#šļ½general-with-images
1 messages Ā· Page 126 of 1
Itās smart
You donāt really have to do much to let it construct stuff that you want.
well, if you really want to take control, you do
but it's certainly a lot of fun to just throw stuff together like this
Yup
same spiral pattern combined with this car gives this
The main intention is containt, and then merging image two to the constraint
Like video mapping
What model is this?
juggernaut
Bro I remember when AI art generators used to be straight crap when it came to prompting
that's not really "real" juggernaut though cuz i used ipadapter which patches the model weights on the fly
preciate you bro
Oh I see, where do you use the model on?
ComfyUI
I use a juggernaut model version on MonAI. That app has been getting hot lately
gotcha
yeah i run local
i'm on a 4090 / 5950x system with 64gb ram
local is where it gets really exciting
Damn bro ik you're prepared to snag SD3 any moment now with that much GPU power
hell yeah
second the 5090 comes out, if it's got 32gb or more vram, i'm buying that too and setting up another
@festivalman just bought a fleet of 4090s so he can pump out silly images even faster š
most ppl on here are running local
Shiii bro y'all doing serious business
Is it possible to reference API model work from Stability AI, or it's impossible to do it without them sharing actual internal information @nimble mason
idk, haven't looked into it at all
I was thinking if I could build an AI art generator referenced from DALL-E 3 and SD3 combine. I know it will take a long time and a lot of money but yk I'm willing
I might have to get to know people that know people
Sheesh that's scary realistic
Training a bigass lora is the way to go
Only problem is it takes 15 FUCKING HOURS
I'm going to go insane
Ok team⦠My mission for the night is to generate what an alien abduction would look like.
is your gpu low vram or something
Never try to install comfyui from scratch on 2 machines at the same time. It's pretty clear those servers will throttle you. currently getting 15k/second on a gigabit line.
Letās GO son. I get to go home in an hour⦠and literal alien abduction is whatās gonna start spitting out in this channel.
So my bad people
Im trying to go mildly dark, bad acid trip.
holy hell
good to know
THAT makes me happy.
I was trying to grow faces from tree bark earlier
I got real faces⦠on treesā¦
Where you're going, I can't follow, but godspeed.
lol alien abduction?
Better value
Just buy a datacenter.
Yeah true
Itās gonna be fun
Fuck it. Buy nVidia.
test render on one of the new nodes with a cascade clipvision. muscle man, with a prompt of an iphone that's charging. I like the placement of the magnetic charger it did.
@eternal comet what you said about Lora is true but find a way to combine multiple models together like Stability AI did
not the cat you asked for, but if you know anything about clownshark style........... š
Sora AI?
Sora? Hell no
Lmao
Gorgeous system⦠but someone is smoking if Iām gonna take 2 days to render poop resolution
I believe that they are improving it in more areas
Ok ok š
Not too hard
I can do that
Combining checkpoints
I read thatā¦. And noped it. Siiiiiiiick looking system for someone that wants to paint in 2 second increments a few times a week.
How much money you need to make it happen will vary, but it's worth it. Research and grab the best models out there
That CANāT be real
Lmao
i do it on the fly all the time
there's even a comfyui node for it built into the software called "Merge Simple" and a few others
8 hour render times are kinda my breaking point.
i need another system badly
That's INSANE
i have a 3080 12gb just sitting on the shelf but that will prolly just be a backup inference card whenever i build a home for that
Old animation stuff took forever..
We thought shiney orbs were king shi* of fu** mountain
I needa look more into that
Nah facts tho!
Reflections? GOD TIERā¦. Nowā¦. āPaint me a midget riding shrekā
I wouldn't mind merging an sdxl checkpoint with a 1 gig lora that andrea75c put out. will that comfyui node do that as well?
yeah, you can add i think
i think that's the way... it's been a while but iv'e done it
So I get my giggles out of pushing a system to see what the future is. Right now, thats animations at quality.
yeah, consistency in composition in particular
if you happen to remember the node, drop me a line, no worries if not.
You humor is hilarious but you speaking straight facts
I love the 3d cute render lora he has, it looks amazing, but realistically it drops the prompt adherence by a noticeable amount.
