Google is rolling Nano Banana 2.1 into the Gemini app today, plus Search, AI Studio, Flow, and Stitch. The pitch is leaps in visual design, mask editing, and keeping the same perso

Google is rolling Nano Banana 2.1 into the Gemini app today, plus Search, AI Studio, Flow, and Stitch. The pitch is leaps in visual design, mask editing, and keeping the same person in the frame. I read the model card before the demo. Their own side-by-side puts the thinking version at 1050 on overall preference, versus 990 for Nano Banana 2 and 935 for Nano Banana Pro. That is Google scoring Google. The same card lists the misses they have not fixed. Small text still blurs. Character match is not always perfect. Left and right still get swapped. Infographic factuality on their auto-rater is 0.521 with thinking on, up from 0.179 on the last Flash image model. Better, and still a coin flip. What you can actually try today is a mask edit on a photo you already have. Paint one region, ask for one change. If the face drifts, the launch number did not matter. Has the mask edit held a real face better than Nano Banana 2, or did it still slide?