#🏞｜general-with-images
hasty nova
oak osprey
nice
pink floyd / alan parsons project vibes
uhhhhhhhhhh
that's not an isopod 😭
oak osprey
what about go farts in korests
slow motion gravy sloshing with drops glinting in the light, 3D photoreal UHD 8K
😁
smoky oak
you are usually throwing around the link lmao
hasty nova
I got you
smoky oak
oh, you're here!
hasty nova
Also the huggingface has the safetensors anyway
smoky oak
thanks haha
smoky oak
thank you!
oak osprey
wispy nest
I thought I'd get more triangles, or any at all :P
hasty nova
I thought I'd get more triangles, or any at all :P
What model are you using then?
oak osprey
1.5 is really, really bad at abstract stuff unless you extensively train it for that
wispy nest
What model are you using then?
base 2.1
Can't remember the names of any 2.1 I have :P
wispy nest
hello
oak osprey
lol fuckin gravy boat's ready for lift-off
hasty nova
https://images-ext-1.discordapp.net/external/eI8Q3CHIfONennzWXZtOQEEcYx5vAKVylCV...
looks more like caramel with coffee coming out the top
which would honestly be pretty good
wispy nest
Very good
oak osprey
ty
smoky oak
@hasty novaHey, si there a reason why you have 2 of the same model uploaded to civit?
these show as the same model
hasty nova
<@800501037793280030>Hey, si there a reason why you have 2 of the same model upl...
One is probably v1?
Wait no
I have no clue, must have been a mistake.
smoky oak
no, V1 is separate
hasty nova
I don't know if I can take it off or not.
smoky oak
I have no clue, must have been a mistake.
ah, alright, wanted to make sure I didn't mess up lol
hasty nova
ah, alright, wanted to make sure I didn't mess up lol
Yeah, they are the same thing I think. Thanks for pointing it out though! Ill try to delete one of them
smoky oak
no worries
oak osprey
their UI sucks
they allowed me to upload a whole zip of the wrong crap and call it a model
hasty nova
I've deleted it and also realized I never put tags on my model
which may be why it was never getting many downloads
oak osprey
yeah yours was hard to find
smoky oak
As cool as digital diffusion is in terms of styling and stuff, it unfortunately still has that 2.x "crust" to it that really puts me off from 2.x models ._.
spiral birch
Hey there, there is no tool here to create?
crystal deltaBOT
smoky oak
@spiral birch
hasty nova
As cool as digital diffusion is in terms of styling and stuff, it unfortunately ...
All good, but could you share an example maybe? I would like to see what you are talking about.
smoky oak
its in all of your example images as well
a sort of compression and grittiness that just makes images look really gross IMO
just really compressed and full of artifacts
spiral birch
Thnx
oak osprey
just really compressed and full of artifacts
prompting is super important. try UHD 8K Sharp
smoky oak
you can see the 1.5 image is 1/2 the res of the 2.x image, but the 1.5 image is much clearer
oak osprey
smoky oak
and I am not going off realism, i am going off detail
hasty nova
you can see the 1.5 image is 1/2 the res of the 2.x image, but the 1.5 image is ...
I see, maybe v3 can help fix that, maybe not. I am trying new training settings and all that. I will let you know if I stop seeing it!
oak osprey
i wear glasses, you know. but i don't see any issues with those images
smoky oak
when you put out 512x512 out of a 768 model?
this is a 768x768 image gen
wispy nest
that's also what I'm going off of, cartoon styles can be detailed just like realistic photos. And to me, I just can't get past the lack of details, or blurriness when it comes to ai images. I like it sharp and focused. Blur can still be there, it should just be because of focus :P
oak osprey
that's also what I'm going off of, cartoon styles can be detailed just like real...
realistic photos, most of them these days, are from cell phones
smoky oak
I think the 512x512 one actually looks better in this case
wispy nest
realistic photos, most of them these days, are from cell phones
probably, but that doesn't change what I like, and don't ;P
smoky oak
but yeah, unfortunately this model has the exact thing that deters me from using SD 2.x
hasty nova
I think the 512x512 one actually looks better in this case
nah the dress looks really grainy
oak osprey
it looks like it has ringing artifacts
smoky oak
nah the dress looks really grainy
I think its somehow less grainy than than the 768
both are super mesys tho
hasty nova
I think its somehow less grainy than than the 768
both are super mesys tho
well whats the prompt? Maybe one of us could try and change some things in the positive/negative and see if it helps. If it doesnt then oh well, if it does then yay.
smoky oak
these are some of my favorite realistic images I have made
oak osprey
well they're sharp but if you're going for period pieces, they ain't realistic 😛
smoky oak
well whats the prompt? Maybe one of us could try and change some things in the p...
My prompts are not something I am looking to give out at the moment, cause they are making me money, but your images on your civit page have the exact same issues
oak osprey
Brahms
hasty nova
My prompts are not something I am looking to give out at the moment, cause they ...
Oh, good for you. Have fun with that!
smoky oak
thank you!
hasty nova
And I don't mean that in a sarcastic way, it can be hard to tell through text sometimes
smoky oak
you can see that your images on civit also have huge compression and banding artifacts for whatever reason
they are great images, I like them a lot, but they just have that 2.x messiness that I stay way from 2.x because of
hasty nova
yeah :/ I was also not great at prompting for 2.1 back then, maybe I could try to remake that image with some prompting changes rq and see if anything changes.
oak osprey
bro you're using 1.5 models that are fine-tuned to oblivion, you can't hold a 2.9 million image model or whatever, against his digital art centric 300 image dataset model.. and you're not also willing to put the effort in... it's kinda frustrating that you endlessly shit on 2.1
smoky oak
I have never used a 2.1 model that didn't have these issues, so its fair to say its probably connected
hasty nova
I have never used a 2.1 model that didn't have these issues, so its fair to say ...
probably, but its worth a shot
at any rate I could just make a neat image anyway
smoky oak
bro you're using 1.5 models that are fine-tuned to oblivion, you can't hold a 2....
I am not shitting on 2.x, and I really hate that you keep trying to push that message. It has problems, of which 1.5 does not. that is legit a reason to not use it over 1.5. I want to see this model get better, and you and jungle are working on it, so if you could kindly stop talking crap about my valid criticisms, I would appreciate it.
Not everything has to be me saying 2.x is bad, FFS
oak osprey
please show me how you can reproduce your prompt using base 1.5 from runwayml
without the 2.1 noise
smoky oak
ok, so now we are doing base SD vs a finetuned SD, alright then, is that not what you just criticized me for?
IDGAF if I am comparing base to finetuned, I am comparing my options
oak osprey
you keep saying 1.5 this, 1.5 that
you're not even using 1.5 when you have 2.9 million images with thorough tags added into it
smoky oak
because its another option, so what?
oak osprey
do that with 2.1 and then see
smoky oak
you're not even using 1.5 when you have 2.9 million images with thorough tags ad...
oh my god, one model, just one model and apparently thats a problem
hasty nova
guys its just some numbers we are changing bit by bit to use in conjunction with random characters to produce cool images from random noise for fun, its nothing you need to fight over
oak osprey
it's your basis for comparison is skewed, friend. these results take a lot of work to get, and you know that, which is why you do LoRAs instead of fine-tuning
it's literally your own words
smoky oak
it's your basis for comparison is skewed, friend. these results take a lot of wo...
