#🍥|anime
1 messages · Page 157 of 1
hard one?
wouldn't say hard, just have to get creative with tags since it's not much of a done-before concept surprisingly
i mean, it took me some serious prompt engineering to even get possible
this is where I'm at rn
very aesthetic, high quality, [dryad], {{upper body}}, {subject focus, facing viewer, standing}, {{nature tree 1girl}}, branch arms, leaves on head, foliage, {{wood skin}}, wood woman, gorgeous, dynamic pose, brown and green skin
[anime:0.1], (anthromorphic:1.6) (tree:2.0), [(lewd:0.2):0.4] dryad
and that's for this one( actual upscale)
[anime:0.1], (anthromorphic:1.6) (tree:2.0), [(lewd:0.2):0.4] dryad
was trying to get that working too, but it's FCKN hard for real
puts on ad voice
well with the power of naid v3... 
and honestly, this is so much fun too!!! [anime:0.1], creative {absurdism|surrealism|realism|fantasy|symbolism|abstract| } , {2-3$$ BREAK $$ __songmaster__}
This Dryad looks lovely.
here's an interesting one
honestly never expected branch arms to work as well as it has 
damnit Eface you've sent me down a rabbit hole
@short coral nr2 tho 😮
oooo close
oh man, it's so close 😮
so cute tho
yeah
fun, pushing a model to it's limits 😄
much more fun than propmting stuff it'll handle for sure
I was looking at her stomach and noticed the lil branch veins
cute
@short coral you probably have a bit more power available than i do, probably
not to toot our own horn ofc, but that's something I've always liked about really flexible general models
which is why it makes me sad to see how much things have gone towards things that are less controllable
things in v3 you'd like to see different?
i mean, given the tags i see you throw in, it looks like it's handling those pretty well
when we were first playing with SDXL you'd enter a prompt and SDXL would sorta do what it wanted half the time
now it still does but in a way that enhances the prompt you send rather than trying to outdo it
LOL
I took out the girl too much
in my book, this is actually really good
it's a perfect anthromorphic tree
I'm playing around with not actually prompting for a girl and just bringing it out with other human-related features
that's the trick with the latent space -> you can't reference girl
or antyhing
it'll just warp the image to "generic waifu"
also put human in neg prompt
lmao, at least it got the branch arms
I love finding irregular tags that just "work"
ahh, I added looking at viewer and it brought back the face
yup, can't have that
looking at viewer = WAY overtuned at females imho
thumbs up?
not entirely sure, model was gettin a lil quirky w/ it I think since I didn't say a specific pose
aww
sorry for the low quality -> using the lcm lora so i have some speed in my generations
that's cool
poor guy is stuck to that tree
the men are actually pretty decent
oo
completely wholesome 
surely nothing unholy will happen to her
oh man 😮
I think I'm reaching a pretty good spot with this prompt
i'll just post the ones that ican show
my tree girl not turning out as cool
the one on the right is awesome
I think I have to use SDXL for such a thing
put "woman, female" in neg prompt (finally) and now these dudes starts showing up to wreck shit for real!
ok. this one IS really cool, even if not waifu
my dude be staring at ur waifu tree
oh my
holy shit
woahhhh
okay this prompt is sick when you flip it around to a guy
very aesthetic, [dryad male], man, {cowboy shot}, {subject focus, facing viewer, standing}, {{nature tree}}, branch arms, leaves on head, foliage, {{wood skin}}, wood man, wooden face, brown and [green] skin, vine veins, [[[looking at viewer]]], muscular
girls/female warp it to a VERY specific "wanted" outcome
DAMN! GOT IT!
ayyy nice
now, lets see if it still survives without the lora setup
this one was also really cool, forgot to send it
very cute!
v3 destined for great shit!
very different methods to get there, but we both got there!
awesome stuff @short coral
creative {absurdism|surrealism|realism|fantasy|symbolism|abstract| } ,[{2-3$$ BREAK $$ __songmaster__}|{2-3$$ BREAK $$ __songmaster__}]
if v3 handles wildcards...
no but we do have a random prompt generator releasing alongside v3, it's pretty solid
completely random prompt
1boy, sunburst background, full body, black eyes, very long hair, sidelocks, hip bones, animal on head, short jumpsuit, nervous, praying, {ribbon choker}, bare arms, armband
this one is better
it helps that the model is so good at creating something from anything
yeah, for me, personally, those don't fly. at all xD
throw in RANDOM bullshit, and then see what the model does!
sure gimme an example lol
creative , I got voices in my head again, tread carefully. And I don't medicate, it helps me temporarily I got problems, I got issues, yeah, apparently Trauma that I'm burying, I think I need some therapy I battle depression, I'm back with a message I'm asking the question that, if you hate me, why you acting obsessive? I'm past the point of no return, fuck being passive-aggressive I'll brandish a weapon, teach all you motherfuckers a lesson, yeah
haha, alright
that gets the really cool stuff, imho
this might be sketch tho, model is def best with tags
I put in some Anime artist names in v2 and was blown away by the style change--even lesser known names on booru. Very impressive. If v3 keeps that, then it'll be one of the best if not THE go-to.
yeah, my model was mixed to survive that onslaught of bullshit 😄
LMAO, about what I expected
oof, that's rough
yeah, seems your training data was not prepared for that kind of thing
designed around tags
can see if there's a bit more stable one in between them 🙂
does the anime tag help/do antyhing for v3??
good. shit. !
added our aesthetic tag to it 
damn ❤️
I'd watch/read that.
