- I-SynthID Detector iyithuluzi lamahhala elivela ku-Google elihlonza ama-watermark edijithali angabonakali kumafayela e-multimedia.
- Iyahambisana nokuqukethwe okudalwe yi-Google kanye nabalingani bezobuchwepheshe njenge-OpenAI, i-NVIDIA kanye ne-Kakao, kanti i-Apple ihlelelwe ikusasa.
- Ukutholakala kwayo kuyamelana nokuhlelwa okuyisisekelo, yize kungesona isimiso esijwayelekile ngoba iqaphela kuphela imikhiqizo ehambisanayo.
Namuhla, ukuhlukanisa phakathi kwalokho okungokoqobo nalokho okukhiqizwe umshini kuyindaba ebuhlungu ngempela. Ngokusetshenziswa okukhulu kobuhlakani bokwenziwa obukhiqizayo, i-inthanethi isigcwele okuqukethwe okubonakala kuyiqiniso kodwa okuwukusungula kwe-algorithm okumsulwa, okusiphoqa ukuthi sifune amathuluzi okusisiza sigweme ukugwinya yonke inkohliso.
Ukuze kuxazululwe lokhu, i-Google inqume ukuvula i-SynthID Detector yayo emphakathini jikelele , ipulatifomu yewebhu eyenzelwe ukuthola ama-watermark angabonakali edijithali kumafayela e-multimedia. Lokho okwakuyisizinda sezintatheli kanye nochwepheshe bokuhlola amaqiniso manje sekuyatholakala kunoma ubani ofuna ukwazi ukuthi leyo vidiyo esakazekile noma isithombe esithakazelisayo sihlungiwe yi-AI.
Iyini ngempela i-SynthID futhi isebenza kanjani?
Ngokungafani ne-metadata ejwayelekile, engasuswa yinoma yimuphi umsebenzisi onolwazi oluncane lobuchwepheshe, i-SynthID isebenzisa ubuchwepheshe obuthuthukiswe yi -Google DeepMind obufaka amasignali angabonakali ngqo esakhiweni sedatha yefayela. Lokhu kusho ukuthi uphawu lwe-watermark alushintshi ikhwalithi yokubuka noma yomsindo yokuqukethwe, kodwa luhlala yisiginesha yedijithali.
Uhlelo luqinile impela, njengoba lezi zimpawu zivame ukumelana nokuguqulwa okuvamile njengokunqunywa kwesithombe, ukusetshenziswa kwezihlungi ezithile, noma ukucindezelwa kwamafayela lapho kwabelwana ngazo ezinkundleni zokuxhumana. Kodwa-ke, kubalulekile ukuqonda ukuthi uma ifayela lidlula ezinguqukweni ezinkulu noma licutshungulwa ngokuphindaphindiwe yi-AI ehlukahlukene, uphawu lwe-watermark lungalimaza futhi lungabonakali.
Izinyathelo zokusebenzisa isitholi sokuqukethwe okwenziwe
Ukusebenzisa ithuluzi kulula futhi akudingi ukufakwa kwesofthiwe okuyinkimbinkimbi, njengoba kusebenza ngqo kusuka kusiphequluli sakho ku-synthid.com . Inqubo ilula kakhulu: uma isifikile, umsebenzisi umane angene futhi alayishe ifayela afuna ukulihlaziya, kungaba ngokulihudula aliyise ebhokisini elihambisanayo noma ngokulikhetha kudivayisi yakhe.
Ukuze ufinyelele isevisi, kudingeka i-akhawunti ye-Google, i-Apple, noma i-ChatGPT . Uma izinto sezilayishiwe, uhlelo luskena ifayela ukuze luthole isiginesha yedijithali bese lubuyisela umphumela obonisa ukuthi isiginesha sitholiwe noma cha. Kubalulekile ukuqaphela ukuthi, okwamanje, isikhombimsebenzisi sepulatifomu sisebenza kakhulu ngesiNgisi.