So what you're saying is... they take a lot of work, but giving feedback/critsism is unfair. How does that make any sense.
I said his model has a dope style, but it struggles from the grittiness of 2.x, where is the lie?
oak osprey
guys its just some numbers we are changing bit by bit to use in conjunction with...
technically i think we change like 65,000 or more bits at a time 😄
jesus wept 
some of these pre-negative images
wispy nest
for me I would say that 2.1 has a much greater upper limit for details, but 1.5 has the upper hand on everything because whatever you might want, I'm quite sure there's something out there trained on it. Be it embeddings, loras, etc, that 2.1 don't have. basic 1.5 is worse than basic 2.1 in most situations. But these aren't really in a vacuum.
That's why I look at the outcome and not what it's made with. For example, I just want a specific thing from my images. Crispness, sharpness, and details. And those things, in the way I want them. Can't be done easily in 2.1, I can get good results using embeddings etc in 2.1 but it's so much work for slightly better results than when using stuff I learned and got used to in 1.5 :P
oak osprey
for me I would say that 2.1 has a much greater upper limit for details, but 1.5 ...
people use 2.1 without even turning on clip skip lol
wispy nest
turning on? I thought it was on by default :O
and doesn't 2.1 have the same results on each layer there?
oak osprey
it is now that they have config files mandatory for models
smoky oak
My only reason for not using 2.x is its huge amount of artifacts and "grittiness", so if you want me to use it, fix those issues. People did it with 1.5, and now we have better tools
oak osprey
the model declares whether it is 2.x or 1.5
i don't care if you or anyone else want to use it to make money with, that's not what i'm about, i'd rather just make fun images or extend this into making fun videos
smoky oak
and I would rather have good results, personally
oak osprey
i'm thankful my gritty images lack commercial appeal
smoky oak
i'm thankful my gritty images lack commercial appeal
your images are not gritty
oak osprey
Samus be thicccc
smoky oak
yours don't seem to have that problem, like I said a while ago
frozen socket
these are some of my favorite realistic images I have made
Reminds me (after a fashion) of some I made around 1920's/1930's Hollywood glamour style earlier this month
smoky oak
Reminds me (after a fashion) of some I made around 1920's/1930's Hollywood glamo...
very nice :>
also, here is 1.5 base with far less artifacts
its also at half the res of the 2.1 image
is it perfect? absolutely not lol
wispy nest
And I don't really feel all that comfortable in saying if this or that is better. What I like is not what anyone else like and I don't really have much to say for this or that than, "I like this more because of X." Art is so subjective that there really isn't any use to try to say what's best, or rather, in a way not to sound like I'm being mean :P
smoky oak
I like the "vibe" of the 2.x one more
oak osprey
I like the "vibe" of the 2.x one more
ayyye that's why i've decided to ditch 1.5, there's something about ... i don't know what it is. even when 2.x is completely destroyed it can make such cool stuff
smoky oak
hm it seems to have sent the same image twice
they are different
oop
they were different lol
hasty nova
they are different
I know, but I meant to send a different alternate one
oak osprey
hm it seems to have sent the same image twice
the ultimate troll
smoky oak
the two you sent had less artifacts, but there are still some in it
hasty nova
smoky oak
it was a big step in the right direction
smoky oak
The first one looks considerably better
oak osprey
it was a big step in the right direction
my understanding on this is that they put a lot of negatively labelled images with poor quality into the dataset to ensure that these negative prompts work better than they did in 1.5 but a side effects is they became essentially mandatory
frozen socket
And I don't really feel all that comfortable in saying if this or that is better...
It's like being at an opticians. " better on the left or better on the right?"
hasty nova
The first one looks considerably better
nice, is there like a specific area you can screenshot to show what might still be wrong with it?
oak osprey
It's like being at an opticians. " better on the left or better on the right?"
"I don't need new glasses. I've been wearing these for 24 years, and they work fine"
my dad
he's friggen always asking me to read stuff for him
smoky oak
the "crunchy" areas of the hair seem to have a lot of chromatic aberration and distortion
the gradients also still have some banding and compression, but they are far better than they were before
And then the rest I would say is just up to the fact that its a small finetune
smoky oak
ayyye that's why i've decided to ditch 1.5, there's something about ... i don't ...
Thats not 2.x doing that, thats Jungles finetune
The style he has gotten into it is dope
I like it a lot
oak osprey
yeah the way 2.x adopts styles is incredible
smoky oak
a sort of moody offset esque look
oak osprey
my LOTR model was really really epic but i had issues with it i needed to resolve
i still plan on making it happen...
the same data (broken data tbf) went into 1.5 base and it came out all mangled
just horses with many legs and heads, levitating wizards in the wrong ways
frozen socket
for me I would say that 2.1 has a much greater upper limit for details, but 1.5 ...
my biggest issue with 2.1 models is that they really mess up on nipples unless you turn the CFG scale down to 2 or 3.
A really odd thing to go wonky but hey ho there you go 🙂
smoky oak
I have a feeling that the reason jungle rally's model still has the grit is cause there haven't been enough images to finetune on to get rid of it. But it is headed in the right direction for sure
oak osprey
ask for a puppy, get a couch cushion
he might have some artifacts in the training images, i used MJ 5.1 for its clarity
all of my images are downsampled before training on them
his are downsampled but they are different aspect ratios
so i found out, a 16:9 image ends up padded on the top and bottom
you know 'letterboxing'?
smoky oak
he might have some artifacts ***in the training images***, i used MJ 5.1 for its...
I doubt it is in the training data, as again, all of these issues are just hugely prevalent in 2.1 as a whole on its base
the reason that 2.1 is so far behind in refinement is cause it wasn't until semi recently that people started figuring out how to tune it really
oak osprey
it's not in his folder of training data. it's the way it is being processed before training.
wispy nest
my biggest issue with 2.1 models is that they really mess up on nipples unless y...
heh, for me I just can't get anything I really like. I can get somewhat good results with a heavy use of embeddings but it's just so much work and trial and error that it just fatigues me out.
That's why I'm still 99% of the time using 1.5 stuff. It's a lot easier to get what I want and like :D
smoky oak
but these are big steps forward
oak osprey
i looked at the source code of the dreambooth extension for sd-webui to see how it's working for his resolutions and it does some messed up stuff to them
hasty nova
smoky oak
I have confidence that 2.x can be really good if trained well, its just not there yet is all
smoky oak
Are you inpainting these, or what are you doing?
hasty nova
I am also generating directly ay 768x1024 which 1.5 can struggle with, but I suppose hires fix is there for that
hasty nova
Are you inpainting these, or what are you doing?
Just changing the prompt
smoky oak
I think all the other issues are just the underlying model itself
oak osprey
but also UHD is used a lot in my captions on very clear images
so maybe that works better for me
hasty nova
did you try my UHD thing?
