#🏞|general-with-images
1 messages · Page 77 of 1
yup its greed, plain and simple
When you get away from the CPU/GPU makers I don't see near the greed
Actually, I am surprised at the rest of the stuff's pricing. AM5 being new I see higher prices as I did when AM4 was new but AMD/Intel/Nvidia is just beyond greed.
This amount of greed goes into hurting humanity/society. Greed is alright but there is a breaking point.
This only empowers Jensen to finally do what he has been signalling he wishes to do for a while now and that is leave the consumer market and concentrate on AI/ML solely as that is where the money is. I bet dollars to doughnuts the 5k series will be the swan song for consumers.
@smoky oak so apparently sd-webui dreambooth just stops tuning the TE after a certain % of the run, and LoRA instead uses multiple learning rates
its sadly what makes the world go round. smdh
I always lived a simple life and when I ran a software company in the 80's I did then while my competition was greater. Simple, let me make enough to pay all bills + 10% and I am happy.
you, as a customer, are happy too
the 80s? Damn General you got some real life xp
this, Jensen takes a 10% pay cut to still make 20 million per year blows my damn mind.
yeah poor guy lol
I never, ever, was down with the mindset that a CEO should make 200-400 times the wage of the average worker under them NOR when the company does bad they still get their bonuses while everyone else gets nothing.
During the 2008 great recession I saw companies going out of business and the workers chained out from their jobs the next day to have to file for their last pay cheque while the CEO was getting millions in their bonuses. It was in their contract is utter shite.
yeah and then we have to cut the CEO's taxes again too
Company utterly collapsed so cue in the great Willy Wonka Gene Wilder meme here
and make sure he has a 0.0 inherent tax while he tells the rest of us to pick ourselves up by our bootstraps
but I guess we getting a little ot for images lol
Well, don't get me wrong I am not a communist, or the AoC eat the rich type but there is a breaking point.
im not gonna fight for the wealthy they have more than enough
since reagan we cut taxes to an insane level for the people already with enough to live 20 lifetimes
I don't have a problem with that at all. Where I have a problem is they have loopholes, legal, that they take to pay nothing. Get rid of those legal loopholes but never gonna happen.
maybe we can cut their taxes again at least
Did you know if I work for someone I pay less taxes than if I work for myself because I have to pay their share too so my tax rate shoots up? It is pretty repressive.
people like dipshit trump talk about making america great again, well when was that? post ww2? then lets change the tax code to that to give the revenue
look at eisenhower, nixon, ford tax rate, they must not have been republicans but hardcore commies
i dont think your tax rate should shoot up if youre making under like 5 million a year
the bush tax cuts or regans trickle down was for the top 1 percent
I almost typed something so I am stepping back as this is getting political.
and it killed the econmy in 92 then nearly caused a second depresion in 2008
but we better do it again, im sure trickle down will work this time
yeah little ot for images lol
yep, but we surely would have a lively discussion over this I can tell.
btw, I lived through communism at the tail end. No thanks, but I keep being told that was not real communism so what do I know. shrug
lol as if democratic policies are communist
like i said, nixon musta been a commie then
his tax rate was a s hit ton higher than clinton, obama or biden
fucking general eisenhower was a red, who knew
How did you just jump over the cavern connecting what I said with democrats?
Ive lived among white racist fascit religious fucks, so I know how you feel bro
well what were you talking about then, in context of this convo
USSR
youre implying not giving the one percent another tax cut is tantamount to communism
and anytime we talk post regan about raising the riches taxes, but but its communism to do that is the argument
so like i said, what was eisenhower, nixon and ford?
because their top rate is higher than clinton or obama or biden
I can tell you. You ready?
sure ive heard it all before
somehow my new model can make snakes better
Because a president isn't in power for that it is Congress who holds that power and if you look most of those were of the other party in control. Same thing happened when the debt ceiling was raised more times under a rep president because Congress is the ones that did it and they were Democrat controlled.
not, uh, perfectly, but
All a Pres can do is sign or veto
the debt ceiling is to pay bills bro it doesnt actually raise the debt
youre like a walking fox news talking point
fox news isn't even conservative these days. More like MSNBC or The National Enquirer
youre out here so close to figuring out how the rich get all the breaks but then youre still lockstep to make sure they get another tax cut. or how about 2017 when we reduced corporate taxes from like 35 to 20. yeah they needed that
thank god bill gates got another tax cut
anyway, I am an environmental classical liberal, social moderate, and a hard line fiscal conservative.
in other words, a republican
actually, until the aoc types got in, I voted more for Dem candidates than Rep ones.
the aoc types lol. shes a congresswoman from ny. theres 538 of them
but she is a woman
and that brings something out of people
aoc type, iow a damn progressive not a true liberal (hence my usage of the term classical liberal).
whats a true liberal believe
imagine the size of the popcorn
'classical liberal' smh
simple - I may not like what you say, but I will fight for your right to say it. Progressive says "Words hurt and shut up".
fuckin cornado
ive yet to meet a 'classical liberal' who doesnt vote rebublican. just own it
yeah like Ron DeSantis
banning books, whining about trans people existing
Good Bushite that one
what did AOC ban?
btw, good bushite is a slur
@smoky oak my v2 model ^
iow, I don't care much for Ron
someone post a desantis gen lol
does it do him?