Let it shine! I will never give it up All the spirit is gonna let it fly And you know every tear is gonna dry And the moon will shimmer in the sky Shimmer shimmer for me! Running circles around what is mine I'm getting closer as I move in time Just shine all you want! Just shine all you can!
There's a day when all courage collapses, And our friends turn and leave us behind, Creatures of darkness will triumph, The sun won't rise
But this bitch ain't stacked in my favor We race towards the end of days And while you pave way for your savior I fear the rationality of the faithfully insane
same.
The translation though
, Will tomorrow ever come? Will I make it through the night?. Will there ever be a place for the broken in the light? Am I hurting, am I sad? Should I stay or should I go? I've forgotten how to tell, did I ever even know?
lmao
XD
v3 wants to gen text so bad it's crazy
The hiragana looks so good.
ie
it wants to soooooo bad
I'd be surprised if our next model version doesn't have good text
We're getting closer
seriously, the way the tech evolved just a year is insane
in 2 years, can't even imagine
falling behind
and i kept it to 1 verse...
creative , The best of intentions will not see the road paved. The end of illusions, who could ever be saved What's left behind in the storms that we've braved The troubles we find and the chances we waived BREAK An invisible heart with a body I can clearly see Gotta take oath to what I really want to believe in!
(BREAK is moving on to next set of 75 tokens, dunno how v3 would handle that)
but i just can't get enouhg of these kinds of images
lmao, sure ig
I will say the lack of punctuation probably effects it
creative absurdism, Our scars tell a story of who we are. We are the Warriors Raise your hands to the sky ready to light up the world tonight We live in the fault lines surrounding the Earth We are the Twilight Children of our Age, We are the product, the Final Stage
THE HORROR PROMPT
creative absurdism, I may look peaceful on the surface but underneath there is a torrent I’ll be the cause, you be the purpose 'cause all I feel now is dissent BREAK So we live small lives in the moment. doesn't mean we don't burn bright Accept no blame for our torment But we're the ones who need to fight BREAK Through the smoke and mirrors. we see clear as day and light the way Raise your hands to the sky We gonna light up the world tonight
this one was good 
anime, abstract creativity, BREAK (No) More forgiveness (No) I'll bring you death and pestilence I'll bring you down on my own I have returned
would probably be -> abstract creativity, (No) More forgiveness (No) I'll bring you death and pestilence I'll bring you down on my own I have returned
awesome stuff!
notable that coherency is definitely down w/o tags, but that's what happens when you try to mess with prose in a tag-purposed model ig 
I'm honestly surprised it's generating anything coherent at all so it's fine by me
^^ fair point
my model also sucks WITH tags
so yeah xD
i mean, it responds mildly at bast at some tags
but stuff like yandere -> no dice
but i'd take this over a yandere any day 😄
Your model really loves to give women lipstick. 
must be those lyrics that mention ketchup
Nah, in my own prompts. Even with neg prompting "makeup. lipstick."
nr3: [anime:0.1], creative , Now here we stand with their bloodKETCHUP on our hands We fought so hard, now can we understand? I'll break the seal of this curse if I possibly can For freedom of every man BREAK Can you hear the silence? Can you see the dark? Can you fix the broken? Can you feel, can you feel my heart?
just casually clipping her thumb through the camera
not a wrong hand, just clipping through
I think one of my favorite parts of v3 is being able to get that weird stereotypical anime lora look with just a tag
Nah, it's a reflective part that's just mirroring her pinkie right in front. 
i would be afraid. my first thought went to the cursed song.
Rip off this lace
That keeps me imprisoned
But beware of the enchantment
For my eroticism is your oblivion
Old love lies deep you said
Deeper shall be the wound between your legs
Within Temptation:
"Endless War"
You're fighting an endless war
Hunting a miracle
And when you're reaching for the stars
They just cut you down
Looking through the glass
Cannot recognize the ghost that you are seeing
Every step you take on the burning sand
Slowly sinking deeper
All the memories remain all the pain from your broken home
And the walls 'round your heart grew so strong
You can't let go
You keep crawling on
Don't wanna let it go
So you keep holding on
To feel whole
'Cause you're fighting an endless war
Hunting a miracle
And when you reach out for stars
They just cut you down
'Cause you're fighting an endless war
Hunting a miracle
Is it worth dying for?
Or are you blinded by, blinded by it all?
Here they come again
Hear them whispering when your heart is bleeding
Like a symphony playing on repeat
Unravelling your feelings
All the memories remain all the strain, you feel broken up
And the weight that you feel pulls you down
You can't let go
You keep crawling on
Don't wanna let it go
You keep holding on
To feel whole
'Cause you're fighting an endless war
Hunting a miracle
And when you reach out for stars
They just cut you down
'Cause you're fighting an endless war
Hunting a miracle
Is it worth dying for
Or are you blinded by, blinded by it all?
