Google’s Imagen 4 text-to-image mannequin guarantees ‘considerably improved’ boring photos

Google has unveiled its newest text-to-image mannequin Imagen 4 with the standard promise of “considerably improved textual content rendering” over the earlier model, Imagen 3. The corporate additionally launched a brand new deluxe model known as Imagen 4 Extremely designed to comply with extra exact textual content prompts should you’re keen to pay additional. Each arrive to a paid preview within the Gemini API and for restricted free testing in Google AI Studio.

Google describes the primary Imagen 4 mannequin as “your go-to for many duties” with a value of $.04 per picture. Imagen 4 Extremely, in the meantime, is for “once you want your photos to exactly comply with directions” with the promise of “robust” output outcomes in comparison with different picture mills like Dall-E and Midjourney. That mannequin boosts the worth by 50 p.c to $.06 per picture.

The corporate confirmed off a spread of photos together with a three-panel comedian generated by Imagen 4 Extremely displaying a small spaceship being attacked by a large blue… area lizard? with some sound results like “Crunch!” and inexplicably, “Had!!” The picture adopted the listed immediate beat for beat and appeared okay, not in contrast to a toon rendering from a 3D app.

One other immediate learn “entrance of a classic journey postcard for Kyoto: iconic pagoda below cherry blossoms, snow-capped mountains in distance, clear blue sky, vibrant colours.” Imagen 4 output that to a “T,” albeit in a generic type missing any appeal. One other picture confirmed a mountain climbing couple waving from atop a rock and one other, a pretend “avant garde” style shoot. The photographs have been undoubtedly of fine high quality and adopted the textual content prompts exactly however nonetheless appeared extremely machine generated.

Imagen 4 is ok and does appear a gentle enchancment from earlier than, however I am not precisely wowed by it — significantly in comparison with the market leaders, Dall-E 3 and Midjourney 7. Plus, following an preliminary rush of enthusiasm, the general public appears to be getting sick of AI artwork, with the primary use case apparently being spammy adverts on social media or on the backside of articles.