Google’s Imagen 4 text-to-image model promises ‘significantly improved’ boring images

Google has unveiled its latest text-to-image model Imagen 4 with the usual promise of “significantly improved text rendering” over the previous version, Imagen 3. The company also introduced a new deluxe version called Imagen 4 Ultra designed to follow more precise text prompts if you’re willing to pay extra. Both arrive to a paid preview in the Gemini API and for limited free testing in Google AI Studio.

Google describes the main Imagen 4 model as “your go-to for most tasks” with a price of $.04 per image. Imagen 4 Ultra, meanwhile, is for “when you need your images to precisely follow instructions” with the promise of “strong” output results compared to other image generators like Dall-E and Midjourney. That model boosts the price by 50 percent to $.06 per image.

The company showed off a range of images including a three-panel comic generated by Imagen 4 Ultra showing a small spaceship being attacked by a giant blue… space lizard? with some sound effects like “Crunch!” and inexplicably, “Had!!” The image followed the listed prompt beat for beat and looked okay, not unlike a toon rendering from a 3D app.

Google

Another prompt read “front of a vintage travel postcard for Kyoto: iconic pagoda under cherry blossoms, snow-capped mountains in distance, clear blue sky, vibrant colors.” Imagen 4 output that to a “T,” albeit in a generic style lacking any charm. Another image showed a hiking couple waving from atop a rock and another, a fake “avant garde” fashion shoot. The images were definitely of good quality and followed the text prompts precisely but still looked highly machine generated. 

Imagen 4 is fine and does seem a mild improvement from before, but I’m not exactly wowed by it — particularly compared to the market leaders, Dall-E 3 and Midjourney 7. Plus, following an initial rush of enthusiasm, the public seems to be getting sick of AI art, with the main use case apparently being spammy ads on social media or at the bottom of articles.

Google

This article originally appeared on Engadget at https://www.engadget.com/ai/googles-imagen-4-text-to-image-model-promises-significantly-improved-boring-images-111711517.html?src=rss