I did include ultra hd in the prompt, yes, but I mostly changed the negative
smoky oak
i told you to change your prompt Sytan <:KEK:747148843434377316>
If my 190 tag negative isn't good enough, then there is something wrong with 2.1 lmao
hasty nova
smooth skin, messy face, teeth, grainy, sandy, poorly drawn, low quality, ((messy)), distorted, (chromatic aberration), extra limbs, extra fingers, blurry, burnt, high contrast, pattern, repeating, tiles, mosaic, jpeg compression
This is mine
for this image specifically
oak osprey
you should probably try and eliminate the excessive negative tags
smoky oak
I tried it with my 30 tag, 75 tag, 130 tag, and 190 tag, all of them had the issues
oak osprey
ah no 'artifacting'?
smoky oak
what?
oak osprey
in his neg
smoky oak
oh
hasty nova
ah no 'artifacting'?
guess not
oak osprey
noisy
hasty nova
my negative usually starts out as 3 or 4 words then expands from there when I see stuff I dont want
oak osprey
noisy, noise, static
i would try
i'm surprised none of you use the Compel embeddings instead, i have had better luck with that
it uses ++ and -- to weight terms and () to group attention
when i go from a basic pipeline to simply just processing my prompt into embeds first via Compel, i see like 10-20x improvement alone
hasty nova
?
smoky oak
here, I am trying with a drastically reinforced pos and neg in digital diffusion
oak osprey
i love how it's unclear where the sun is coming from
hasty nova
yeah that's not good really in my eyes
wispy nest
How people use negative prompts tells me a lot of how people style things :P
I have my own opinions on it, but that's just that—an opinion hehe
oak osprey
that would work as an ad for AdultFriendFinder
smoky oak
same prompt in 1.5 at 768x768
hasty nova
Try this exact prompt though:
Pos photograph, woman, red hair, peasant, cold lighting, outdoors, fantasy, cinematic, bright, midmorning, overcast, grass, ultra HD, RAW image
Neg smooth skin, messy face, teeth, grainy, sandy, poorly drawn, low quality, ((messy)), distorted, (chromatic aberration), extra limbs, noisy, noise, static, blurry, burnt, high contrast, pattern, repeating, tiles, mosaic, jpeg compression
you don't need a huge positive
oak osprey
what the hell is that
hasty nova
that actually usually is worse in dd for me
smoky oak
<:KEKL:842966965752037376> what the hell is that
don't go pointing out small issues lol
smoky oak
I am saying its funny to see you point out the small issues, but the 2.1 image has 3 focal planes and no consistent blur direction lmao
smoky oak
its peach fuzz, on a non cherry picked gen lmao
oak osprey
what
frozen socket
although 2.1 models can be nice, cant beat a nice mature Japanese lady. Even the crooked hands and eye can be excused by age 🙂
oak osprey
she's basically a wizard! hairy!
frozen socket
although 2.1 models can be nice, cant beat a nice mature Japanese lady. Even the...
and please feel free to critique
oak osprey
and please feel free to critique
oh trust me they do 
smoky oak
although 2.1 models can be nice, cant beat a nice mature Japanese lady. Even the...
I like her style :p
she looks like she is wise
wispy nest
and please feel free to critique
it's too blurry.
I'm so, so sorry, but that's my humble opinion 😭
frozen socket
it's too blurry.
I'm so, so sorry, but that's my humble opinion 😭
you know the earlier conversation about opticians...................
smoky oak
I think it looks like it is straight out of 2.1, and thats all I will say lol
oak osprey
it's too blurry.
I'm so, so sorry, but that's my humble opinion 😭
you wanna deep dive into those cracky folds eh
frozen socket
@wispy nest in fainess some of that may be due to the prompt
*(Raw Photo)
nsfw an erotic watercolor of a photorealistic 90 year old Japanese woman wearing a kimono, skin wrinkles, skin blemishes, age marks,grey hair
(art style by (Utagawa Hiroshige:0.9) ),
(soothing tones, insane details, intricate details, hyper detailed,photorealistic, 8k, ultra realistic, volumetric lighting,(film grain:1.4)),*
wispy nest
<@456226577798135808> in fainess some of that may be due to the prompt
*(Raw P...
is everything in the normal prompt? or are the last paragraph the negatives?
frozen socket
is everything in the normal prompt? or are the last paragraph the negatives?
no thats all of the normal prompt
The negative is
Negmutation-200, (grainy:1.3), low-res, error, cropped, worst quality, (JPEG artifacts:1.2), duplicate, out of frame, (blurry:1.3), dehydrated, (low quality:1.4), amateur work, disjointed, overexposed, underexposed, (pixelated:1.3), (compression artifacts:1.2), noise, (oversaturated:1.2), (undersaturated:1.2), unrealistic, unbalanced, inconsistent, unoriginal, cliché, poorly designed, damaged, worn, outdated, uncoordinated, disproportionate, warped, misshapen, impractical, (unscientific:1.2), unimaginative, illogical, chaotic, distorted, unappealing, emotionless, (disfigured:1.5), mutated, (deformed:1.6), poor anatomy, poor posture, incorrect, poor lighting, poor color balance, poor contrast, poor shading, poor texture, poor perspective, poor geometry, poor layout, (inaccurate:1.2), unathletic, poor form, uncoordinated, lack of skill, unrealistic, incorrect technique, unbalanced, exaggerated, implausible, (unsafe:1.4)
smoky oak
Try this exact prompt though:
Pos `photograph, woman, red hair, peasant, cold li...
same prompt in DD and 1.5
wispy nest
my prompts can look really weird as well, I don't really care about that as the ai doesn't see words exactly as most humans I think :P
frozen socket
erotic water color of a 90 year old woman
it floats my boat and gets results ;o)
hasty nova
same prompt in DD and 1.5
Interesting. Im aware the model still isn't great. I am really hoping for v3 to be better, training starts tonight! I have improved the dataset up to 500 images. Im retraining from scratch. Lower learning rate, more steps. This should be good hopefully.
smoky oak
here is the same prompt in 1.5 with one of my heavily tuned negatives
frozen socket
fair enough haha
just remeber, that grey haired old granny sat on the bus next to you has probably had more sex than you've had hot dinners
smoky oak
just remeber, that grey haired old granny sat on the bus next to you has probabl...
I choose not to lol
smoky oak
tweaked the prompt more, and I am very happy with this
hasty nova
well, you are welcome
hasty nova
go ahead and sell it for all I care, just glad you got something you like
smoky oak
just the theme
smoky oak
go ahead and sell it for all I care, just glad you got something you like
oh I don't sell like that lol
hasty nova
oh
is it like people ask for what they want generally and you make a prompt out of that
smoky oak
I sell LoRAs mostly lol
hasty nova
ohh
smoky oak
But I feel commissions on contract work for a large company
frozen socket
some summery ones I generated last week
hasty nova
gtg
frozen socket
my personal favourite is the top LH one
marsh oracle
some summery ones I generated last week
how tf do you get stuff this clean lmao
frozen socket
how tf do you get stuff this clean lmao
I use a cheap 1080Ti and learnt the art of patience young Padawan
marsh oracle
I use a cheap 1080Ti and learnt the art of patience young Padawan
i also have a cheap 1080ti
in my ubuntu server
full reef
/help Why can't I use image creation bots anymore?
marsh oracle
speaking of which I need some help
I can't get Torch to install, and I'm getting an error code, would someone be able to help me if I move the conversation to #🤝｜tech-support
wispy nest
<@456226577798135808> in fainess some of that may be due to the prompt
*(Raw P...
