Google updates its chatbot Gemini using a new AI image model, which provides users with more accurate control over editing of photos, which aims to catch up with popular OPENAI image tools and attract users from CHATGPT.
The update, called the Flash Gemini 2.5 image, begins with Tuesday for all users in the Gemini application, as well as programmers via API Gemini, Google AI Studio and Vertex AI.
The new AI Gemini image model has been designed to make more precise editing of images – based on natural language demands from users – while maintaining facial consistency, animals and other details, something that most competing tools struggle with. For example, ask chatgpt or grok XAI to change the color of someone's shirt in the photo, and the result may contain a distorted face or a changed background.
The new Google tool has already noticed. In recent weeks, social media users Claimed By an impressive AI image editor on the CrowdSourced, Lmarena rating platform. The model appeared anonymously under the pseudonym “Nano-Banana”.
Google claims that he is behind the model (if not so obvious Already Of all tips related to banana), which is really a native image ability in its flagship model AI Flash 2.5. Google claims that the image model is the most modern in Lmarena and other comparative tests.
“We really push the visual quality forward, as well as the model's ability to follow the instructions,” said Nicole Bichtova, Keeping a product in visual generation models in Google Deepmm, in an interview with TechCrunch.
“This update performs a much better edition of the edition more smoothly, and the model's results are useful for what you want to use them for,” said Bichtova.
AI image models have become a critical battlefield for Big Tech. When OpenAi launched a native GPT-4O image generator in March, he drove the use of chatgPT through the roof thanks to the madness of the studio generated by AI Ghibli memes, which, according to the general director of the OpenAI, Altman herself, left the company's GPS “melt. “
To keep up with OpenAi and Google, Meta announced last week that it would licensed AI painting models from the startup midjourney. Meanwhile, German Black Forest laboratories in Black Lass, supported by A16Z, still dominate in test testing with the AI stream image model.
Perhaps an impressive AI Gemini image editor can help Google closing the gap of users from Opeli. CHATGPT now logs in over 700 million weekly users. During the merger with Google earnings in July, the general director of the Sundar Pichai technological giant revealed that Gemini had 450 million monthly Users – suggesting that weekly users are even lower.
Bichtova claims that Google has specially designed a picture model for cases of consumers, such as helping users in visualization of their home and garden projects. The model also has better “global knowledge” and can combine many references in one prompt; For example, a combination of a sofa image, photos to the living room and color palette into one coherent rendering.
While the new AI Gemini image generator makes it easier for users to create and edit realistic images, the company has security that limits what users can create. Google has been struggling with AI image generator security in the past. At one point, the company apologized for Gemini to generate historically inaccurate photos of people and completely withdrew the AI image generator.
Now Google thinks he has achieved a better balance.
“We want to provide users with creative control so that they can get what they want from models,” said Bichtova. “But it's not as everything goes.”
The generative section AI Google services prohibits users to generate “non -consumer intimate images”. The same types of security do not seem to exist in the case of grocs, which allowed users to create generated AI clear images Reminiscent of celebrities such as Taylor Swift.
To solve the growth of Deepfake images, which can make it difficult for users to recognize what is real online, Bichtova claims that Google uses visual aquatic signs for images generated by AI, as well as identifiers in metadata. However, someone changing the image in social media may not look for such identifiers.