Thatās what makes it great lol
i think it was just ModelMergeAdd
should be built into base comfyui
and it's sloooow loading a 1 gig lora every time.
are you generating tons all at once?
like large batches? or flipping through a bunch of loras
i guess i would be if i made it available on the discord thing.
i think i remember noticing a1111 was loading and unloading loras with every generation, i don't think comfy does that
even with a bunch of boxes rendering, because I'm doing light upscaling on every image, I forgot how long things still take. š
i am going to take a dump in my pants over this new ipadapter
People are freaking out about SD being pulled back
I wasn't paying attention to the times too much because I'd submit and come back later. when I'm waiting for it, then seconds equal minutes etc.
Sorry⦠thinking it will be
I use Proteus as base Loras as quality increases
gotcha makes sense
Prompt following is insane
took me a while to notice too
Thatās pretty
Without IP adapter and controlbet
try darkarts
The refractions this puts out is nutty
I've been paying attention to the proteus datavoid discord server. I'm getting the impression he's taking the next model private. He just started charging $10 a month to use his render bot.
Doesn't matter
lame
I can get better quality
With just Loras
Same shit
Loras are a lazy way to increase performance
Ok, are Loraās something you start with or add into the node network?
basically the same thing as ipadapter in terms of what it does
Im trying to maintain my grasp
Add to in stable diffusion tbf
loras are basically a patch on a base model
It loads it after the base image has been processed a certain percentage
think of it as an expansion pack like for a game
I don't use comfy
They can also increase understanding and prompt following
Lora does what? Motion model? Light model?
loras are for characters/ artist styles/objects that model doesnt know so u only add a lora and add that thing to the picture
Having two people in an image doing things may also help.
yep, and they can wreck it, etc, it's a patch on a model
Interesting take
Ok.. so Lora is kinda like an outline for the āpaintingā
think of a LORA as the same way you think of a finetuned model
it's just a smaller version, with less content so to speak
that doesn't have all the data needed to stand alone
so it's an expansion pack for a model
Kinda likeā¦. Copy this style?
yeah, that's one thing you can do, or a character
tihnk of all the times you try to generate something and can't
usually that means either your prompts suck, or the model wasn't trained on it
Yeah makes sense, I know it's an extension
Checkpoint - marvel - Lora = super heroās?
so, you can then gather up some images that are of what you're looking for... 20-60 of em
Like pinpointing specifics?
then train a LORA, and load it with a model when generating images, and boom it's there
Or more like checkpoint - marvel - Lora = Superman specific
It's like a fucking gun attachment adding extra mag cartridges to an AR
yeah, the simplest example is... well, for example, i recently noticed most checkpoints weren't generating lauren boebert realistically when i prompted for her eating cat shit out of a litter box
so i trained a LORA on 60 images of her, and now i can generate her eating cat shit out of a litter box
ya know what dark arts images checkpoint will do?
kinda yes, more or less
That'd be funny to see ngl. Dm that shii to me š
imagine if you're an artist that grew up in north korea and you were asked to draw a comedian named stephen colbert
what would that drawing look like? you'd be fn guessing
Sigma behaviour
Only issue is imagine if you train one off your ex
š
probably would draw a korean guy with traditional north korean crap, maybe military guys around
We making it into the jail cell wit this one š„
Im finally understanding this madnessās
The get back is crazy lmao
Do image models have quantisation
That's the only reason id think they can't do text
We should focus on training cascade
The architecture has good potential
now, if you are that same NK artist, and you are then shown an hour of colbert videos
Great power comes great responsibility
you'd now know what he looks like, how he acts, you've seen him from multiple angles, under different lighting, and you can draw him
Damn that's dope
FACTS
š£ļø š„
Oh I had no idea what I was doing when I made thatā¦
We making it into the jail cell with this one
The STUPID madness is on the pc at home.
Imagine some red pilled mf makes a lora of his ex and posts it on civai
It's so over
And sometimes I have to fight with the machine not to try and flirt with me cuz dangā¦
That girl gonna be fucked for the rest of her life
Imagine people genning pics of your face
Especially large language models
Yeah he's going to jail somehow he catching some weird charge
Defamation
I've used llms with sexual bias
Openhermes mistral 7b
It deadass immediately resorted to that shit
Ye
Also indecency
Maybe
Now this⦠is art
I don't know
you still have that error? š
Looks like a tiny fairy that lost its wings
Aaaaa xd
Thank F
u cant go to jail for defamation šæ
You would still get arrested
The only error I have no is me being stupid with steps and how far to push
I tought another week later you still havent fixed it š
True just saying
u can get sued
Ain't no grown ass man doing that
Yeah
Which would make you bankrupt
Because of what you did
ā ļøš
NGL oh shit I might be guilty of this
I've been training my models off discord egirls
Some things are worse than prison. Just remember that.
