Summary
Paul J Lipsky puts Google’s new Nano Banana 2.1 through a series of side-by-side tests against Nano Banana Pro, Nano Banana 2, and ChatGPT Image 2.5. Google claims the new model beats all of its earlier image models, including Pro, at a quarter of the price, with better instruction following, stronger in-image text rendering, grounding in Google image search, and improved character consistency. It is available in Google AI Studio, the Gemini API, and the Gemini app.
The tests run in OpenArt, which sponsored the video and lets several models generate from one prompt. The first is a busy, photorealistic coffee shop street scene packed with text to stress-test rendering and reflections. Results are judged blind before the models are revealed, looking at text legibility, color saturation, and anatomy such as hands.
A second test is a cinematic low-angle gas station scene with specific requirements: which hand holds the umbrella, which side of the sedan faces the camera, and where the pump sits. None of the models meets every instruction. A further test covers motion with a person jumping on a trampoline, and the video also compares 2.1 directly with Nano Banana 2.
📺 Source: Paul J Lipsky · Published October 06, 2026
🏷️ Format: Comparison
