February is here and Google’s Bard AI has some neat new features added to it. Similar to the implementations of DAL-E you can now give it a prompt to work on and it can draw results, when it wants to.
The image generator is, I would guess about 4x as fast as Bing’s DALL-E implementation, but due to that it tends to return one or two pictures at a time with a little button to request more.
I haven’t got the quality of scary photos that I managed to get with Bing back a while ago, but it does a somewhat decent job without much description of making undead Muppets.
It seems to not be interested in letting you do anything with celebrities or dead presidents.
Evidently were a machine to look at either of the photos above it would know they’re created by AI by some watermarking that we cannot see.
A double check feature is better (click the Google logo after the text is generated to have a different section of the AI verify with online sources that it wasn’t hallucinating)
And Gemini Pro is available in all supported languages (40) and places (230) now.
It appears the main non-pro version actively is faster than Microsoft’s implementation and is solving my particular queries better, but Bing’s ChatGPT seems to have more creativity. Options on Bard are “more casual” and “more professional” whereas options on Bing are “more precise” to “dropped acid two hours ago let me tell you about feet and how they ground us to this mortal plane.”
It’s cool, still seems to have a way to go and the locking of even dead presidents from photo manipulation seem like a bit of an overreaction… I can’t even get someone dressed like Abraham Lincoln.
Fun times.[Google blog]