The team behind the 3D design app Rooms, a group of former Googlers from Things Inc., has introduced a new project. It is a playful AI photo editor for iOS called Mixup. This app allows anyone to create new AI-generated photos using a system of “recipes.” These recipes function like Mad Libs, providing fill-in-the-blank prompts for your own photos, text, or sketches.
For example, you could use Mixup to transform a simple sketch into a beautiful Renaissance painting. You could reimagine your pet in a funny Halloween costume or use a selfie to see yourself with a different hairstyle. The app can also create sillier concepts, like envisioning a friend transformed into the form of Italian brainrot.
The technology is built on Google’s Nano Banana model. The unique recipe format offers a new way to interact with the AI, turning a generative tool into something resembling an online party game. The founder and CEO of Things Inc., Jason Toff, noted that Nano Banana’s key innovation was its ability to use your image and maintain it in a convincing way that was not creepy. Toff’s background includes working on experimental apps at companies like Google and Meta, as well as managing product at Twitter.
A particularly fun aspect of Mixup is that its user-generated AI prompts, the recipes, are shareable. The team identified a common problem with generative AI tools: they present a blank text box and expect the user to be instantly creative. To address this, Mixup lets users see a prompt that worked and simply fill in the blanks.
After creating a new image in Mixup, users can choose to publish it with its recipe to a public feed or just download it for personal use. From the feed, other users can view the photo and tap a button to “Try recipe.” This lets them reuse the same prompt with their own photos, text, or doodles made with the app’s simple drawing feature.
This system of showing the photo alongside its recipe also helps address the unpredictable nature of generative AI. The team referred to this unpredictability internally as a “slot machine” problem, where you push a button and get a random result with little control. In Mixup, users can see both the final photo and the prompt that created it, giving them a better idea of what to expect from their own output. They can also toggle a button to see the before and after image if the creator allows it.
Similar to other AI apps, users can upload their own photos to Mixup for use in AI images. If you do this, any person you follow in the app can then create AI images using your likeness, a feature called “mixables.” The company imagines friends would follow each other to use this feature, but a creator class could also emerge for those who do not mind seeing themselves mashed up in bizarre ways. If you prefer to keep your image private, you simply do not upload it or do not follow anyone.
The app uses OpenAI technology to handle some moderation concerns around AI imagery. The CEO also acknowledges that Mixup relies heavily on Google’s built-in controls within its image model to restrict content like sexual material or violence.
At its launch, Mixup is optimized for iOS 26 but is supported on iOS 18 and up. If the app becomes popular, a web version or Android app may be developed later. Free users receive 100 credits, which is equivalent to four dollars, as each image costs nearly four cents to produce. Once the initial credits are used, users can subscribe to different tiers that offer 100, 250, or 500 credits per month.
The app launches globally on the App Store at midnight on November 20. It will initially require an invite to access. Mixup is available for pre-order ahead of its official launch.