{"id":62275,"date":"2022-06-01T17:10:16","date_gmt":"2022-06-01T08:10:16","guid":{"rendered":"https:\/\/smilegate.ai\/?p=62275"},"modified":"2022-06-07T05:51:42","modified_gmt":"2022-06-06T20:51:42","slug":"text-to-image-multimodal-ai","status":"publish","type":"post","link":"https:\/\/smilegate.ai\/cn\/2022\/06\/01\/text-to-image-multimodal-ai\/","title":{"rendered":"\uc0c1\uc0c1\uc744 \ud604\uc2e4\ub85c \uadf8\ub9ac\ub294 AI"},"content":{"rendered":"

\uc378\ub124\uc77c \ucd9c\ucc98: Google Imagen<\/a><\/strong>\ub97c \ud1b5\ud574 \uc0dd\uc131\ub428(text: \uc0ac\ud558\ub77c \uc0ac\ub9c9\uc5d0\uc11c \ubc00\uc9da\ubaa8\uc790\uc640 \ub124\uc628 \uc120\uae00\ub77c\uc2a4\ub97c \uc4f4 \uc791\uc740 \uc120\uc778\uc7a5)<\/p>\n\n\n\n


[\uac00\uc0c1\uc778\uac04\uc5f0\uad6c\ud300 \uc2ec\ud64d\ub9e4]<\/p>\n\n\n\n

\ud14d\uc2a4\ud2b8, \uc774\ubbf8\uc9c0, \uc74c\uc131 \ub370\uc774\ud130 \ub4f1 \ub2e4\uc591\ud55c \ud615\ud0dc(modality)\ub85c \uc774\ub8e8\uc5b4\uc9c4 \ub370\uc774\ud130 \ud2b9\uc9d5\uc744 \ud6a8\uacfc\uc801\uc73c\ub85c \ud559\uc2b5\ud558\uae30 \uc704\ud55c \ubc29\ubc95\uc744 <\/strong>Multimodal learning\uc774\ub77c\uace0 \ud558\ub294\ub370\uc694,  \ucd5c\uadfc \uad6d\ub0b4\uc678 \ud559\uacc4 \ubc0f \uc5c5\uacc4\uc5d0\uc11c\ub294 \uba40\ud2f0\ubaa8\ub2ec \uc778\uacf5\uc9c0\ub2a5(AI)  \ubaa8\ub378 \uad00\ub828 \uc5f0\uad6c\ub4e4\uc744 \ud65c\ubc1c\ud788 \uc9c4\ud589\ud558\uace0 \uc788\uc73c\uba70 \uc774\ubbf8 \uc5c5\uacc4\ub4e4\uc5d0\uc11c\ub294 \uc6b0\uc218\ud55c \uc131\ub2a5\uc758 \uc11c\ube44\uc2a4\ub4e4\ub3c4 \ucd9c\uc2dc\ud558\uace0 \uc788\uc2b5\ub2c8\ub2e4. \uc624\ub298 \uadf8\uc911 \uba87\uac1c\uc758 \ub300\ud45c\uc801\uc778 Multimodal \uc5f0\uad6c\uc0ac\ub840\uc5d0 \ub300\ud558\uc5ec \uac04\ub2e8\ud788 \uc18c\uac1c\ud574 \ub4dc\ub9ac\ub3c4\ub85d \ud558\uaca0\uc2b5\ub2c8\ub2e4. <\/p>\n\n\n\n

[Nvidia GauGAN]<\/strong><\/p>\n\n\n\n

2019\ub144\uc5d0 Nvidia\ub294 GAN(generative adversarial networks)\uc744 \uae30\ubc18\uc73c\ub85c \ub300\ub7b5\uc801\uc778 \uc2a4\ucf00\uce58\ub97c \uc0ac\uc2e4\uc801\uc778 \uc774\ubbf8\uc9c0\ub85c \ubc14\uafd4\uc8fc\ub294 \uc2e4\uc2dc\uac04 AI \ud398\uc778\ud305 \uc560\ud50c\ub9ac\ucf00\uc774\uc158\uc778 GauGAN <\/a>\uc744 \ubc1c\ud45c\ud558\uc600\uc2b5\ub2c8\ub2e4.  2\ub144 \ud6c4 NVIDIA\ub294 GauGAN\uc758 \ud6c4\uc18d \uc81c\ud488\uc778 GauGAN2<\/a>\ub97c \uacf5\uc2dd \ucd9c\uc2dc\ud588\uc2b5\ub2c8\ub2e4. GauGAN2\ub294 \uadf8\ub9bc\ub9c8\uc800 \ud544\uc694 \uc5c6\uace0 \uac04\ub2e8\ud55c \ud14d\uc2a4\ud2b8\ub97c \uc785\ub825\ud558\uba74  \uc774\uc5d0 \ubd80\ud569\ub418\ub294 \uc774\ubbf8\uc9c0\ub97c \uc54c\uc544\uc11c \ub9cc\ub4e4\uc5b4 \uc90d\ub2c8\ub2e4. <\/p>\n\n\n\n

\uacf5\uc2dd \ube14\ub85c\uadf8\uc5d0 \uae30\uc7ac\ub418\uc5b4 \uc788\ub294 \uc608\uc2dc\uc640 \uac19\uc774 \u201cocean waves hitting rocks on the beach\u201d\ub97c \uc785\ub825\ud558\uba74 \ubaa8\ub378\uc740 \uc810\uc9c4\uc801 \uc870\uc815\uc744 \ud1b5\ud574 \uc785\ub825 \ub41c \ud14d\uc2a4\ud2b8\uc640 \uc77c\uce58\ud558\ub294 \uc774\ubbf8\uc9c0\ub97c \uc0dd\uc131 \ud560 \uc218 \uc788\uac8c \ub429\ub2c8\ub2e4<\/p>\n\n\n\n

\"\"
[\uadf8\ub9bc 1] “ocean waves hitting rocks on the beach” \uad00\ub828 \uc0dd\uc131 \ub41c \uc774\ubbf8\uc9c0 (\ucd9c\ucc98: Nvidia GauGAN2<\/a>)<\/figcaption><\/figure><\/div>\n\n\n\n

