[新手包] 窮撚玩 AI - 用 stable diffusion 砌本寫真畀自己 J

淚痣應該用咩prompt
我好似打乜都冇咁

我跟呢個
https://github.com/AUTOMATIC1111/stable-diffusion-webui/discussions/5461?sort=new#top

佢有講用GPU加速好食RAM
如果慢，你試下跟佢troubleshooting改下

不過我都無改過嘢


買咗部M1max咁耐，真係第一次GPU長期用盡風扇又開盡

當然無問題啦

想問下點打先出到全身？
我搞極都出唔到

prompt "full body"..etc
change resolution ratio will help, like 512*768, 512*1024

prompt 打 full body shot
大小 set 512 x 1024
最好用埋 contronet openpose 夾硬控制埋

佢話我gpu usage limit 過咗係咪無得玩?

Google colab?
樓主之前答係，佢有10個 google account

正到咇精

wide shot, model shot ,或者用controlnet都出到全身

留名學嘢

我好似無打淚痣，應該係因為用左 korean doll less lora

試完
無乜起色 都係上返colab 玩算

我整親都同原本身體皮膚唔同色 inpaint 應該要點set 好？

畫動漫畫風記得要dl返vae去補返個色,唔係就會同你dl人地果d圖好大出入


事旦上網鳩搵D Leona既圖,出來既效果都唔錯

Denoising strength 較高少少
Masked content 記住揀 original

我會喺 negative prompt 加 abrupt color transition, 唔知有無用

Batch count 較高啲，一次過 gen 幾十幅相
然後慢慢揀嚟執

有無人整inpaint嘅時候遇到呢個
OutOfMemoryError: CUDA out of memory.

搵d ig 女試下刀

感覺上 inpaint 大食 ram 過 txt2img
較細啲幅圖，或者 batch size 細啲會好啲

係civitai抄人分享既Prompt
好短好簡單但效果好好
隨便gen20張(只係改左俄妹樣)大部分都唔錯

Prompt
best quality, masterpiece, (Realistic:1.2), beach, swimsuit, upper body, sunlight, huge water splash, sun flare

Negative
(low quality, worst quality:1.4),

CFG scale 6
Steps 20
Euler a