Image model releases are driving growth for AI mobile apps, generating 6.5 times more downloads than traditional model updates, according to a report from Appfigures. This shift signals a change from earlier trends, when new models for conversational experiences and features like voice chat interfaces were the primary drivers of demand.
ChatGPT and Gemini each saw tens of millions of new downloads following the release of their image models, Appfigures reported. Google’s Gemini, with its Nano Banana image model launched in August 2023, achieved over 22 million additional downloads in the 28 days after its launch, increasing the app’s total downloads by more than four times during that period.
ChatGPT’s introduction of the GPT-4o image model in March 2023 resulted in more than 12 million additional installs in the same 28-day timeframe. This figure represents approximately 4.5 times the downloads compared to the previous releases of GPT-4o, GPT-4.5, and GPT-5.
Other releases followed similar patterns but with lesser impacts. Meta AI’s AI video feed, Vibes, released in September 2025, added an estimated 2.6 million downloads within 28 days after its launch. While this model focuses on video, it is fundamentally part of a broader trend towards visual content that attracts user interest.
Despite the increase in downloads from image model releases, Appfigures stated that this does not always correlate with higher mobile revenue. The report emphasizes that while these new models attract users to install the app, they do not guarantee conversion to paid subscriptions.
For example, despite generating more downloads, Gemini’s Nano Banana only led to an estimated $181,000 in gross consumer spending within 28 days of its release. Similarly, Meta AI’s Vibes increased downloads but did not contribute meaningful revenue.
Among these examples, only ChatGPT managed to translate increased user engagement into revenue. The GPT-4o image-generation model generated an estimated $70 million in gross consumer spending in the 28 days following its launch compared to prior performance metrics.
DeepSeek’s R1 also gained 28 million downloads after its January 2025 release, but the surge was driven by innovative techniques rather than an image model release. This instance illustrates how curiosity can fuel downloads, even when the excitement is not directly tied to image capabilities.
