In short
Google upgraded Gemini with 2.5 Flash Picture, taking goal at OpenAI’s dominance.
Builders can remix template apps in AI Studio and deploy customized picture initiatives immediately.
Google has expanded entry via OpenRouter and fal.ai, widening distribution to coders worldwide.
Google launched Gemini 2.5 Flash Picture on Tuesday, delivering a brand new AI mannequin that generates and edits photographs with extra precision and character consistency than earlier instruments—trying to shut the hole with OpenAI’s ChatGPT.
The tech large’s push to combine superior picture enhancing into Gemini displays a broader push amongst AI platforms to incorporate picture technology as essential characteristic. The brand new software, now obtainable throughout Gemini apps and platforms, lets customers edit visuals utilizing pure language—dealing with advanced duties like pose adjustments or multi-image fusion with out distorting faces or scenes.
In a weblog submit, Google stated the mannequin permits customers to “place the identical character into completely different environments, [and] showcase a single product from a number of angles… all whereas preserving the topic.”
The mannequin first appeared below the pseudonym “nano-banana” on crowdsourced testing web site LMArena, the place it drew consideration for its seamless enhancing. Google confirmed Tuesday it was behind the software.
Google stated the system can fuse a number of photographs, preserve character consistency for storytelling or branding, and combine “world data” to interpret diagrams or mix reference supplies—all inside a single immediate.
The mannequin prices $30 per million output tokens—about 4 cents per picture—on Google Cloud. It’s additionally being distributed through OpenRouter and fal.ai.
OpenAI launched the GPT-4o mannequin in Could 2024 and added picture technology in March 2025, which helped push ChatGPT’s utilization above 700 million weekly lively customers. Google reported 400 million month-to-month lively Gemini customers in August 2025, which might point out weekly utilization that significantly trails OpenAI.
Google stated all outputs will embrace an invisible SynthID watermark and metadata tag to mark them as AI-generated to handle considerations round misuse and authenticity.
Typically Clever E-newsletter
A weekly AI journey narrated by Gen, a generative AI mannequin.
