#🏞|general-with-images
1 messages · Page 63 of 1
oh god, that makes me feel sick
That's disgusting :/

That one actually looks pretty decent
first person to get sued by StabilityAI instead of the other way around
fucking legendary
i'm happy about that. you say it like it's a bad thing.
messing around with it a bit myself lol
I would count it a plus lol
pretend I did not just type "Plus" as "pless" OMFG
I can't lmfao
a little bit of Gwyneth Paltrow mixed up in the Hildawg photo folder there, Sytan?
lmao
i thought you said Prius, so
we're all a little brain damaged
i'm like, "that's a car
"
oh, do any of you have a model that can do Beavis and Butthead?
some more basic low res tests
i'm talking about 2.1 where you can do both in one shot

see my danny devito and robin williams

something is off
bam
i win
Provided to YouTube by ML Music
صعب تزاملني · Muslim · Mohamed Muslim · Mohamed Muslim
صعب تزاملني
℗ ML Music
Released on: 2023-05-02
Auto-generated by YouTube.
you know, I have some EXRETEMLY cursed gens lol
i wonder if i could fix it to be the original art style
OH MY GOD NOOOOOOOOOO
@smoky oak 🚓 ?
LMFAOOOO
I did surreal spongebob lol
I made a LoRA of Raihan from Pokemon for a client a few days ago
I am sure I could do spongebob too
Tupac!
Raihan in show
My Raihan LoRA
I have an idea for a LoRA style lol
i don't even know why it tried
it should just have black squared me
i would have understood
OMG, I have to make this style lmao
if I can pull this off, it would be insane lmao
I might try it sometime soon
I made a LoRA for toothless a while ago
still happy with the results here
SHBK
i am gonna nuke you
hahahaha that face
unfreeze the model?
freezing the model at first, keeps the original model layers in-tact and adds new ones and trains those
and then you unfreeze it and fine-tune the base model layers
im talking about avoiding merging
it improves the final result. the reason you freeze it is to stop from modifying the base weights and to speed up training, as you don't have to keep doing gradient accumulations/passes over those layers
because you want to fine tune the base model layers once you have two or more fresh layers on top, with your newly applied weights
it just helps it all fit together better
yes
train some more, but not as much.
does it mark epochs in DB?
you just need one to bring the model weights up
aye
so this is because when you're putting two new layers (or more) on top you're just optimizing a subset of the feature space
doing one epoch of unfrozen training will improve the accuracy
if you have enough time, you should train the whole thing unfrozen
it's just way more expensive
because it continually optimizes the whole base model
but this needs a lot of compute
it's hard to say
you could make a copy of one, and try
how many layers did you add
hmm it might have some rather untrained epochs in its top layers
you might have to unfreeze just those
so the way training is done you gradually freeze the model so that the base layers have very fundamental feature extraction and the top layers are specialised
if you go over duplicate images is the only issue
do you just loop the whole folder again and again?
reusing the same images
It is critical to only do this step after the model with frozen layers has been trained to convergence. If you mix randomly-initialized trainable layers with trainable layers that hold pre-trained features, the randomly-initialized layers will cause very large gradient updates during training, which will destroy your pre-trained features.
very low learning rate for training some of the base layers
well then dreambooth sucks
in Keras* you just do:
assert inner_model.trainable == False # All layers in `model` are now frozen
assert inner_model.layers[0].trainable == False # `trainable` is propagated recursively
oh well
it shall remain forever just out of reach
"Freezing a layer of a neural network during training is equivalent to having a random nonlinear projection in your algorithm" - poetry
here's how the Huggingface Tutorial for DistilBERT writes:
Because DistilBERT’s pre-trained weights will serve as the basis for our model, we wish to conserve and prevent them from updating during the initial stages of training when our model is beginning to learn reasonable weights for our added classification layers. To temporarily freeze DistilBERT’s pre-trained weights, set layer.trainable = False for each of DistilBERT’s layers, and we can later unfreeze them by setting layer.trainable = True once model performance converges.