I wouldn't sully my computer with his likeness 🤮
scene of carrots with cinematic undertones
I like the vibrant colours. How does it do with portraits?
oh you mean, of people? 
just make one of those trolls with the wild fuzzy hair
old ronny boy
this model has some kind of bias i can't place my finger on
is that good? lmao
he has a basket of potatoes and i don't know why
oh. maybe that's supposed to be the spaghetti he should be eating
those last few are a little over saturated
hard to control the machine 😛
try with lower cfg
it switched seeds too, lol
apparently this text encoder is a lil whacko now. i neeeeeed to add the layer freezing code
what does it mean
its so weird
chill vibes
pretty close
interesting they never have pupils

Brad Pitt in The Hobbit
bradd pitt as frodo photoreal photoreel
Ok so I almost just nuked my PC
Wow, that was one of the biggest oversights I have ever had
I seriously hope I didn't damage anything
they watch over you
dual psychedelic beefcake godsters
My PC is in over temperature protection
I'm letting it cool down, I can't believe that happened
how it happen. did you stuff a pillow behind it?
My mobo was apparently 70C, GPU was 94, and CPU was at 102, but I got no warning
how
The new graphics card I have is extremely bright when I'm trying to sleep, so I have a sheer curtain that I put over the glass panel when I'm sleeping, but because of the way it was sitting it got sucked over the intake, and when I opened my window I managed to push it behind the exhaust as well, so my computer was Hot boxed severely, and I didn't see it
called it kinda
It also doesn't help that I had my MIDI keyboard on top of it, so it blocked all of those exhausts as well
age progression of a random star did not go as he expected
why is this model so fond of making like, collages 
It doesn't help that I was in the middle of playing cyberpunk at maxed out settings when that happened
That really was just a perfect storm of thermals
are you Sheriff Andrew Jones's lil friend?
MSI should have an RGB light control application to turn off the leds
a truck with balls
Yeah, if I wanna leave 15% of my PC's performance on the table lol
I used to use the Gskill RAM RGB controller, and found that thing would use nearly 60% CPU and 7GB RAMhen you woke it from sleep
imagine my surprise when this came out and i was like "thats not a parrot" and then i googled what a parrot was and realised i meant pigeon
i know 😄
its super pretty too
LOL did it make it 4-legged
i love it
researchers: "in the future, people will forget what real animals and objects look like, as AI images proliferate"
me, now: "is that what a parrot really looks like when it walks?"
watermark takes a few points off, so, 7 / 10
That was part of the image.
Generated with it @oak osprey
11/10
yeah you can put that in your negative prompt, watermark
I don't think it meant to do a watermark...
eh it's ok
lol
it's just that they're very common
😛 i was just being picky. it's the only thing i could find
lol
YOU ASKED
i am but a mere shrek salesman, sir
Lol
would you like a jar of our finest liquid shrek
grown on site, right on shrek
Nice
it does it, not me
can't tell if on ocean or moon
my first img
lol thats awesome
8/10, not enough visual context
yeah fair
Is there a way for me to input this into sd through img2img and get it to make it higher quality?
the iron giant + phoenix rising
@oak osprey ?
Iron Giant reborn
ControlNet Tile Upscaling
Is that an extension?
it's controlnet extension yeah
How to install?
hmmm ok
interesting faces but ok lol
it took like 2500 gens to find one of the simpsons this close
elmo goes to aleppo, syria the movie cover, works pretty well
science
"Don't worry! ELMO brought Backup!"
annoying I have to restart A1111 when I create a new wildcard file
just copy the part where it loads the wildcard at the beginning, into the place where it reads the cached version
bam
i wish it wouldn't do that to him
use a square AR
better
oh no, one of gpt35's prompts was Elmo++ joyfully explores Aleppo++'s alleyways--while seeking++ a new cause++ to join--
somehow it really just sucks at doing elmo
like it clearly knows him super well but not well enough
it's so.... beautiful sniff
was i supposed to do something more specific than just spongebob
spongebob squarepants cartoon?
spongebob, happy, hanging out in bikini bottom
spongebob squarepants cartoon
spongebob is basically a no-go lmao
nicolas cage buying groceries
irons are hard to use
asked gpt what it wanted to see nicolas cage do and it said a cooking janitor
I mean... yeah
@smoky oak you know someone mentioned this but I think our one saving grace from the evil greed of Nvidia in AI/ML is the cheaper ASIC market.
