[ AI Art ] - stable diffusion 討論 分享 02

有冇辦法知自己整個lora 出黎用唔到嘅話，係錯啲咩？
搞左成日真係好無癮

over train， 再唔係試下set 0.2 0.5咁試下會見到效果

幾好用

Machine learning 呢樣野真係好鬼神奇

感謝，試吓先

lol 直頭好核突
其實用control net 咩曳曳圖都幾乎出到

有冇d cloud 推出用
見大陸仔有個papermoon介紹，比一百幾十唔洗下扯痕自己部機搞training好似抵玩

last year? what!
suddenly AI audio exploding!

https://www.youtube.com/watch?v=57KYN-q2fig&t=32s

Train果時最好每段epoch都出個model, 咁樣就唔怕練完成日先發現overtrain咗。

點解我用bra v5 gen出嚟啲女好似全部都下巴兜兜 唐氏咁

paperspace?
雖然佢 claim 有免費 tier，但基本上一定要俾錢

haven't try yet, share first.
https://github.com/volotat/SD-CN-Animation

有無圖睇下有幾兜

已經del晒啲嘢了遲啲得閒down返再試

機娘IRL

just text2promt? any guide?

*txt2img and promts

我自己再remix 過個Brav5 model,
本身model係好靚不過有少少overtrained唔夠general，
所以我再mix咗啲Rev anime 同Anything入去 (主要mix 喺 Unet 底層幫加啲concept)

同埋Brav5好重白光，用咗呢個方法，ControlNet加咗啲綠色filter 入去，等佢光線柔和啲同有少少舊相feel
https://twitter.com/Flamonster_SD/status/1655087622744834051

用colab train成日都train到一半斷逼我俾錢

wow, I followed you twitter, lots of great works!

Po 得太多甜野俾Twitter shadowban緊， 等睇吓幾時解ban再post

如果佢唔解ban就直接flag 都唔flag 放甜野算鳩數