Tiny Tostones
75 years ago a disc jockey in LA, NY or Chicago could latch on to an unknown record and play it over and over as part of the regular rotation. Folks would call the station asking who that was and where could they get it, and within a few days the record was a local hit. As the record industry grew more sophisticated promotion became a faster and more expansive way to gain a following and local hits became regional hits. Today it is not uncommon for promotion to start far in advance of the actual release. Then there is social media, where almost anything can gain cult status if it happens to appeal to the ‘share’ crowd.
While Ai is certainly a topic on social media, AI models have yet to reach social media in that same ‘share’ mode, or have they? As avid users of image generators, we are constantly looking at the improvements made in consumer image generators, and on a practical basis we test those models almost daily. Most are free and easy to use, but few make headlines and rarely do image generators become the topic of conversation on social media. But that has changed recently with Nano-banana, the code name for the most recent consumer oriented image generator imbedded in Google’s Gemini AI.
Nano-banana’s real name is Gemini 2.5 Flash Image, the successor to Gemini 2.0 Flash which was the successor to Gemini Imagen 3 and Imagen 2. We note that Imagen 2 was released at the end of 2023, Imagen 3 in August of last year, Flash 2.0 in February of this year and Flash 2.5 (Nano-banana) in June, so the releases are getting closer together and are constantly being leapfrogged by their own developers. We also note that other than Flash 2.5 (the banana) none of these releases or the models behind them became social media sensations, most likely as when it comes to a discussion about image generation models or what your BFF said to the head cheerleader at the last big game, the models don’t usually win out.
So we flash back to the old days and how a mixture of timing, a good beat, and repetitive promotion could take a song to the top of the charts. All three had to be working for the record to click. So what made nano-banana a hit, and by hit we mean usage far surpassed any of the other Gemini image generator releases in just weeks? It didn’t cost anything, but neither did the previous models. It was accessible as part of Gemini, but so were the others, and it was a natural language model making it easy to use, as were the others. Nano-banana had one feature that changed it from a footnote in the lore of image generation models to a social media hit with a funny name and that was consistency.
Before Nano-banana when one entered a prompt into an image generator, say “Create an image of a man and a woman sitting beside a pool with a large mansion behind them” that’s what you got. But if you realized that the mansion needed some foliage and perhaps some outdoor furniture, when you added that request to the prompt, you got the foliage and the furniture but the man and woman were usually different, sometimes even after you said, “keep everything else the same”. The models lacked consistency from one image to the next and that was a significant challenge for the average user.
Nano-banana has the ability (most of the time) to maintain things like faces across multiple iterations of an image, which allows even an unsophisticated user to place their face on a runway model image or keep the man and woman by the pool consistent while you adapt the background details. When images using the Flash 2.5 model began to appear on social media, they created a sensation that not only made Nano-banana the most popular image generator to date, but also brought new users to Gemini.
Does this mean that a new industry will develop dedicated to the promotion of new AI models? Stranger things have happened, but what it really means is that a unique function and promotion (in this case social media) can create a hit in the same way local disc jockey’s could in the 1950’s and 60’s. So rather than always searching ‘faster than the competition’ or ‘smarter than the competition’, model builder should be searching for ways to remove the simple bottlenecks in applications that make using them a chore or a ‘learning process’. Make it easy to use and it will generate its own media attention with just a bit of gentle social media stimulation, and if it is really good, its name will change from Image generator 3.65b to a flasher “Pixel Pear” or “Binary Blueberry”. It’s a strange world…
While Ai is certainly a topic on social media, AI models have yet to reach social media in that same ‘share’ mode, or have they? As avid users of image generators, we are constantly looking at the improvements made in consumer image generators, and on a practical basis we test those models almost daily. Most are free and easy to use, but few make headlines and rarely do image generators become the topic of conversation on social media. But that has changed recently with Nano-banana, the code name for the most recent consumer oriented image generator imbedded in Google’s Gemini AI.
Nano-banana’s real name is Gemini 2.5 Flash Image, the successor to Gemini 2.0 Flash which was the successor to Gemini Imagen 3 and Imagen 2. We note that Imagen 2 was released at the end of 2023, Imagen 3 in August of last year, Flash 2.0 in February of this year and Flash 2.5 (Nano-banana) in June, so the releases are getting closer together and are constantly being leapfrogged by their own developers. We also note that other than Flash 2.5 (the banana) none of these releases or the models behind them became social media sensations, most likely as when it comes to a discussion about image generation models or what your BFF said to the head cheerleader at the last big game, the models don’t usually win out.
So we flash back to the old days and how a mixture of timing, a good beat, and repetitive promotion could take a song to the top of the charts. All three had to be working for the record to click. So what made nano-banana a hit, and by hit we mean usage far surpassed any of the other Gemini image generator releases in just weeks? It didn’t cost anything, but neither did the previous models. It was accessible as part of Gemini, but so were the others, and it was a natural language model making it easy to use, as were the others. Nano-banana had one feature that changed it from a footnote in the lore of image generation models to a social media hit with a funny name and that was consistency.
Before Nano-banana when one entered a prompt into an image generator, say “Create an image of a man and a woman sitting beside a pool with a large mansion behind them” that’s what you got. But if you realized that the mansion needed some foliage and perhaps some outdoor furniture, when you added that request to the prompt, you got the foliage and the furniture but the man and woman were usually different, sometimes even after you said, “keep everything else the same”. The models lacked consistency from one image to the next and that was a significant challenge for the average user.
Nano-banana has the ability (most of the time) to maintain things like faces across multiple iterations of an image, which allows even an unsophisticated user to place their face on a runway model image or keep the man and woman by the pool consistent while you adapt the background details. When images using the Flash 2.5 model began to appear on social media, they created a sensation that not only made Nano-banana the most popular image generator to date, but also brought new users to Gemini.
Does this mean that a new industry will develop dedicated to the promotion of new AI models? Stranger things have happened, but what it really means is that a unique function and promotion (in this case social media) can create a hit in the same way local disc jockey’s could in the 1950’s and 60’s. So rather than always searching ‘faster than the competition’ or ‘smarter than the competition’, model builder should be searching for ways to remove the simple bottlenecks in applications that make using them a chore or a ‘learning process’. Make it easy to use and it will generate its own media attention with just a bit of gentle social media stimulation, and if it is really good, its name will change from Image generator 3.65b to a flasher “Pixel Pear” or “Binary Blueberry”. It’s a strange world…
Figure 1 - Nano-banana clips from social media - Source: Various
Figure 2 - Pixel Pear and Binary Blueberry - Source: SCMR LLC, Nano-banana
RSS Feed