Probably Nvidia will buy them up to prevent that or something. :/
Isn't there an opensource ASIC programming model?
I know the best miners were ASIC driven not GPUs back when that was a thing.
Scalpers are out as I went to Nvidia and it showed a MSI tri frozen for 1599 since they no longer make a FE. I click the link that takes me to Newegg and I am hit with
Cowherd and Weaver Girl Flying Kites on Rainbow Bridge
do you guys like this? SD 1.5 and dreamlike photoreal 2.0
bart simpson sewing a scarf
the camera lens appears
having fun with inpainting 🙂
guys i found this image in leonardo ai and it said its made via SD v1.5 but when i tried to make it with stable diffusion v1.5 it gave me bad results , did they use other models or something ?
They have their own model they trained as well as the base model you just need to see which one was used.
it says dreamsharper , i found the model online but it has an extension of .safetensor but i am not sure if it can work with automatic1111
oh i will download it and try then 🙂
🙂
ckpt files allow remote code execution upon load. it's quite devastating
https://github.com/facebookresearch/xformers/issues/631
seems like xformers has a few issues with causing NaNs
New style unlocked. This should keep me busy for a bit.
I genereated this thing by accident recently, so beautiful...
holy 
the new driver is so fast
The AI researchers are mostly on Linux platform but they are releasing the support for MS Windows proprietary tools. Corporates help each other to build their own closed ecosystem for oligopoly. I expected this kind of crap would happen one day when MS embraced Linux in the form of WSL.
i love this guy, haha. takes zero shits from A1111 or CivitAI
it's not done yet
can't get good results with text in IF
ah there we go
trying to reproduce the magic of my pseudo-journey model in DF is hard
"corn field" -_-
🙂
Me when me when me when
When me when me when tensorrt
I dont feel a difference
did you finish training it?
Made this cuz i was bored
what, the original pseudo-journey model is finished as it is. the v2 is a WIP. it is a test bed for my new fine-tuning code changes. so far, i've now got the ability to freeze text encoder layers:
for name, param in text_encoder.named_parameters():
pieces = name.split('.')
if pieces[1] != 'encoder' and pieces[2] != 'layers':
print(f'Ignoring non-encoder layer: {name}')
continue
layer_level = int(pieces[3])
if layer_level < 21: # choose whatever you like here
print(f'Freezing text_encoder layer {layer_level}')
param.requires_grad = False
What
expose & apology video
Enchancr - Open Source Full AI Upscale & Interpolation Suite
[Mafiosnik's Patreon] https://www.patreon.com/mafiosnik
[Enhancr Discord] https://discord.gg/yBXTeVBrqD
[Enhancr Source Code] https://github.com/mafiosnik777/enhancr
[Sudo] https://github.com/styler00dollar
[Hubert] https://github.com/HubertSotnowski
My Links...

When I go to img2img to upscale, I have a screen which looks different than the screen on the upscaling tutorial I'm following. I don't have any of the upscaling settings! (I uploaded screenshots of both my screen and the tutorial... someone help!
my screen is the one with the car images
@tropic shelladvertising
@oak osprey help?
🙏
In the future, you can just use ⚠️ like you well know, or file a ticket with #1010934719455707218
Mb, didn't know, will do
i dont use that

oh... ok :(
Np! All good 🙂
I'm still recovering from surgery, so in case I'm not around, etc, it's just best to file a ticket, etc!
Sounds like he had surgery, not that hes a surgeon. dudes in recovery mode
I actually had back surgery! Spinal f u s i o n. As for my job description, that ladies and gents, you'll only find out by gettin' to know me. Thank you, though.
I was born with a fused L4 L5. I'm trying to formulate a joke with this information, but cannot come up with anything.
i knew a spinal surgeon, he had such a good sense of humour they gave him his own column
You're infused with same kind of spinal comedy, then.
oh man
Too much fun lol
i just started a new training run on my A100 with unet compiled by torch2
wow
cool
i get 3it/sec instead of 3s/it
the improvement in speed is enormous with torch2
so 100,000 steps of training now takes like 15 hours
wow

yeah it also seems to be freezing the text encoder layers properly now that i've added that in
when you rush so hard to clean that you accidentally
and that's the last time i bring my bike to HU ocom to clean it
even the capybara is improving
😮
Howdy
who is the guy under the table?
or that is an alien
cosmic horror
isn't BLIP the interrogation model?