GauGAN2\ub294 \ub2e4\uc591\ud55c \ud14d\uc2a4\ud2b8 \ud504\ub86c\ud504\ud2b8\uc640 \uc2a4\ucf00\uce58\ub97c \ud1b5\ud574 \uc0ac\uc6a9\uc790\uc5d0\uac8c \ubcf4\ub2e4 \uc138\ubc00\ud55c \ucee8\ud2b8\ub864\uacfc \ube60\ub978 \uc7a5\uba74 \ucee4\uc2a4\ud130\ub9c8\uc774\uc9d5\uc744 \uc81c\uacf5\ud574\uc8fc\uace0 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n\n\n\n

\"\"
[\uadf8\ub9bc 2] \uc0ac\uc6a9\uc790\uac00 \uac04\ub2e8\ud55c \ud14d\uc2a4\ud2b8\uc640 \uc2a4\ucf00\uce58\ub85c \uc0ac\uc2e4\uc801\uc778 \uc774\ubbf8\uc9c0\ub97c \uc0dd\uc131 \ubc0f \ubcf4\uc644 \ud568 (\ucd9c\ucc98: Nvidia GauGAN2<\/a>)<\/figcaption><\/figure><\/div>\n\n\n\n

\uc0ac\uc6a9\uc790\ub294 \uc0c1\uc0c1\ud55c \uc7a5\uba74\uc758 \ubaa8\ub4e0 \uc694\uc18c\ub97c \uc9c1\uc811 \uadf8\ub9ac\uc9c0 \uc54a\uc544\ub3c4 \ub429\ub2c8\ub2e4. “\ud638\uc218 \uc704\uc758 \uc740\ud558\uc218”\uc640 \uac19\uc740 \uc9e7\uc740 \ubb38\uad6c\ub97c \uc785\ub825\ud574 \uc774\ubbf8\uc9c0\uc758 \uc8fc\uc694 \ud2b9\uc9d5\uacfc \ud14c\ub9c8\ub97c \ube60\ub974\uac8c \uc0dd\uc131\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uc2a4\ucf00\uce58\ub97c \ud1b5\ud574 \ud638\uc218\ub97c \ub354 \ud06c\uac8c \ub9cc\ub4e4\uac70\ub098 \uc804\uacbd\uc5d0 \uba87 \uadf8\ub8e8\uc758 \ub098\ubb34\ub97c \ucd94\uac00\ud558\uace0 \ud558\ub298\uc5d0 \ubcc4\uc744 \ub354 \uc0dd\uc131\ud558\ub294 \ub4f1 \uc0ac\uc6a9\uc790 \uc815\uc758 \ub610\ud55c \uac00\ub2a5\ud569\ub2c8\ub2e4. \u2018\uace0\uac312\u2019\ub294 \ud604\uc2e4\uae30\ubc18\uc758 \uc774\ubbf8\uc9c0\ubfd0 \uc544\ub2c8\ub77c \ub2e4\ub978 \uc138\uc0c1\uc758 \ud48d\uacbd\ub3c4 \ubb18\uc0ac\ud560 \uc218 \uc788\ub2e4. \uc608\ub97c \ub4e4\uc5b4 \uc601\ud654 \u2018\uc2a4\ud0c0\uc6cc\uc988\u2019 \uc2dc\ub9ac\uc988\uc5d0\uc11c \ub450 \uac1c\uc758 \ud0dc\uc591\uc774 \uc788\ub294 \uc0c1\uc9d5\uc801\uc778 \ud589\uc131 \ud0c0\ud22c\uc778(Tatooine)\uc758 \ud48d\uacbd\uc744 \uc7ac\ud604\ud558\ub824\uba74 \u2018\uc0ac\ub9c9\u2019, \u2018\ubaa8\ub798\uc5b8\ub355\u2019, \u2018\ud0dc\uc591\u2019\uc774\ub77c\ub294 \ud14d\uc2a4\ud2b8\ub9cc\uc73c\ub85c \uc2dc\uc791\uc810\uc744 \ub9cc\ub4e4\uace0, \uc774\ub97c \ubc14\ud0d5\uc73c\ub85c \uc0ac\uc6a9\uc790\uac00 \ub450 \ubc88\uc9f8 \ud0dc\uc591\uc744 \uc2a4\ucf00\uce58\ud574 \ub123\uc73c\uba74 \ub429\ub2c8\ub2e4. <\/p>\n\n\n\n

GauGAN2\ub294 1,000\ub9cc\uac1c\uc758 \ud48d\uacbd \uc774\ubbf8\uc9c0 \ub370\uc774\ud130\ub97c \uc0ac\uc6a9\ud558\uc600\uae30\uc5d0 \ud48d\uacbd \uc0dd\uc131\uc5d0\ub9cc \uc911\uc810\uc744 \ub461\ub2c8\ub2e4. nvidia \uc5f0\uad6c\ud300\uc740 \ub610\ud55c \uc774\ubbf8\uc9c0\uc5d0 \uc0ac\ub78c\uc774 \ud3ec\ud568\ub41c \uc7a5\uba74\uc774 \uc5c6\ub294\uc9c0 \ud655\uc778 \ud558\uae30 \uc704\ud558\uc5ec \uc774\ubbf8\uc9c0 \uac80\ud1a0 \uad00\ub828 \uc791\uc5c5\uc744 \ud558\uace0 \uc788\uc2b5\ub2c8\ub2e4.\u201d\ub77c\uace0 \ubc1d\ud614\uc73c\uba70 \uc774\ub807\uac8c \ud558\uba74 GauGAN2\uc758 \ud3b8\uacac\uc744 \uc904\uc774\ub294\ub370 \ub3c4\uc6c0\uc774 \ub420\uc218 \uc788\uace0,  \ube44\ub85d \uc544\uc9c1 \ub370\ubaa8 \ub2e8\uacc4\uc774\uc9c0\ub9cc \ucd94\ud6c4 \ud48d\uacbd\ubfd0\ub9cc\uc774 \uc544\ub2c8\ub77c \ub2e4\uc591\ud55c \uc774\ubbf8\uc9c0 \uc0dd\uc131\uc744 \uc704\ud574 \uad00\ub828 \ub41c \ud3b8\uacac\uc131\uc744 \uc904\uc774\uae30 \uc704\ud55c \uc791\uc5c5\ub3c4 \ud568\uaed8 \ud560  \uc608\uc815\uc774\ub77c\uace0 \uc785\uc7a5\uc744 \ubc1d\ud614\uc5c8\uc2b5\ub2c8\ub2e4.<\/p>\n\n\n\n