Same prompts but removed a couple NSFW words. I also removed all the negative prompts as I don't see much use to them :P
I didn't add any prompts as well because I don't even know where to start with such a image 🤔
frozen socket
Same prompts but removed a couple NSFW words. I also removed all the negative pr...
I also used a custom merge of various models ;o)
frozen socket
I can't get Torch to install, and I'm getting an error code, would someone be ab...
you can try 🙂
wispy nest
I also used a custom merge of various models ;o)
I assume everyone does that. I'd rather hear if someone only used the normal 1.5 or 2.1 without added stuff :P
oak osprey
omfg LMAO
i can't stop laughing
i prompted something about goldie hawn stealing a baby pacifier and this came out
why is she so blinged out
ahahahahaha
marsh oracle
you can try 🙂
i just put larger explanation of what the issue is in #🤝｜tech-support
frozen socket
I assume everyone does that. I'd rather hear if someone only used the normal 1.5...
Same Orginal Prompt but in the two raw models, ArtiusV21 & Providence Anchor. Easy to see where most of the influence comes from 🙂
wispy nest
no idea what ArtiusV21 & Providence Anchor means :P
wispy nest
Same Orginal Prompt but in the two raw models, ArtiusV21 & Providence Anchor. Ea...
there we go, I lowered the number of years, as well as wrinkles in one swift swoop! ;P
If it only worked like that in real life >:I
hasty nova
no idea what ``ArtiusV21 & Providence Anchor`` means :P
they are just models, artius is 2.1 I know, not sure what the other one is but its probably also 2.1
frozen socket
there we go, I lowered the number of years, as well as wrinkles in one swift swo...
nice , you may want to take out or tweak down the weight og things like sknin blemishes and wrinkles etc
oak osprey
honestly there's so much style applied to every damn image this makes that it's impossible to see the noise
wispy nest
nice , you may want to take out or tweak down the weight og things like sknin bl...
hehe, if I were to change stuff to what I want, then the next "old woman" you'd see from me is this
oak osprey
I should be in bed, just noticed the time
sucks to be you, living in the future
is this a guava? 
dense tapir
oak osprey
dense tapir
wispy nest
this is actually not that bad in my eyes when it comes to blurriness, almost :P
smoky oak
oak osprey
that gave me this
smoky oak
Looool
oak osprey
i know
no negative, no positive
that's what made me ask it 'who are you?' because i was laughing and loving that dog image
self portrait
ahahahaha
smoky oak
Giga chaddy
oak osprey
@smoky oak man i really got what i wished for with this model
pastel mango
A look back to the 90s dial up. Image Loading..........Loaded
smoky oak
<@296720689468145664> man i really got what i wished for with this model
Yeah, its super cool. I hope I can get my hands on a working version sometime haha
oak osprey
i keep getting distracted on making that happen 
it's just so pretty
my system has crafted something like 60,000 class images already
wispy nest
what model is it
oak osprey
ptx0/pseudo-journey
smoky oak
If you can get me a working copy, I would be down to try and make some LoRA's for it
Could try a realism LoRA for it
oak osprey
what would you make them to do? offset noise would be super interesting
smoky oak
I have no idea how to do offset noise, as it's a whole process
Or well, actually, I bet I could do it
I have an idea on how I could do it
hasty nova
Here we go. First sample image of the night at 500 steps (1 epoch.) I am doing 150 epochs for 75k steps, so itll be about 12 ish hours. Gn yall! Im feeling sick so imma go to bed early.
smoky oak
Sleep well! Looking forward to the new version
oak osprey
wispy nest
and this might be my last for this nightshift... maybe, I don't feel that tired yet somehow :/
dense tapir
wispy nest
lets see what you guys can do with it
i dont even have auto installed now, i got bored i gotta admit waiting for a new stable model. but man i just checked out dalle for the first time in like 6 months and it looks ass like they must not have updated at all
dense tapir
Nothing from me since it is 1.5 based
wispy nest
it seems like it regressed even
there's nothing but stable and midjourney is there?
yeah it is. the image generator in bing seems better than dalle2 idk
I can remember waiting and being all hyped to get that dalle2 invite last year lol
dense tapir
wispy nest
bing dalle is definitely better than dalle2
but i know thats ot so pardon me lol
i think the one in bing is like dalle3 or something. i tried the same prompt and its much better there than dalle2
wispy nest
i bet dalle will rise again though, i mean its by the chatgpt people who are basically becoming the name in the game of ai
dense tapir
I dunno I find GPT from them to actually suck.
It refuses to follow my orders no matter how many times it appologizes it does it again.
wispy nest
gpt4 is dope imho. and i dont know any language model better than it atm. certainly not stables flop
well i mean if you wanna be all toxic and what not on it or insult people yeah maybe its not for you
yeah that gets old, its still dope as hell imo
dense tapir
Bing was the olny free GPT4 then, last week, they shut it down so now it responds with something such as "I can only do searches so frame your input as a search request".
I don't want a gpt search
I don't know but I used it for a couple of days and it was better in a lot of ways and in some not as good as 3.5
same prompts no longer worked so I went back to 3.5
worked for 2 days then I got the above so I guess Bing was massively hit, or something.
dense tapir
wispy nest
quite a mix there general
rain gazelle
cool dragons
dense tapir
Thanks. That second one has seen a lot of action I think.
wispy nest
third one is dope as hell looks like some movie still
dense tapir
third one is dope as hell looks like some movie still
You mean the woman?
wispy nest
whoevers in the mask
dense tapir
Yep
Not sure what she is doing with those glowing orbs she is stroking.
Maybe seeing someone in a distant land?
Snooping/spying
One thing SDXL is supposed to do, unless last minute changed is any rez you want so 540 vs 536 or 544
dense tapir
oak osprey
wispy nest
see, 1.5 can also add wrinkles ;P
wispy nest
In my eyes the skin is better than what 2.1 can make, it feels more "natural" somehow. I wonder why 🤔
zealous geyser
#🏞｜general-with-images s golden fish
wispy nest
heh, that's what I see on images made with 2.1, they are so, leathery or extreme :P
gritty lichen
No
smoky oak
Same prompt and seed. No negative, vs standard/general negative, vs realism/cinematic negative
oak osprey
i hope you get a watermark
oak osprey
in the gen
smoky oak
I am confused
oak osprey
in the corner
oak osprey
smoky oak
summons your cat to walk on your keyboard while training
wispy nest
no one deserves a watermark in their image :(
smoky oak
AGAIN
dense tapir
oak osprey
this will accidentally press a button your mouse when it steps on it. and click delete on the folder! 