it sometimes throws nonsense
i don't care
I'm pretty sure there is a version of it that does negative interrogation
it's not actually throwing nonsense, it's just telling you how that image is represented in OpenCLIP
and if you're fine tuning a style into a model this is perfect
yeah, im not expecting anyone to go and describe thousands of images
i mostly care if it can find some important subject in a caption and have that tag in it. i can also add those if they aren't there, it's a small dataset
well that's why scraping reddit is perfect for this
people caption things there
ahahaha
licks finger to turn the page
the eyes are an abomination, is this with restore face?
dude, i still don't get all the beef with automatic
really?
what the hell did he do? for all i know, he plays a major roll in most of this community

the outrage toward this prompted him to make "BlackOnly" which has a logo of a bushman with a spear on it. the one i linked to has smiling happy white people on a sci-fi background
oh, that's bad
it gets worse, but i don't want to keep talking about him, he's a really fucking bad dude
technically speaking he also has allowed remote code execution vulnerabilities to exist in his project
i heard he had to do with what happened with novelAI
he has political reasons for not adopting a superior filesystem layout. as far as i'm concerned, he's holding us all back
then why is his WEBUI the only UI that has as much features as it does?
because so many people use and contribute to it. trust me, that's not because of him, it's despite him
huh, so i think if the same people that helped improve the A1111 UI would help improve somthing like InvokeAI, then it would be way better
i don't know why it matters, they're all open source and you can improve them too
i don't use any of it
BLIP properly labelled all of my images, using the Salesforce fork
like, for all i know- controlnet is only available for A1111, and controlnet it one of the main advantages running AI locally has over paid AI
isn't blip made by openAI? Also it's good until you do different aspect ratios
the source for controlnet is right there on the controlnet page, and it is standalone
why would you ever do different aspect ratios? the training data has to be 512x512 or 768x768 square, without exception
the implementations that deal with other aspect ratios do really fuckin terrible things to the images to make it work
like random crop, centre crop. it's no good. they might even downsample and pad to make it square. meaning the image becomes tiny 512x512 with letterboxing
just crop everything to the proper resolution and know you have good data.
like, how would BLIP interrogate this? it might not even get close to the original prompt
it literally doesn't need to be exact
i think you're looking at BLIP from the perspective of someone who expects to use it to create a perfect prompt for reproducing an image
that is not what it is for
@smoky oak ^ lol my model's photoreal toddlers
babies kinda always look all weird so it's okay
Its definitely no realism model lol
but its not going for that, so thats valid
now that i'm training with mostly frozen text encoder, it really wants to make things realistic
capybara before, vs now
First contact with @oak osprey ‘s pseudo-beta established. Over.
the art generator
gpt
first visually decsribe a really creative but abnormal scene using keywords separated by comma in less than 10 words, then write multiple completely unrelated and random less known visual properties, 3 words for objects, 3 shapes or forms, 3 artstyles and 1 color schemes. Write them together in one block separated by comma and mix their order up
Giant toothbrush, fluorescent banana, rusty harmonica,
Geometry, fractals, biomorphism,
Monochrome, neon bright.```

With diffusers on colab. Will explore pseudo-journey as well soon and update
I wish Automatic1111 could do diffusers
this dude is having the time of his life, growing, and evolving, but i think someone stole his mic
freezing 20 text encoder layers means i get very consistent foundations to the images and they don't change in as radical ways, even 2000 steps into training
damn that looks crazy
what model is this
pseudo-journey
huh
just throw it in gpt and make some
Objects: Stuffed alligator, rusty tricycle, antique globe
Shapes/Forms: Amorphous blob, hexagonal prism, S-shaped curve
Artstyles: Art Deco, Art Nouveau, Cubism
Color Schemes: Monochromatic blues, earthy browns
Art Nouveau, rusty tricycle, Cubism, hexagonal prism, S-shaped curve, mermaids lounge on jellyfish, Monochromatic blues, stuffed alligator, earthy browns, Floating neon coral reef, antique globe.```
i have my own auto-prompter in the bot i use
ah
Hey thats my secret sauce!
Generate an++image of a flamingo with a top hat and monocle, balancing on one leg on a pile of--unicycles while eating spaghetti with a fork.
GPT pulled out all the stops for that one
Here's my box of oddities 🙂
Ooh this is nice. Love the contrasting sky. I throw cactuses in collages and montages mostly.
i want to make a prompt that morphs a regular scene using a metaphor
2. Make a list of abstract concepts related to the object .
3. Pick a random abstract concept and make a list of related visual objects or properties to that concept.
4. Pick one of the visual properties from that list and reimagine the first visual description but creatively mixing it with the picked abstract concept.
ideas ?
