Stable Diffusion 2.0 Image Synthesis System I whakaurua

Kua whakaputahia e Stability AI te putanga tuarua o te punaha ako miihini Stable Diffusion, e kaha ana ki te whakahiato me te whakarereke i nga whakaahua i runga i te tauira i whakaarohia, i te whakaahuatanga tuhinga reo maori ranei. Ko te waehere taputapu mo te whakangungu whatunga neural me te hanga whakaahua ka tuhia ki te Python ma te whakamahi i te anga PyTorch ka whakaputaina i raro i te raihana MIT. Kua tuwhera nga tauira kua whakangungua i raro i te raihana whakaaetanga Creative ML OpenRAIL-M, e taea ai te whakamahi arumoni. I tua atu, kei te waatea he kaihanga whakaahua ipurangi demo.

Nga whakapainga matua i roto i te putanga hou o Stable Diffusion:

  • He tauira hou mo te whakahiato whakaahua i runga i te whakaahuatanga kuputuhi — SD2.0-v — kua hangaia, e tautoko ana i te whakatipu whakaahua me te taumira 768x768. I whakangunguhia te tauira hou ma te whakamahi i te kohinga LAION-5B o te 5.85 piriona whakaahua me nga tuhinga tuhinga. Ko te tauira e whakamahi ana i nga huinga tawhā rite te tauira Stable Diffusion 1.5, engari he rereke i te whakawhiti ki te whakamahi i tetahi momo momo momo rereke OpenCLIP-ViT/H, i taea ai te whakapai ake i te kounga o nga whakaahua ka puta.
    Stable Diffusion 2.0 Image Synthesis System I whakaurua
  • Kua whakaritea he putanga ngawari o te SD2.0-base, i whakangungua i runga i nga whakaahua 256×256 ma te whakamahi i te tauira matapae haruru puāwaitanga me te tautoko i te hanga whakaahua me te taumira o 512×512.
    Stable Diffusion 2.0 Image Synthesis System I whakaurua
  • Ko te kaha ki te whakamahi i te hangarau o te supersampling (Super Resolution) e whakaratohia ana ki te whakanui ake i te whakataunga o te ahua taketake me te kore e whakaiti i te kounga, ma te whakamahi i nga algorithms mo te taatai ​​mokowhiti me te hanga ano o nga korero. Ko te tauira tukatuka whakaahua kua whakaratohia (SD20-upscaler) e tautoko ana i te 2048x upscaling, ka taea te whakaputa whakaahua me te taumira o 2048xXNUMX.
    Stable Diffusion 2.0 Image Synthesis System I whakaurua
  • Ka whakaarohia te tauira SD2.0-depth2img, e whai whakaaro ana ki te hohonutanga me te whakatakotoranga mokowhiti o nga mea. Ka whakamahia te punaha MiDaS mo te whakatau tata hohonutanga. Ma te tauira ka taea e koe te whakakotahi i nga whakaahua hou ma te whakamahi i tetahi atu ahua hei tauira, he rereke te rereke mai i te taketake, engari ka mau tonu te hanganga me te hohonu. Hei tauira, ka taea e koe te whakamahi i te ahua o te tangata i roto i te whakaahua hei hanga i tetahi atu ahua i roto i te ahua kotahi.
    Stable Diffusion 2.0 Image Synthesis System I whakaurua
    Stable Diffusion 2.0 Image Synthesis System I whakaurua
    Stable Diffusion 2.0 Image Synthesis System I whakaurua
  • Ko te tauira mo te whakarereke i nga whakaahua kua whakahoutia - SD 2.0-inpainting, e taea ai e koe te whakakapi me te whakarereke i nga waahanga o te ahua ma te whakamahi i nga kupu akiaki.
    Stable Diffusion 2.0 Image Synthesis System I whakaurua
  • Kua whakatikahia nga tauira mo te whakamahi i runga i nga punaha tikanga me te GPU kotahi.

Stable Diffusion 2.0 Image Synthesis System I whakaurua


Source: opennet.ru

Tāpiri i te kōrero