It's an endless war
But then they cut you down
They cut you down once more
Don't let them cut you down
Cut you down once more
this one is already in
THE cursed song is meganeko - milkshake
eye see what you did there!
aight, imma go to bed, cya later ❤️
lobsterpunk?
lobster western
it's hard to get a good consistent claw
G'night!
That hand tho. 
Yeah, it's pretty great. I went and downloaded some choice anime models to stack up alongside Eface's BPN, and am running comparisons with (relatively) simple prompts. So far, I'm very surprised at the level of non-standard SDXL I'm getting out of them. Some are very anime--like 1.x or better.
oh?
I tried nekoray but it seemed worse
The main problem I'm running into is cohesion issues. Not much stacks up to BPN.
Of course, I'm testing everything with upres. And face/person detailer
I think that I'm gonna order another 32 or 64gb ram when I buy my 2tb nvme m2 ssd
now open her mouth 👄
@untold glacier nah no money rain, will be like 250 dollars total. I've saved for awhile so even with my poverty life I can afford it 😛
tatsumaki
Hides RPG inside camera
i see a lot of video articles about using LCM with SDXL on comfyui, but anyone have used LCM with SD 1.5 on A1111?
how do you like the idea of making these sort of images with spectacular scenic view for calendars as well?
how much you paying? 😛
thanks. I wouldn't make it for free though, if I can make money then the little capitalist Chibi with a fat belly doubling as a wallet, will take this opportunity
nubby....
i like your previous kakigori version better
yess
not over, i like v2 better
more perky than the current one
yeah i still have it
but where would I sell em? Who makes calendars for people?
locals and online both maybe
perky = lively
try etsy for online
you could also hit up local shops for this idea
you know like thrift stores
they wouldn't know what to do with the idea, or care
basically its all your idea from creating arts to printing at press
and you give the shops a cut for sales / each
i like this idea... i might come up with a plan to sell
a thrift store wouldnt really stock this sort of thing
maybe there's some like you know those who make t-shirts
I dunno, but the problem is that legally, I don't own AI generated images
hmmm
good point
but there are also some of the mods that allow you to publish commercially
and printing tshirt is also another good idea
this economy is such a pos you'd have to go around a long way to get things done
what do you mean?
the EU laws say that telling an AI what to make, doesn't give you the right to it as your property as you didn't draw it or anything
you just told something instead of someone to draw it
this is the law, I don't agree with it
but this is it
so what's wrong with writing your ideas into prompts to create something unique with Ai as a tool?
dude I'm not the one you have to convince
lol ... but you get what im saying
for example that scenic view up there, i dont think there is any issues with printing in on a calendar or tshirt and sell them
if you create your own model, if you photoshop it after
etc
then you will own most of it
since it couldn't have been done without your sketch and pose
no but that's a different kind of issue with copyright
in order to obtain copyright from AI generated, you need to have become an integral part of the creation
no its not but anyone can use them after they see them
since I dont own them unless I do what I said
anyone can just reprint it
hmmmm that makes me think of NFTs
the laws are being contested and stuff and might change
its a new market
but if I sketch out some trees, a cliff or whatnot, maybe pose 3d models. If I create a lora out of photographs I've taken or licensed
you know how NFT would give you the rights over a creation
why do you say that? when it can give you rights on what you've uniquely produced?
NFTs give you the rights over a redirect address to a host. If the host changes, you have no more image
just an empty address
i mean the NFT market might be getting scammed but the concept of it is different
and it has no value
well...
im kinda new to all these new techy stuff.. but we are heading for web3 and blockchain soon enough
just because there's good tech behind all the bullshit, doesn't mean the bullshit is good
i think we are talking about two different things, the concept and the market
redditors when they think their american copyright law applies to the whole world
no the concept of an nft is also bullshit
in what context?
nft can't work w/o blockchain
yes, you buy the blockchain but there's no image directly attached to it
there's a redirect
but once that image disappears from that host
block chain allow you to reserve a unique token for your items right?
you own the token address, or chain or whatever
the image that's on it can change or disappear
its linked to you with your rights alone
blockchain itself is not hosted in a single host thats why its called a block chain aka theres multiple hosts for something to dissapear the whole chain would have to dissapear like peer2peer
and there is no disappearnce from block chain
the images aren't hosted on the chain
not usually anyway
and we all know chains can break rather quickly also
hold on .. lol we are getting into the whole component of it all
as an item of value, NFTs hold none
yea like when the us government broke the monero chain
blockchain nft web3
oh wait they didnt
that's not what I meant
block chain isn't fully implemented yet, far from it, and those who are using it are at an early stage
if a chain is abandoned because it goes flat in value, the potential hosts of anything are slim
if you're looking for that hosting
im under the impression when you put something up on web3 that stays forever and web3 is integral to blockchain
I also say that NFTs have no value and I stick by it. The value is too abstract, the market becomes so niche that it can't maintain
yea the only way to make money with nft's is either be famous and release some expensive bs or be rich buy some famous guy nft and flip it while praying ppl buy it and u dont lose money
yes, in other words you gotta screw people over
thats the only way NFTs can hold value
before they no longer do
case in point how they've all nose dived from 100k too 1 dollar
when they were new u could sell em because everybody was buying them but now hype is gone
why dont you think of NFT as unique signature to an item of yours? can we keep market value aside?