[OpenAI DALL\u00b7E]<\/strong><\/p>\n\n\n\n

2021\ub144 Open AI<\/a>\uc5d0\uc11c\ub294 \uc6d0\ud558\ub294 \uac83\uc744 \ud14d\uc2a4\ud2b8\ub85c \uc124\uba85\ud558\uba74 \uc774\ubbf8\uc9c0\ub85c \uc0dd\uc131\ud574\uc8fc\ub294 DALL.E<\/a>\ub294 \ub97c \uc18c\uac1c\ud588\uc5c8\uc2b5\ub2c8\ub2e4.  \uc774 \uc11c\ube44\uc2a4\ub294 \uace0\ud574\uc0c1\ub3c4\uc758 \uc774\ubbf8\uc9c0\ub97c \uc9e7\uc740 \uc2dc\uac04 \ub0b4\uc5d0 \uc0dd\uc131\ud558\uace0 \ud3ec\ud1a0\uc0f5\uc774\ub098, \ub514\uc9c0\ud138 \uc544\ud2b8 \uae30\uc220 \uc5c6\uc774\ub3c4 \uac04\ub2e8\ud55c \ub3c4\uad6c\uc640 \ud14d\uc2a4\ud2b8 \uc218\uc815\uc73c\ub85c \uc774\ubbf8\uc9c0\ub97c \uc27d\uac8c \ud3b8\uc9d1\ud560 \uc218 \uc788\ub2e4\ub294 \uc7a5\uc810\uc774 \uc788\uc2b5\ub2c8\ub2e4. 1\ub144 \ud6c4 Open AI\uc5d0\uc11c\ub294 DALL.E \uc758 \ud6c4\uc18d \ubc84\uc804\uc778  DALL.E 2<\/a>\ub97c \ubc1c\ud45c\ud558\uac8c \ub418\uc5c8\ub294\ub370\uc694, \ub2f9\uc2dc \uc0dd\uc131\ub41c \uc774\ubbf8\uc9c0\ub4e4\uc758 \ub180\ub77c\uc6b4 \uc131\ub2a5\ub4e4\ub85c \uc778\ud574 \uba87\uc8fc\uac04 \ud2b8\uc704\ud130\uc5d0\uc11c \ud070 \ud30c\uc7a5\uc744 \uc77c\uc73c\ucf30\uc5c8\uc2b5\ub2c8\ub2e4.  \uae30\uc874 DALL.E\ub294 \uc885\uc885 \uc77c\ubc18 \ubc30\uacbd\uc5d0 \ub300\ud574 \ub9cc\ud654\uac19\uc740 \ubc29\uc2dd\uc73c\ub85c\ub9cc \uc774\ubbf8\uc9c0\ub97c \ub80c\ub354\ub9c1\ud560 \uc218 \uc788\uc5c8\uc73c\ub098 \uc0c8\ub85c\uc6b4 DALL-E 2\ub294 \ubcf5\uc7a1\ud55c \ubc30\uacbd, \ud53c\uc0ac\uacc4 \uc2ec\ub3c4(depth of field) \ud6a8\uacfc, \uc0ac\uc2e4\uc801\uc778 \uadf8\ub9bc\uc790, \uc74c\uc601 \ubc0f \ubc18\uc0ac\uac00 \ud3ec\ud568\ub41c \uc0ac\uc9c4 \uac19\uc740 \uace0\ud574\uc0c1\ub3c4 \uc774\ubbf8\uc9c0\ub97c \uc0dd\uc131\ud560 \uc218 \uc788\uac8c \ub418\uc5c8\uc2b5\ub2c8\ub2e4. \uae30\uc874 DALL-E\uac00 256×256 \uc774\ubbf8\uc9c0\ub97c \uc0dd\uc131\ud55c \uac83\uc5d0 \ube44\ud574 DALL-E 2\ub294 \uc774\ub97c 4\ubc30 \uc5c5\uc2a4\ucf00\uc77c\ub9c1\ud558\uc5ec 1024×1024\uae4c\uc9c0\uc758 \ud574\uc0c1\ub3c4\ub97c \uc9c0\uc6d0\ud55c\ub2e4\uace0 \ud569\ub2c8\ub2e4. <\/p>\n\n\n\n

\"\"
[\uadf8\ub9bc 3] 1. vaporwave \uc2a4\ud0c0\uc77c\ub85c \uc6b0\uc8fc\uc758 \uc5f4\ub300 \ub9ac\uc870\ud2b8\uc5d0\uc11c \ud734\uc2dd\ud558\uace0 \uc788\ub294 \uc6b0\uc8fc \ube44\ud589\uc0ac 2. \ud138\uc2e4\ub85c \uc9e0 \uad34\ubb3c\ucc98\ub7fc \ubcf4\uc774\ub294 \ud55c \uadf8\ub987\uc758 \uc218\ud504 (\ucd9c\ucc98: Open AI DALL.E 2<\/a>)<\/figcaption><\/figure><\/div>\n\n\n\n

\ub354\ubd88\uc5b4 DALL-E 2\ub294 \uc790\uc5f0\uc5b4\ub85c\ubd80\ud130 \uc774\ubbf8\uc9c0\ub97c \uc0dd\uc131\ud560 \ubfd0\ub9cc \uc544\ub2c8\ub77c, \uae30\uc874 \uc774\ubbf8\uc9c0\uc5d0 \ub300\ud574 \uc6d0\ud558\ub294 \ubc29\ud5a5\uc73c\ub85c\uc758 \uc218\uc815\uc744 \ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uc6d0\ub798 \uc0c1\uc5c5\uc6a9 \uc774\ubbf8\uc9c0\ub97c \ub9cc\ub4e4 \ub54c\ub3c4 \uc2dc\uc548\uc744 \ubf51\uc740 \ud6c4 \ub514\uc790\uc774\ub108\uc640\uc758 \ud611\uc758\ub97c \ud1b5\ud574 \uc774\ub97c \uc810\ucc28 \uc218\uc815\ud558\ub294 \uacfc\uc815\uc744 \uac70\uce58\ub294\ub370, DALL-E 2\uc5d0\ub3c4 \uc790\uc5f0\uc5b4\ub85c \uc774\ub7f0 \uc694\uccad\uc744 \ud560 \uc218 \uc788\ub2e4\ub294 \uac83\uc785\ub2c8\ub2e4. <\/p>\n\n\n\n

