Do you have a consistent spread in your pocket? “Draw Things” Brings AI Images to the iPhone
Liu Liu, a San Francisco-based developer, released Draw Things: AI Generation, a free App Store app that allows iPhone users to run the popular static diffusion AI image generator. Enter a description, and the app will generate an image within a few minutes. This is a significant step toward making image synthesis more accessible to a wider audience, with the added privacy of running it on your hardware.
Stable Diffusion (SD) is an AI image generator model that generates novel images from text descriptions. People typically run the SD through a commercial DreamStudio service, on a remote cloud machine with a rent calculation time, or locally on a PC through a custom open-source implementation. SD requires a fairly beefy GPU to generate images quickly when running locally, but some developers adapted the model to run on older GPUs with less VRAM. Similarly, Liu Liu has managed to adapt static props to run on the iPhone, a difficult process described in a blog post by the developer.
The main challenge is running apps on iPhone devices with 6GB RAM, writes Liu Liu. While 6GB sounds great, if you use more than 2.8GB on a 6GB device and more than 2GB on a 4GB device, iOS will begin destroying your apps. When you launch Draw Things for the first time, it downloads several necessary files to your iPhone, including the stable Diffusion 1.4 model. To use it, type a prompt at the top of the screen and then press the “Generate” button. Tap the number in the top center of the screen to randomize the seed, which is a number that partially guides the image’s generation. Enlarge a screenshot of the “Draw Things” app’s Generation (left) and Settings (right) screens on the iPhone. A 384384 image took slightly more than two minutes to create on our iPhone 11 Pro. It’s quick on the iPhone 14 Pro, according to Liu Liu, producing an image in about a minute. SD is computationally demanding in either case. Our iPhone became noticeably hot to the touch after each generation.
It should be noted that with constant diffusion, 384384 images frequently produced relatively poor, low-detail results because the SD creators trained the model using 512512 images. When we attempted to generate a 512512 image on our iPhone 11 Pro, we received a warning but continued anyway, but the app crashed to a black screen. Draw Things also supports InPainting, which allows you to replace a portion of an image with AI-generated imagery and load additional image synthesis models, such as the unofficial “Modern Disney Diffusion” model (which generates Disney-looking characters) and anime-powered “Waifu Diffusion” models (we tested, and the app can generate NSFW content, so be warned). Because of the inclusion of those features, “DrawThings” may not remain in the App Store for as long if it becomes widespread, as the content it generates may violate Apple’s terms of service.
In any case, Draw Things appears to be an important proof-of-concept, demonstrating that static propagation can be run locally on the iPhone, albeit slowly. If the app remains, perhaps with some filters or modifications, Liu Liu describes potential future optimizations that could speed things up: “I still have 50% of my performance left.” This isn’t the first time the iPhone has experienced sound diffusion. Matt Waller, the developer, ran Stable spread locally on their iPhone XS in September, but they did not provide an App Store app that would allow others to replicate the feat. As device computational power increases, we will see more local AI image synthesis on smartphones.