Apple¤¬²èÁüÀ¸À®AI¡ÖStable Diffusion¡×¤Ë¤Þ¤µ¤«¤ÎÀµ¼°Âбþ¡¢³«È¯¼Ô¤¤¤ï¤¯¡Ö²èÁü¤ò1ÉðÊÆâ¤ËÀ¸À®²Äǽ¡×
ʸ¾Ï(¥×¥í¥ó¥×¥È)¤òÆþÎϤ¹¤ë¤À¤±¤Ç²èÁü¤òÀ¸À®¤·¤Æ¤¯¤ì¤ëAI¡ÖStable Diffusion¡×¤Ï¡¢2022ǯ8·î¤Ë°ìÈ̸ø³«¤µ¤ì¤Æ°Ê¹ß¡¢Í»Ö¤Ë¤è¤Ã¤Æ´Êñ¤ËÆ°¤«¤»¤ëUI¤ä³Æ¼ï³ÈÄ¥µ¡Ç½¤¬Â³¡¹¤ÈÀ¸¤ß½Ð¤µ¤ì¤Æ¤¤¤Þ¤¹¡£¤½¤ó¤ÊStable Diffusion¤Ë¤Ä¤¤¤Æ¡¢Apple¤¬µ¡³£³Ø½¬¥Õ¥ì¡¼¥à¥ï¡¼¥¯¡ÖCore ML¡×¤Ø¤ÎºÇŬ²½¤òȯɽ¤·¤Þ¤·¤¿¡£Æ±»þ¤ËMacBook¤äiPhone¤Ê¤É¤ÎApple¥·¥ê¥³¥óÅëºÜ¥Ç¥Ð¥¤¥¹¸þ¤±¤Î¥³¡¼¥É¤â¥ª¡¼¥×¥ó¥½¡¼¥¹¤Ç¸ø³«¤µ¤ì¤Æ¤¤¤Þ¤¹¡£
https://machinelearning.apple.com/research/stable-diffusion-coreml-apple-silicon
GitHub - apple/ml-stable-diffusion: Stable Diffusion with Core ML on Apple Silicon
https://github.com/apple/ml-stable-diffusion
Stable Diffusion¤ò»È¤¦ºÝ¤Ï¡¢¡Ö¥í¡¼¥«¥ë¤Ë¼Â¹Ô´Ä¶¤òÍÑ°Õ¤·¤Æ¼Â¹Ô¡×¤È¡Ö¼Â¹Ô´Ä¶¤ò¹½ÃۺѤߤΥµ¡¼¥Ð¡¼¤Ë¥¢¥¯¥»¥¹¤·¤Æ¼Â¹Ô¡×¤È¤¤¤¦2Ä̤ê¤ÎÊýË¡¤¬¤¢¤ê¤Þ¤¹¡£¥í¡¼¥«¥ë¤Ë´Ä¶¤òÍÑ°Õ¤·¤Æ¼Â¹Ô¤¹¤ëÊýË¡¤Ë¤Ï¹¥¤¤Ê²èÁü¤ò¹¥¤¤Ê¤À¤±À¸À®¤Ç¤¤ë¤È¤¤¤¦¥á¥ê¥Ã¥È¤¬¤¢¤ë¤â¤Î¤Î¡¢¡ÖNVIDIAÀ½¤Î¹âÀǽGPU¤òÅëºÜ¤·¤¿Windows¥Þ¥·¥ó¤¬É¬ÍספȤ¤¤¦¥Ï¡¼¥É¥ë¤¬Â¸ºß¤·¤Þ¤¹¡£¤½¤³¤Ç¡¢Apple¤ÏApple¥·¥ê¥³¥óÅëºÜ¥Ç¥Ð¥¤¥¹¤ÇStable Diffusion¤ò¥í¡¼¥«¥ë¼Â¹Ô¤µ¤»¤ë¤Ù¤¯AppleÀ½µ¡³£³Ø½¬¥Õ¥ì¡¼¥à¥ï¡¼¥¯¡ÖCore ML¡×¤Ø¤ÎºÇŬ²½¤ò¹Ô¤¤¡¢¤½¤ÎÀ®²Ì¤¬2022ǯ12·î1Æü¤Ë¥ê¥ê¡¼¥¹¤µ¤ì¤¿¡ÖmacOS Ventura 13.1 Beta 4¡×¤È¡ÖiOS and iPadOS 16.2 Beta 4¡×¤Ë¼è¤ê¹þ¤Þ¤ì¤Þ¤·¤¿¡£
Stable Diffusion¤Ç´üÂÔÄ̤ê¤Î²èÁü¤òÀ¸À®¤¹¤ë¤Ë¤Ï¡¢²èÁüÀ¸À®¤ò²¿Å٤ⷫ¤êÊÖ¤¹É¬Íפ¬¤¢¤ê¤Þ¤¹¡£¤³¤Î¤¿¤á¡¢1²óÅö¤¿¤ê¤ÎÀ¸À®¤Ë¤«¤«¤ë»þ´Ö¤Ï¤Ê¤ë¤Ù¤¯Ã»¤¯¤¹¤ë¤³¤È¤¬µá¤á¤é¤ì¤Þ¤¹¡£Apple¤¬¸ø³«¤·¤Æ¤¤¤ë³Æ¼ï¥Ç¥Ð¥¤¥¹¤Ç¤Î²èÁüÀ¸À®»þ´Ö¤ò³Îǧ¤¹¤ë¤È¡¢¡ÖM1¥Á¥Ã¥×¤òÅëºÜ¤·¤¿iPad Pro¤ÇStable Diffusion 1.4¤ò¼Â¹Ô¡×¤È¤¤¤¦´Ä¶¤Ç¤Ï1ËçÅö¤¿¤ê38ÉäȤ¤¤¦¼ÂÍÑŪ¤Ê®Å٤DzèÁü¤òÀ¸À®¤Ç¤¤ë¤³¤È¤¬Ê¬¤«¤ê¤Þ¤¹¡£¤Þ¤¿¡¢¡ÖM2¥Á¥Ã¥×¤òÅëºÜ¤·¤¿MacBook Air¤ÇStable Diffusion 2.0¤ò¼Â¹Ô¡×¤È¤¤¤¦´Ä¶¤Ç¤Ï1ËçÅö¤¿¤ê18ÉäȤ¤¤¦¤«¤Ê¤ê¹â®¤Ê²èÁüÀ¸À®¤ò¼Â¸½¤Ç¤¤ë¤è¤¦¤Ç¤¹¡£
