6.5x. That's how much more download activity a new image AI model launch generates compared to a chatbot upgrade, according to new data from Appfigures published in May 2026. The finding flips a common assumption: that large language model (LLM) improvements - the text-based AI that powers tools like ChatGPT - are the primary engine of user acquisition.
They aren't, at least not anymore.
Downloads Are a Visual Problem Now
The pattern makes sense when you think about how people discover apps. Screenshots, app store previews, and social sharing all favor visual output. A new image model that produces stunning portraits or photorealistic scenes shows up in feeds and screenshots in a way that a better text summarizer never will. Someone showing off an image gets shares. Someone showing off a cleaner summary does not.
This is part of why tools like DALL-E 3 and similar image generators have consistently driven outsized spikes in sign-ups relative to their underlying model improvements. The output is inherently demonstrable.
Chatbot upgrades, by contrast, tend to appeal most to existing power users who already know what they're looking for. They improve retention among people who stayed. They rarely pull in new audiences at the same rate.
The Revenue Gap That Makes This Finding Complicated
Here's where the Appfigures data gets more useful: most apps riding an image model launch don't convert that download spike into revenue. The download bump is real. The monetization isn't.
This is the gap that should concern product teams more than the growth numbers. A 6.5x download multiple sounds good until the cohort data shows those users churning within two weeks because the novelty wore off and there was no deeper hook to keep them paying. Image generation is inherently bursty - users arrive for a specific creative need, finish it, and leave. Sustaining revenue requires either a workflow reason to return (editing, brand consistency, bulk production) or a broader product that image generation feeds into.
The companies making money from image AI right now aren't primarily standalone generators. They're tools that embedded image generation inside a broader creative or marketing workflow - where the image is one step in a process, not the entire product. That context drives the kind of repeated usage that justifies a subscription.
The takeaway for anyone building or evaluating AI tools in this category: download metrics from an image model launch tell you about awareness, not product-market fit. The two things feel connected during a launch cycle and diverge almost immediately afterward.