Suprisingly they help the model with text
Because they draw over their hands and shit
Fansign
It teaches the model how to write
Cascade takes 90 seconds to gen an image for me
Torture in Islamic countries is worse than prison
ouch
Mexican words
Cartel
Cartel torture is way worse
Deep dark Forrest woods, endless stars, hell on earth, lights in the sky, the lights through the trees, gone
Oh yeah that's bad
Beheading is seen as mercy š bro cartel will rip off your fingers
Religiously Isis and shit do beheadings because apparently it says it's the most humane way to kill someone
In the quran
Or something
Yeah over the pettiest things
Fair is fair game⦠just force the returned favor.
That ⦠is a prompt
Am I wrong saying cascade is ass
Without fine tuning
It's so bland
It is decent at prompt following but needs more data
I think sd3 is just cascade with an llm and more data
Deadass
I thought about stable swarmā¦. But I also think it might be pointless vs just running comfy
it's not at all
transformers
Yeah with an llm
@south temple @eternal comet how do models work with gore I never tried generating gore images before as long as Ive worked
That's what I meant
that actually looks...pleasing in a way
I've done a few up there
Let me find them
that's not the same
Likeā¦. Nightmare fuel gore?
Yeah
If you ever have them around dm some to me
.
What has been seen cannot be unseen
That's something at least. Needs improvements
Yeah
If you tell it "beheaded" probably could do it
It wastes a shit ton of gram
Vram
With all my Loras model uses 10gb of vram
Yeah, I bet SD3 is gonna have those prompts downpack
If it gets released
And doesnāt fall behind a paywall
I think Emad is the one who wanted to close source it
Well he left
He deadass said Microsoft is "leading open source" on his twitter
š
Mf sold his sold
Now we know why
I pray for it to be free
Not to pursue open source decentralised but the opposite
I pray for it to be open source
Microsoft / free = never
Open source is ultimate freedom
Idgaf about licenses on open source shit being deadass
Nowā¦. If they feed tech down the pipeline to fuel creators while keeping the brand new new to enterprise level eliteā¦. Maybe
Bro have u seen those people
Selling loras
Of personalized influencers
ā ļøā ļøā ļøš
@south temple When godly models like DALL-E 3 and MidJourney v6 are in the hands of stingy people. I wanna go on a rampage
dalle3 and MJ6 are trash
Someone will always crack it⦠someone will always make it free.
Knowledge should be free
Commercial use is commercial use
AND WHY IS THAT?
I can't wait. I've seen uses of the versions on different websites already
cuz you can't use them locally
you're using an app so you haven't tasted what you can do locally
Proteus meets that level anyway
š
no censorship is the most obv, but the custom tools give you the ability to do shit you couldn't even dream of with dalle3 or mj
I'll look into Proteus
text prompts have their limits and even a model with perfect comprehension would be pretty limited
i mean... does a graphic artist use photoshop, or write instant messages on discords telling a bot how to use photoshop
Yeah that makes it suck tbh you right
it's faster and more efficient and gives you far more control when you can just grab the reigns directly and that often means NOT using text
Yeah I want those models to go god mode
Releasing all restraints to generate things we want
that's called stable diffusion ^^
you can generate literally anything you want with it if you're willing to do a bit of extra work
So SD3 is the one we're all waiting for basically
How long would you give til Stability AI releases SD3? I say give or take about 3 months at the least
Interesting
Those bubbles
@nimble mason I will tell you this tho: 4Chan users are squeezing the most out of DALL-E 3 to get what images they want, few of them are even nude if they're fortunate enough
can't do what we do routinely
you can even train your own models like i was saying earlier
Just⦠build your own machine
dalle3 is for people who are lazy or don't have the money for a good nvidia gpu with 12+gb vram
I agree
Same thing with sora
Itāll be cool
But youāre gonna pay for itā¦
One way or another
Until they need people interested
I just broke my pecs
50 pushups I go down to do 10 more feels like some mf put a metal bar through them
try just being lazy it's safer
Nice image!