\"\"
[\uadf8\ub9bc 4] \uc67c\ucabd \uadf8\ub9bc\uacfc \uac19\uc774 \uc6d0\ubcf8 \uc774\ubbf8\uc9c0\uc5d0 \ucf54\uae30\ub97c \ucd94\uac00\ud560 \uc704\uce58\ub97c \uc120\ud0dd\ud558\uba74 \uc2e4\uc81c \ubc30\uacbd\uacfc \ucf54\uae30\uac00 \ud569\uc131\ub41c \uc774\ubbf8\uc9c0\uac00 \uc0dd\uc131 \ub428.(\ucd9c\ucc98: Open AI DALL.E 2<\/a>)<\/figcaption><\/figure><\/div>\n\n\n\n

\uc624\ud508AI\ub294 GPT-3 \ub54c\uc640 \ub9c8\ucc2c\uac00\uc9c0\ub85c \uc2e0\ub8b0\ud560 \uc218 \uc788\ub294 \uc18c\uaddc\ubaa8 \uc774\uc6a9\uc790\uc5d0\uac8c DALL-E 2\ub97c \uc120 \uacf5\uac1c\ud55c \uc774\ud6c4\uc5d0 \uac1c\uc120 \uc791\uc5c5\uc744 \uac70\uccd0 \uc77c\ubc18 \ub300\uc911\uc5d0\uac8c \uacf5\uac1c\ud560 \uacc4\ud68d\uc774\ub77c\uace0 \ud569\ub2c8\ub2e4. GPT-3\ub294 \uacf5\uaca9\uc801\uc774\uac70\ub098 \uc720\ud574\ud55c \ud14d\uc2a4\ud2b8<\/a>\ub97c \uc0dd\uc131\ud558\ub294 \ubb38\uc81c\uac00 \uc788\uc5c8\uc2b5\ub2c8\ub2e4. \ud558\uc5ec \uc624\ud508AI\ub294 GPT-3 \uc774\uc6a9\uc790\ub4e4\ub85c\ubd80\ud130 \ud53c\ub4dc\ubc31\uc744 \ubc1b\uc544\uc11c \ub354 \uc548\uc804\ud55c \ubc84\uc804\uc778 \u2018\uc778\uc2a4\ud2b8\ub7ed\ud2b8GPT(InstructGPT)<\/a>\u2019\ub97c \ub9cc\ub4e4\uc5c8\uc2b5\ub2c8\ub2e4. DALL-E 2\uc5d0\uc11c\ub3c4 \uc774\uc6a9\uc790 \ud53c\ub4dc\ubc31\uc744 \ud65c\uc6a9\ud574\uc11c \uc81c\ud488\uc744 \uac1c\uc120\ud558\uace0\uc790 \ud55c\ub2e4\uace0 \ubc1d\ud614\uc2b5\ub2c8\ub2e4. \uc624\ud508AI\ub294 \uc6b0\uc120 \ucd08\uae30 \uc774\uc6a9\uc790\ub4e4\uc5d0\uac8c \ub2e4\uc591\ud55c \ubc29\uc2dd\uc744 \uc774\uc6a9\ud574 DALL-E 2\uac00 \uacf5\uaca9\uc801\uc774\uac70\ub098 \uc720\ud574\ud55c \uc774\ubbf8\uc9c0\ub97c \uc0dd\uc131\ud558\uac8c \ub9cc\ub4e4\uc5b4 \ubcf4\ub77c\uace0 \uad8c\uc7a5\ud558\ub294 \ubc29\uc2dd\ub4e4\uc744 \ud1b5\ud574 \ubc1c\uacac\ub41c \ubb38\uc81c\ub4e4\uc744 \ubaa8\ub450 \ud574\uacb0\ud558\uace0 \ub098\uba74 DALL-E 2\ub97c \ub354 \ub9ce\uc740 \uc0ac\ub78c\ub4e4\uc5d0\uac8c \uacf5\uac1c\ud560 \uc608\uc815\uc774\ub77c\uace0 \ud569\ub2c8\ub2e4.<\/p>\n\n\n\n

[Google Imagen]<\/strong><\/p>\n\n\n\n

\ucd5c\uadfc \uad6c\uae00\uc5d0\uc11c\ub3c4 \ud14d\uc2a4\ud2b8 \uc785\ub825\uc744 \uae30\ubc18\uc73c\ub85c \uc0ac\uc2e4\uc801\uc778 \uc774\ubbf8\uc9c0\ub97c \uc0dd\uc131\ud560 \uc218 \uc788\ub294 \ud14d\uc2a4\ud2b8-\uc774\ubbf8\uc9c0 \ud655\uc0b0 \ubaa8\ub378(Diffusion Model)\uc778 Imagen<\/a>\uc744 \uacf5\uac1c\ud558\uc600\uc2b5\ub2c8\ub2e4. \uad6c\uae00 \uc5f0\uad6c\ud300\uc740 \uae30\uc874\uc758 \uc720\uc0ac AI\ub4e4\ub3c4 \ud658\uc0c1\uc801\uc774\uace0 \ud765\ubbf8\ub85c\uc6b4 \ub2e4\uc591\ud55c \uc774\ubbf8\uc9c0\ub97c \uc0dd\uc131 \ud560 \uc218 \uc788\uc9c0\ub9cc, Imagen\uc774 \uc0dd\uc131\ud55c \uc774\ubbf8\uc9c0\ub294 \uc815\ud655\uc131<\/strong>\uacfc \uc774\ubbf8\uc9c0\uc758 \ucda9\uc2e4\ub3c4(fidelity<\/em>)<\/strong>\uc5d0\uc11c \ub354 \ub6f0\uc5b4\ub098\ub2e4\uace0 \uc8fc\uc7a5\ud558\uace0 \uc788\uc2b5\ub2c8\ub2e4. Imagen \uacf5\uc2dd \uc0ac\uc774\ud2b8\uc5d0\ub294 \uc785\ub825\ub41c \ud14d\uc2a4\ud2b8\uc640 \uadf8\uc5d0 \ub530\ub77c \uc0dd\uc131\ub41c \uc774\ubbf8\uc9c0\uc758 \uc0ac\ub840\ub4e4\uc774 \uc18c\uac1c\ub418\uc5b4 \uc788\uc2b5\ub2c8\ub2e4. \uc774\ub7ec\ud55c \uc774\ubbf8\uc9c0\ub4e4\uc740 \ub9c8\uce58 \uc0ac\ub78c\uc774 \uc9c1\uc811 \ud3ec\ud1a0\uc0f5\uc744 \ud55c \uac83\ucc98\ub7fc \ucc29\uac01\ud560 \uc815\ub3c4\uc758 \uae54\ub054\ud558\uace0 \ub192\uc740 \ud574\uc0c1\ub3c4\uc758 \ud004\ub9ac\ud2f0\ub97c \uac16\uace0 \uc788\uc2b5\ub2c8\ub2e4. <\/p>\n\n\n\n