so when the first half of the model is frozen, it impacts your new weights a lot. and then you unfreeze it and fine tune each layer one epoch at a time if you wish to backpropagate some of those weights through the earlier layers of the model, to assist in more coherent classification and embedding
but it seems like dreambooth is set up to just way overfit a new layer
it's not theorizing. i can do it. right now, in python.
using the diffusers library from huggingface
they make everything pretty damn easy tbh
that's what DB uses internally, but just doesn't expose all the knobs, to keep it as simple as possible for the express purpose of overfitting a model on purpose
yeah so go use BlueWillow 

no one
i look at the fuckin source code to solve disagreements
the source code is the documentation
hurry up, ya old geezer
crack what
looks like it's just alternating attention between steps
define a new one what
these are most likely being passed in as callbacks. how are you browsing this source code?
well, for one thing, string interpolation is a unique fuckin hellscape
it IS simple
you're using notepad++
use VSCode and Pylance
naw. right click those and click "Go to references"
it'll bring you to where those are being registered as callbacks
well my point is, it's hard to do when you're not used to the tools
but vscode is like, night and day better than Notepad++
@smoky oak
portrait art of amorphic solids in the ferrofluid-infused lava pour, suspended animation, superposition structures within the geometric complexity of intricate positive and negative space, highly detailed, epic, cinematic, dark, twisted, impossible! mystical fog, rotational inertia, quantum feedback superposition
Unfortunately, it seems like it's too low to be of much use
Those would look dope with noise offset
was trying to get a photoreal sphinx morphing into tiger
"Who dares.. disturb.. my slumber..."
damn, got those words wrong
that 1994 CGI is pretty fucking lit
Young Marty McBrown, LMAO
looks just like the old one
kinda looks like charlie bucket
age progression lmao
using that prompt plus some embeds
yeah its wild, i can get this exact marty mcbrown out of Blue Willow
but i can't get it out of 1.5 or any fine-tunings i've tried
so to me, they're using 2.1 with text embeddings and CTU / UU. they might even be running everything through the Automatic1111 SDK
selecting embeds or loras based on the prompt maybe
yeah
they seem to have a BERT model that does that
it's actually not hard at all to train a BERT model on some text data
have some labeled prompts for each LoRA and teach it what the most probable result is for a given prompt
can return 5 possible results with the probability for each
that would need a lot of prompts but they're running on like, 100,000 servers now?
they have the data
craploads of tflops
they seem to have some deep pockets
everyone can do 5 images concurrently and it seems like they have the ability to fulfil those requests for like 20 people at a time
in a single channel, on a single guild. and i think they're running a few hundred inference servers, if not a thousand of them
A100's can split into 7 Multi-instance GPUs, running 7 inference concurrently, you can get 8 of those on a box, so, 56 concurrent jobs for one server in the best case
spitting out an image in less than a second
it takes about 20-30 seconds and it does a batch of 4 so it seems like they're running on fairly decent stuff or a highly optimised model layer set and never changing the base model out, or, keeping them all loaded at once
the upscaler is almost immediate though
for sure they would have the models in vram
well yeah but every thing you do like that, limits the overall capacity of the server for the other models. you can only run 7 concurrent inferences and each of those need to hold a separate copy of the same model
you can technically run as many inferences as you can store models in vram for, and there's ways to break a models' processing down using torch.multiprocessing but i've not gotten into that
like, i imagine you can interpolate the processing of two requests one step for each at a time
jumping between the two states
if they're doing all that, it's no wonder it breaks all the damn time.
they would have a number all linked sharing the same vram pool. I imagine
yeah they had some major capacity issues at some point last month and redistributed their resources for each model
they probably do, or at least do now, one system per model type
the thing is, sharing the vram of a single model is really hard to do
i thought it should be simple, but it just deadlocks when you simply run two pipes in two threads
the pipeline is a mutable object state and it doesn't do "Copy on Write", which would allow you to have essentially the same VRAM use at the beginning, and then only the updated changes for the pipeline, would increase the VRAM use.
