Making that an open-source CLI tool that other stable-diffusion-web-ui can choose as an alternative backend. I also committed to have everything supported in the app available in repository. That means batch mode (with prompt variations), prompt emphasizing, face restoration, loopback (if one can suffer the extended time), super resolution (possibly high-res fix, but that could be too long (5 to 10 mins) on mobile), image interrogation, hypernetwork + textual inversion (Dreambooth is not possible on device) and more to come! The goal is to have more refined interface and feature-parity with AUTOMATIC1111 when it is possible on mobile (I cannot match its development velocity for sure!). More technical details were discussed in this accompanied blog post: I cross posted on PH: Please upvote there! There is also a thread on HN: It also is fully featured, meaning that comparing to other mobile apps that does this on the server, it supports txt2img, img2img, inpainting and use more models than default SD one. It is fully offline, download about 2G model, and takes about a minute to generate a 512x512 with DPM++ 2M Karras sampler at 30 steps. Hi all, as teased in now the app is available in AppStore, you can check it out in
0 kommentar(er)
