Google Reveals Gemini-Flash 2.5 Image To Be Nano-Banana, Smashes Image Benchmarks

Google already had the best video generation models around, and it is now head and shoulders above the competition in the image generation space as well.

Google has announced the launch of Gemini-Flash 2.5 Image, which it revealed to be the viral nano-banana model that had been generating buzz on platforms like LMArena. “The new Gemini 2.5 image model is by far the best out there with a whopping +180 ELO point lead in image editing & it really excels at character consistency,” wrote Google DeepMind CEO Demis Hassabis on X. “Available for free in the GeminiApp right now. Try uploading an image & playing around with it, it’s pretty amazing!” he added.

On the LMArena, Gemini 2.5 Flash image was far ahead of competitors in overall preference, character, creative, infographics, object/environment and in particular the product recontextualization benchmarks. On LMArena, users try out prompts on two different models, and pick the one they like in a blind test. Users overwhelmingly seemed to choose Gemini 2.5 Flash’s outputs over those of other models.

The model is currently free to try on the Gemini app. On the API, the model is priced very competitively, at $0.039 per generation. In comparison, OpenAI’s image model costs $0.19 per image.

As users had noticed, the model excels in consistent character and product tasks. “A fundamental challenge in image generation is maintaining the appearance of a character or object across multiple prompts and edits,” Google wrote on its blog. “You can now place the same character into different environments, showcase a single product from multiple angles in new settings, or generate consistent brand assets, all while preserving the subject,” it added.

Google has even created a tool in which users can simply drop in one image into another, and it gets integrated into the image.

Google Gemini 2.5 Flash Image does seem to be a step change over existing image generation models. It seems to be much better than other approaches in creating consistent characters, and can be particularly useful for marketing teams that are looking to create product shoots. Google has also priced the model very competitively, and it can take up a whole lot of market share from existing models. Google had been slow off the blocks with AI, but with a flurry of top-quality releases in recent weeks, seems to be coming into its own in a race that was always its own to win.

Posted in AI