AND THEN BREAKS THE MOUSE because it's enormously heavy
oak osprey
*summons your cat to walk on your keyboard while training*
jokes on u, entropy is GOOD
so thanks!
dense tapir
@smoky oak Sen came back earlier and released his model on civit.
smoky oak
<@296720689468145664> Sen came back earlier and released his model on civit.
great, hope his annoying ass got banned again
dense tapir
hehehehe
smoky oak
o yeah and got banned in like 4 min
good lmao
oak osprey
George Nepson
dense tapir
this time it was Sen the Hydra
smoky oak
Never been so proud of this server
smoky oak
this time it was Sen the Hydra
He has come back as sen the hydra a couple times
dense tapir
yep
wispy nest
what happened? what about getting banned?
smoky oak
you know, if he wasn't such a dumbass, he could come back without making it painfully obvious its him
dense tapir
With the new name change thing about to happen I dunno
smoky oak
what happened? what about getting banned?
Annoying and problematic person who keeps getting banned keeps coming back expecting a different outcome lol
wispy nest
Annoying and problematic person who keeps getting banned keeps coming back expec...
they did something wrong? or just something like spamming?
dense tapir
yes, and yes
smoky oak
both
they did everything wrong, were extremely rude, and are over all just a gross human being IMO
dense tapir
oak osprey
smoky oak
<@296720689468145664> https://huggingface.co/ptx0/pseudo-journey/resolve/main/ps...
does it work now? owo
dense tapir
This one is yeah
oak osprey
but... this converter has way more error handling
wispy nest
they did everything wrong, were extremely rude, and are over all just a gross hu...
doesn't sound like a good thing, I've noticed a couple "weird" people when it comes to ai art from time to time
smoky oak
a few is an under statement
smoky oak
like, its one thing to be into or do weird things in privacy, but then there are some freaky ass people that decide to blast their degeneracy
oak osprey
mi model a thiccy boi 😁
dense tapir
like, its one thing to be into or do weird things in privacy, but then there are...
Nep, which has devolved into just being a pest.
smoky oak
Nep, which has devolved into just being a pest.
yeah, just waiting for nep to get banned like sen did
dense tapir
He comes on at odd times like last night
oak osprey
The one you linked to is only 4.8
smoky oak
both are problematic and hateful people. Nep being actually worse for intolerance and all forms of "phobia"
smoky oak
oak osprey
GiB vs GB maybe
dense tapir
both are problematic and hateful people. Nep being actually worse for intoleranc...
Nep sure does like their school girl aged Anime though.
wispy nest
like, its one thing to be into or do weird things in privacy, but then there are...
yeah, I've noticed that there's people who don't really have any breaks on what to show to random people.
There was a person who I helped with some AI art last year and they just got weirder and weirder. Stopped responding when they started talking about super private things :/
oak osprey
one of the most annoying things in the world, Sytan.
Windows measures in Gigabytes.
Most of the world measures in Gibibytes.
smoky oak
Most of the world measures in Gibibytes.
Whatever you SS's also uses GB not Gib
dense tapir
yeah, I've noticed that there's people who don't really have any breaks on what ...
Yep, it does weird things to some, but those people were already mentally unstable and on an edge.
oak osprey
they did the fuckin suffix wrong which is a battle of its own
oak osprey
it does
smoky oak
@oak ospreyModel is broken
oak osprey
you're broken
smoky oak
you didn't provide a config, again
smoky oak
wait, was the model a ckpt or safetensors?
wispy nest
Yep, it does weird things to some, but those people were already mentally unstab...
I agree on that. It's a really weird and awkward situation whenever it happens :/
oak osprey
wait, was the model a ckpt or safetensors?
ckpt is what the script calls it
smoky oak
you provide it
they make them for you when they are needed
smoky oak
ckpt is what the script calls it
and its 4.8 GB?
oak osprey
who is they 
smoky oak
who is they <:chibiYell:1009552154480087101>
all of the training programs
oak osprey
no
dense tapir
I agree on that. It's a really weird and awkward situation whenever it happens :...
Tell you what if it isn't straight up fed's busting down doors it is Anime. Not sure what Waifu Anime does to the mind, but it sure does mess with a lot of people.
smoky oak
is your model supposed to be 4.8GB? Cause what you linked to is 4.8GB, which is the same size as the last model
smoky oak
ok so did you link to the wrong file?
oak osprey
no
smoky oak
how does that make any sense
oak osprey
windows measures in GiB
oak osprey
i said it backwards earlier
why do people like
read the words i say
and then
take them seriously
smoky oak
ok, I was about to say lol
smoky oak
One of my first LoRA's was of him lol
oak osprey
ok imagine i just posted that roll eye gif again
im not doing it twice, but
i would
smoky oak
back when I used protogen and it murdered all celebs
smoky oak
you are obnoxious sometimes, you know that? lol
Side note, I shared this yesterday at a very weird time.
This a new little song demo I whipped up off of a random spurt of inspiration.
The vibe is supposed to be more like high energy exploration in a kind of video game I guess
oak osprey
when u download the file but u already had the file and ur like wheres the file but u need the file and the internets slow
and it says 4.81 here
smoky oak
slow internet moment
oak osprey
yours is corrupt
sytan's hardware is broken
its the PSU
draining and hurting the bits of storage
smoky oak
@oak ospreyYou absolute swine
your model does have a config
its literally on your hugging face
oak osprey
thats different i think
oak osprey
its literally on your hugging face
you think i have a hugging face? 
thats so nice
dense tapir
oak osprey
@smoky oak i was commissioned to do some photo edits for a wedding couple and they didn't give me any specific instruction and they really hated it so i made this video to show them how much work it was but the video only ended up being 20 seconds long so i had to give the money back
oak osprey
i tried to sell it to them as a metaphor of their unity through the pastor
i needed that money so bad
i really wanted to buy another funko pop and i had to wait an extra week
smoky oak
I wanna see if your model is finally the one that fixes the terrible artifacts in 2.1
cause it looks like it does from what you have sent
oak osprey
it's not without like a shitton of UHD, 8K, Sharp, blah blah and negatives
smoky oak
I know, but still
oak osprey
but at least without negatives and all that i don't hate the output so it's promising
smoky oak
I just need that damn proper yaml
oak osprey
like your Handsome Men model, okay, it outputs all of its class images as handsome men
smoky oak
although, it seems like your VAE is what has broken it
smoky oak
2.1 doesn't allow you to apply other VAE's
smoky oak
alright, fine, then A1111 and all of the other GUI's don't let you
smoky oak
your model generates, but all it outputs is NaN's
oak osprey
idk i have 8 minutes to download it and then i'll load it in to my broken-ass copy of poop stain tasters v4.0 or whatever sd-webui calls itself
fuckin hate that thing
mossy garden
have a question; is it possible to reliable replicate an art style(sprite) using something like lora + dreambooth to train it? heres an example of the art:
new to sd
smoky oak
you could do a LoRA of it if you have enough high res images
Or you could use ref only in controlnet to make some changes but maintain most of it
oak osprey
@smoky oak i saw a post on the sd-webui page from a dev asking for Diffusers model support and they gave up after 6 weeks of arguing and wrote their own basic Gradio application instead
mossy garden
theyre low res by default because they are sprites
oak osprey
have a question; is it possible to reliable replicate an art style(sprite) using...