2. Writing, communication, creativity, education, expression, artistry
3. Expression -> Emotion, body language, facial expressions, tone of voice, symbols
4. Emotion + Pen -> The pen seemed to be filled with every emotion in the spectrum and with each stroke, it poured out onto the page like a visual symphony of feeling.```
2. Abstract concepts: warmth, nourishment, comfort, routine, efficiency.
3. Visual objects/properties related to 'routine': clock, timer, calendar, calendar dates, clock hands.
4. Mix 'routine' and 'toaster': reimagine the toaster as a giant clock with a conveyor belt, where the bread goes in and comes out toasted perfectly every time. The clock hands represent the toasting time and can be adjusted for different types of bread. The toaster-clock gives a sense of routine and efficiency, providing warm and nourishing breakfast every day.```
on pseudo-journey, add UHD 8K sharp best quality to prompts
it is nearly essential to get good outputs


cursed
Generate an image of a forest where the trees are made of smoke and the ground is covered in ash. The sunlight casts a haunting glow across the misty landscape.
yo where do i get pseudojourney
or is it private
Queued: "Create an image of a --talking llama wearing a top hat++ in a --flower garden that is --floating in the clouds++. Bonus points if the llama is also juggling flamingos."
it is on huggingface under ptx0
i asked for a thought-provoking image showing a banana for scale against the world's problems
this is what i got
lmao
Damn you are on a roll. Great stuff!
i think i may have created a prompt to just make normal images

2. Make a list of abstract concepts related to the object .
3. Pick a random abstract concept and make a list of related visual objects or properties to that concept.
4. Pick one of the visual properties from that list and reimagine the first visual description but creatively mixing it with the picked abstract concept.
5. Write a prompt for an image generator that describes the imagined scene in 4. using keywords sorted by descending visual importance separated by comma```
Kinda getting the hang of prompting with pseudo-beta @oak osprey Was getting noisy outputs before.
Nah you are on the right track I'd day. Just need to experiment more.
*say
2. Make a list of abstract concepts related to the object .
3. Pick a random abstract concept and make a list of related visual objects or properties to that concept.
4. Mix the visual description of the object from 1. with the Pick from 3. creatively
Carry out the above algorithm with the object in 1. being "[OBJECT]"```
this works
that model is burnt pretty badly and serves one sole purpose lmfao and anything outside that, it fails almost spectacularly but if you're enjoying it, well, everyone has their vices
every now and then it's funny to get a lil tipsy n look at some malformed gens from a burnt model
It seems to be doing that 80s analog scifi look that I'm kinda into really well. Will experiment more and see what it can do
This for example
certain things it can do really well because they were nowhere near the latent tensor space i focused on, or, they're right next to it, and were also overtrained
idk exactly
it also seems to take a lot of work to prompt well
Empty Nest++: Two birds-- of a feather-- flock together as they navigate the-- bustling streets of Miami++ on their unlikely quest for love and laughter
80s Sitcom Madness++: Recreate an iconic 'the-situation-is-getting-out-of-hand' moment from your favourite 80s sitcom, but add an unexpected twist-- a cast of animals replaced the human characters! Imagine a rambunctious family of chimpanzees run amok in "Family Ties" or a gang of mischievous raccoons turned the set of "Growing Pains" on its head.
I don't think the ai really painted all those things in that prompt :P
gpt35 is out of its league with pseudo-beta. the text encoder on this model is broken lmao
don't know what a "pseudo-beta" is :P
2. Make a list of abstract concepts related to the object from 1.
3. Pick a random abstract concept from 2.
4. Make a list of objects which share the same abstract property from 3 but are distinct from the object in 1.
5. Pick a random object from 4.
6. Describe the object from 1. but it is only composed of the Pick from 5 .
7. Imagine an artwork that is a visual metaphor based on 6.
8. Make a prompt for an image generator that describes whats seen in the image from 7. using only keywords separated by comma sorted by descending importance```
getting there
try this
pls 
Contraception, David Lynch edition.
because i had vaping in the dataset
some of the characters just start like, belching smoke

heh, never wrote david lynch before :P
Need to it out better after me sleep :D
clock mug
hello. digital clock
Scaramouche
:P
1. Pick a random object.
2. Make a list of emotions related to the object from 1.
3. Pick a random emotion from 2.
4. Make a list of objects which share the same concept from 3. The objects have to be unrelated to the object from 1.
5. Pick a random object from 4.
6. Make a list what the Pick from 5. is made of.
7. Describe the object from 1. composed of the objects or materials from 6.
8. Imagine an artwork visual metaphor based on 7.
9. Describe whats seen in the Artwork of 8. using only keywords separated by comma, sort them by descending visual importance
Carry out the above algorithm with the pick of 1. being "Jungle"```
@dense tapir
??