i really cannot say enough how incredible the new comfyui ipadapter nodes are
total gamechanger
give me the prompt
lol
it is good at prompt following
but meh
man and women hugging. both wearing blue clothes. with a cloud containing the text "money" above them with money bills floating down from the cloud. hyper realistic
lol, what model were you using?
lets use the magnus opus
of models
that i have
š¤
thius shit takes 5 hours to fucking load tho
all and all
that was cascade
now im gonna try sdxl
THIS
IS THE BEST MODEL
TBF
its so over
the vram requirements
ARE INSANE
I SWEAER
VRAM IS CRAZY
I downloaded it and deleted because i didn“t know how to use xd
what model
dude sdxl models all use the same vram lol
get a 4090
not this one
i have a shit ton of loras
running on this
are you using a1111
yes
there's your mistake
use forge or comfyui
a1111 is just so bad with speed and vram use
so so bad
Yeah forge is just a1111 with comfy as a backend or sth like that
yep^^
comfy is the shit
needs some stuff added to smooth some things out but hey it's one person holding that shit together, it really is very impressive
also
my gpu has half the vram of a 4090
id deadass be better off buying two of my current gpu
then buying a 4090
its cheaper
no lol
yes lol
it's about vram on a single card
you mean like youtube peeps who's using AI avatar?
go ahead and buy 2 3080s i'll laugh at you
you can offload to seperate cards
š youve never offloaded
you can use two gpus
its not like sli
yeah but then the OOM shit is still there and training sdxl is a nightmare
have you trained an SDXL lora
how long did it take you
Lmao, this is in sd forge
yuuuup^^
15 hours
on 500 pictures
i didnt change the resolution to 1024
i can do it in about an hour
cuda cores yeah
Lora training or dreambooth? š
it's not just that
you're gradient checkpointing and oding all kinds of other shit to avoid OOM
tbf if you wanna use llms
aswell
youre better off getting two gpus
id rather buy a p100
or a p40 for interfacing
lets try cascade now
cascade wins in text
to be fair to it
ok it wins in prompt
lol
In the center of the picture, there is a brave firefighter. He is wearing protective gear and holding a fire hydrant, ready to tackle the fire. The fire hydrant resembles a water dragon, and the flowing water sprays out like the lively body of a dragon, fighting against the flames. The scene of the fire is filled with intense firelight and smoke, contrasting sharply with the courage of the firefighter. Buildings and trees in the background are surrounded by flames, highlighting the urgency and danger of the fire. Other firefighters around are also working nervously, dedicating themselves to extinguishing the fire and rescuing those trapped. The entire scene is portrayed with dynamic lines and colors, expressing a tense and heroic atmosphere, showcasing the bravery and fearless spirit of the firefighters.
Here you can get it.
It's a non published node of mine
#1047610792226340935 bots currently offline, you can run it locally or use online services like https://seaart.ai
Thicc thighs
NGL you have good taste in women bro
i promt it spicial
Nice
you like asian woman?
Yeah
jill from resident evil 3 after week on burgers
Did my image get deleted?
why
hm is queastion
It mostly depends on model too
I like how photorealistic certain 1.5 models look
š SDXL has a very cinematic-like style if you know what I mean
ah hah! you're the guilty party, shoulda known š
thanks for the attachment sir
was going nuts in searches know it had to be somewhere lol
Looks very realistic!
so cuteee š
@eternal comet most models biggest flaws is rotated/flipped images. For some reason they can't be stable with a person on a handstand upside down or a person laying in bed and the view is flipped. Hopefully SD3 covers that issue
an example on reddit seemed to prove that it may not
handstand does sounds something that it could do
as it's something that would be easily caught by CogVLM's captioning
buuut, there are some counter examples:
@cyan shoal how you doing excited for sd3?