\"\"
[\uadf8\ub9bc 5] \uc785\ub825\ub41c \ud14d\uc2a4\ud2b8 \uae30\ubc18\uc73c\ub85c google Imagen\uc744 \ud1b5\ud574 \uc0dd\uc131\ub41c 1024 \u00d7 1024 \uc774\ubbf8\uc9c0 \uc608\uc2dc (\ucd9c\ucc98: google Imagen<\/a>)<\/figcaption><\/figure><\/div>\n\n\n\n

Imagen\uc740 \ud14d\uc2a4\ud2b8\ub97c \uc774\ud574\ud558\ub294 \ub300\ud615 \ubcc0\ud658\uae30(transformer) \uc5b8\uc5b4\ubaa8\ub378\uc758 \uc131\ub2a5\uc744 \uae30\ubc18\uc73c\ub85c \uc815\ud655\ub3c4 \ub192\uc740(high-fidelity ) <\/em>\uc774\ubbf8\uc9c0\ub97c \uc0dd\uc131\ud558\ub294 \ud655\uc0b0 \ubaa8\ub378\uc744 \uacb0\ud569\ud558\uc600\uc2b5\ub2c8\ub2e4. \uad6c\uae00 \uc5f0\uad6c\ud300\uc740 Imagen\uc744 \ud1b5\ud574 \ud14d\uc2a4\ud2b8 \uc804\uc6a9 \ub9d0\ubb49\uce58\ub85c \uc0ac\uc804 \ud6c8\ub828\ub41c \uac70\ub300 \uc5b8\uc5b4\ubaa8\ub378\uc758 \ud14d\uc2a4\ud2b8 \uc784\ubca0\ub529\uc774 \ud14d\uc2a4\ud2b8-\uc774\ubbf8\uc9c0 \ud569\uc131(text encoding)\uc5d0 \uc0c1\ub2f9\uc774 \ud6a8\uacfc\uc801\uc774\ub77c\ub294 \uac83\uc744 \ubc1c\uacac\ud588\ub2e4\uace0 \ud569\ub2c8\ub2e4.<\/p>\n\n\n\n

\"\"
[\uadf8\ub9bc 6] Imagen \uc2dc\uac01\ud654: 1. Imagen\ub294 \uba3c\uc800 frozen\ub41c T5-XXL \uc778\ucf54\ub354\ub97c \uc0ac\uc6a9\ud558\uc5ec \uc785\ub825 \ud14d\uc2a4\ud2b8\ub97c \uc77c\ub828\uc758 \ud14d\uc2a4\ud2b8 \uc784\ubca0\ub529\uc73c\ub85c \uc778\ucf54\ub529\ud568. 2. \uc870\uac74\ubd80 \ud655\uc0b0\ubaa8\ub378\uc740 \ud14d\uc2a4\ud2b8 \uc784\ubca0\ub529\uc744 64\u00d764 \uc774\ubbf8\uc9c0\uc5d0 \ub9e4\ud551\ud568. 3. \ub2e4\uc74c \ud14d\uc2a4\ud2b8 \uc870\uac74\ubd80 \ucd08\uace0\ud574\uc0c1\ub3c4 \ud655\uc0b0 \ubaa8\ub378\uc744 \uc774\uc6a9\ud558\uc5ec 64×64 \uc774\ubbf8\uc9c0\u2192  256×256 \uc774\ubbf8\uc9c0\u2192 1024×1024 \uc774\ubbf8\uc9c0\ub85c \uc5c5 \uc0d8\ud50c\ub9c1 \ud568.<\/figcaption><\/figure><\/div>\n\n\n\n

\uacf5\uac1c\ub41c \ub17c\ubb38\uc5d0 \uc758\ud558\uba74 Imagen\uc758 \uc131\ub2a5 \ud3c9\uac00\ub97c \uc704\ud574 \ud14d\uc2a4\ud2b8-\uc774\ubbf8\uc9c0 \ubaa8\ub378\uc5d0 \ub300\ud55c \ubca4\uce58\ub9c8\ud06c\uc778 DrawBench\ub97c \uc0ac\uc6a9\ud574 Imagen\uc744 VQ-GAN+CLIP, Latent Diffusion Models \ubc0f DALL-E 2\ub97c \ud3ec\ud568\ud55c \ucd5c\uc2e0 \ubc29\ubc95\ub4e4\uacfc \ube44\uad50\ud55c \uacb0\uacfc, [\uadf8\ub9bc 7]\uacfc \uac19\uc774 \ud3c9\uac00\uc790\ub4e4\uc774 \ub2e4\ub978 \ubaa8\ub378\ubcf4\ub2e4 Imagen\uc744 \uc120\ud638\ud55c\ub2e4\ub294 \uac83\uc744 \ubc1c\uacac\ud588\ub2e4\uace0 \ud569\ub2c8\ub2e4. <\/p>\n\n\n\n

