Gemini has enhanced its image editing features with a recent update from Google DeepMind. The model, referred to as “nano banana” on the LMArena leaderboard, is officially named Gemini 2.5 Flash Image. It has been designed to ensure subject consistency across different AI-generated image modifications, tackling a frequent challenge faced by AI image models.

Since April, users have had the ability to upload and edit photos directly in Gemini without the necessity of technical photo editing expertise. With this updated model, users can modify a subject’s clothing and setting while preserving their appearance. Multiple images can be merged, and particular details can be adjusted, such as altering room colors or furniture.

In a trial, Gemini tried to place a dog into a downward dog pose within a yoga studio. Although the resemblance was upheld, the pose was not correctly executed. Google DeepMind recognizes possible inaccuracies concerning fine details and text. Images feature both visible and invisible watermarks to identify AI-generated content. The update is currently accessible in the Gemini app.