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 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |