid type conditioning
- Advanced inpainting and outpainting
Let’s unpack every of those options.
Improved image-caption understanding
To generate higher-quality, extra correct pictures that higher match prompts, Imagen 2’s coaching dataset had extra detailed picture captions. This helps Imagen 2 higher grasp the connection between pictures and phrases, enhancing its understanding of context and nuance.
Take this immediate for example:
An picture of: Consider the subtleness of the ocean; how its most dreaded…