Amafomethi asekelwayo kanye nobubanzi bethuluzi
Ukuguquguquka kwe-SynthID Detector kuyamangalisa, njengoba ingagcini nje ngezithombe kuphela. Emnyangweni wamavidiyo , ingacubungula amafayela ngamafomethi e-MP4, MOV, kanye ne-WEBM. Uma kukhulunywa ngomsindo , uhlu lubanzi kakhulu, lwamukela i-WAV, MP3, OGG, FLAC, AAC, kanye ne-M4A. Ngokuphathelene nezithombe , isekela cishe yonke into: kusukela ku-JPG yakudala kanye ne-PNG kuya kumafomethi esimanje njenge-WebP, AVIF, HEIC, HEIF, TIFF, kanye ne-GIF.
Iphuzu elibalulekile ukuthi i-Google ayifuni ukwenza yodwa kulo mzamo futhi ihlose ukudala indinganiso yemboni. Ngakho-ke, leli thuluzi aligcini nje ngokubona okuqukethwe kwe-Gemini kodwa futhi liqaphela nemikhiqizo evela ku -OpenAI, NVIDIA, kanye ne-Kakao . Ngaphezu kwalokho, sekuqinisekisiwe ukuthi i-Apple izojoyina le ndlela yokuqinisekisa maduze nje, okuzokwandisa kakhulu ukusebenza kwe-detector.
Ukulinganiselwa okubalulekile okufanele ukwazi ngakho
Yilapho kudingeka khona ukuqapha ukuze kugwenywe ukushesha kufike eziphethweni. I-SynthID Detector ayiyona i -AI detector esetshenziswa yiwo wonke umuntu. Lokhu kusho ukuthi uma okuqukethwe kudalwe nge-AI engasebenzisi lolu hlelo lokumaka ama-watermark, ithuluzi ngeke libike i-watermark, kodwa lokho akusho ukuthi ifayela liyiqiniso.
Ngamafuphi, umphumela ongemuhle awusibo ubufakazi bobuqiniso, kodwa umane nje ukungabikho kwesiginesha ehambisanayo . Ngaphezu kwalokho, uhlelo aluchazi ukuthi yiziphi izinguquko ezithile ezenziwe kufayela; luqinisekisa kuphela ukuthi ukungenelela kwe-AI kwenzeke yini. Endabeni yamavidiyo, i-Gemini ingakhomba izingxenye ezithile lapho isiginesha sitholakale khona, okucacisa ukuthi yonke i-clip ayiyona ngempela into yokwenziwa.
Amanye amasu okuthola ama-deepfake nokuqukethwe okungamanga
Njengoba kungekho thuluzi elingenaphutha, kuyalulekwa ukuhlanganisa ukusetshenziswa kwe-SynthID neso elibucayi. Enye indlela ephumelelayo ukubheka izinto ezibonakalayo , njengombhalo ongafundeki, iminwe ekhubazekile, izithunzi ezingahambisani nokukhanya, noma amaphethini abonakala ephindaphinda kakhulu futhi enziwe ngokwenziwa.
Kuwusizo kakhulu futhi ukusebenzisa usesho lwezithombe olubuyela emuva lwe-Google ukulandelela umsuka wangempela wesithombe. Okokugcina, ukuhlaziya idatha ye-EXIF yefayela lokuqala kungaveza imodeli yekhamera esetshenzisiwe noma isofthiwe yokuhlela esetshenzisiwe, kunikeze izinkomba ezengeziwe mayelana nokuba semthethweni kwezinto.
I-Google ihlanganise lobu buchwepheshe ngezinga elikhulu, kakade ihlanganisa izithombe namavidiyo angaphezu kwezigidigidi eziyi-180.000 , kanye nenani lomsindo elilingana neminyaka engu-240.000. Ngokuhlanganiswa kwalezi zici ku- Chrome naku-Google Search , ukulwa nolwazi olungelulo kanye nama-deepfake kuthola umngane onamandla ocubungula izicelo ezingaphezu kwesigidi nsuku zonke, okusiza ekubuyiseleni izinga elithile lokuphepha endaweni yedijithali.