**圖多慎入**

來(lái)畫(huà)我家看板娘阿燁來(lái)練練手好了(嗯是手?

最近在碰Stable Diffusion，受限於目前的遊戲顯卡，目前測(cè)試我家的顯卡最穩(wěn)定能跑 544*728 的圖，超過(guò)了就很容易跳出錯(cuò)誤 CUDA out of memory...

不知道是不是詠唱太長(zhǎng)，模型支援的詞彙太少還是我的英文詞彙過(guò)於貧乏，即使嘗試控制權(quán)重，有的時(shí)候產(chǎn)出的圖會(huì)些詞不達(dá)意

即使加上了Negative prompt，有的時(shí)候是會(huì)多出十字架跟不明所以的物品。在理想與現(xiàn)實(shí)的拉扯下，最終畫(huà)出了這些圖。之後再來(lái)找找跟面容細(xì)節(jié)有關(guān)係的詞彙吧...

set COMMANDLINE_ARGS=--opt-split-attention --lowvram --precision full --no-half

Prompt:

cinematiclighting,(((8k wallpaper))), best quality, ultra-detailed, illustration,official art,dramatic angle,(lridescent light refraction), (1girl), cute, finely detailed beautiful eyes, close up, blonde long straight hair, blunt_bangs,  (((very long straight sidelocks))), red eyes, ((white maid headdress with 2 orange bow tie)), femininity, standing pose, ((gigantic red and black stripe bow tie on the chest)), (fireworks background),  medium breasts, focused eyes, (((sleeveless red lolita with cross Wide Waistband))), (((red Crinoline))),barefoot

Model:

animePastelDream_softBakedVae, Version: v1.2.1

Size=544*728, batch=2, Steps=25

用時(shí)：4m 24.23sTorch active/reserved: 1505/2110 MiB, Sys VRAM: 3760/4096 MiB (91.8%)