¤µ¤é¤Ë¡¢Stable Diffusion 1.4¤Ç¥¹¥Æ¥Ã¥×¿ô¤ò50¡¢²òÁüÅÙ¤ò512¡ß512¥Ô¥¯¥»¥ë¤ËÀßÄꤷ¤Æ³Æ¼ïAppleÀ½¥Þ¥·¥ó¤Ç²èÁü¤òÀ¸À®¤·¤¿ºÝ¤Î¼Â¹Ô»þ´Ö¤ò¤Þ¤È¤á¤¿É½¤¬°Ê²¼¡£Mac Studio¤Ç¤Ï9ÉäDzèÁü¤òÀ¸À®¤Ç¤¤Æ¤¤¤Þ¤¹¡£Èæ³Ó¤Î¤¿¤á¤Ë¡¢ÊÔ½¸Éô¤Ë¤¢¤ëNVIDIA GeForce RTX 3060¤òÅëºÜ¤·¤¿Windows¥Þ¥·¥ó¤ÇƱÍͤÎÀßÄê¤Ç²èÁü¤òÀ¸À®¤·¤¿·ë²Ì¡¢²èÁüÀ¸À®¤Ë¤«¤«¤Ã¤¿»þ´Ö¤Ï9¡Á10ÉäǤ·¤¿¡£
Stable Diffusion 1.4¤Ç¹âÉʼÁ¤Ê²èÁü¤òÀ¸À®¤¹¤ë¤Ë¤Ï¥¹¥Æ¥Ã¥×¿ô¤ò50¤Ê¤É¤Î¹â¤¤ÃͤËÀßÄꤹ¤ëɬÍפ¬¤¢¤ê¤Þ¤·¤¿¤¬¡¢Stable Diffusion 2.0¤Ç¤Ï¥¹¥Æ¥Ã¥×¿ô¤ò1¡Á4¤ËÀßÄꤷ¤Æ¤â¹âÉʼÁ¤Ê²èÁü¤òÀ¸À®¤Ç¤¤ë¤³¤È¤¬Êó¹ð¤µ¤ì¤Æ¤¤¤Þ¤¹¡£Apple¤Çµ¡³£³Ø½¬´ØÏ¢¤Î³«È¯¤ò¹Ô¤Ã¤Æ¤¤¤ëAtila Orhon»á¤Ë¤è¤ë¤È¡¢M2¥Á¥Ã¥×¤òÅëºÜ¤·¤¿¥Ç¥Ð¥¤¥¹¤Ç¥¹¥Æ¥Ã¥×¿ô¤ò1¡Á4¤ËÀßÄꤷ¤ÆStable Diffusion 2.0¤òÆ°ºî¤µ¤»¤ë¤È¡¢1ÉðÊÆâ¤Ë²èÁü¤òÀ¸À®¤Ç¤¤ë¤È¤Î¤³¤È¤Ç¤¹¡£
For distilled #StableDiffusion2 which requires 1 to 4 iterations instead of 50, the same M2 device should generate an image in <<1 second: https://t.co/raA8RggRrU— Atila @NeurIPS (@atiorh) December 1, 2022
Apple¥·¥ê¥³¥ó¤ÏCPU¤äGPU¤Î¾¤Ëµ¡³£³Ø½¬ÀìÍÑ¥Á¥Ã¥×¡ÖNeural Engine¡×¤Ê¤É¤¬¤Ò¤È¤Þ¤È¤á¤Ë¤Ê¤Ã¤¿SoC¤Ç¡¢Apple¤¬³«È¯¤·¤¿µ¡³£³Ø½¬¥Õ¥ì¡¼¥à¥ï¡¼¥¯¡ÖCore ML¡×¤ÈÁȤ߹ç¤ï¤»¤ë¤³¤È¤Ç¡¢¹âÀǽ¤Êµ¡³£³Ø½¬¥·¥¹¥Æ¥à¤ò¼Â¹Ô²Äǽ¡£Apple¥·¥ê¥³¥ó¤ÏÎòÂåiPhone¤äiPad¤Ë²Ã¤¨¤Æ2020ǯ11·î°Ê¹ß¤Ëȯɽ¤µ¤ì¤¿Mac¥·¥ê¡¼¥º¤ËÅëºÜ¤µ¤ì¤Æ¤¤¤Þ¤¹¡£Apple¤ÏStable Diffusion¤òApple¥·¥ê¥³¥óÅëºÜ¥Ç¥Ð¥¤¥¹¤ÇÆ°ºî¤µ¤»¤ë¤¿¤á¤Î¥³¡¼¥É¤òGitHub¤Ç¸ø³«¤·¤Æ¤¤¤ë¤¿¤á¡¢AppleÀ½¥Ç¥Ð¥¤¥¹¤Ç¼ê·Ú¤ËStable Diffusion¤ò»È¤¨¤ë¥¢¥×¥ê¤¬Â³¡¹¤ÈÅо줹¤ë¤³¤È¤¬´üÂԤǤ¤½¤¦¤Ç¤¹¡£