maybe in the future it will be like that
It's not an item though
that background imatge up on top?
it is an item
you created it but you are not reserving rights to it
thats an item too
if eface wrote that prompt to create it he can think of reserving rights as well
digital products are an item too
gun skins in games also dont do anything for u yet they are regulated and sold as real items
This isn't even kakigori. It's a merge that someone else made with Kakigori. Each version of Kakigori evolves with my current tastes at the time that I make them, so they will look different. I leave previous versions up so that people can use whichever one they like best. I'd get bored trying to make the same style on each new version.
you can assign emotional monetary value, but that's also an abstract
image can hold values by representing ideas, concepts of values
i saved v2 in my portable drive... just in case
no...not like that. Not so that a single address holds more value of ideas, concepts etc than another address of it that nobody paid for
youll never convince me a digital image like that holds monetary value
It's not going anywhere. As long as either HF or CivitAI are still running you'll be able to find it again if you ever need to.
cryonicus we are litereally discussing physical world and meta physical world values, they are about to merge
if you say so
this goes beyond into spirituality and mind
You're talking like we're nearly at singularity, lol.
lol
kind of we are .. we are headed there
Headed there and close-to are completely different things. It's a long way off.
some phiolophers are claiming we can achieve siguarlity by 2030 or near
yes after ww3
hope not .. no ww3
ww3 will set us back into old ages
nothing says ww3 has to be nuclear
it will likely be fought way differently though, cyber attacks. drones, robots
depending who wins we either become an ai technology dictadorship or a religious dictadorship
good grief
idk just gimme my sexy robots
we'll all be starving together
people are already starving kinda lol
its their own fault for not becoming warlords 😔
By the way, here's a preview of the big put-together I'll be posting tomorrow. Did an analysis and comparison on all SDXL models (only 9 made the cut) that can put together an actual, traditional anime style without western influences. Some neat finds.
cryptocoins are also basically like a light form of gambling unless it's locked in value. Since people don't hold them to do transactions etc, they hold them to sell them
if you wanted to only do transactions, then having a fixed value is beneficial
i should start mixing again to make a brightprotoanime 😮
Your BPN model made the cut as well, with a caveat. I'd love to see a pure anime model mixed by you though. That'd be awesome.
@wispy canopy earlier today I got into a rabbithole of the whole elemental spectrum alongside the tree people
I'm giving you the blame
everyone should have access to a button that generates one of these once it happens 😄
Detail
now that everything is calm u can hold my balloons
anyway...maybe I should sell calendars, dinner plates, glasses, backpacks, clothes etc with nice designs. Not all AI designs
Aight. I'm heading off for the night. Enjoy your convos and try not to party at the end of the world too hard now. Goodnight peeps.

i have a technical question ... should i even consider using LCM with sd 1.5 on a1111?
but I dunno where or how but still, whereever it's cheapest for above average quality and I have the biggest profit margin
the lora is an sdxl lora, dunno if there's an lcm lora for 1.5
oooh, that explains it then
and thanks, cause i was searching for guides to get it working with sd 1.5
anime, sexy robot
her legs are too sexy nubby
we need more armpits here
that's why she covered them up 
she can't cover it up .. its visible
all the right bends
lol... dont let censoring ruin natural expressions
@native halo these the robots you're looking for?
yes i wanna be crushed by skynet 😍
im pretty sure the way some of the AI chatbot interacts with people, if they were put into synthetic bodies people will fall in love with them
I'll be surprised if there aren't aren't reported cases of that even if they aren't ever put into a synthetic body.
only if they look like the girls from chilloutmix
well there are some reports on people falling for dolls
i think i read somewhere some people actually married life size dolls
and given how Ai is progressing and are capable of interacting on emotional level that will add more depth
wow .. the updated ab blocker that im using, i see youtube trying tp show ads while video is streaming, and the adblock just blocks it and i also dont see google not blocking with timed message to disable ad blocker anymore
I think you may be overestimating how difficult it is for robotics to perform the tiniest of motor skills and do it as naturally as a human being
maybe at current stage
google had to stop doing that cause they were breaking EU law
what EU law was that, sounds sweet btw
forget exactly, but it has to do with privacy and freedom to repair and modify etc
i like it that made google stop the annoying block message
the ad blockers im using is pretty good altho its free
I never use sponsors and they get nothing from me watching it
same
so I use an addon that skips em on youtube since they have to be marked as "ad"
They haven't made a ruling or taken any action to stop doing it. It may be illegal under EU law.