oh nvm sorry for tag then
No problem
1. Pick a random object.
2. Make a list of emotions related to the object from 1.
3. Pick a random emotion from 2.
4. Make a list of objects which share the same concept from 3. The objects have to be unrelated to the object from 1.
5. Pick a random object from 4.
6. Make a list what the Pick from 5. is made of.
7. Describe the object from 1. composed of the objects or materials from 6.
8. Imagine an artwork visual metaphor based on 7.
9. Describe whats seen in the Artwork of 8. using only keywords separated by comma, sort them by descending visual importance
Carry out the above algorithm with the pick of 1. being "TEXT"```
try this
3. Chosen emotion: mysterious
4. Objects sharing the concept of "mysterious": fog, moon, shadows
5. Chosen object: fog
6. Fog is made of: water droplets, air particles
7. "Alien female" composed of: water droplets and air particles
8. Imagined artwork: A portrait of an "Alien female" made entirely of water droplets and air particles, with a misty background of fog, moonlight, and shadows.
9. Keywords: misty, watery, ethereal, formless, moonlit, mysterious, shadowy, abstract```

could probably add a line that just spams fine detail keywords and then i do prompt editing so it adds those fine details after it made the main composition
mmm avocado toast
what is that extension that allows different parts of the prompt to be separated?
like "red coat, brown hat" will apply the colors to the right stuff
avocado crossed with a blobfish and some ketchup from the black lagoon
Bollywood + 80s + South Indians with pseudo-beta
Realism Engine for 2.1
I’m stealing this 😂
avocado crossed with a blobfish and some ketchup from the black lagoon
that's the stuff
A Series I’m working on using SD
I've been playing around quite a bit with block weight merges. Finally managed to rip the pretty asian woman bias out from my favourite mix. Then did some more merges to try and restore the quality again
Yeah I've been using the supermerger extension
I didn't care for that one as much as MBW
Hi folks (again) ! I really need help here. I installed the plugin to my photoshop - but it says that "Failed to load stable art"
I did all that was in the instruction
I have few versions of photoshop. And the plugin doesn't work in any of them.
I had one that would merge in real time just hit the gen button, then it ceased to work a few weeks later I can't find it
I got rid of the asian woman bias by calculating the difference between the BRA model and Realistic Vision model, and subtracting -0.9 from my model
SuperMerger can merge and generate without saving. I usually try all the different presets and pick the one that gave the best result
I got rid of the heavy heavy bias of naked women in one and added more men
No, I mean on the fly
so much easier than the one you are talking about but now it is gone
yeah it is a bit fiddly to get running
I guess he took his github down as my ticket is not even on there
the faces are a bit odd 😅, but it's so nice to not have every 2nd picture a portrait of a woman I didn't ask for
it's particularly good at traditional painting, very sharp details
SD has always had an issue with medium to long shots for faces
99 out of 100 times img2img mask the face and regen and it is perfect so beats me why txt2img doesn't work
I think the issue is 64x64 that SD uses.
It is weird it can do so much stuff in detail, but then really struggles with other things you would think be straight forward
naw, we had this issue in 2018 with deepfakes and it was helped when we went to 256x256.
That is why I know this problem is related to that 64x64 noise base or not
even just 1 notch up to 128x128 would help faces in the medium shot but long shots would still be funky
This did not like Indian women as it made some really weird eyes I couldn't correct.
Interesting.
I like this one more
this looks amazingg. wow. good job
Happy Saturday!
I'm going for a "sleep music" vibe and these images are great but they're all way too bright. Who can sleep in lighting like this? 😛
I've tried all kinds of "dim lighting" and "dark room" and "very super dark" type keywords but no luck.
Any ideas?
What would you do, bring in the image and just prompt "dark bedroom" or something?
Pretty much, and in Controlnet you use a depthmap for lighting. I haven't had much success with that method though.
Alright! Time for another rabbit hole! 🤩
Thanks!
lol, why is it so hard! 😅
because we do not YET have true darkness in SD
it is coming though and all models have to be retrained. All we have now is a bandage approach
darkest one so far
see, not dark
anyway, it is a known issue with diffusion method but a new technique will take care of it at training time
i mean it's dark compared to the fluro white images i'm getting. something about head phones or eyes closed or sleeping is making the images super bright
Naw, that's pretty dark, I'd go for that 😆
that is dark? LOL
Well it's not full night but it's definitely not full noon either!
😅
Compared to the general brightness of my samples ^
your prompting must be awry
give a pic and the prompt used people in here could help
I always get exactly what I want on the first try!