Honestly same, it will likely be impossible to use on my GPU
how much vram
12 on a 4070 Ti š¤
12GB is more on the possible side, but I understand
I also have 12GB and I don't know if we'll be able to run the models
always rdy to make better booba and cocka
Give prompt
I'll try
You will be able to
8b models not that hard to run
Unless they are not quantsized
Oh shit
I forgot you can't quant checkpoints
Ye we might be fucked
so fp16 on 12GB is out of the question
fp8 makes sense on 40 series as it may not hurt the quality nearly as much (bigger models quantize better ā)
Bfp16
Yes
hmm I never knew this
I thought bf16 was supposed to be MORE precise
oh you are right
Brain float (BF16) and 16-bit floating point (FP16) both require 2 bytes of memory, but in contrast to FP16, BF16 allows to represent a much larger numerical range than FP16, so under-/overflows won't happen as often. However, this comes at the cost of reduced precision.
misremembered lol sorry
16gb will be chilling, right?
in tech support dont answer to me maybe you know something

Error -3 while decompressing data: invalid block type when i install whl
[19:18]
its bad file?
If you have a poor connection as described in #š¤ļ½tech-support , try using a download manager. Usually they allow you to resume downloads š
how do i download 8 gb of additional vram?
I've seen it mentioned a bunch of times that since comfy loads and u loads models between steps, 16 gig should be full powered for sd3
It would only be an issue for stuff like a1111 that tries to keep all steps in vram.
if you've got 16gb 40 series, you've also got access to the hopper transformer engine so you can load models in fp8 for free. half the memory usage of half-precision.
doing it in software is slow but ada got you covered
and from what i can tell, comfyui poorly implements fp8, using the raw api switches as command line launch options. while it works for all the stock nodes, its a crap shoot of compatibility in the custom node sphere.
wtf
yeah the thickness of that last I letter doesn't match the rest of the font-face at ALL. WTF seriously
yeahhh š
training data is abit goofy
i've seen bad keming before but never like this
detail is a coincidence. speed increases... i'm not sure. maybe its showing you the speed per image, which would be the time / batch size
be wary of confirming your biases. its a trap
today free tutorial from skorchekd DISKORD E GIRLS
i cba to train another dataset
so im stuck generating egirls to test lora quality
it took me 8 hours to train
meh i'm bored with -with-images again. y'all are weird
can you whow images what is it
its bias it keeps generating asian women
you get only picture woman?
they want comliments?
oh lol
DONT GIVE IT FOR FREE
if they know that you train lora with photo they very happy
you give them big attention
you choice wrong way bro!
many black girls
these images are depressing
imagine your daughter doing this
š
i aint having kids
i want too
maybe their dad cukold š¤£
yuh
the culture is so different
these girls are being taught to basically turn themselves into fucking objects bro š
i was taught to train myself to be strong not weak
is bad
Not really, you just have to activate bucket resolution (i think its activated by default on kohya)
skorchekd you see my yoga girls today?
@amber arrow I use this one to bring all incoming images into a maximim size spec. you can keep it proportional.
Hii!! The node is part of a custom pack? I couldn't found it on my standard comfy nodes š¬
got another new weight type with the new ipadapter... style transfer
Oh, just found it. It's part of the ComfyUI Essentials pack
Like up adapter, I'm easing in-out of lunch break. š Back later.
these are legit OF girls
or at least 99% of them
Imagine that someone from the civitai discord server said that he heard some girls which had "tentacle f*tishes" and wished to generate images with them with "tentacles" - you know what I mean.
doomed apes
OF generation
And then they are bragging about being objectified while they appear half naked on the street
real
no role modelz
breakup culture for a women is to hookup with another guy
breakup culture for a man is to work on himself and go gym
just shows you how society is
no wonder why modern women especially on internet are like this
exactly
That's why I think that all these social medias and OF doomed the modern girls/women
because they only care about being "liked" (numbers) and they would do whatever for those likes - the like button is like a "confirmation" for them for how good looking they are :)))
so they start posting half naked photos and so on
to attract more and more likes
it's a vicious cycle unfortunately
the online culture for men is much better
to be fair
because of fitness influencers etc
but that can lead to arrogance
we have to make peace with our own demons
people do that other ways
being too emotional is a bad thing
in my opinion
yep, because everyone will use you
yeah
work on yourself
and respect yourself
i feel bad seeing thse women
it really does hurt me in some way
its very sad
but theres nothing we can do
they choice to be like that unfortunately
cant help people who dont want to be helped
true
all we can do is use them as training data
and it's really hard to change a girl/woman after she's already 18
and work on ourself
i use them as fuel
use this current social state as fuel tbf
when youre working out
if their parents couldn't teach them, then I won't waste my time
if they hate your dad
they're more like lost causes to me
that's why the passport bros are a thing nowadays
the Western girls are legit no wife material anymore
in 90% of the cases
yeah
tbf i aint gonna get married
i dont see myself having kids or doing anything like that tbf
i dont feel "alone" i feel content
some people have to be with someone others dont mind being on their own
im not "alone" because im not sad im by myself and im happy
because i feel ive made peace with myself and whatever happens happens
Imagine that there are lonely men who waste money for an "illusion" relation with OF girls. They lie themselves that these girls really care a bit about them. :)))
I think they're legit braindead
yeah
thats very sad
a man has to go through alot of shit
before he makes peace with himself
paying for photos :))))))
down bad
did you know that men are more likely to have depression
and men commit suicide more then women
can we take heavier topics such as above to dms rather than air put here in this server pls and ty- also none of the above talk about objectifying women and scraping selfies of them whilst looking down on them is cool or should be aired out here either
...?