i guess they understood they could get sued
well I don't receive one, haven't for days
or over a week
or two
it was for a short while then it stopped
i notice the goodl changes the past few days
it literally had timed block message .. you'd have to wait it out before you could resume playing
https://sponsor.ajay.app/ this is another one I use
this is what im using works universally for all the ads https://getadblock.com/en/
It doesn't work for sponsors that are integrated into the video
not ads that run, but actually in the mediafile sponsor
i see some sponsored ads try to show up and they get blocked insntantly
those that are built into the video, hosted by the video creator?
like "check out our sponsor SKILLSHARE!" or "It's RAID SHADOW LEGENDS!"
i think those are what google tries to show in the middle of a video playing, not creator embedded
no they are not
although i dont usually play any videos where the author puts in a bunch of ads
they dont put in a bunch of ads
they have a sponsor for the video
that pays them
don't tell me you've never seen a creator with a sponsor?
i see but i dont bump into them with the ad blocker im using
I ran adblock and it never skipped them for me
odd
if you started watching it say 4:40, it skips the sponsor=
in 2023 people are still using adblockers that aren't ublock origin?
it's not an ad
not in the traditional sense
it's not google adsense
so adblock skipped it in its entirety? regular adblock?
starts off here
thats the beginning
I made the second link for a reason
though maybe it doesnt wor
https://youtu.be/8q_HpSq95bw?t=282 now it should
cryonicus is talking about skipping the actual sponsors sections within videos, like vanced/revanced can do.
that was from clicking the seond link btw
click that and wait 10 seconds or so
sponsor should come on
the new link that is
Rocket Money sponsor should come on
yep
how would ad block understand an ad when its embeded into the video
i see a guy talking about money ad
but thats not external
.......................
I give up, if you don't want something that works on it
that's on you
pretty sure they used crowdsourced timestamps basically
no they use youtube timestamps
any sponsor or ad has to be marked
in the video
sponsorskip looks for that
and skips
i think those timestamps are yellow ones?
try it? ^^
well if youre using a supported broswer
chromium, firefox, safari, edge,
sweet .. it skipped also see how it marked it with green bar
thanks lol.. nice one
this is now a thing
np
that's a lot of hair...must have gained a hell of a power up
undeniably
better space images than my space witches lol
@
looks very cute, is that new model of yours?
it's a merge that someone else made with Kakigori
interesting, doesnt look anything like kakigori tho
characters are very different
thanks
this image you posted now and the profile images on the civitai links are also very different
Yeah, the kakigori parts are pretty unrecognizable. I wish they had included the names of the other models they used for it, there are probably some other interesting ones in there.
oh i see what they did, they merged your v3
yeah, they used kakigori_v3, but I have no idea what else they used.
btw @cyan burrow the concept of copyright content is outrageous, its linked to the scarcity driven monetary economy, considering we achieve singularity as big thinkiers in science coumminty are striving for, ideas have to be sharable as common heritage to everyone who exist
I'm a capitalist, as I said
capitalism is unsustainable
it's been doing fine for 10,000 years or so
its been in pratice that long and heading for a big collapse
it can't keep up with technological revolutions and even besides that its wrecking havoc ecologically
well good luck coming up with a working alternative
cryo that's a challenge for everyone
we have to come up with a sustainable solution
Who's we? I don't have the energy for this, especially when I don't agree with you. There's always something wrecking havoc somewhere every generation and it's always the biggest thing ever and nobody has ever seen or experienced anything worse
id say canibalize redditors like him
I'm too old and too jaded and too fatigued from the past 3 years to care
we as the whole society
yeah i know what you mean steven.. some people may not live up that long for changes to take place
yeah that's not gonna happen @nova remnant , there's too many differing cultures with different values and ways of life, tradition, religion and so on. You'll never find a universal system that speaks as neutral as capitalism
its ok dont worry we will always remember you as a pillar who helped to forge humanity way upwards so i salute u for that
🫡
aye aye
exactly. thats the challenge we all face
as we move towards communism™ (the real one this time, not the other two dozen attempts)
we don't face it cause I don't agree with you
even if we do or not its not like u will see it
hey smol sus where have you been?
ohh..didnt seen you in a while, wondering if you got too busy with real life errand
now this is the type of robot id pay for 🤠
we have human beings for that kind of attraction 🙂 we often overlook that cause of negative social conditions
did you just start psychology 101 in high school or college @nova remnant ?
right
i graduated at trump university
there is always some conflict going on somewhere that doeesnt mean tho we have to accept that
I'd say 95% of all activism is about self-indulgence and peer acceptance
a lot of the problems derive from economic model we have
and there is little to say about that unless there is technological abundance to automate most of human tasks
or we agree internationally to slavery
there is more to that, human task are prone to error that can be overcome through mechanical precision
dont pop the balloons
talking about socialproblems in this channel? interesting lol
some thoughts, it started out with copyright policies
yeah, i pretty much feel the same.