\"\"
[\uadf8\ub9bc 7] Imagen\uc758 Drawbench \ud3c9\uac00 \uacb0\uacfc (\ucd9c\ucc98: Google Imagen<\/a>)<\/figcaption><\/figure><\/div>\n\n\n\n

\uc544\ub798 [\uadf8\ub9bc 8]\uc740 \ud3c9\uac00\uc5d0 \uc0ac\uc6a9\ub41c \ud14d\uc2a4\ud2b8 \ubc0f \uac01 \ubaa8\ub378\uc5d0 \ub300\ud55c \uc774\ubbf8\uc9c0 \uc0dd\uc131 \uacb0\uacfc\uc785\ub2c8\ub2e4. \uadf8\uc911 “A horse riding an astronaut.” prompt \uc5d0 \ub300\uc751\ub418\ub294 \uacb0\uacfc\ub97c \ubcf4\uba74 “\uc6b0\uc8fc \ube44\ud589\uc0ac\ub97c \ud0c0\uace0 \uc788\ub294 \ub9d0”\uc778\ub370 \ubaa8\ub450 “\ub9d0\uc744 \ud0c0\uace0 \uc788\ub294 \uc6b0\uc8fc \ube44\ud589\uc0ac”\uc5d0 \ub300\uc751\ub418\ub294 \uc774\ubbf8\uc9c0\ub97c \uc0dd\uc131\ud55c \uac83\uc744 \ubcfc \uc218 \uc788\uc2b5\ub2c8\ub2e4. \u201cA panda making latte art.\u201d\uc5d0 \ub300\ud574\uc11c\ub294 Imagen\uc740 \ubd80\ud569\ub418\ub294 \uc774\ubbf8\uc9c0\ub97c \uc0dd\uc131\ud588\uc73c\ub098 DALL.E 2\ub294 “\ud39c\ub354\ub77c\ub5bc \uc544\ud2b8”\uc5d0 \ub300\uc751\ub418\ub294 \uc774\ubbf8\uc9c0\ub97c \uc0dd\uc131\ud55c \uac78 \ubcfc \uc218 \uc788\uc2b5\ub2c8\ub2e4. \ub610\ud55c \uceec\ub7ec \uce74\ud14c\uace0\ub9ac \uad00\ub828 \ub41c \uc2e4\ud5d8\uc5d0\uc11c “A black apple and a green backpack.” prompt\uc5d0 \ub300\ud574 Imagen\uc740 \uac80\uc815 \uc0ac\uacfc, \ucd08\ub85d\uc0c9 \ubc31\uc5d0 \ubd80\ud569\ub418\ub294 \uc774\ubbf8\uc9c0\ub97c \uc0dd\uc131\ud588\uc73c\ub098, DALL.E 2\ub294 \ucd08\ub85d\uc0c9 \uc0ac\uacfc, \uac80\uc815 \ubc31\uc774\ub780 prompt \uc5d0 \ubd80\ud569\ub418\uc9c0 \uc54a\ub294 \uc774\ubbf8\uc9c0\ub97c \uc0dd\uc131\ud55c \uac78 \ud655\uc778 \ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uc774\ucc98\ub7fc \uc5ec\ub7ec \uc608\uc2dc\ub4e4\uc744 \ubd24\uc744 \ub54c Imagen\uc774 \ub354 \uc0ac\uc2e4\uc801\uc778 \uc774\ubbf8\uc9c0\ub97c \uc0dd\uc131\ud558\uc9c0\ub9cc color , aligned \uad00\ub828\ud574\uc11c \ubaa8\ub378\ub4e4\uc740 \uc5ec\uc804\ud788 \uc5b4\ub824\uc6c0\uc744 \uacaa\uace0 \uc788\uc74c\uc744 \ud655\uc778 \ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n\n\n\n

\"\"
[\uadf8\ub9bc 8] Drawbench \ud3c9\uac00 \uc791\uc5c5 \uc2dc \uc0ac\uc6a9\ub41c Imagen , DALL.E 2 \uc0dd\uc131 \uacb0\uacfc (\ucd9c\ucc98: Google Imagen Paper<\/a>)<\/figcaption><\/figure>\n\n\n\n

DALL-E\uc640 \ub9c8\ucc2c\uac00\uc9c0\ub85c Imagen\ub3c4 \uc544\uc9c1\uc740 \ub300\uc911\uc5d0\uac8c \uacf5\uac1c\ub418\uc9c0 \uc54a\uc744 \uac70\ub77c\uace0 \ud569\ub2c8\ub2e4. \uad6c\uae00\uc740 \uc5ec\ub7ec \uac00\uc9c0 \uc774\uc720\ub85c \uc77c\ubc18 \ub300\uc911\uc774 \uc0ac\uc6a9\ud558\uae30\uc5d0 \uc544\uc9c1 \uc801\ud569\ud558\uc9c0 \uc54a\ub2e4\uace0 \ub9d0\ud588\uc2b5\ub2c8\ub2e4. \ud55c \uac00\uc9c0 \uc608\ub85c, \ud14d\uc2a4\ud2b8-\uc774\ubbf8\uc9c0 \ubaa8\ub378\uc740 \uc77c\ubc18\uc801\uc73c\ub85c \uc6f9\uc5d0\uc11c \uc218\uc9d1\ud558\uace0 \uc120\ubcc4\ub418\uc9c0 \uc54a\uc740 \ub300\uaddc\ubaa8 \ub370\uc774\ud130 \uc138\ud2b8\uc5d0 \ub300\ud574 \ud559\uc2b5\ub418\ubbc0\ub85c \ub9ce\uc740 \ubb38\uc81c\uac00 \ubc1c\uc0dd\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. <\/p>\n\n\n\n