DeepFloyd is trained on 64x64 images and upscales them strategically to 1024x1024
mossy garden
would controlnet be a better option
oak osprey
i think DeepFloyd is your best option
smoky oak
I am honestly not too sure, never worked with such low res images
oak osprey
it's fantastic at pixel art
smoky oak
i think DeepFloyd is your best option
just casually name one of the most extreme and non accessible methods lol
oak osprey
pfft don't listen to the haters
mossy garden
how so?
oak osprey
10GB 😤
mossy garden
i have a 3080ti, will that suffice
oak osprey
dude please do, it's pretty interesting
wispy nest
well, pixelart is possible, but I haven't really tested it more than this :P
oak osprey
i think you will appreciate the photoreal
smoky oak
although that seems suspiciously low VRAM for such a huge model
static tusk
The main difference I see with Pseudo is the config.json has this extra name_path entry for vae. Models like DD or illuminati dont have that
realism engine and artius dont either
oak osprey
idk why that's there
smoky oak
@static tuskDo you have any idea how one would go about getting his model to run is a standard GUI?
I am actually willing to try 2.1 as his model doesn't seem to be riddled with the issues 2.1 ships out of the box with, and I really wanna try it
static tusk
I think he has configured the model to refer to a huggingface vae. Just a guess
mossy garden
from what im reading deepfloyd is for going from low res toa higher res, what if the output images dont need to be a higher res? would this still be the best approach for sprites?
smoky oak
where does one even get the deep floyd model
oak osprey
unless you have a really good downsampler that works in your format.
smoky oak
I am lookin for it
oak osprey
it's on huggingface
i am not sure how you'll run it
do you just want to try the Space first?
smoky oak
All I can find is the demo for it on hugging face
I wanna use the actual model, since you said I can
smoky oak
how did you
smoky oak
all that popped up for me was the IF
smoky oak
Ah, so that is the only way to run deep floyd locally?
mossy garden
how do you train it with images?
oak osprey
Ah, so that is the only way to run deep floyd locally?
i mean, there's other ways, but this is straightforward
oak osprey
how do you train it with images?
in the diffusers repo there is a scripts folder with dreambooth_if.py in there
smoky oak
Alright, I think I will pass for now actually. I don't wanna go through setting up even more stuff
smoky oak
oh god-
wispy nest
well, now I know more about ai pixel art :P
smoky oak
the deep floyd demo is terrible lmao
what in the hell happened lmfao
it started so good, until it upscaled it
oak osprey
import os
import torch
os.environ["FORCE_MEM_EFFICIENT_ATTN"] = "1"
import sys
from deepfloyd_if.modules import IFStageI, IFStageII, StableStageIII
from deepfloyd_if.modules.t5 import T5Embedder
from deepfloyd_if.pipelines import dream, style_transfer, super_resolution, inpainting
import torch.nn.functional as F
import random
import torchvision.transforms as T
import numpy as np
import requests
from PIL import Image
import torch
import re
device = "cuda:0"
if_I = IFStageI("IF-I-XL-v1.0", device=device)
if_II = IFStageII("IF-II-L-v1.0", device=device)
if_III = StableStageIII("stable-diffusion-x4-upscaler", device=device)
t5 = T5Embedder(device=device)
prompt = (
"ultra close-up color photo portrait of rainbow owl with deer horns in the woods"
)
count = 1
result = dream(
t5=t5,
if_I=if_I,
if_II=if_II,
if_III=if_III,
prompt=[prompt] * count,
seed=42,
if_I_kwargs={
"guidance_scale": 7.0,
"sample_timestep_respacing": "smart100",
},
if_II_kwargs={
"guidance_scale": 4.0,
"sample_timestep_respacing": "smart50",
},
)
if_I.show(result["I"], size=3)
if_I.show(result["II"], size=6)
if_I.show(result["III"], size=14)
ah yes, looks lovely
oak osprey
this one isn't great because it relies on their crappy deepfloyd module
hold on
i have a way that links it into CTU
mossy garden
the notebook one?
smoky oak
there is a huge difference between using 10GB VRAM, and being able to run on a 10GB GPU
mossy garden
what if you dont need upscaling?
smoky oak
god, these are nightmare fuel lmao
oak osprey
what if you dont need upscaling?
it uses very little vram then
the problem is it loads three models
you can load one, delete it. and then load the next.
but most of the scripts are not written to do that
oak osprey
god, these are nightmare fuel lmao
sounds like a skill issue
wispy nest
heh those results are about what I thought it would be. It's always like that when people talk about a model than showing it more often than not :P
oak osprey
smoky oak
these are so cursed haha
smoky oak
lmao
mossy garden
kind of new to the sd space, however i am a software dev. I just need a step-by-step on how to run this. I have jupytr and loaded the df notebook, and I have a dataset with a few hundred images of the sprites I will use. you said the script for training is in the huggingface repo?
smoky oak
I thought deepfloyd was a lot better than this lmao
oak osprey
size mismatch for model.diffusion_model.output_blocks.11.1.transformer_blocks.0.attn2.to_k.weight: copying a param with shape torch.Size([320, 1024]) from checkpoint, the shape in current model is torch.Size([320, 768]).
wat the hell is that
smoky oak
no idea
oak osprey
cuz it thinks it's 1.5?
smoky oak
your model throws out dozens of those
oak osprey
======================================================================================================================
The most likely cause of this is you are trying to load Stable Diffusion 2.0 model without specifying its config file.
See https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features#stable-diffusion-20 for how to solve this.
======================================================================================================================
Stable diffusion model failed to load
sounds like people when 2.0 was released ;P
oak osprey
😭
smoky oak
this one is pretty cool
although, I have a feeling that normal SD models will do better 😅
yeah, I would have to say that SD won here haha
and it was way faster
oak osprey
although, I have a feeling that normal SD models will do better 😅
gotta compare it against DALL-E 2, and Imagen, or other Pixel diffusion models which are far worse and less performant, and more bloated
you don't have to beat the bear. you just need to run faster than your friends
smoky oak
gotta compare it against DALL-E 2, and Imagen, or other Pixel diffusion models w...
You know what, good point
They are very different types of AI image generators
dense tapir
oak osprey
They are very different types of AI image generators
try more text prompts in there to try and get it to arrange letters using objects
the candy on the plate one looks amazing
smoky oak
The more I mess with it and get text out of it, the more I just want to use SDXL lnao
oak osprey
man you and i are built differently, i can't wait to fine-tune it
you just throw beautiful toys in the trash
ok, try your redhead girl prompt but try to have her standing in front of a marquee like in front of a movie theatre that says "CINEMA"
smoky oak
man you and i are built differently, i can't wait to fine-tune it
But you don't wanna finetune SDXL?