š
What GPU?
4090
time to sell a kidneyong
can you try to train cascade
you could run even 5 controlnets at the same time
Already to much to do and learn here ... if I don't need it I won't learn it ^^
you're kinda sitting on a gold mine if you venture into the blackhat world for making money with AI
kidding
That's why I bought it ...
Already had an exhibition about A.I. Art ...
but wait, you really use blackhat or you're fully whitehat/greyhat user?
Don't know these hat-thingy ... I'd say I'm a crazy guy doing creative stuff ...
I see
Only sold one picture so wrong audience or bad artist š
The art you're talking about is more inclined to the "NSFW" field?
Well ... if I want a naked male I will create one ... but don't care about porn ... nudity would be part of the story or if you like the word: art ...
A lot of stuff is more about fooling around ... that's why I don't say I'm an artist ....
hey yall, I'm finally am going to purchase a system that I can use with Stable Diffusion. Can anyone tell me what you used, how, running on what, and so on, so I can get a good system an set it up smoothly.
Heard that the furry artists are making banks from this type of art
AMD graphic and a lot of VRAM š
Just buy RTX 4090
haha, can you give me specifics please.
What's your system though?
I have a bad PC for AI, that's why I use cloud GPU providers
4090 is still more than enough for me ... I render pictures in energy safe mode ... still fast enough for me
But I know what works best with AI nowadays
so a 4090 will be your best bet
- a processor from AMD like Ryzen 7 or Ryzen 9
heresy
i'd power mine with coal if it got me +20% it/s
For animations it's different ... sure ...
MOJO! That plant girl is MONEY!
For me it's fun š
SVD ...
Sure ... with some tricks
Anything you wanna share with the class?
You dont have to give away your secret sauce so to speak. But Ill take any advice thrown
I use SVD wit 6 FPS increase the Motion Bucket a bit and use Topaz later for 2x SloMo and 24 FPS ...
Increase the motion bucket⦠I was decreasing it
I got some jarring motion with a decrease
Topaz is doing a damn good job ...
Dang
Topaz Video AI a program ...
Dang dang dang
Can you share what the output of that plant girl looks like without topaz?
I don't think so ...
Haven't found an upscaler as good as the magnificant (correct?) one for Stable Diffusion yet
SVD is sickā¦. Has a big future if it has refined controls
lol
There's already an A.I. with control over the kind of animation ....
Runway? To much input here ... š
Runway is fun. I just want to runs stuff local
I need the control over the creations.
Same here ...
There has been an artist called āanymaā that played live with a really interesting set of animations
And I got sucked right tuhfuq in
It's a crazy world ... learning A.I. for 2 years now ...
Ive only been deep diving for about 4 days
Pretty proud of my self for how much Iāve crammed into my head in a short time
It will never end! It will never end!!! Don't end up here ....
lol
ups ... to much blood?
About that picture, what I want to do is use the constraints like the āroomā in that picture
And create animations within those bounds.
Almost like a glass cage
Bad luck I can't find pictures easy .... need an A.I. to sort them š
Do whatever you like ... but a few days later it can be boring stuff from yesterday ^^
Not often ... to much to learn for me ^^
I my eyes comfy would need a second interface for "Users only"
Otherwise you can mess everything up to easy ....
