AI graphics are now driving app growth, beating the rise of chatbots

Graphics are driving the growth of AI mobile apps, generating 6.5x more downloads than traditional apps, according to a new report from software intelligence provider Appfigures.

This shows the change in the early days, when the release of new models empowering conversational experiences drove more demand, along with innovations such as a voice chat interface.

For example, ChatGPT and Gemini each added millions of new downloads after releasing their images, App icons found.

For Google’s Gemini, the release of its Nano Banana icon has resulted in an additional 22+ million downloads in the 28 days following its launch. Image of Gemini 2.5 Flash last August. This implementation boosted the app’s downloads by more than 4x during that period, it showed.

Meanwhile, ChatGPT added over 12 million installs within 28 days of launch its GPT-4o image in March of last year. That’s more than 4.5x more downloads than it saw in its GPT-4o, GPT-4.5, and GPT-5 versions, Appfigures showed.

Other model outputs followed suit, albeit to a lesser extent. Meta AI’s launch of the AI ​​video feed Vibes added over 2.6 million downloads in 28 days. September 2025 release. (Yes, technically, this is a video clip, but it’s about visuals, not just text.)

However, the report cautioned that some downloads do not necessarily mean more mobile money.

In fact, the release of new images gives people a reason to install the app and test its image-making capabilities. This does not mean they will convert to paying subscribers. For example, Appfigures reports that Nano Banana only drove $181,000 in consumer spending in the 28-day window after its release, even though it generated more downloads than the 4o version of ChatGPT.

The implementation of Meta AI for Vibes also resulted in some downloads, but to no avail.

Of the three, only ChatGPT turned more interest into real dollars.

OpenAI’s 4o graphics model generated $70 million in revenue in the 28 days after launch, compared to its previous record, Appfigures reported.

The company also looked at DeepSeek in its analysis, but it didn’t fit the plan.

While DeepSeek R1 downloaded 28 million later its January 2025 releaseit was not a model to compare an event. This was DeepSeek’s breakthrough, when it went from obscurity to overnight buzz as tech companies learned about the methods they used to train their AI models at a fraction of the cost of their competitors. This article shows how the attention can be downloaded – although in this case, the attention was not related to the image.