24 Like
8 Dislike
CyberTrump
2023-04-24 14:45:14
同你講,我部i9 Mac 都係咁,用 Mac就預左無得玩Stable Diffusion
Un4v5s8bgsVk9Xp
2023-04-24 14:58:47
Thanks Ching
Un4v5s8bgsVk9Xp
2023-04-24 15:00:27
咁你轉PC定點 ?
iMac有冇裝到邊款
moon_sun(不短炒)
2023-04-24 15:02:56
用colab咪得
CyberTrump
2023-04-24 15:35:46
Colab 已經唔係local 做啦
moon_sun(不短炒)
2023-04-24 15:37:14
咁有人問有方法做到
咪就係用cloud
佢又冇指定一定要用local
WE_CONNECT
2023-04-24 16:41:07
一定會有deviation ,唔同prompt 就, 算同一粒seed 出黎唔同樣正常 (同一個prompt ,同一個lora generate 十次都應該幾乎一樣) , 不過理論上輪廓唔會差太遠,或者係你個source image 張張個樣都好唔同, apply Lora 果陣就自然會次次變
試下gen 左出黎再用imgtoimg mask 塊面, apply 個lora 睇下有咩變化。
醉愛小姐的奶兔
2023-04-24 17:23:43
lm學野
半張紙皮
2023-04-24 18:30:15
Cls 啲光影
Moon-River
2023-04-24 18:58:59
lm
喜聞曲腥
2023-04-24 19:00:17
same prompt same output -〉yes
但點解 same seed same prompt with batch size 大過 1,每張相會唔同?有咩變數唔同左?
迷途小股民
2023-04-24 19:48:58
試左發現用動漫人物lora黎整會好好多
但真人lora整既話都係個樣變得唔太似
CyberTrump
2023-04-24 20:33:01
Colab 技術上我386都得啦
Moon-River
2023-04-24 21:03:56
lm
茄牛飯炒底
2023-04-24 21:11:29
有無人用過SD2.1, 用佢定繼續1.5好?
moon_sun(不短炒)
2023-04-24 21:11:39
colab我用iphone
moon_sun(不短炒)
2023-04-24 21:14:17
梗係等多d support先啦
而家1.5咁多extentions
柱現(18碎)
2023-04-24 22:14:52
試咗 似乎依個checkpoint同其他lora唔係好夾 成日爛面
武吉士
2023-04-24 22:23:07
等之後直接跳SDXL
天之庭
2023-04-24 23:30:59
因為兩張相個seed係唔同囉
你當同一粒seed就好似同一張畫紙
上面用一樣既prompt 同setting 就大致會出同一張相
同一粒seed 唔同resolution 畫出黎都會唔一樣
我就是個垃圾
2023-04-24 23:37:06
你哋個個都有張勁嘅display card?
我部垃圾電腦用on board vega 8
自己裝automatic1111 gen512x512圖都not enough memory
Google colab又話paid user only ... 雖然偷雞run都得,但係每次start up都好慢,又有time limit...