"Google said Gemini 2.5 Flash's particular strength is in keeping a person or animal's likeness the same from image to image."
"Based on the feedback so far, Google claims that users prefer the new model over those from OpenAI and other rivals, based on its Elo score - a method for calculating the relative skill levels of generative models."
"It's not perfect - it sometimes struggled to replicate faces when combining two different photos - but it's especially strong at making small tweaks to existing pictures."
Google DeepMind developed Gemini 2.5 Flash, tested publicly as "Nano Banana," and integrated the model into the Gemini app and website for general use. Evaluations on LMArena and crowdsourced feedback indicate particular strength in preserving a person or animal's likeness across images. Google reports user preference for the model over competitors based on Elo scoring. Practical testing shows especially strong performance for small edits and detail retention, though the model can struggle when combining two different photos and replicating faces precisely. The model's strengths present a competitive challenge to existing image-editing software such as Adobe Photoshop.
Read at Business Insider
Unable to calculate read time
Collection
[
|
...
]