but try to understand, most people live like zombies, and let things be even when they are about to die, without opposing a real resistance
just a minority can't live accepting to let their soul rot, living without really improving what they don't like
most of the people just wake up one day, realizing they have wasted their life and than returning dormient, by the shock
well asking people for social change isn't gonna work i dont go there, just a waste of energy, but i see good changes developing by itself technoogically that will reshape society
nah, means change nothing.
people will just use tech to consolidate their life style
only the will cange really things
lets put it this way, society by majoirty is incapble of making changes, the changes will come throught technology and Ai is a step towards that
how cute is this ....
i just want my sexy robots
playing in the snow with my family
You can't assume people are just living with what they don't like. Just because you don't like it. I guess they're living with what you don't like ^^
i'm talking about what i saw during my life.
i saw many times the light in the eyes of the people perish, dragging themselves in a meaningless life
many were people dear to me.
both of you have legitimate perspective
its easy to differ becaue everyone has their own takes in life, and none of them should be invalidated
yeah, the point is that everyone have his values and priority
look the gens, everyone have his preferences
the problems we face or willing to admit are more systemic than humanistic
people can truly put their heart in just one thing.
more often than ever, such things are meaningless vices like smoking, or drinking
the only systemic thing in this, is that the actual society, induces people in such vices, to squize them dry
kind of
i see that as flawed economic model that creates artificial scarcity and pushes eco destructive consumerism
torqx, is even worse than this, but i don't think this is the right place to talk about it
im not talking about it just replying to you
so, let's close this question her, or talk it in pm
yeah lets do that... not a very happy topic for discussion anyway besides one can't make the change by himself
yep
background looks good!
guys, did you know about lcm lora?
What's LCM?
Yes, you read it right - a LoRA which actually helps you generate images FASTER! In just 1 SECOND (4 steps) you can generate a 1024x1024 image, PLUS - being a LoRA - it works with any model, IMG2IMG, Inpainting, ControlNet, etc, etc!
Works with stable diffusion 1.5, SDXL and SSD 1B.
Enjoy :)
Huggingface Blog Post:
https://huggingface.co/blog/...
Latent Consistency Models (LCMs) have achieved impressive performance in accelerating text-to-image generative tasks, producing high-quality images with minimal inference steps. LCMs are distilled from pre-trained latent diffusion models (LDMs), requiring only ~32 A100 GPU training hours.
does that work with SD 1.5? 👀
it says it does work on a1111 and sd 1.5, so i just download the lcm lora, place it in lora folder and put that in positive prompt and render as usual?
nah
LCM-LoRA is supported in 🤗 Hugging Face Diffusers library from version v0.23.0 onwards. To run the model, first install the latest version of the Diffusers library as well as peft, accelerate and transformers. audio dataset from the Hugging Face Hub:
The adapter can be loaded with SDv1-5 or deviratives. Here we use Lykon/dreamshaper-7. Next, the scheduler needs to be changed to LCMScheduler and we can reduce the number of inference steps to just 2 to 8 steps. Please make sure to either disable guidance_scale or use values between 1.0 and 2.0.
can you share that link please?
thanks
going to read up and follow to instruction to install it
it should work with all the models too, right?
just got this updated
pip install --upgrade pip
pip install --upgrade diffusers transformers accelerate peft
transformers accelerate are not for win
idk i got these ...
Successfully installed accelerate-0.24.1 diffusers-0.23.0 huggingface-hub-0.17.3 peft-0.6.1 psutil-5.9.6 safetensors-0.4.0 tokenizers-0.14.1 transformers-4.35.0
got the lora downloaded too, time to test
this is the file i got pytorch_lora_weights.safetensors
when using hires fix .. do i set its value to 0 ?
cause im gonna use samplign steps to 10 now with that lora
:/
dont think that lora is working right
is it?
that hot atm?
It's bad
Just a constant 30+ c
Even at midnight
And my house doesn't have the best time cooling off the rooms
Sum old stuff
so weird, when I try to generate SDXL in auto1111, it takes forever. In comfyui, almost instantly 😮
with my 3070 ti 8gb
using upscaling and LoRAs?
sounds strange then, cause i use 3060 rtx and it takes me about 40s to render with just the basic and adetailer
btw i only have --xformers in webui bat file
no vram tags
me too
I'm so sick of SDXL....either way though, when 4070 Super comes out, if they stay true to rumors and it costs the same as a 4070 atm....I will probably upgrade to that 😛
then I can use some more 3d software too, get great performance in games etc
im staying away from sdxl cause of its poor anime quality
and yes lack of community inputs
I like that it understands prompts better
thats the only reason i kept coming back to it
but i also realized its pretty cool to just use booru tags on sd 1.5
I just work on improving my current 1.5 model, it kind of understands natural language
the reason i like booru tags cause it can give you wide room of putting in ideas and have creative outputs, and you could also incorporate controlnet for more precise results
yeah, but i feel that tha precision is somehow limitating 😦
using both is better
well, after all i prefer be surprised by the results lol
maybe the booru prompting is good to make retouches
so here is an example... i wrote up a prompt booru tag style but also descriptive fo the elements i want in the output and then i chose to load up a random photo in controlnet and used the depth feature to creete images based on my tags while the photo in controlnet is quite differnt with the background
ok I think it kinda generates now, way different from sd 1.5
they look like the eyes of frodo of lord of the ring
yes
this is the same in 1.5
weird though, I can't see the Lora's I'm supposed to see, probably a setting
the stuff they are creating on civitai page with such limiting animatediff, this evolving tech will blow the minds in few short years
yep
we will get very high detailed assets in a click
with enough time, even making entire high detailed visive simulations, will be possible
like the difference between dos and windows
a totally new way to interact with data
yeah, im imagining all the possibilities and also sugar and spice
this may not seem obvious yet due to physics nature of limitations, but Ai have the potential to create utopia
found something else about lcm
Note that LCMs are a completely different class of models than Stable Diffusion, and the only available checkpoint currently is LCM_Dreamshaper_v7. Due to this, this implementation uses the diffusers library, and not Comfy's own model loading mechanism.