\uc774\uc678\uc5d0\ub3c4 \uad6d \ub0b4 \uc678\uc5d0\ub294 \ub2e4\uc591\ud55c \uba40\ud2f0\ubaa8\ub2ec \uc11c\ube44\uc2a4\ub098 \uc5f0\uad6c\uc0ac\ub840\ub4e4\uc774 \ub9ce\uc740\ub370\uc694, \ucd5c\uadfc \uc911\uad6d \uce6d\ud654\ub300\uc5d0\uc11c\ub294 \ud14d\uc2a4\ud2b8\ub97c \uc785\ub825\ud558\uba74 4\ucd08 \ubd84\ub7c9\uc758 32 frame \ub3d9\uc601\uc0c1\uc744 \uc0dd\uc131\ud560 \uc218 \uc788\ub294CogVideo<\/a>(Large-scale Pretraining for Text-to-Video Generation via Transformers)\ub780 \ubaa8\ub378\uc744 \uacf5\uac1c\ud588\uc2b5\ub2c8\ub2e4. \uc544\uc9c1 \ub17c\ubb38\uc740 \uacf5\uac1c\ub418\uc5b4 \uc788\uc9c0 \uc54a\uace0 \uacb0\uacfc \ubb3c\ub9cc \uc815\ub9ac\ud574\ub454 \uc0c1\ud0dc\uc785\ub2c8\ub2e4. \ud574\ub2f9 \uae30\uc220\ub3c4 \uc544\uc9c1\uc740 \uc131\uc219\ub418\uc9c0 \uc54a\uc558\uc9c0\ub9cc, \uae30\ud0c0 \uae30\uc220\uc758 \ubc1c\uc804\uc18d\ub3c4\ub97c \ucc38\uace0\ud55c\ub2e4\uba74 \ud574\ub2f9 \ubaa8\ub378\ub4e4\uc758 \uc131\ub2a5\ub3c4 \uae09\uc131\uc7a5 \ud560 \uac83 \uac19\uc2b5\ub2c8\ub2e4. \uadf8\ub54c \ud574\ub2f9 \uae30\uc220\uc744 \uc798 \ud65c\uc6a9\ud55c\ub2e4\uba74 \ub2e8\uc21c\ud788 \ub9d0\ub85c \uc9c1\uc811 \uc560\ub2c8\uba54\uc774\uc158\uc774\ub098, \uc601\ud654, \ubba4\uc9c1\ube44\ub514\uc624 \ub4f1\ub3c4 \ub9cc\ub4e4 \uc218 \uc788\uc9c0 \uc54a\uc744\uae4c \uc870\uc2ec\uc2a4\ub7fd\uac8c \uc0dd\uac01\uc744 \ud574\ubd05\ub2c8\ub2e4.<\/p>\n\n\n\n

[Conclusion]<\/strong><\/p>\n\n\n\n

\ud14d\uc2a4\ud2b8-\uc774\ubbf8\uc9c0 , \ud14d\uc2a4\ud2b8-\ube44\ub514\uc624 \ub4f1\uacfc \uac19\uc740 \uba40\ud2f0\ubaa8\ub2ec\uc740 \ud655\uc2e4\ud788 \ud658\uc0c1\uc801\uc778 \ucc3d\uc758\uc801 \uc7a0\uc7ac\ub825\uc744 \uac00\uc9c0\uace0 \uc788\uc9c0\ub9cc \ubb38\uc81c\uac00 \ub9ce\uc740 \uac83\uc740 \uc0ac\uc2e4\uc785\ub2c8\ub2e4. \uc608\ub97c \ub4e4\uc5b4 \uac00\uc9dc \ub274\uc2a4, \uc2a4\ud338 \uba54\uc77c \ub610\ub294 \uad34\ub86d\ud798\uc5d0 \uc0ac\uc6a9\ub418\ub294, \uc6d0\ud558\ub294 \uac70\uc758 \ubaa8\ub4e0 \uc774\ubbf8\uc9c0\ub97c \uc0dd\uc131\ud558\ub294 \uc2dc\uc2a4\ud15c\uc744 \uc0c1\uc0c1\ud574 \ubd05\uc2dc\ub2e4. \ud559\uacc4\uc640 \uc5c5\uacc4\uc5d0\uc11c \ubaa8\ub450 \uc6b0\ub824\ud588\ub4ef\uc774 \uc774\ub7ec\ud55c \uc2dc\uc2a4\ud15c\uc740 \uc0ac\ud68c\uc801 \ud3b8\uacac\ub3c4 \uc778\ucf54\ub529\ud558\uba70 \uadf8 \uacb0\uacfc\ub294 \uc885\uc885 \uc778\uc885 \ucc28\ubcc4\uc801\uc774\uac70\ub098 \uc131 \ucc28\ubcc4\uc801\uc774\uac70\ub098 \ub2e4\ub978 \ubc29\uc2dd\uc73c\ub85c \uc548 \uc88b\uc740 \uacb0\uacfc\ub97c \ucd08\ub798\ud558\uac8c \ub420 \uc218\ub3c4 \uc788\uc2b5\ub2c8\ub2e4. \uc9c0\uae08\ub3c4 \uc9e7\uc740 \uc2dc\uac04 \ub0b4\uc5d0 \uace0\ud574\uc0c1\ub3c4 \uace0 \ud004\ub9ac\ud2f0\uc758 \uc774\ubbf8\uc9c0\ub97c \uc0dd\uc131\ud560 \uc218 \uc788\ub294\ub370 \uc55e\uc73c\ub85c \uc0c1\uc5c5\uc801\uc73c\ub85c \uc0ac\uc6a9\ud560 \uc218 \uc788\ub294 \ud488\uc9c8\uc758 \uc774\ubbf8\uc9c0\ub97c \uc0dd\uc131 \ud560 \uc218 \uc788\ub2e4\uba74, \uc9c0\uae08\ubcf4\ub2e4 \ub354 \uc27d\uac8c \ub525\ud398\uc774\ud06c \uc774\ubbf8\uc9c0\ub97c \ub9cc\ub4e4\uc5b4 \ub0bc\uc218 \uc788\uace0, \ud574\ub2f9 \uc774\ubbf8\uc9c0\ub97c \uc774\uc6a9\ud558\uc5ec \ub3d9\uc601\uc0c1\uae4c\uc9c0 \uc81c\uc791\uc744 \ud55c\ub2e4\uba74 \ub354 \ub9ce\uc740 \uc0ac\ub78c\ub4e4\uc5d0\uac8c \ubc94\uc8c4\ub97c \uc800\uc9c0\ub974\ub294\ub370 \uc545\uc6a9\ub418\ub294 \uac74 \ubd88\uac00\ud53c \ud558\uac8c \ub420 \uac83\uc785\ub2c8\ub2e4.<\/p>\n\n\n\n

