Descriptions:
Google’s Nanobanana 2 image generation model is put through a series of deliberately challenging tests comparing it against the previous Nanobanana Pro. Rather than showcasing straightforward use cases, the video constructs prompts specifically designed to stress character consistency, world knowledge breadth, and compositional accuracy.
Tests include generating a single coherent group photo featuring 19 named celebrities — among them Johnny Depp, Taylor Swift, The Rock, Michael Jackson, and Barack Obama — as well as a mixed group shot of anime and cartoon characters spanning Naruto, Rick and Morty, Gojo Satoru, Bart Simpson, and Snow White. A 4×4 grid of 16 distinct facial expressions (including subtle states like nostalgia, jealousy, and anticipation) is also evaluated. Nanobanana 2 performs noticeably better than its predecessor in group composition realism, character likeness, and finger accuracy — notably rendering Bart Simpson and Kenny with their canonical four-fingered hands.
Additional tests cover rare and critically endangered species rendering (with mixed accuracy for both models), filling in an animal cell biology worksheet with correctly labeled anatomy, and solving a math assignment in convincingly handwritten style. Across most categories Nanobanana 2 leads, though both models show limitations in precise scientific diagram labeling. The video is a practical reference for evaluating Google’s latest image generation capabilities against real-world creative and educational use cases.
📺 Source: AI Search · Published February 27, 2026
🏷️ Format: Review