i think i also saw some ppl talk about any models that work with lcm
also lcm sampler is not required
its fresh, still under testing by users i guess
Basically you have to tune the steps/model/cfg
But once you get it dialed in, the speed gained VS quality loss makes up for it bigtime
can you give an idea about steps and cfg and how do you setup hires fix ?
In comfy or a1111?
a1111
I'm using lora strength 0.25, cfg2, dpm2s a karras
basically i use 30 steps for samplers, so i could lower that to say 4? and what should be the hires fix ?
when i use the lcm, i can run 8 steps
and you skip hires fix?
i can use it, if i want
and then highres as 4 steps for example
this is 8 steps (testing a new merge already lol xD
im not getting good results
gonna use 8 sampling steps, cfg 2, hires 4
and ofc with lora:pytorch_lora_weights:1
wait
which lora would that be?
cartoon, drawn by __artists/unfiled/comicbooks__, epic action scene, <lora:LCM_LoRA_Weights:0.25> this is all i have
yeah that might be it
weight 1 is prolly too highg and over baking it
this is with weight 0.4
k trying one now
good gosh
i mean the overall quality could be better.. but this is with 8 sampling steps, 4 hires steps, cfg 2
yeah, that's the thing
taking out lora:pytorch_lora_weights:0.25 and see what i get
quality does suffer a bit, but the speed gain is just... you can hardly argue with that
all the same settings but no lcm lora
looking at these 2 images i just posted ...
the one with lora looks richer in quality
has more detail
look at the face
a lot more is resolved, so yeah, it does speed up generation
lora:pytorch_lora_weights:0.3.. sampling steps 15, hires steps 5, cfg 3, while also using adetailer
is there a technical reason to keep sampling steps down to 8?
guidance_scale is cfg?
Please make sure to either disable guidance_scale or use values between 1.0 and 2.0.
because that's how the speedup happens -> you need less steps
less steps for a 'similar' result
cfg 1, steps 8, hires 2
missing a lot of detail tho
oh and used controlnet depth for it as well
w/o controlnet
found this to be a fair balance ...
sampling steps 8
hires steps 4
cfg 2
LoRA weight 0.25
@nova remnant should use this LCM
What is that
And why should I 
a lot faster image gen
good thing i can set custom steps and cfg for adetailer
is it bad in any way
shouldn't be bad, altho i had to update a few files but that went ok no bugs
epic ... for so little steps
i wonder about this LoRA if this has anything to do with color consistency
no BREAK here
1girl. dark red hair. blue eyes. sitting next to a pool. dipping her feet into the water. wearing a white crop top and blue skirt. bright sun. blue sky. <lora:pytorch_lora_weights:0.25>
yet it maintains the right colors for different area as prompted
consistently w/o missing one render
i had to use BREAK before for this prompt portrait of a girl, red hair, yellow eyes, dynamic expression, expressive mood, maroon lipstick, short black dress, bare shoulders
but not now
and this with the model i used before to be using break but not now
somewhere something changed .. im guesing it could be the LoRA
i say that praisingly
besides using this LCM LoRA i updated > pip install --upgrade diffusers transformers accelerate peft
not a single BREAK
that's LCM? looks great
yeah
i just changed my default settings for config json file to use lowe steps
i have no clue how color context improved this much
didnt change any extension
portrait of a girl, silver hair, red eyes, dynamic expression, expressive mood, pink lipstick, short black dress, bare shoulders.
for me, LCM quality really suffers (without LCM vs with LCM)
will play with # of steps
Safest place outside of earth
what steps are you using?
oh ok here we go
first is 8 steps dpm++2m karras, then highres phase is 12 steps
41s vs 73s
you should keep hires to half of sampling steps
yeah? will try
this is my settings for those images i posted
vae taking a lot of time overall
I don't think 4 highres steps is enough, will try more with fixed seed
let me have your prompt so i can run a comparison with my settings
a statue of a cat made out of coal, cracked, textured, depth of field, camera RAW and neg blurry, lowres, boring
lowres because I don't want a lowres output?
its not that, the idea behind LCM is speed with lower steps
4, 8, 12 highres steps
if you use higher steps you'll over cook
usually rec sampling steps for LCM is 2-8
i use 8
ok im gonna try your prompt but you didnt answer why the lowres
I did
oh
fi you dont want lowres that goes in negative not positive
you put that all in one line lol
separated by code formatting 😛
can you give me the positive only?
a statue of a cat made out of coal, cracked, textured, depth of field, camera RAW
the character of the gen is different too
what the model would do at 20 steps vs 8 lcm steps
I asked for quartz
though maybe there's more at play, I think the right is actually closer to real quartz
ah the CFG difference
definitely still a neat technique though
make sure to either disable guidance_scale or use values between 1.0 and 2.0.
