//=time() ?>
現状、pytorch-CycleGAN-and-pix2pixも万能ではなく、SDXL+自作ControlNetの方が優秀ととれる場面もあった
一枚目が自作ControlNet、二枚目がpytorch-CycleGAN-and-pix2pix
これは線画から疑似3D画像に変換するタスク
I haven't drawn it very often, but the dark Mai, along with the marks on the skin are an effect of the Azusa arc. She hasn't used it in the novel yet but it's something she doesn't control well.
Stable Diffusion の実験中
Control Net の「Reference_Only」が優秀すぎたので共有
元絵を参照して生成するというもので試しに過去絵でやってみたけどLoraを作らずとも結構近い絵が生成された
差分とかの案出しに活用できそうだけど元絵より上手い絵が数秒で出てくると心折れそうになるな…
New pvz oc dropped!
Presenting....Raptus, a Tangle Kelp controlling the corpse of a Sunflower.
良すぎて…
some said if jiggy succeeded to control dead-NMJ, he may put stuff not to identify Chifengzun and go out https://t.co/fX7LbAYs3S
👀 OOOHHH? what's this?
did you know my doll vtubers could use controllers??
Look familiar?🤭
Also, I'm having so much trouble with this tablet, lol.
I need to keep reinstalling drivers because it keeps stopping to register. My pen control is awful, and my hands are cramping drawing on this small surface( I use my wrist to draw).
@exedlore Kolath will summon her "inifnite" model and destroy the planet with her hands if needed. Then will consume the closest star. The single model Nine can control gravity. She doesn't need a real "finisher"
素出しした左の画像をsoftedge_pidinet + bdsqlsz_controlllite_xl_softedgeで右にする。
左のラノベの表紙感も良いが、右の闇っぽい剣を描きたかったけど途中で力尽きてまぁこれでええやろ感も良い……
Commission for @Obsidium2 and @Rendal22 .
I guess foot pump is good tool to control. 🤔
@meeSOPsArt You CAN use liquify but I find it hard to control, so what I do is put the pattern over the area I want to cover, click on Mesh Transform (on edit menu), add the number of points I think are necessary and go wrapping them around trying to make it look 3D. 👍
Welp here it is. Happy Valentines Day
It was supposed to be happier but stuff happens and it's out of my control
Whether you support them or not, I mainly made this for the lewds, so look at the lewder variations out on my alt @Cipher6969_
#Artsuki #絵リーラ #pendorART
ControlNet用のデータセットを改善したら露骨に結果がよくなったな
1枚目:Line2Normalmap(データセット改修前)5000epoch
2枚目:Canny2Normalmap(データセット改修後)5000epoch
目視によるデータセットチェックって大事かも(cannyにしたのが影響大きいかもだけど
うおおおお!!!!
pix2pix(50epoch)微妙だった!!!!!
生成AI(ControlNet)の方が多分良い!!!!!
試してみないとわかんないものだな・・・!!!
@__Rudeus Oh my goddddddd
I must control myself