Stable Diffusion 2.0 Image Synthesis System Qhia

Stability AI tau luam tawm tsab thib ob ntawm Stable Diffusion tshuab kev kawm, uas muaj peev xwm ntawm kev sib txuas thiab hloov kho cov duab raws li cov qauv qhia lossis cov lus piav qhia ntawm cov lus ntuj. Cov cai ntawm cov cuab yeej rau kev cob qhia neural network thiab tsim duab yog sau hauv Python siv PyTorch lub moj khaum thiab luam tawm raws li MIT daim ntawv tso cai. Twb tau kawm ua qauv tau qhib raws li Creative ML OpenRAIL-M daim ntawv tso cai, uas tso cai rau kev lag luam siv. Tsis tas li ntawd, lub tshuab hluav taws xob duab demo online muaj nyob.

Kev txhim kho tseem ceeb hauv tsab tshiab ntawm Stable Diffusion:

  • Tus qauv tshiab rau cov duab synthesis raws li cov lus piav qhia - SD2.0-v - tau tsim, uas txhawb kev tsim cov duab nrog kev daws teeb meem ntawm 768 Γ— 768. Tus qauv tshiab tau raug cob qhia siv LAION-5B sau ntawm 5.85 billion dluab nrog cov lus piav qhia. Tus qauv siv cov txheej txheem tib yam li Stable Diffusion 1.5 qauv, tab sis txawv los ntawm kev hloov mus rau kev siv qhov sib txawv ntawm OpenCLIP-ViT/H encoder, uas ua rau nws muaj peev xwm txhim kho qhov zoo ntawm cov duab tshwm sim.
    Stable Diffusion 2.0 Image Synthesis System Qhia
  • Ib qho yooj yim SD2.0-piv version tau npaj, kawm ntawm 256 Γ— 256 dluab siv cov classical suab kwv yees qauv thiab txhawb cov duab tsim nrog ib tug daws teeb meem ntawm 512 Γ— 512.
    Stable Diffusion 2.0 Image Synthesis System Qhia
  • Qhov muaj peev xwm ntawm kev siv thev naus laus zis ntawm supersampling (Super Resolution) yog muab los ua kom qhov kev daws teeb meem ntawm cov duab qub tsis txo qhov zoo, siv algorithms rau spatial scaling thiab reconstruction ntawm cov ntsiab lus. Tus qauv muab cov duab ua haujlwm (SD20-upscaler) txhawb nqa 2048x upscaling, uas tuaj yeem tsim cov duab nrog kev daws teeb meem ntawm 2048 Γ— XNUMX.
    Stable Diffusion 2.0 Image Synthesis System Qhia
  • Tus qauv SD2.0-depth2img yog npaj, uas yuav siv sij hawm rau hauv tus account qhov tob thiab spatial kev npaj ntawm cov khoom. MiDaS system yog siv rau kev kwv yees qhov tob monocular. Tus qauv tso cai rau koj los tsim cov duab tshiab siv lwm cov duab ua tus qauv, uas tuaj yeem sib txawv ntawm qhov qub, tab sis khaws tag nrho cov ntsiab lus thiab qhov tob. Piv txwv li, koj tuaj yeem siv tus cwj pwm ntawm ib tus neeg hauv ib daim duab los tsim lwm tus cwj pwm hauv tib lub pose.
    Stable Diffusion 2.0 Image Synthesis System Qhia
    Stable Diffusion 2.0 Image Synthesis System Qhia
    Stable Diffusion 2.0 Image Synthesis System Qhia
  • Tus qauv hloov kho cov duab tau hloov kho tshiab - SD 2.0-inpainting, uas tso cai rau koj los hloov thiab hloov qhov chaw ntawm cov duab uas siv cov ntawv qhia.
    Stable Diffusion 2.0 Image Synthesis System Qhia
  • Cov qauv tau ua kom zoo rau kev siv ntawm cov qauv siv nrog ib qho GPU.

Stable Diffusion 2.0 Image Synthesis System Qhia


Tau qhov twg los: opennet.ru

Ntxiv ib saib