Text-to-Image Imagen from Google Paints Some Bizarre but Realistic Pictures

June 16, 2022

Google Research gives us a new entry in the text-to-image AI arena. Imagen joins the likes of DALL-E and LDM, tools that generate images from brief descriptive sentences. TechRadar’s Rhys Wood insists the new software surpasses its predecessors in, “I Tried Google’s Text-to-Image AI, and I Was Shocked by the Results.” Visitors to the site can build a sentence from a narrow but creative set of options and Imagen instantly generates an image from those choices. Wood writes:

“An example of such sentences would be – as per demonstrations on the Imagen website – ‘A photo of a fuzzy panda wearing a cowboy hat and black leather jacket riding a bike on top of a mountain.’ That’s quite a mouthful, but the sentence is structured in such a way that the AI can identify each item as its own criteria. The AI then analyzes each segment of the sentence as a digestible chunk of information and attempts to produce an image as closely related to that sentence as possible. And barring some uncanniness or oddities here and there, Imagen can do this with surprisingly quick and accurate results.”

The tool is fun to play around with, but be warned the “photo” choice can create images much creepier than the “oil painting” option. Those look more like something a former president might paint. As with DALL-E before it, the creators decided it wise to put limits on the AI before it interacts with the public. The article notes:

“Google’s Brain Team doesn’t shy away from the fact that Imagen is keeping things relatively harmless. As part of a rather lengthy disclaimer, the team is well aware that neural networks can be used to generate harmful content like racial stereotypes or push toxic ideologies. Imagen even makes use of a dataset that’s known to contain such inappropriate content. … This is also the reason why Google’s Brain Team has no plans to release Imagen for public use, at least until it can develop further ‘safeguards’ to prevent the AI from being used for nefarious purposes. As a result, the preview on the website is limited to just a few handpicked variables.”

Wood reminds us what happened when Microsoft released its Tay algorithm to wander unsupervised on Twitter. It seems Imagen will only be released to the public when that vexing bias problem is solved. So, maybe never.

Cynthia Murrell, June 16, 2022

Comments

Comments are closed.

  • Archives

  • Recent Posts

  • Meta