Google DeepMind has launched the Nano Banana model for AI-powered image editing within the Gemini app, bringing significant improvements in maintaining visual identity and achieving high-accuracy edits. The model enables a wide range of changes such as outfit swaps and image blending, and has outperformed competing models in recent evaluations. Nano Banana is now available to users and developers, with visible and invisible watermarks to ensure authenticity.
اضافة اعلان
This new model marks a major step forward in Google’s AI-generated image capabilities, following earlier image-editing tools introduced in Gemini at the beginning of the year.
The model focuses on preserving a subject's visual identity even during substantial alterations, such as changing clothing styles, haircuts, or adding pets into new scenes. It addresses the common issue of "imperfect likeness," where the realism of portraits breaks down due to inconsistent rendering of a person across multiple edits.
Key Features:
Users can now change locations and outfits while preserving facial details.
Multiple images can be merged into one scene.
Edits can be performed in stages (e.g., repainting a wall, adding furniture).
Design elements from one image can be applied to another (e.g., turning flower petal colors into shoe designs).
Performance Evaluation:
According to LMArena, the Gemini 2.5 Flash Image model powered by Nano Banana received the highest overall preference score compared to models like ChatGPT-4o, FLUX.1 Kontext, and Qwen Image Edit.
While Gemini 2.0 Flash Image achieved higher raw speed (pixel output per second), Gemini 2.5 balanced quality and user satisfaction more effectively.
In category-based testing (e.g., creativity, environmental design, product context), Gemini 2.5 also ranked at the top, with ChatGPT-4o performing well in visual layout and Qwen Image Edit trailing in multiple areas.
Availability:
Nano Banana is now live in the Gemini app for free and premium users.
It is also rolling out to developers via Gemini API, Google AI Studio, and Vertex AI.
All images generated or edited with Gemini will include a visible watermark and an embedded digital tag (SynthID) to indicate they were AI-generated.
This launch follows Google's earlier announcement of Imagen 4 at I/O 2025, reflecting the company’s growing focus on high-fidelity AI image generation.
