finite zinc
Not much to add, except it tends to do nsfw images (despite being almost no nsfw pics in the dataset), so overload the prompt with clothing and sfw tag, and the negative with nsfw in you don't want any surprise. Overall this one turned out much better than Nami and Robin, probably because the dataset was bigger and had far less variance.
UPDATE: 7000 version #1060257969704546325 message
lime osprey
Embedd seems quite horny but so am i so it's not a problem. Works great with realistic models,maybe the collar pattern show a tad too much, used the 5500 one.Horns are hit or miss but that's to be expected.Keep these sweet one piece embedds coming, doing lord's work
finite zinc
yeah, I'll train some more focusing on the horns
finite zinc
Embedd seems quite horny but so am i so it's not a problem. Works great with rea...
let me know if this one is better:
lime osprey
let me know if this one is better:
i'll try it later with the new nami one,I think it might be the model and not the embedding because it clearly gets them reliably with more anime-esque models. Thanks for your work
finite zinc
Yeah, it makes sense. Not many people in real life have horns 😄
lime osprey
Yeah, it makes sense. Not many people in real life have horns 😄
i mean metaphorically a lot do. Real life is a big NTR doujinshi
guess the Ai hasn't caught up to inter personal relationships
lime osprey
I'll drop them here for science. It seems the new yamato does the horns better and picks up her hair color pattern more reliably, the new nami understands that tattoo should be just on one arm. Yamato still has the clothing problem, new nami has some weird eyes and looks more anime-ish,might be mitigated with lower cfg or lower values in the prompt
old yamato is more prone to nsfw compared to the new one
they both have their pros for using newer ones
nami surely will help with impainting he right tattoo on a smaller area
Are you referring to the one i used or the one Lykon used?
finite zinc
I'll drop them here for science. It seems the new yamato does the horns better a...
wow nice!
the embedding here is yamato_5000 or yamato_7000. That one in the negative prompt is an embedding you find around that should improve the final results. Has nothing to do with nsfw.
finite zinc
I'll drop them here for science. It seems the new yamato does the horns better a...
Forgot to ask, what's this model anyway?
lime osprey
Forgot to ask, what's this model anyway?
Anyv3Ga+ElysiumV3 with a sakimichan HN at 0.4 str.
I found that any semi-realistic HN will work, another user uses another one with the same model and prompts and results come out a little different but still comparable
As i use it more i start finding some limitations with the model...
I'd suggest you try with Luna's model which seems far more reliable with less prompt black magic
look here for link and what it can do with anime embedds
still need to try it but it mimics the effect without the prompts and more reliably and with a little bit more detail
for anyv3ga+elysiumv3 you can find it here
magnet:?xt=urn:btih:d4d5343b1ee8bc11a0b284c2bdc56d9bf1770153&dn=models%20safetensors
for prompts just check my images (not grids) and play with them
really great with closeup portrait, a bit iffy with anything where the face isn't 60% of the image
finite zinc
nice
finite zinc
Are you working on some new embeddings?
yes, Ahegao. Also where can I get the sakimichan hypernetwork?
lime osprey
Here
I saw an ahegao embedd on civitai but it was...lets say funny lol
curious to see how it comes out
i reckon also a wlop HN/embedd could work btw
you just need that little help to make it more cohesive
finite zinc
Here
thanks, I only have to put it into the hypernetworks folder and just use it like if it was a TI, right?
finite zinc
curious to see how it comes out
same
