I ve been working on PhoneDiffusion, a local AI image generator for Apple Silicon devices and wanted to share a demo with people who actually use Stable Diffusion.
The main idea is that you can run image generation directly on iPhone, with no cloud inference. I tested it in airplane mode with Wi-Fi and mobile data disabled so you can see it's 100% offline.
clearspace
This is such a cool launch - running Stable Diffusion fully on-device on an iPhone is so impressive, and privacy-first is so refreshing when everything hits a cloud GPU these days. Love that it actually works in airplane mode.
Curious: how do you handle the model selection for different iPhone hardware? Is it a hard cutoff (e.g. A17 Pro gets SDXL, older chips get SD 1.5), or is there some dynamic profiling that figures out what the device can handle? And are you planning to let power users sideload their own fine-tuned models at some point?
@anteloper Thanks Oliver, yeah the model selection is tiered right now but for first time use we auto download SD 1.5 class model because of good UX for first time users.
Custom/fine-tuned model import is something I’m interested in but I don’t want to ship it until we would figure out a good UX. Raw model sideloading is not really that simple on iOS and we ruthlesslty optimize all the models.
Mailwarm
Very interesting approach!
@othman_katim appreciate it!