Image Model Updates Drive 6.5x More AI App Downloads

A better answer is useful. A better reasoning model is powerful. But on mobile, a picture may be worth a thousand downloads. OK, many thousands, if not millions, because image and video tools are much easier to understand, try, and share.

That may sound reductive, but the data keeps pointing in that direction. Model releases can move the needle, especially when they become a media event, but visual tools are becoming the most reliable way to make consumer AI apps feel new, useful, and worth trying.

According to Appfigures Intelligence, ChatGPT added an estimated 12M incremental downloads in the 28 days after OpenAI introduced 4o image generation in March of 2025. Incremental downloads here means downloads above the app's pre-launch baseline, not total downloads. Google Gemini did even more with Nano Banana, adding an estimated 22M incremental downloads in the 28 days after Google introduced Gemini 2.5 Flash Image in August.

Across the launch events I measured, image and video model updates drove 6.5x more incremental downloads than traditional model releases, excluding DeepSeek's awareness-driven spike. ChatGPT's image update added roughly 4.5x more incremental downloads than its average GPT-4o, GPT-4.5, and GPT-5 model releases, while Gemini's Nano Banana launch was in a completely different league from Gemini's other model updates.

That's a lot of curiosity turning into installs.

DeepSeek Is the Outlier

There is one very obvious exception: DeepSeek.

DeepSeek R1 generated the biggest spike in this analysis, with an estimated 28M incremental downloads in the 28 days after its January 2025 release, according to our estimates. But I wouldn't treat that as a normal model-launch comp.

DeepSeek R1 was a global awareness event. The app went from relatively unknown to the center of the AI conversation almost overnight, helped by a wave of coverage about whether DeepSeek could challenge OpenAI, Anthropic, and the rest of the AI leaderboard.

So DeepSeek belongs in the analysis, but as the ceiling for curiosity-driven downloads, not as the benchmark for how a typical model release performs.

Image Models Have the Cleaner Pattern

Once DeepSeek is separated from the pack, the visual launches stand out.

ChatGPT's 4o image generation release added more incremental downloads than GPT-4o, GPT-4.5, or GPT-5 did in their comparable windows. Gemini's Nano Banana launch was even more dramatic, lifting the app's post-launch download baseline by more than 4x in the following 28 days.

Meta AI showed the same kind of shape, just on a smaller base. After Meta introduced Vibes, its AI video feed, the Meta AI app added an estimated 2.6M incremental downloads in 28 days. That's not ChatGPT or Gemini scale, but it supports the same idea: visual AI gives regular people a reason to try the app.

A better chatbot answer is hard to see in a screenshot. A generated image or video is not.

Downloads and Money Are Splitting

But here's the thing: visual attention and mobile revenue are not the same thing.

ChatGPT is the cleanest proof. Appfigures Intelligence shows 4o image generation added roughly $70M in estimated gross consumer spending over the 28 days after launch compared with its prior baseline. That's the biggest spending lift of any event in this analysis.

Gemini's Nano Banana went the other way. It produced the bigger download spike, but only about $181K in estimated gross consumer spending lift over the same kind of 28-day window. That's not a typo. Gemini found a mainstream consumer hook, but it didn't turn that hook into app-store spending the way ChatGPT did.

Meta AI is an even cleaner version of the same split. Vibes moved downloads, but the app generated no meaningful mobile revenue in this dataset.

ChatGPT Still Converts Better

This is where the story gets more important than the spike.

Visual AI looks like the best way to create mobile demand, 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.

That's not a small distinction. Consumer AI apps are all fighting for the same thing now: a reason to be opened directly instead of accessed through a search box, browser tab, operating system, or another app. Image generation gives them that reason because the output is obvious and shareable.

And once users arrive, monetization becomes the real test.

For now, the data says visual AI is the better mobile hook. The next question is whether anyone other than ChatGPT can turn that hook into a business.

App Intelligence for Everyone!

The insights in this report come right out of our App Intelligence platform, which offers access to download and revenue estimates, installed SDKs, and more! Learn more about the tools or schedule a demo with our team to get started.

Are you a Journalist? You can get access to our app and market intelligence for free through the Appfigures for Journalists program. Contact us for more details.

All figures included in this report are estimated. Unless specified otherwise, estimated revenue is always net, meaning it's the amount the developer earned after Apple and Google took their fee.