Nano Banana Allias Gemini 2.5 Flash Image allows you to create and edit images with disconcerting ease, from a simple textual prompt.
After the success of Gemini 2.0 Flash Image in March 2025, a model that the JDN had tested, Google confirms its test with a new more stable and more agile version. Presented on August 26, Gemini 2.5 Flash Image allows you to create high quality images in a wide range of styles. But its most impressive abilities remain the editing and modification of already existing visuals. Gemini Nano Banana stands out as a simple alternative (although somewhat basic) with image editing tools. To understand its abilities, the JDN was able to test the model.
A MOE model, of Deepmind manufacturing
Designed by Deepmind teams, Gemini 2.5 Flash Image is based on MOE architecture (Mixture-OF-Experts). He is able to treat text, image and even audio at the start of the text, like the whole of the Gemini family. Instead of activating all of its vast parameters for each data processed, the MOE allows the model to dynamically router the input tokens to a specific subset of parameters, or “experts.” An approach that allows it to be very efficient while being rather economical in resources, which gives it a fairly exceptional speed.
The model was drawn to a large dataset including texts, images, videos and audio content. The AI ”knows” a wide range of visual styles (until January 2025). When an image is generated with Gemini 2.5 Flash Image, an invisible digital watermark (with Google Synthid technology) is affixed, to allow its traceability.
A relatively low cost
In benchmarks, the model excels in image generation and remains preferred to Imagen 4 Ultra (Google) and GPT IMAGE 1 (OPENAI) and even flow. 1 KONTEXT (MAX), depending on the LMARENA classification. Likewise when it comes to editing images, the LMARENA classifies the model in first position. In general, Nano Banana offers much higher quality standards, especially for the visual coherence of the characters, the generation of infographics, the fine management of objects and the environment or even the ease of control of the stylization of images.
The model is directly available in the Gemini application or from Google AI Studio. It is enough in Google AI Studio to select the Nano Banana model. In the Gemini application (web or mobile application), simply prompt AI with the default model. A feature available for free, for professionals and the general public.
For developers using the API, Google GEMINI 2.5 Flash Image 30 Dollars The million tokens out. Knowing that the generation or publishing of an image generates on average 1,290 tokens, the cost is approximately $ 0.039 per image generated. A relatively low cost if you consider the time saved in particular for the edition of Visuel.
Gemini 2.5 Flash Image: The JDN test
In image editing mode, the model notably makes it possible to merge several images into one, to generate variations in an image by maintaining the coherence of objects or people, or to carry out targeted transformations (blurring, cutting, deletion of an object / of a human, modification of luminosity and colorimetry parameters). As part of this test, we will focus only on the publishing features and not generation which remain classic.
Merge two images into one
This is one of the most stunning cases with Nano Banana. The tool makes it easy to transpose the object of a first image in another. Take for example the image of a ski track and the image of a car. It is then possible to ask Gemini to merge the two images so that the car is directly integrated on the ski track.
The result is visually very correct and would have taken several minutes with Photoshop type editing tool. The most impressive being that the shadow of the car is perfectly consistent.
Delete elements
Simpler but still interesting, the new Google model can isolate and delete a particular element of an image. Ideal for quickly reworking photographs.
Here we asked the AI to delete all the horse – cavalier couples except that of the center with the yellow bib. The result is still very good. The model not only manages to identify the elements to be deleted but also to recreate material to fill their disappearance, while remaining coherent.
Modify image parameters (bokeh, contrast, temperature)
Even more subtle, Gemini Flash Image allows you to modify the exhibition, colorimetry and even depth of field parameters in an image.
First example, let’s take a portrait without bokeh (the background blur). Gemini can virtually reconstruct a shallow field depth to detach the background subject. The functionality is already available on many smartphones but its integration into a generative model is a first.
The result is quite good but still perfectible. The AI cannot quite reproduce the natural bokeh with a shallow depth of field (presence of flare, lack of homogeneity).
Nano Banana can also subtly modify the contrast or temperature parameters of a photograph. Perfect for quickly touching up an image and giving it a particular grain.
The accentuated contrast reinforces the visual presence of the image and gives it a more immersive atmosphere. The same way, it is possible to change the temperature of an image to obtain warmer or colder tones.
We thus obtain an image with a completely different colorimetric atmosphere.
Camera view change
Another creative possibility offered by the Gemini model: change the point of view of the camera to offer a different perspective of a scene. The AI will then imagine and build the new decorative elements as faithfully as possible.
We then discover a whole new scene, certainly but realistic.
Industrialize the photography edition
Gemini 2.5 Flash Image therefore stands out as a small revolution in the editing of images by democratizing photo retouching capacities so far reserved for experts. Thanks to its interface based on simple textual prompts, anyone can now make complex modifications without any technical skills. Beyond this accessibility, the model opens the way to an industrialization of photo retouching with a cost of approximately $ 0.039 per image, making viable the processing of large volumes for companies.
