Close-up of a smartphone screen displaying app icons for DeepSeek, ChatGPT, Gemini, Copilot, Claude, and an AI-related app. The icons are clear with distinct logos and names visible.

The still image is incredibly powerful; photographers know this better than anyone. And even in AI, it is the picture side of the product that is turbo-charging their gigantic businesses, according to a new report.

Per Android Authority, the fastest growth for apps like ChatGPT and Google Gemini comes when there is an AI image trend, like the Studio Ghibli one last year.

App intelligence provider Appfigures provided the data, essentially finding that image model releases correlate with large increases in downloads far more than upgrades to the chatbots.

Google Gemini has benefited from Nano Banana, which gained the AI app an extra 22 million downloads when the viral AI image generator was released. That was four times the usual rate of Gemini downloads.

It was a similar story for when OpenAI announced ChatGPT-4o, which was a big step forward in terms of the realism that AI imagery can achieve. The excitement surrounding the image generator resulted in approximately 4.5 times the normal number of downloads, far more than when the text generator was updated.

Bar chart titled "The Biggest Install Moments" shows estimated worldwide downloads above baseline for various AI model launches. Plugins and Poe/Claude have the highest downloads; image and chat models are shown in orange and gray bars, respectively.

But there is a major difference between OpenAI and Google’s viral AI image generator movement: one was able to monetize it, and one didn’t as much. TechCrunch notes that ChatGPT generated $70 million from customers after it launched ChatGPT-4o’s images. In contrast, Google’s Nano Banana only grossed $181,000 — despite getting more downloads than ChatGPT.

The outlier is the Chinese AI firm DeepSeek, which blew up at the beginning of 2025. The release of DeepSeek R1 and its AI image model, Janus-Pro-7B, drove 28 million downloads. But the difference between DeepSeek and the others is that it was a relatively unknown app and exploded in popularity seemingly overnight. Part of the intrigue was that the app had been created at a fraction of the cost of traditional models. It prompted OpenAI to say that DeepSeek “may have inappropriately distilled our models.” This was ironic given that OpenAI built its entire company using copyrighted content it never asked permission to use.

Bar chart showing consumer spending above baseline for various AI models. Most models show low spending, except "ChatGPT 4o Images," which has a much higher orange bar labeled "Image Model.

Even Meta got a bump after it released its AI video feed Vibes, adding roughly 2.6 million downloads.

“Visual AI looks like the best way to create mobile demand,” writes Appfigures. “But ChatGPT is still the app converting that demand into subscription spending. Gemini can make people curious. Meta can make them browse. DeepSeek can turn global attention into a download rush. But ChatGPT is the one turning a visual feature into money.

Image credits: Header photo licensed via Depositphotos.