Stability AI ya buga bugu na biyu na tsarin koyo na inji Stable Diffusion, wanda ke da ikon haɗawa da gyaggyarawa hotuna dangane da samfuri da aka ba da shawara ko bayanin rubutun harshe na halitta. An rubuta lambar kayan aikin don horar da cibiyar sadarwar jijiyoyi da tsara hoto a cikin Python ta amfani da tsarin PyTorch kuma an buga su ƙarƙashin lasisin MIT. An riga an buɗe samfuran horarwa a ƙarƙashin lasisin izini na Creative ML OpenRAIL-M, wanda ke ba da damar kasuwanci. Bugu da ƙari, ana samun janareta na hoto akan layi.
Maɓallin haɓakawa a cikin sabon bugu na Stable Diffusion:
- An ƙirƙiri sabon samfuri don haɗin hoto dangane da bayanin rubutu - SD2.0-v - wanda ke goyan bayan tsarar hotuna tare da ƙudurin 768 × 768. An horar da sabon samfurin ta amfani da tarin LAION-5B na hotuna biliyan 5.85 tare da bayanin rubutu. Samfurin yana amfani da saitin sigogi iri ɗaya kamar samfurin Stable Diffusion 1.5, amma ya bambanta ta hanyar canzawa zuwa amfani da maɓalli mai mahimmanci na OpenCLIP-ViT/H, wanda ya ba da damar haɓaka ingancin hotunan da aka samu.
- An shirya sigar tushe mai sauƙaƙan SD2.0, horarwa akan hotuna 256 × 256 ta amfani da ƙirar tsinkayar amo na gargajiya da tallafawa tsarar hoto tare da ƙuduri na 512 × 512.
- Yiwuwar yin amfani da fasahar supersampling (Super Resolution) an ba da ita don haɓaka ƙudurin hoton asali ba tare da rage ingancin ba, ta amfani da algorithms don sikelin sararin samaniya da sake gina cikakkun bayanai. Samfurin sarrafa hoto da aka bayar (SD20-upscaler) yana goyan bayan haɓakar 2048x, wanda zai iya samar da hotuna tare da ƙudurin 2048 × XNUMX.
- An gabatar da samfurin SD2.0-depth2img, wanda ke la'akari da zurfin da tsarin sararin samaniya na abubuwa. Ana amfani da tsarin MiDaS don ƙididdige zurfin ƙima. Samfurin yana ba ku damar haɗa sabbin hotuna ta amfani da wani hoto azaman samfuri, wanda zai iya bambanta da gaske daga asali, amma riƙe gaba ɗaya abun da ke ciki da zurfin. Misali, zaku iya amfani da matsayin mutum a cikin hoto don ƙirƙirar wani hali a cikin wannan matsayi.
- An sabunta samfurin don gyara hotuna - SD 2.0-inpainting, wanda ke ba ku damar maye gurbin da canza sassan hoto ta amfani da saƙon rubutu.
- An inganta samfuran don amfani akan tsarin al'ada tare da GPU ɗaya.
source: budenet.ru