Google Gemini AI Image Model Gets “Banana” Upgrade
Google upgrades the Gemini Chatbot with a new AI image model, giving users more control over photo editing. It aims to catch up with OpenAI’s popular image tool and pull users out of ChatGPT.
Called Gemini 2.5 Flash Image, this update will be deployed to all users of the Gemini app from Tuesday, as well as developers via the Gemini API, Google AI Studio and the Vertex AI platform.
Gemini’s new AI image model is designed to provide more accurate editing of images based on natural language requests from users. For example, ask ChatGpt or Xai Grok to change the color of someone else’s shirt in the photo. The results may change the distorted face or background.
Google’s new tools are already attracting attention. Over the past few weeks, social media users have captivated the impressive AI image editor of crowdsourcing evaluation platform Lmarena. This model was displayed to users anonymously under the pseudonym “Nano-Banana”.
Google says that’s behind the model (if it wasn’t already clear from all the banana-related tips). This is a native imaging feature within the flagship Gemini 2.5 Flash AI model. Google says the image model is cutting edge in Lmarena and other benchmarks.
In an interview with TechCrunch, Nicole Brichtova, product lead for Google Deepmind’s visual generative model, said:
“This update does a much better job of editing your edits more seamlessly. The output of the model can be used for what you want to use,” Brichtova said.
The AI image model is an important battle foundation for big technology. When Openai launched its native image generator for the GPT-4o in March, it used ChatGpt on the roof thanks to the studio Ghibli meme, which was an AI-generated studio Ghibli meme that left the GPU “melt” GPU.
To catch up with Openai and Google, Meta announced last week that it would be licensed to the AI image model from Startup Midjourney. Meanwhile, the German unicorn black forest lab supported by the A16Z continues to dominate the benchmark with its Flux AI image model.
Perhaps Gemini’s impressive AI image editor can help Google close user gaps in Openai. ChATGPT records over 700 million weekly users. In a revenue call from Google in July, Tech Giant CEO Sundar Pichai revealed that Gemini has 450 million users each month.
Brichtova said Google specifically designed the image model using consumer use cases, including helping users visualize their home and garden projects. This model also has excellent “world knowledge” and allows multiple references to be combined into a single prompt. For example, you can fuse a sofa image, a living room photo, or a color palette into a single cohesive rendering.
Gemini’s new AI image generator makes it easy for users to create and edit realistic images, but it has taken safeguards to limit what users can create. Google has struggled with safeguarding AI image generators in the past. At one point, the company apologised for generating historically inaccurate photos of people and completely rewind the AI image generator.
Now, Google feels it’s a better balance.
“We want to provide users with creative control so that they can get what they want from the model,” says Brichtova. “But that doesn’t seem to go anything.”
Google’s Terms of Service Generation AI section prohibits users from generating “intimate and indifferent images.” These same kinds of protection measures did not appear to exist in Grok, allowing users to create explicit images generated in AI, similar to celebrities such as Taylor Swift.
Brichtova says that to address the rise of Deepfake images, which can make it difficult for users to identify what is actually online, Google applies visual watermarks to AI-generated images and their metadata identifiers. However, people scrolling past social media images may not be looking for such an identifier.
Source link