oak osprey
ima fine-tune all the shit
i haven't messed with SDXL enough to know what i can even bring to the table
smoky oak
it does really good text, just like deep floyd
wispy nest
redhead girl prompt but try to have her standing in front of a marquee like in front of a movie theatre that says "CINEMA"
I, erm... maybe it says cinema in another language. Can't rule it out yet! :P
smoky oak
SDXL base model beats some of the best 1.5 finetunes right out of the box
ok, as bad as this looks, the fact that it knew what I wanted is pretty dope haha
oak osprey
well i'm glad you can still see some value in it lmao
i love prompt coherence of the new models which is what drew me into 2.1
i asked it for something really really stupid and it did it
smoky oak
well i'm glad you can still see some value in it lmao
oh, I didn't say that deepfloyd isn't cool
oak osprey
ugh this dumb ass automatic1111 is downloading torch again because it installed the AMD version the first time
smoky oak
you seem to always think that me not praising something means I hate it, like you keep saying with my view on 2.1
when thats not the case
deepfloyd is neat
I just think that SDXL will be considerably more useful and versatile when we get access to it
oak osprey
i was surprised to hear that the stage 3 upscaler for DF will be heavily trained and improve the stage 2's output monumentally
dense tapir
oak osprey
so that is still something i'm also looking forward to 😄
pixel diffusion is kinda neat because of how efficiently it can integrate into existing art workflows
smoky oak
i think you would be super interested in SDXL
oak osprey
oh yeah i'm already excited for it but it's not out yet
dense tapir
Only after I get my AMD card will I be interested in any of the newer stuff.
smoky oak
so you are going AMD
oak osprey
start a gofund me and say it's for medical issues
smoky oak
you keep switching lmao
I am genuinely considering making a shop for AI image commissions and LoRA commissions on Etsy, filing an LLC, and riding my whole new PC/laptop off on taxes lmao
dense tapir
I would love to get an Nvidia but #1 I can't take Jensen any longer. #2 the sheer price for anything with the ram I want is beyond hideous, and #3 Jensen is saying you want it then you pay what I say or get lost. Which resorts back to issue 1.
oak osprey
fucking dumb thing. man. i can't get this working locally at allllll
dense tapir
then that is you
oak osprey
it wants to install AMD torchvision no matter what
its these newfangled automatic1111's
at this point, i don't care if a single other person never uses this fuckin model
if you all want it working, figure it out
smoky oak
bruh, installing A1111 is like... so easy lmao
dense tapir
The only issue I have with Nvidia is actually two things. Lisa Su acts like Jensen's little ho and they take so long to release ROCm for their cards. Estimation for the 7k cards (5.6.0) is 3 to 4 more months.
oak osprey
it's so easy even an engineer can't do it
it installs, it runs, it installs ALL THE WRONG DEPS
i have the nvidia version of torch and it goes to grab the AMD version of torchvision
smoky oak
bro, i legit installed A1111 for like, a dude in his 60's yesterday, and all I did was send an article to him lmao
smoky oak
i have the nvidia version of torch and it goes to grab the AMD version of torchv...
only reason it would do that is if you copied the AMD repository instead of the NVIDIA one
They are two different SD A1111 links
oak osprey
no, it's because i'm on a laptop, with an AMD APU, and a NVIDIA discrete GPU
and it chooses the wrong fuckin thing
smoky oak
still, you choose which by what link you use
dense tapir
yep
oak osprey
look
there is ONE command for Linux users
not two repos
find me the nvidia specific one
smoky oak
my friend has an AMD GPU and CPU, and it still installed NVIDIA cause he used the wrong link
oak osprey
there's not even a command line arg to disable one or the other
smoky oak
because they are two separate installs
smoky oak
there's one repo
two people who actually use this daily are telling you that you are wrong, i don't know what else you want
dense tapir
Sytan, I saw the best AMD from last gen and the best this gen and honestly for SD stay away from last gen. This gen doesn't evne have rocm yet and it was already 14it/s. For a card with no transformers, and no rocm that is actually surprising.
oak osprey
i'm sick of you telling me that i'm wrong, i'm a damn software developer, and i'm telling you. scroll down to the fucking install section.
static tusk
there's one repo
https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Install-and-Run-on-AMD-GPUs#install-on-amd-and-arch-linux use those instructions
oak osprey
Automatic Installation on Linux
Install the dependencies:
# Debian-based:
sudo apt install wget git python3 python3-venv
# Red Hat-based:
sudo dnf install wget git python3
# Arch-based:
sudo pacman -S wget git python3
Navigate to the directory you would like the webui to be installed and execute the following command:
bash <(wget -qO- https://raw.githubusercontent.com/AUTOMATIC1111/stable-diffusion-webui/master/webui.sh)
Run webui.sh.
Check webui-user.sh for options.
Is it working? No? You did something wrong lmao
oak osprey
i need nvidia not amd
smoky oak
I have installed SD probably 2 dozen times across my PC and other PC's, and had 0 issues, AMD, or NVIDIA
oak osprey
you don't use linux, and windows has no amd support, please stop talking
smoky oak
So now you're just gonna be a dismissive ass? Fine, screw it then, keep having problems
see if I give a damn
I'll go back to my WORKING one click install
oak osprey
you have literally been ragging on me about how "you're doing it wrong, you're wrong", and i'm just following the "automatic install" and i'm telling you, that thing sees "AMD" in my lspci output and then goes full balls-deep on ROCm
and it's a LAPTOP APU, not a AMD GPU, it's broken, because it is not taking that into account
ROCm doesn't support laptop APUs.
oak osprey
i doubt i ever will because this is honestly frustrating and then the documentation is shit. it references there, webui-user.sh for more arguments, and that file doesn't even exist
smoky oak
I wanna experiment with some alternative ways of training a style LoRA, just to F around and find out
dense tapir
yep, can do that with fast local cards
smoky oak
Yeah, my new GPU can do BS24 no problem at 512x512 lol
oak osprey
lmfao i don't even think anything on the nvidia gpu install page applies
there is no linux section
just windows
smoky oak
lol
dense tapir
Even on AMD the page is for windows, wtf?
oak osprey
i guess i could do the WSL2 instructions
smoky oak
Could always waste 7 hours trying to get VoltaML to work only for it to be slower than A1111 if you want
oak osprey
so like, the only point of me getting this to work, is so that i can validate whether the safetensors / ckpt file are working
i could care less if i ever actually got this to run
smoky oak
news flash, they aren't working 😅
oak osprey
i will probably delete it once it does verify that it worked
dense tapir
you could care less but actually do care?
oak osprey
i'm surprised so many people are doing this on windows
dense tapir
I dual boot and I really don't like Linux all that much. In the CLI days I loved it.
oak osprey
i'm a kernel developer, so, i am biased toward it
but, any other OS at this point feels like a straitjacket
can't do like, deep dynamic tracing on windows like i do here. performance is better, no PowerShell to worry about.
wispy nest
well, being a dev is all about getting something that doesn't work, to work :P
oak osprey
i don't want to ever use this software because i fucking hate its developers
wispy nest
that's fair, I'm like that when I hear someone say regex
shudders :P
smoky oak
I am not a developer, yet I still managed to get the massively less documented Volta ML running in ubuntu through WSL2 on windows lmao
tho, it was not worth it
very much not worth it
oak osprey
oh, if i care i will stop at nothing to figure it out and make it work, open a pull request, whatever. but i see repeatedly these people interacting with their users in a manner that i would never do a thing to help them out
static tusk
I would disable the AMD APU
oak osprey
"disable" it?
i have to have it enabled for power efficiency
nvidia blows chunks in that dept
static tusk
in windows you can easily
oak osprey
i was looking for some kind of env var to set but AMD also lacking in documentation
damn, there's 7 different launch scripts and they're all undocumented
i could understand if this were an unpopular toy project but it's a massively popular piece of crap
dense tapir
I would disable the AMD APU
At least until installation is done.