Silhouette against
Just a rando, not my "super top favorite"
what was the prompt for it?
in the PNG info, but....
k
POS:
dark bedroom, long shot of a woman lying on her back, short hair, mini headphones, eyes closed, (neutral colors:1.3), cross process, thick pillow background, intricate details, wearing t-shirt
NEG:
3d, 3d render, 3dcg, amputation, animated, bad anatomy, bad art, blurry, body out of frame, border, brothers, cartoon, clones, copyright, crossed eyes, cross-eye, deformed, desaturated, disfigured, doll, double, duplicate, extra arms, extra legs, extra limbs, frame, hat, illustration, logo, long neck, lowres, make up, mutated, mutated hands and fingers, mutation, out of frame, over saturated, photoshop, poorly drawn, poorly drawn face, poorly drawn feet, poorly drawn hands, repeated person, same face, signature, sisters, split image, text, tiling, twins, ugly, video game, watermark, wrong anatomy
what the
I'm MUCH happier with this new model, than I was with my earlier attempts...
Yeah, that's ... almost! getting close...
now I lost the headphones
Nice though
Plausible lighting
I think I saw her on a recent RuPaul episode...
hahaha
Good lighting though!
Yeah if I can just get an image that's sufficiently dark I don't think the pose matters as much
😆 close the curtains before you...
Plausible night-time makeup, yes...
My wife always comments on that
"wow she really did her hair nice before bed"
Can I bug you for the prompts? Not seeing any info in the PNG 🤓
I did that earlier ... can try again
hehe
no idea what happened
hehe, oops
Oh, your dimensions is different than mine so that matters
That's a nannycam shot, clearly
What I'm getting with "moon" ... ha, not quite!
what is with the moon in each?
Painted on the wall 😄
Nice!
I tried thinking of other "night" keywords, like moon ... stars, galaxy ... I guess I could always go surreal but I'd prefer a bedroom background
That makes me think of Trek:TNG for some reason
Maybe 'cause she looks like Ensign Ro? 😛
she's wearing headphones, you just can't see them 😄
hehehe EXCELLENT, PERFECT DARKNESS 😆
except for the globe of sun at the bottom. Ruins AI astrophotography pics.
The kind you go to montana or Alaska to get irl.
happy now????
nope
We have achieved ... DARKNESS
Funky!
I can't wait for the new technique to be available for more trainers than just Joe Penna's
um... better get that looked at
headphones have been passed down for generations
hehehe marvelous!
she's really feeling this song
She either is passed out or dead. ngl, could go either way.
busted
So you just typed "moon" to get these clearly-night-time shots?
Starting my prompt with "total darkness, complete darkness" was not taken seriously 😆
aahhh
Think she prefers Bach or Beethoven?
Anyway! Thanks for the good times! Daddy duties calling...
kk, take care
cya!
that's how she wanted to be buried
she's trying to summon the darkness, but it just got worse
Yep, dead
Sort of reminds me of that ship in the Chronicles of Riddick
looks like trump
These are amaze
Awesome, but not him.
Img 2 img how
@oak osprey just found a cool 2.x model for scifi esque gens, and while it does look dope, it still has that 2.1 gritty/dirtiness that makes 2.1 not the move IMO
I love gritty images as the sterile ones just turn me off
no, I mean like, artifacty
Oh, ugh
like blobby and distorted
like her eyes and stuff, all messy and the whole image just looks cooked
the eyes are always the worst offenders
Yeah, I disagree about the yes but that damn overcooked is my bane
even cfg 1 overcooked
I think the overcooking comes at the training level from my own training attempts.
looking at some gens from the original creator, they don't look as bad, but still have the signature 2.x artifacts and deformations
I didn't realize until semi recently that 2.1 struggles harddd with single focal planes in images with bokeh
deformations is because damn SAI decided to do 75% of 2.1 at 512x512 and the last 25% 768x768
reason we were getting elongated necks
choose one and train all of it at that and so many issues would never have happened
I did not know this until yesterday but AMD is over taking Nvidia in gaming.
Nvidia lost 38% gamers and AMD lost 6%
the biggest drop for Nvidia was for productivity stuff so I suspect CAD/CAM
AI/ML was a huge increase
Thanks to their greed it is now spilling over into pc parts not selling, and even Lenovo took a 75% hit or something in their laptop sales. Basically, a lot of people buy a new GPU and CPU and RAM and MOBO at the same time. No buy GPU no buy the rest so a lot of deals happening right now on mobo, cpu ddr5 ram.
