Follow ZDNET: Add us as a preferred source
Gemini 2.5 Flash Image also specializes in making minor edits to images based on natural language instructions (“Remove that marinara stain from my shirt, please”) and fusing multiple images together, among other technical abilities.
Want more stories about AI? Sign up for AI Leaderboard, our weekly newsletter.
Before its official debut — and before Google had even publicly claimed credit for it — the model went by the codename “nano-banana” and quickly became a superstar preview model on LMArena. On the same day that it was publicly introduced in late August, Adobe announced that it could be accessed through its Firefly and Express tools.
Watermarking and deepfakes
Many developers are investing heavily in image-generating AI tools, aiming to sell them to creators and businesses as cost-efficient alternatives to lengthy and expensive in-person photoshoots. Just this week, OpenAI released Sora 2, the latest model behind its Sora AI video generator, which showed massive improvement in photorealism.
Also: Meta gives advertisers new AI personalization tools – while using your chats to target content
As a result, technology has advanced rapidly, producing a flurry of tools that can generate photorealistic images in seconds — as well as a torrent of deepfakes. Without comprehensive federal regulation, tech companies have had to take responsibility for building transparency measures into their image-generating tools to make sure their users know when they’re seeing something generated by a machine — or not.
Any images that have been created or edited using Gemini 2.5 Flash Image include an invisible Synth ID watermark, according to Google, which means they can be identified as AI-generated by another model specifically trained for that purpose — but probably not by the human eye.