oak osprey
@dense tapir when it starts up, it finds it, and begins installing dependencies
i went and manually removed torch and installed the correct versions
it mangled the venv trying to undo that
i'm just giving up, sorry
dense tapir
In linux I forget how but you can tell it which is the primary device.
yours may be set as 0 AMD and 1 Nvidia
oak osprey
naw in the webui script if it sees AMD at all, it takes precedence. one of the last comments in that bug report is suggesting that it be changed to give NVIDIA precedence and only AMD if that's there without NVIDIA
which is, honestly, how you'd expect it to be
don't wanna have to some how find a way to hide OS libraries from this thing
static tusk
naw in the webui script if it sees AMD at all, it takes precedence. one of the l...
you arent using Auto permanently. just disable to test the model then revert
oak osprey
like, literally anyone of you can use diffusers to do this conversion too if it's so easy
there's probably even an extension for it
dense tapir
Auto has been asked for over six month to allow diffusers but no
oak osprey
@static tusk i can't disable it, the laptop's display is always wired to it. i would have to uninstall the drivers for that to work, and installing them and getting it working to begin with was a challenge, as, this is Gentoo.
the discrete GPU can be disabled entirely, but the APU can not. and i don't see the value in that
a year ago i experimented with different setups to see if the OPTIMUS driver stuff that nvidia does adds latency and i managed to yank all the AMD init stuff from my kernel and force everything over to nvidia, but the laptop's LCD stopped working, and it didn't improve things, other than the baseline power consumption which goes up by 30W
dense tapir
Yeah, APU disable is in bios if at all
dense tapir
then fucked
oak osprey
yes
i wish it were sunny out, i would just start the 4090 system but it'd just kill my power storage for no reason
and like Sytan says the latest ckpt didn't even work for them, so what's the point in even getting a1111 working
that little boat is a bit close to the shore
dense tapir
oak osprey
another 9800 regularization images made now lol
damn it, too far.
work it, work it, work it
orphan keyword -> daft-punk 😁
cyberpunk -> steampunk
dense tapir
dense tapir
static tusk
lol
dense tapir
static tusk
looking at this guy 😆
dense tapir
looking at this guy 😆
Damn, he saw the SS and Flounder and just froze in shock.
static tusk
yep
dense tapir
oak osprey
static tusk
dense tapir
smoky oak
lmao
oak osprey
imagine having to make kitten mittens for all them feets
it nailed the feet shape tho
she has the cutest little round ones
oak osprey
oh i thought that was tail
i mean the front two
@smoky oak i'm up to 20gb of reg images
smoky oak
@oak ospreyI just made my first successful style
I am gonna upload it to civit semi soon
oak osprey
DARKNESS, like your SOUL?
smoky oak
I am shocked at how well it worked lmao
oak osprey
ugh i need sleep. adios
wispy nest
lots of cool stuffs. love the funkos lol
frozen socket
i hope you get a watermark <:KEK:747148843434377316>
is that the visible or invisible watermark option ?
/me wanders away whistling
dense tapir
frozen socket
ignore this post
smoky oak
who?
dense tapir
Thank you as search the web via google, nor on civit brought that up for me.
I use it sometimes and was wondering if it had a trigger word as I just downloaded one of his others that did.
frozen socket
Thank you as search the web via google, nor on civit brought that up for me.
Note how I changed the search term used based on your original input.
think ouside the box , file names are often concatanated versions of friendlier names
So starting with sdartCompleteEdition_v2Base21 its a simple case of breaking it out into it's constituent parts and deleting some bits
sd art Complete Edition v2
dense tapir
I am finding that as time advances these search engines are getting dumber. :/ an actual file name, even down to the extension used to find all kinds of stuff on Google.
frozen socket
I am finding that as time advances these search engines are getting dumber. :/ ...
nah I think users are getting dumber but thats just a cranky old git speaking ;o)
dense tapir
No idea but I used to find all kinds of programs by knowing its exact filename and extension. Findmenow.exe etc... now it is harder and harder to do it.
frozen socket
a lot of that is down to the hosting websites and the way they store files these days and whether or not they have robots.txt files
This is the way
dense tapir
I need to find a search site that disregards that robots.txt as it is not mandatory you adhere to it.
frozen socket
after all it sayas a lot that you couldnt find it on Civitai directly so if a website cant find its own file how can a search engine?
wispy nest
websites use more javascript than it did before, harder to find what they hide :P
dense tapir
I agree with both of you
smoky oak
after all it sayas a lot that you couldnt find it on Civitai directly so if a we...
I have to use google search to find civit models all the time, cause civits search is trash
dense tapir
this is why I did the exact filename hoping google scraped it
frozen socket
you just have to think like a tangerine
dense tapir
btw, Guizmus's model seems to have a trigger word
wispy nest
because if you really want to find the download for that sdart thingie, then you'd need to know its "real" name :P
dense tapir
makes the end result much better
I have always hated javascript
Lexica is even worse
wispy nest
what is it you like? can't say I ever heard what you enjoy :P
wispy nest
yeah, I don't think I ever heard, or rather, read that you enjoyed or want/liked something at all :P
dense tapir
I have said it numerous times but you may have not been around. Dark, macabre, makes you wonder what is about to happen, or with the next step. Dark horror tones but not gorey.
if you notice a lot of my stuff I post is darker almost to a burnt state. I like other stuff too though
wispy nest
you mean something like this or?
quartz sun
you mean something like this or?
This is incredible. Reminds me of Beksiński!
dense tapir
wispy nest
This is incredible. Reminds me of Beksiński!
by John Bauer was my artist prompt for it. I believe it was the strongest prompt as well
quartz sun
pretty new to AI art but I'm really looking forward to playing around with it
dense tapir
Oh, I forgot to add that I even went to the author and clicked models and nothing happened.
CivitAI is a really poor site
quartz sun
I'm getting almost the same issue! I posted in #🤝｜tech-support about it
cunning hornet
I'm getting almost the same issue! I posted in <#1002602742667280404> about it
thx let me check for it
wispy nest
this is as much burn I can handle :P
dense tapir
That is about it. A little lighter but see, for me at least, who is that over there? Is that a murderer in the shadows? What happens next?
Feel the burn, errr crunch.
wispy nest
what happened next was me trying the same thing in a 2.1 model ;P
dense tapir
Needs more cowbell.
@hard wagon Remember we talked about Smolones with the author and how Civit did him? I just alerted him that his model is back up so his appeal, that they did not answer, must have worked.
wispy nest
no cowbell >:I
wispy nest
and this is where I left off with one of my embeddings. It's two of my embeddings on the normal 2.1 model
wispy nest
I wanted 2.1 to be darker, and have some more "horror" elements
dense tapir
Well, there is a new technique for all of the versions to actually get real black. Sadly, it means to get it the models have to be retrained because it isn't offset noise, which is just a bandage approach.
I personally don't care much for offset noise because it is akin to just jacking up the contrast.
This new techinque to solve a problem SAI gave us is brilliant.
Joe Penna's DB has it now