Google Unveils “Nano‑Banana”: A New AI Image Editing Model

Google has launched a new image‑editing model officially called Gemini 2.5 Flash Image, popularly nicknamed “Nano‑Banana.” The model is now live in the Gemini app and is accessible to both paid and free users.

Performance and Ranking

Listed as gemini-2.5-flash-image-preview (Nano‑Banana) , it leads the community leaderboard with a 0.855 score and 1362 points, outperforming other top‑tier image models.

Core Advantage

The standout feature is likeness preservation : facial features remain consistent even after drastic edits such as changing clothing, background, or adding props. Generating a 1024×1024 image via the API costs roughly $0.04 per picture.

Example Workflow (Flux Labs AI Integration)

Using the Flux Labs AI platform, the model is selected as Gemini 2.5 Flash. After uploading a reference photo, the prompt “switch scene to sunset” is entered. The generated result keeps the subject’s face unchanged while transforming the sky and overall tone to a sunset, as shown in the side‑by‑side comparison images.

Virtual Try‑On Demonstration

By uploading a person image together with a clothing image and issuing the prompt “dress female model with reference clothing,” the model produces a realistic composite where the garment fits the model perfectly, demonstrating strong understanding of shape and texture.

Key Functionalities

Subject‑consistent editing: swap outfits, hairstyles, or scenes while preserving facial features.

Multi‑image fusion: combine pets and people into a single cohesive scene.

Step‑by‑step editing: apply changes incrementally (e.g., paint wall → add sofa → place coffee table) with automatic state memory.

Cross‑domain design: extract color palettes from flowers for dresses or transform butterfly wings into shoe designs.

Prompt‑driven iteration: unlimited edits via concise, clear prompts yield the best results.

Text‑to‑Image Capability

Switching to the text‑to‑image tool in Flux Labs AI, selecting Gemini 2.5 Flash, and using a detailed prompt about a black Labrador swimming in a pool produces a high‑quality image in about ten seconds. The current limitation is support for only square aspect ratios and a lack of diverse output formats.

Conclusion

Nano‑Banana blurs the line between image generation and editing, allowing a single model to handle multiple tasks with high quality. This consolidation challenges niche startups that rely on specialized models for virtual try‑on, image fusion, or consistency preservation, as the same capabilities are now available directly within Gemini.