Google DeepMind is launching Gemini 2.5 Flash Image, also called “nano-banana,” for the Gemini app and for developers through the Gemini API, Google AI Studio, and Vertex AI. This update addresses a frequent complaint with AI image tools: small edits that end up changing the entire image. Google describes this release as an improvement in quality and control compared to earlier versions.
The main feature is character consistency. You can keep a person, pet, or product looking the same across different scenes, even as you change outfits, hairstyles, time periods, or settings. The model can also combine several images into one, make specific changes using natural-language instructions, and use Gemini’s world knowledge during image creation and editing.
You can use this tool to put the same character in different settings, show a product from various angles, or keep brand images consistent across campaigns. Multi-turn editing lets you keep making changes, such as adding furniture and decor to a room to try out ideas. You can also mix designs, add a pattern from one image to an object in another, or blend a person and a pet into a new scene.
Pricing is transparent for developers: Gemini 2.5 Flash Image costs $30 per one million output tokens. Each image counts as 1,290 output tokens, which translates to approximately $0.039 per image. Other input and output modalities follow the standard Gemini 2.5 Flash pricing.
For safety, generated images include a visible AI mark and an invisible SynthID digital watermark. Google says SynthID can still be detected after common edits, which should help verify image origins as synthetic media becomes harder to recognize.
Google says early previews rate this as a leading image-editing model. The Gemini app’s built-in editing now keeps subtle details in your photos. You can upload a photo, request changes, combine images with your pet, swap backgrounds to try new wallpaper, or place yourself in different scenes. You can also use the edited image in Gemini to create a short video.
Source(s)
Google (in English)