I need to drop this into scribble and see if it works
Only⦠if itās a stage design dementosn are off⦠but its proof of concept
Trial and error ... best way to learn ... and sometimes youtube videos š
Rendering videos at the moment ...
Comfy is cool but I'm not sure whther it's good for first contact with A.I. ...
Im a little beyond first contact lol
All this node stuff can distract from prompting for example ...
Funny I mention that⦠I get too meet someone from Microsoft on April 3rd
Maybe you can teach them A.I. š
lol
Well.
We are meeting for responsible Ai generative use discussions
Whatever that means
Censorship ... š
They are gonna hate me the
Then
The only thing I would scrub from it is the use of human children.
Have a look at all those A.I. companies ... some block the word: "invisible" cause you could try to make clothes invisible ....
For ALL the reasons of exploitation
Nudity not really my topic but always been part of art and sometimes it's good for a story ...
German here ... we have FKK ... š
Fkk?
People naked for sunbathing or at the beach ...
Yea, US needs to kinda ⦠relax a little
No nipples but blood is OK ...
But other countries are a little more self regulated.
Americans are a little ā¦. Un-hinged after 2020
I think they are all different ... as long as I can do what I want to ...
Hey guys, I need your help on how to create a Lora. I've watched multiple videos on how to do it and even implemented it myself, but the results don't look good at all. Sometimes the face doesn't match the character I'm trying to train on, and the body ends up all deformed. I followed all the steps, and still, I get trash results. I want to know, if I'm training a character on koyha , should I just use the face of my character or should I include the full-body image? Please, can someone help me?
Never made a lora ^^
Can you still catch a murder charge if the person you murder comes back as a zombie?
generate A natural landscape with snow-capped mountains and a crystal clear lake
No Bot online ...
Here is the image you requested.
Reminds me of surreal 1980s album art
satanic whispers
i love it
Thx ! ...
mic check
Nice work!
hellpppp
I need to convert this image to image with a line art work,
tutorial says to use control net
I tried to configure the parameter as what tutorial told me, but the results definitely looks wrong
not even shapes as the pose at all
So, this may be an unusual question (new here), but am I in the right place to find a mentor?
Iāve messed around with Stable Diffusion through websites like Playground and Leonardo, and I learned enough through them to create this image here (which I touched up in Photoshop), but Iām looking to expand my knowledge and start using Stable Diffusion with Automatic 1111. Got a great machine for it now. There are specific things I want to learn how to do, and I donāt even know where to begin, or what theyāre called. For instance, I wanna redo this image to achieve a more realistic horror theme. I even have samples of the look Iām hoping for!
Plus, I wanna transfer (?) some of my characters made through Photoshop into Stable Diffusion to put them in different looks and whatnot. I wanna learn the roundabout way to do that.
Lots of things I wanna figure out how to do, but I struggle to learn on my own, so Iām hoping to find a mentor to help me along.
img->img
lots of folks here run auto, im a comfy human
my personal opinion is to learn the hard stuff (comfy) because of what it can do
but its not glamerous
and you will hate life once or twice
it gets easier than a1111 pretty quickly
the real problem is how shit the documentation is
actually, poor description, generally it doesn't even exist
well
I found a few sites with great stuff to follow
but you gotta know basics
no such thing as drag n drop in this realm
beautiful girl in room
wtf
Here is the image you requested.
Here is the image you requested.
a dog driving a tiny car on top of a gigantic ant with a train car body that is riding the tracks at night with a flaming comet streaking toward chicago. a blimp with the texture of a pickle, the color of a barbie doll, and polka dots is rocketing with jet engines while dropping nuclear bombs and party streamers on a parade full of evil clowns and maniacal batwing vampires. realistic style.
lol - loaded
my prompts are fuuucked up
im here for it
def check out some of my workflows and prompts... haha
i always leave them embedded
100% pro sharing
your workflows are a lot lol
have you looked
last time this thing barfs out this img
i work like a tornado
just full speed ahead and then after about an hour of carnage and a long trail of damage i disappear back into the clouds and move onto the next workflow
hahahah
ill drop ya this one
its pretty nice
ill share with your prompt once this last chunk of garbage renders
have fun with this if you can run it
ONLY edit the OG img
Wait did that come from my crazy prompt? No way lol