Google Gemini’s AI image model gets a ‘bananas’ upgrade

Google is set to elevate its Gemini chatbot with a revamped AI image model, designed to offer users greater precision in photo editing. This upgrade, known as Gemini 2.5 Flash Image, is being rolled out to all users of the Gemini app starting Tuesday, along with availability for developers through the Gemini API, Google AI Studio, and Vertex AI platforms. The fresh AI image model aims to facilitate more accurate edits based on natural language commands while maintaining the integrity of faces, animals, and intricate details—an area where many competing tools typically falter. For example, when users request a color change for clothing in an image, other tools often produce results with distorted features or altered backgrounds. In contrast, Google's new offering has garnered significant attention, especially after users praised its capabilities on the crowdsourced platform LMArena, where it appeared under the playful pseudonym "nano-banana." Nicole Brichtova, a product lead for visual generation models at Google DeepMind, emphasized the model’s advancements in a TechCrunch interview, stating, "We’re really pushing visual quality forward, as well as the model’s ability to follow instructions. This update makes edits smoother, and the outputs are ready for practical use." The competition among Big Tech companies in the AI image model arena is intensifying. OpenAI's launch of GPT-4o’s native image generator in March significantly boosted ChatGPT's user engagement, leading to a surge in AI-generated content. In response, Meta recently announced plans to license AI image models from the startup Midjourney, while Black Forest Labs, backed by a16z, continues to excel with its FLUX AI image models. Despite the challenges, Gemini’s advanced AI image editor aims to help Google bridge the user gap with OpenAI, which boasts over 700 million weekly users compared to Gemini's 450 million monthly users as revealed by CEO Sundar Pichai during a July earnings call. Brichtova noted that the image model has been tailored for consumer applications, such as aiding users in visualizing home and garden projects. It also features enhanced “world knowledge” and can integrate multiple references into a single prompt, allowing for seamless combinations of images. While the new AI image generator simplifies the process of creating realistic images, Google has implemented safeguards to regulate user-generated content. The company previously faced backlash for inaccuracies in generated images and has since worked to strike a balance between creative freedom and responsible usage. Brichtova reiterated, "We want to give users creative control, but it’s not like anything goes," highlighting the generative AI terms of service that restrict the creation of non-consensual intimate images. To combat the rising issue of deepfake imagery, Google has also incorporated visual watermarks and identifiers in the metadata of AI-generated images. However, Brichtova acknowledged that users scrolling through social media may not always notice these identifiers, underscoring the ongoing challenge of maintaining trust in digital content.