Now I need to find a way to run these AI image generating models natively on Android. Any tips?
Discussion
Unfortunately, that won't be possible in the next few years. Model data is several Gb in size, and cpu / gpu computation is intensive.
Please sir, my poor phone battery.
Don't think we are quite there yet. #gpt4all seems to be the best open source llm software that can be run on CPU. Not sure what exists for stable diffusion or other image generation techniques though.
There is a crazy install-python-as-an-app-dependency and run stable diffusion directly on Android but I don't know if it actually works.
Definitely the future though. Probably closer than many think. Will keep an eye out for any other projects and let you know if I see anything that may work with Android. 🤙
So the plan is to have SD in Amethyst but as remote connection to colab or something? I'm trying to understand in a less techie term and more in christian term lol
No, the plan is to run SD locally, entirely in-device. Like we do with translations
That may be a bit much for a phone to handle. But you can probably run tailscale and make calls to your remote PC running SD pretty easily
Yisas 👀 SD is like between 9 to 12GB only the base more recent mobiles has good storage space, So no problem on that. LoRas and models well depends a lot in the user. But how do you plan to have the computational power? Since 8GB Vram is the barely minimum.
Me personally that's why I depend a lot in Paperspace, Colab and the such. I still hope someone make SD for truly potato pc's lol at least that runs on Nvidia 730GT.
What do people use for neuraul network libraries on android? Can pytorch be compiled to be used in Kotlin? How about web assembly?
Check out tinygrad and what George Hotz is doing with openpilot.
Stable Diffusion?
DALLE
Con Bing se pueden generar imágenes con IA desde Android usando la aplicación y poniendo tu cuenta