Sistimi ea ho ithuta ka mochini bakeng sa ho kopanya litšoantšo le ho fokotsa lerata lifotong tsa bosiu

Stability AI e phatlalalitse mefuta e lokiselitsoeng e etselitsoeng mochini oa ho ithuta oa Stable Diffusion, o khonang ho kopanya le ho fetola litšoantšo ho latela tlhaloso ea mongolo ka puo ea tlhaho. Mefuta e na le laesense tlas'a laesense e lumelletsoeng ea Creative ML OpenRAIL-M bakeng sa tšebeliso ea khoebo. Ho koetlisa sistimi, ho ile ha sebelisoa sehlopha sa 4000 NVIDIA A100 Ezra-1 GPUs le pokello ea LAION-5B, ho kenyeletsoa litšoantšo tse limilione tse likete tse 5.85 tse nang le litlhaloso tsa mongolo. Nakong e fetileng, khoutu ea lisebelisoa tsa ho koetlisa marang-rang a marang-rang le ho hlahisa litšoantšo e ne e bulehile ka tlas'a laesense ea MIT.

Ho fumaneha ha mofuta o lokiselitsoeng hantle le litlhoko tsa sistimi tse itekanetseng tse lumellang motho ho qala liteko ho PC e nang le li-GPU tse tloaelehileng li lebisitse ho hlaheng ha merero e mengata e amanang:

  • textual-inversion (khoutu) - kenyelletso e u lumellang ho kopanya litšoantšo ka sebopeho se fanoeng, ntho kapa setaele. Phapang ea pele e tsitsitseng, lintho tse litšoantšong tse entsoeng li na le tšohanyetso ebile ha li laolehe. Keketso e reriloeng e u lumella ho eketsa lintho tsa hau tse bonoang, ho li tlama ho mantsoe a bohlokoa le ho li sebelisa ha li kopanya.

    Ka mohlala, ka "Stable Diffusion" e tloaelehileng u ka kopa tsamaiso ho hlahisa setšoantšo ka "katse ka sekepeng". Ho phaella moo, o ka hlakisa litšoaneleho tsa katse le sekepe, empa ha ho tsejoe hore na ke katse efe le sekepe se tla etsoa. Textual-inversion e u lumella ho koetlisa sisteme setšoantšong sa katse kapa seketsoana sa hau le ho kopanya setšoantšo ka katse kapa sekepe se itseng. Ka mokhoa o ts'oanang, e ka boela ea nkela lintho tsa setšoantšo sebaka ka lintho tse itseng, ea beha mohlala oa setaele sa pono bakeng sa ho kopanya, 'me ea qaqisa mehopolo (mohlala, ho tsoa ho lingaka tse fapaneng, u ka sebelisa khetho e nepahetseng le ea boleng bo holimo. ka setaele se batloang).

    Sistimi ea ho ithuta ka mochini bakeng sa ho kopanya litšoantšo le ho fokotsa lerata lifotong tsa bosiu

  • stable-diffusion-animation - popo ea litšoantšo tse phelang (tse tsamaeang) tse ipapisitseng le ho kenella lipakeng tsa litšoantšo tse hlahisitsoeng ho Stable Diffusion.
  • stable_diffusion.openvino (khoutu) - koung ea Stable Diffusion, e sebelisang CPU feela bakeng sa lipalo, e lumellang liteko ho litsamaiso tse se nang li-GPU tse matla. E hloka processor e tšehetsoeng laebraring ea OpenVINO. Ka molao, OpenVINO e fana ka li-plugins bakeng sa li-processor tsa Intel tse nang le lisebelisoa tsa AVX2, AVX-512, AVX512_BF16 le SSE, hammoho le liboto tsa Raspberry Pi 4 Model B, Apple Mac mini le NVIDIA Jetson Nano. Ka mokhoa o sa reroang, hoa khoneha ho sebelisa OpenVINO ho li-processor tsa AMD Ryzen.
  • sdamd ke boema-kepe ba AMD GPUs.
  • Ts'ebetsong ea pele ea synthesis ea video.
  • stable-diffusion-gui, stable-diffusion-ui, Artbreeder Collage, diffuse-the-rest - graphical interfaces bakeng sa ho hlahisa litšoantšo ho sebelisa Stable Diffusion.
  • beta.dreamstudio.ai, Ho Hugging Face Spaces, hlky Stable Diffusion WebUI - marang-rang a marang-rang bakeng sa ho kopanya litšoantšo ho sebelisoa Stable Diffusion.
  • Li-plugins tsa ho kopanya Diffusion e tsitsitseng le GIMP, Figma, Blender le Photoshop.

Ho feta moo, re ka ela hloko phatlalatso ea Google ea khoutu ea sistimi ea ho ithuta ea mochini oa RawNeRF (RAW Neural Radiance Fields), e lumellang, ho ipapisitsoe le data e tsoang litšoantšong tse 'maloa tsa RAW, ho ntlafatsa boleng ba litšoantšo tse lerata haholo tse nkuoang lefifing le ka har'a naha. kganya e mpe. Ntle le ho felisa lerata, lisebelisoa tse ntlafalitsoeng ke morero li etsa hore ho khonehe ho eketsa lintlha, ho felisa ho benya, ho kopanya HDR le ho fetola mabone ka kakaretso lifotong, hammoho le ho nchafatsa boemo ba lintho tse tharo ka ho sebelisa linepe tse 'maloa tse tsoang mahlakoreng a fapaneng. fetola pono, laola tsepamiso le ho hlahisa ditshwantsho tse tsamayang.

Sistimi ea ho ithuta ka mochini bakeng sa ho kopanya litšoantšo le ho fokotsa lerata lifotong tsa bosiu
Sistimi ea ho ithuta ka mochini bakeng sa ho kopanya litšoantšo le ho fokotsa lerata lifotong tsa bosiu


Source: opennet.ru

Eketsa ka tlhaloso