be ironically funny if their greed imploded the entire PC market. :/
I heard vendors were forcing pc part shops to buy other components for every GPU they bought
That I do not know about
Nvidia still king tho
Not for long at the rate they are bleeding in all but AI/ML
This was when ethereum mining was popular and causing the GPU prices to spike
ahhh, yeah
for someone like me who does gaming as well as heavy creativity stuff from 2D to especially 3D i dont see a Nvidia alternative
gamedev in the future
Well, I am not joking their sales figures just came out and gaming down 38% and 52% in business productivity. They were down in every single metric except for AI/ML
AMD is a fantastic alernative for gaming, outside of DLSS, really
better performance for the same price
saw the report
Most people don't use RT or DLSS
but sales go down doesnt also mean they are about to die or something. Probably people just didnt upgrade
I use DLSS in everything I can, but RT is usually a limitation somewhere else
DLSS is nice
most people on steam still play in 1080p so DLSS is ugly at 1080p
they released a genuinely terrible generation of GPU's, so I would say its a safe bet they need to fix it before a whole generation of revenue is down the toilet
I agree with Sytan
im still looking at that RTX 4090 and hoping for its price to drop down until next year or eventually 2 😄
looks fine to me, its higher fidelity than native 1080p, so I would say thats a benefit cause of the AI anti aliasing
the RTX 4090 is the only good card, and its a shame
What Jensen is doing is throwing it all in one basket of AI/ML and he is not dumb he has to know datacenters and universities will buy those cards (H100, A100) about 100 each time and use them for a decade or more. Hell, a lot of universities are still using Maxwell cards
1.700-2.300€ price tag tho
arguably, its price is quite reasonable for what it can do compared to previous gens, IMO
id need to spend around 7.000€ for the whole build + keyboard, mouse, monitor and some other extra stuff
although that is region dependent, as most non American or Canadian countries get screwed over by price hikes and government controlled import fees
What happens when he finally sales all those H100/A100 and no one needs more for a decade? He burned down the bridges with gamers and productivity houses.
now that sounds like a huge and unnecessary reach, but to each their own
IDK what country you are in, so maybe thats not as insanely unreasonable as it sounds
It is a bubble but not like the crypto bubble but a bubble nonetheless
Austria
Ah, makes sense then
Yep
also the setup is a beast and monitor alone is something below 1.000€
Here in America you can get great parts for much lower than majority other places
it is a perk to being in this crumbling country lol
yeah, i often have to pay extra VAT for stuff
yeah, the EU screws over most inhabitants with outrageous import and processing fees
Right now, this second I can build a top of the line gaming machine (best Ryzen, DDR5 CL30, etc...) for 1200USD thanks to all these combo deals and that was 2k-2.3k. Sad thing is the best card to put in the best system requires more than the entire PC, even including a case and still need 500 more. wth?
especially in a smaller and economically insignificant country like Austria
it sums up for me since my creative workflow pipeline is pretty long
and expensive
with VAT its more 😄
like often 50€ extra on top of price
My ex boyfriend turned best friend lives in Hungary, and I know how hellish it is in that economy
Yeah, EU is "How much did you make for the year? Remit now."
Austria is not bad, we dont live worse than americans
I mean, that is all subjective in many regards
I don't know if you guys use your own separate currency or not
the housing policy in Vienna is probably the best in the world
Euro
Hungary is so in the gutter cause their gov is worthless, and they still try to have their own region locked currency worth almost nothing by now
You know, I am not going to lie but there are so damn many countries I would love to live in then I see their govt and rath die from gun shots from local thugs in the USA.
Wonderful regular people but their govts makes me so sad
Unfortunately, a lot of smaller European countries are down in the dumps from old local governments and an older population that is too scared to fall behind times and become irrelevant
Cheapest 4090 you can here is about $2,700 AUD inc tax, which is about $1,760 USD
politically i see US as one of the worst
US is sometimes called "Afghanistan of America"
Yeah, as much as I like to rag on about the government here for being a clown show, there are many other countries with way worse governments
bowing out on this convo as I don't need another vacation.
The one that shocked me was how utterly terrible the gov in Canada is right now
He was meaning being muted lol
lol
Forced vacation for 24h
I got about 5 weeks vacation available 😄 Need to take some leave
all i did since i woke up was playing around with Photoshop Firefly and chatting in Discord lol
didnt even eat and drink my tea damn it
I will say this every man, woman, and cat beast on this planet should have the right to speak out against their govts as long as it is done peacefully. If you can't there is something wrong.
General bows out for gov talk, I bow out for Firefly talk lmao
We are both very passionate about those topics in our own way lol
for sure, I think thats reasonable, and a non problematic take
admit it Sytan, you like our salty discussions 
Yep, govt talk or religion talk get you a ban hammer
I just have too many connections in AI at this point to think of firefly as anything but a laughing stock
Well, it is in beta still coming to PS

its integrated in PS beta app now
I just find it funny that I have seen single devs in a couple days do better features than firefly, and its insane to me lol
there are people who use SD or MJ and then continue/finish in Photoshop with generative fill which is Firefly now
and they aren't just copying code from the community?
There is one thing I do like about Firefly...it gets the masses/artists/etc... more used to AI and less "destroy it all"
The evil is Adobe. 'nuff said.
I'd still rather use SD's addon for photoshop, but I don't have photoshop cause I have no goals to support such a shitty company again
in what regards?
they are hunting their niche of users