0.25
ahh yes thats a good number
I think this will have a niche in realtime img2img where you paint + get a gen on the right
then, you click "upscale" and it refines without LCM
i kinda let adetailer run with custom steps
i keep both steps and cfg for adetailer seperate from prompt
my gens don't have a human face
yeah i see that, but im just saying you will need to do that when you render humans
adetailer works with SDXL too if you wanna run a few comparisons
lgtm
thing is I have to run sdxl on comfy and don't know how to adetailer there yet 😄
👀 what do you think im posting so far
lol the 1.5 speed is absurdly fast
yeah i like it
SDXL Anime Model Comparison V2
- V2 Updated with the biggest and best models, Blue Pencil, Kohaku, and HimawariMix. Also newly tested was OOO and RealisticFreedom. Here's the new half-size comparison grid preview: #🍥|anime message
I've put together a comparison of all of the most "anime"-like SDXL models currently out on Civitai. The models were curated by finding and choosing those which had the closest to traditional pop-anime styles in their thumbnails.
The comparison is a grid of 9 prompts and settings of various qualities along the Y axis (each prompt is on a separate seed.) The models (along the X axis) number 14, and every image is a standard SDXL resolution with a 2x upres and face/person detailer applied depending on the prompt. All images were genned with the unaestheticXL Sky 3.1 negative embedding.
I recommend viewing the downloadable full file--as that's the only way you'll see jpeg artifacts and other imperfections with some models. Be warned, this is an extremely large file and will likely need to be viewed in an app that can handle it, like Windows Photos: https://drive.google.com/file/d/1sI97oyvK5pnn49BtQI-N99e3XpjYOLu-/view?usp=drive_link
If you're tagged, you're part of a list of people who have recently expressed an interest or concern with SDXL anime models before, and I thought you might find this useful: @stable crane @nova remnant @wispy canopy @keen hollow @iron star @merry tendon @nova remnant @nova remnant @near elbow @slate narwhal
Findings here: #🍥|anime message
My findings:
Anime Model Ranking:
✅ 1. Blue Pencil v1.0 (Next level anime model.)
- Even with the squished face problem, this model pops: great detail, cohesion, and prompt following. Super aesthetically pleasing.
✅ 2. Nekoray (Best Flat 2D Anime Model) - Excellent cohesion / prompt following. Slightly less contrasty than the rest, but sheer cohesion, aesthetic creativity, and prompt following more than make up for it.
✅ 3. Kohaku Beta 7 Pro (The flattest and best at following prompts.) - The hard flat sketch style is great.
✅ 4. HiwamariMix (Most non-squished faces. Closest to 1.x.)
- BrightProtoNuke 1.2 (Best Generic Model. Can be very difficult to get a good anime style, but when you can, it's fantastic.)
- EndlessCoatingMix (Great flat 2D, but a too haphazard sketch style in every gen.)
- BreakDomain v05g (Cohesion issues, but creative.)
- NekoXL Nsfw Mix v1a (See special note at end.)
❌
9. RealisticFreedom SFW Alpha (Good following the prompt, but has trouble sticking with an anime style and cohesion problems.)
10. Endless Bard_Mix Mix (The same as Endless Coating, but with an inability to gen eyes properly and a profound lack of cohesion.)
11. SDVN8-ArtXL Base (Too non-anime stylized and incoherent.)
- Don't use this unless you want to fight against the model to try and make anything anime.
- AnimagineXL v1 (Very messy. Very incoherent.)
- OOO (Jpeg artifacts, low cohesion, and super messy.)
- SDXL
Special note: NekoXL Nsfw Mix v1a has major jpeg artifact issues and can veer away from your prompts hard. With that being said, it is quite creative, has an amazing anime style, and is very aesthetically pleasing (see the examples.) It's likely best used as a base gen model with a different model for the upres to bypass the quality issues if you can get past the prompt veer.
- FYI: Cohesion/Coherance: The ability to generate an image without mangled/missing limbs, not fully-formed objects, etc.
stylistically, I dig Endless Bard a lot
how about blue pencil?
Yeah, I love many aspects of what it is, but it has a hardness to it that is difficult to escape, along with eye issues and a lack of cohesion in the end resulting gens some of the time.
I looked at and tested the SDXL anime models on Civitai that I felt could accurately and properly represent a true and traditional "anime" style without any western influences. When I looked at Blue Pencil, every face was far too vertically squashed in the example images for my liking. So it was disqualified before the comparison.