\uae30\uc220 \uc790\uccb4\ub97c \ubc1c\uc804\uc2dc\ud0a4\ub294 \ud55c\ud3b8 , \uc704\uc640 \uac19\uc774 \uc0ac\ud68c\uc801\uc73c\ub85c \ud574\uacb0\ud574\uc57c \ud560 \ubb38\uc81c\ub4e4\ub3c4 \ud568\uaed8 \uacf5\uc874\ud558\uac8c \ub418\ub294\ub370\uc694, \uc88b\uc740 \uae30\uc220\ub4e4\uc774 \uc545\uc6a9\ub418\uc9c0 \uc54a\ub3c4\ub85d \ud568\uaed8 \uace0\ubbfc\ud558\uace0 \ub300\ucc45\uc744 \ub9c8\ub828\ud558\uba74\uc11c \uae30\uc220\uc744 \ubc1c\uc804\uc2dc\ud0a8\ub2e4\uba74 \uc55e\uc73c\ub85c\ub3c4 \ub354 \uc0c1\uc0c1\uce58 \ubabb\ud558\ub294 \ubb34\uad81\ubb34\uc9c4\ud55c \uc0ac\ub78c\ub4e4\uc5d0\uac8c \ub3c4\uc6c0\uc744 \uc904 \uc218 \uc788\ub294 \uc5f0\uad6c\uc640 \uc11c\ube44\uc2a4\ub4e4\uc774 \ub098\uc624\uc9c0 \uc54a\uc744\uae4c \uae30\ub300\ud574\ubd05\ub2c8\ub2e4! <\/p>\n\n\n\n

<\/p>\n\n\n\n

<\/p>\n\n\n\n


\u53c2\u8003<\/p>\n\n\n\n

[1] https:\/\/gweb-research-imagen.appspot.com\/
[2] https:\/\/arxiv.org\/pdf\/2205.11487.pdf
[3] https:\/\/blogs.nvidia.co.kr\/2021\/11\/24\/gaugan2-ai-art-demo\/
[4] https:\/\/openai.com\/dall-e-2\/#demos
[5] https:\/\/smilegate.ai\/2021\/11\/24\/let-an-ai-paint-for-you-gaugan2\/
[6] https:\/\/github.com\/THUDM\/CogVideo
[7] http:\/\/www.aitimes.com\/news\/articleView.html?idxno=144897
[8] http:\/\/www.aitimes.com\/news\/articleView.html?idxno=143854
[9] https:\/\/www.cctvnews.co.kr\/news\/articleView.html?idxno=232608<\/p>\n

<\/span><\/div>","protected":false},"excerpt":{"rendered":"

[\uac00\uc0c1\uc778\uac04\uc5f0\uad6c\ud300 \uc2ec\ud64d\ub9e4]
\n\ud14d\uc2a4\ud2b8, \uc774\ubbf8\uc9c0, \uc74c\uc131 \ub370\uc774\ud130 \ub4f1 \ub2e4\uc591\ud55c \ud615\ud0dc(modality)\ub85c \uc774\ub8e8\uc5b4\uc9c4 \ub370\uc774\ud130 \ud2b9\uc9d5\uc744 \ud6a8\uacfc\uc801\uc73c\ub85c \ud559\uc2b5\ud558\uae30 \uc704\ud55c \ubc29\ubc95\uc744 Multimodal learning\uc774\ub77c\uace0 \ud558\ub294\ub370\uc694, \ucd5c\uadfc \uad6d\ub0b4\uc678 \ud559\uacc4 \ubc0f \uc5c5\uacc4\uc5d0\uc11c\ub294 \uba40\ud2f0\ubaa8\ub2ec \uc778\uacf5\uc9c0\ub2a5(AI) \ubaa8\ub378 \uad00\ub828 \uc5f0\uad6c\ub4e4\uc744 \ud65c\ubc1c\ud788 \uc9c4\ud589\ud558\uace0 \uc788\uc73c\uba70 \uc774\ubbf8 \uc5c5\uacc4\ub4e4\uc5d0\uc11c\ub294 \uc6b0\uc218\ud55c \uc131\ub2a5\uc758 \uc11c\ube44\uc2a4\ub4e4\ub3c4 \ucd9c\uc2dc\ud558\uace0 \uc788\uc2b5\ub2c8\ub2e4. \uc624\ub298 \uadf8\uc911 \uba87\uac1c\uc758 \ub300\ud45c\uc801\uc778 Multimodal \uc5f0\uad6c \uc0ac\ub840\uc5d0 \ub300\ud558\uc5ec \uac04\ub2e8\ud788 \uc18c\uac1c\ud574 \ub4dc\ub9ac\ub3c4\ub85d \ud558\uaca0\uc2b5\ub2c8\ub2e4. <\/p>\n

<\/span><\/div>","protected":false},"author":1,"featured_media":62288,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"rank_math_lock_modified_date":false,"footnotes":""},"categories":[16,19],"tags":[188],"class_list":["post-62275","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-tech01","category-tech04","tag-featured","category-16","category-19","description-off"],"_links":{"self":[{"href":"https:\/\/smilegate.ai\/cn\/wp-json\/wp\/v2\/posts\/62275","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/smilegate.ai\/cn\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/smilegate.ai\/cn\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/smilegate.ai\/cn\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/smilegate.ai\/cn\/wp-json\/wp\/v2\/comments?post=62275"}],"version-history":[{"count":12,"href":"https:\/\/smilegate.ai\/cn\/wp-json\/wp\/v2\/posts\/62275\/revisions"}],"predecessor-version":[{"id":62300,"href":"https:\/\/smilegate.ai\/cn\/wp-json\/wp\/v2\/posts\/62275\/revisions\/62300"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/smilegate.ai\/cn\/wp-json\/wp\/v2\/media\/62288"}],"wp:attachment":[{"href":"https:\/\/smilegate.ai\/cn\/wp-json\/wp\/v2\/media?parent=62275"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/smilegate.ai\/cn\/wp-json\/wp\/v2\/categories?post=62275"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/smilegate.ai\/cn\/wp-json\/wp\/v2\/tags?post=62275"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}