#Sink Artstyle.
164 messages · Page 1 of 1 (latest)
it's really hard to post consistently when I can only replicate the artstyle perfectly 1/50 times.

hm
this is just a stunt to get more instagram views
idk if screaping is possible on insta
No I just want to be able to replicate the eyes and hair texture in the posts.
don't rember the model? which eyes do you want to look like?
It's really hard to get this kind of stylized texture when randomly prompting.
Post everything on aibooru and I'll do it
okay
I can't scrape IG because no account so if you give me a zip dataset I'll do it
sink didnt you make the mika plaso lora
I did
thats thing blows all my loras out of the water
i'll do it when i have time tho since i'm moving into a new appartement in a bit
I'm busy with other work stuff I can't spend hours on making LoRA's 
nah i cant even right click and dl the image... bro insta gram xd
luckly im 18, unemplyoed, and have time to spare
hol up, let me just zip all the files and post them here.
unlucky, im arse at making loras
I'll do a LoRA if you gib zip
gud
if you send me now I can train while sleeping
yeah give me like 3 mins
noice
there, I posted it as a zip file.
HAH
I can train LoCon at 768x if you want
What would be a good checkpoint to train this on?
your choice really, it's a semi-realistic style so you should be able to train it on whatever
since the model doesn't have to do THAT much learning
I'll do anylora then
sure, gl
letitsinkin style
lemme put mksks style
Some sample images
token is letitsinkin but samples are never good so idk
I have no idea how good it is
I'll let you try because I'm off to bed for real this time
goodnight my lovely laptop
@spice thicket can you try this out ?
I'm in bed rn 
It has been 5 minutes and you posted nothing 
/j
You were supposed to be asleep 
True true

Seems pretty good?
that jamexX90 kid is a meanie
ok
also the locon blew up on me
its generating non... things
💀
nvm
: ?
thats just my model going haywire
bro... why xd
nah 
hold up
my colab is bing fcky
bing chilling
ok ok
so my colab just died
and killed all the models
and none of the work
ugh
idk what your prompts look like, but i think a realism anime model would be good for this rather than pure realism ones
im using a super simple prompt
The model in this image is literally Base NAI and it's a Kohya Sample image :p

No I haven't actually used the model in the WebUI at all rn
Will boot it up in a while once I'm finished with work.
i mean for training
Yeah
what work 
College Projects, Social Outreach etc
Maybe you have tiling enabled
You know the thing for textures that no one use
okay
I've tested mine and it's bad since I tried something new
:p
anyone got some new ideas ?

as long as it's not overcooked and it's slightly resembles it. I'm okay with that
berry cute

Update : I retried it myself, It worked to an extent till I realized the quality and quantity of the images in the dataset are too varied for me to train a consistent style so I'm not going to work on this further.
Therefore this thread is closed..


glad to have helped 
what model was this trained on?
nai
try another model maybe
like anylora
since ur style was more towards 2.5d
iirc
have you tested the LoRA ?
I wouldn't share it If I didn't think it was good lol
it works fine
ah ok
No I didnt
o

lol did sink ever try it?
Sink artstyle is trash, it does not include sinks in the picture after generating it
Didn't have the budget to make sinks every time
Basically
put 1000 sink images
tag it
with
sink artstyle
Would appreciate donations of sinks
go train
Are you trying to turn us into https://civitai.com/user/uiouiouio
I'll be trying to train this again If I can get some of my old images for a better quality dataset.




