Google's neues Bildgenerierungs-König Nano Banana 2 hat in der späten Nacht einen Überraschungsangriff gestartet. Seine Leistung hat die Ranglisten dominiert, die Geschwindigkeit ist sprunghaft angestiegen und der Preis ist halbiert.
During practical tests, a 4K image could be generated within a minute, and the problem with the clocks has finally been solved.
Zhidongxi reported on February 27th. Just now, Google officially released its most powerful image generation and editing model, Nano Banana 2 (Gemini 3.1 Flash Image). This model has been launched in all Google products such as the Google Gemini application, Search, and the AI Studio.
Google announces the release of Nano Banana 2
Nano Banana 2 combines professional - level functions with flash - level speed and has achieved comprehensive improvements in terms of world knowledge, image quality, inference ability, and subject consistency. In benchmark tests, it has significantly outperformed industry - leading models such as GPT - Image 1.5, Seedream 5.0 Lite, and Grok Imagine Image Pro. Combined with the thinking mode and text and image search tools, it even surpasses Nano Banana Pro.
Benchmark results of Nano Banana 2
Zhidongxi immediately tested Nano Banana 2 and found that the truthfulness of the details of the generated images has been improved, and it exceeds expectations when precisely executing instructions. The ability for text rendering and the knowledge of traditional Chinese culture have improved, and it shows a significant improvement in processing complex scenarios.
For example, we asked Nano Banana 2 and Nano Banana Pro to generate images based on the same hint of a "60 - year - old Asian fisherman". The result of Nano Banana 2 was significantly more realistic, more detailed, and had higher accuracy in following the instructions.
Generated by Nano Banana 2
Generated by Nano Banana Pro. Hint: A high - resolution close - up of an approximately 60 - year - old Asian fisherman, with a blurred wave pattern as the background. His face is covered with age spots and freckles, the skin texture is extremely fine, and fine pores and silver beard stubble can be seen. The sunlight falls from the side at a 45 - degree angle and illuminates half of his face. His eyes are slightly closed, and determination and calmness are reflected in his gaze. The light of the distant horizon is reflected on the surface of his eyes. The image has very high sharpness, and the skin texture is so real that you can almost touch it.
Nevertheless, Nano Banana 2 is still constrained by the classic problem of "clock + full wine bottle". When simultaneously dealing with problems such as multiple objects, physical logic, and light and shadow, the generation is still inaccurate (this will be explained in detail in the first part of the following text). Additionally, some images still have an "AI feeling" and cannot be completely faked as real. However, the advantages outweigh the disadvantages. Compared with Nano Banana Pro, Nano Banana 2 has a significant generational leap.
Although Nano Banana 2 generally delivers better results and is faster, the prices have dropped. On the Google AI Studio platform, the price per input image of Nano Banana 2 has dropped from 2 US dollars to 0.5 US dollars compared to Nano Banana Pro, and the price per output image has been halved from 0.134 US dollars to 0.067 US dollars.
The prices of Nano Banana 2 have dropped
Last August, Nano Banana (Gemini 2.5 Flash Image) conquered the world and redefined image generation. In November, Nano Banana Pro, based on Gemini 3 Pro, was almost a "god" in the AI image generation scene thanks to its stronger intelligence and creative control. Today, Google has combined the advantages of both models for the first time to create a new model.
According to Google's design, Nano Banana Pro is suitable for professional tasks that require the highest factual accuracy, while Nano Banana 2 is suitable for fast generation, precise instruction following, and image search integration.
In the Google Gemini application, Nano Banana 2 has replaced Nano Banana Pro in the Fast, Thinking, and Pro versions. Pro and Ultra subscribers can still use Nano Banana Pro as needed.
Nano Banana 2 is launched in Google Gemini
01. Generate 4K images in one minute, more realistic details, "clock generation problem" solved
Zhidongxi immediately tested Nano Banana 2 and found that it performs well in precisely executing instructions, and the problem with Chinese special characters has been significantly improved. The assessment of traditional Chinese culture exceeds expectations.
As shown in the following figure, when Zhidongxi requested an image of a "giant panda writing with a brush", Nano Banana 2 not only accurately generated the "fine hairs of the panda" and the "realistic texture of pearls and wool balls", but also implemented the landscape outside the tea house and the photo style according to the instruction. In text rendering, the font of "Generative AI" is smooth and has almost no spelling mistakes, but the "Gong" in the lower - left part of the character "Shi" is not written quite correctly.
Image generated by Zhidongxi with Nano Banana 2. Hint: A cute giant panda wearing a traditional Chinese opera costume sits in a modern Chinese tea house and writes the Chinese character "Generative AI" on Xuan paper with a brush. The panda's hairs should be fine and detailed, the texture of pearls and wool balls on the costume should be realistic, and the font of "Generative AI" should be smooth and error - free. Outside the tea house, a blurred future city landscape (skyline of Shenzhen) can be seen. The style should be a mixture of hyper - realistic photography and traditional Chinese painting.
Nano Banana 2 is also quite realistic in perspective rendering from special angles. As shown in the following figure, when generating an image on the theme of "shooting a ballet dancer in mid - jump from a upward view", Nano Banana 2 achieved precise control over the facial proportion and perspective. The chin, the shadow of the chin, and the stretched body lines, including the concentrated gaze, are precisely presented, and the figure is not distorted.
Image generated by Zhidongxi with Nano Banana 2. Hint: Shooting a ballet dancer in mid - jump from a upward view. The camera shoots from bottom to top, and the viewer can see the dancer's chin, the shadow of the chin, and the stretched body lines. His arms are outstretched, the dance skirt is fluttering, his facial expression is concentrated and calm, and he is looking into the distance. The stage light falls from above and creates a strong contrast between light and dark on his face. The perspective from the chin to the forehead should be accurate and have no distortion.
When generating images of multiple people and their emotional interactions, Nano Banana 2 accurately reproduced the scene of a groom and a bride looking at each other with tears in their smiles. The contents of the hint, including "the texture of the suit fabric" and "the blurred meadow and the flower arch", are precisely presented. The movements, facial expressions, and gestures of the groom and the bride are quite natural, but the tears flowing from the corners of the groom's eyes are not quite natural in terms of reflection and the shape of the liquid flow, which is a bit "out of the mood".
Hint: The moment of the first look of a newly - married couple on their wedding day. The bride is wearing a white wedding dress and gently covers the groom's eyes from behind. The groom turns around, and the two look at each other and smile, with tears in their eyes. The sunlight shines through the leaves on them, and the lacework of the wedding dress and the texture of the groom's suit fabric are clearly visible. The background is a blurred meadow and a flower arch, and the image radiates an atmosphere of happiness and affection.
The following example further shows the precise instruction - following ability of Nano Banana 2. When I uploaded three images and asked Nano Banana 2 to replace the cars in the parking lot of the left - hand image with the two cars in the right - hand image, it replaced some cars in the image while maintaining the consistency of the original image. The ratio of the replaced cars to the other cars remains the same, and the sunlight shadows under the cars comply with the laws of physics.
Three photos input by Zhidongxi into Nano Banana 2
Image generated by Zhidongxi with Nano Banana 2. Hint: Replace the cars in the parking lot of Image 1 with the cars from Image 2 and Image 3. The image should have the same style and comply with the laws of physics.
Many users in China and abroad have confirmed the generational leap improvement of Nano Banana 2. Nano Banana 2 has shown significant improvements in processing complex scenarios and detail density.
An internet user shared a generated image of a "complex city landscape". As in the