Howdy, I am back
not sure. Im guessing how they do it lol
me too but i know for certain how this can't work, because i tried it 😄
maybe not with the tools available to you
howdy, giggle farts
oh they're there but i just don't know how to use them, it doesn't quite apply to my needs so i've not dove down that
ah right
but i did cluster design for like 6 years at my old job and it's a pretty fucking hard puzzle to solve. it's quite fun tbh
$200k USD
yup not cheap but in comparison to cpu servers its cheaper
whut lol
as they can do more in less time with less power
and less space
that's like quantum supremacy claims for boson sampling. it completes that task in 30 seconds. a traditional CPU? 9000 years. it's just not an option to do CPUs
stevie nicks?
cyndi lauper
You really can't increase the weighting on some celebrity prompts without their face coming out like crumpled tin foil
which version of karras?
I prefer 2M. I like SDE but its just too slow
I know the name I use for it is different than what a1111 does.
and you just need 32 steps
it's pretty much instant here.
alright I'll try the same prompt and see how it goes
i think it is not possible to set.
Putin Sword Practice
remided me Naked gun 2 and 1/2 🙂 clean energy
Uwu
here we have the zebra-tiger in its natural habitat 😆
Zebra mated with tiger
punk Zebra 🙂
I
...What?
swimming styles 🙂
Zigerbra
Howdy Y'all, anybody know where Styles in SD are stored? I saved one with the wrong name
god, I could never deal with transfers that slow anymore 😅
but thats bound to happen, since you do nothing but dreambooths
main folder styles.csv
i had to deal with 20mb transfer yesterday for 40gb worth of stuff
Aha! Thanks
I thought it was a folder with text files lol
when i was moving my extentions from the sd on my hard drive to the sd on my ssd
I am not sure I will ever own a HDD again
I'll probably just buy a high capacity sata SSD
4k uspcaling on ultimate upscaler is beautiful
yeah, with the caveat of being absurdly slow for anything thats not purely sequential (and still slow then)
i cant even send the results i got for the one i did last night
which would be better to hold an OS a hdd, ssd or an m2?
m.2 is not really any faster for boot, but it can help with getting things done WHILE booting
well itd probably be faster than how im booting right now...
10 minutes to get to windows
and 30 minutes for everything to actually load when IM IN WINDOWS
i got it 500GB and 600TBW
I have 2x 1TB M.2's, a 2TB m.2, and a 540GB SATA SSD
Most people aren't so wasteful with their storage lol
most games are bad now
most game devs seem to rush the release and the games end up buggy
upsettingly true
yeah, too bad nobody cares cause they screwed over their launch
I have 1100 hours in CP77. love it
I enjoyed CP2077 even at launch, but my annoyance with the game comes from knowing how much better it would have been if they waited
they waited actualy quite long i think.
they did, just not long enough
Maybe I will get back into it now that I have my new GPU
tho, my CPU still won't be able to run it, so probably not
marketing team hyped it too much before launch so the delays got a bit ridiculous after that so they had to release
even so I enjoyed it heaps
waiting for the expansion now. thats gonna rock
Ok, that's a HARD disagree lol
I may have enjoyed it at launch, but it was a dumpster fire. Just happened to be a fun one
The number of bugs and glitches in that game was legitimately impressive lol
luckily I didnt have any game breaking bugs, just hilarious things that popped up here and there
But yes, I did enjoy my time in it. Maybe I'll play it again sometime, assuming I don't mind my CPU being a problem
I've been enjoying Against the Storm
I think roguelike might be my fav genre
Against the Storm is like a Settlers roguelike
i knew slimes are hiding something 🙂 the other skeleton is because outpainting
?
Heyy guys, I am using random people's images(100)+1500 reg images in dreambooth with SD1.5 but I am not getting any new results as compared to SD. Any ideas on what I can do?
what
Yeah lol... Any idea why that might be happening?
what
Hello?
hello
do you have an adblocker or privacy extension installed?
yes, ad blocker
is it enabled for localhost?
Can you right click on the image to save it and see what the full thing looks like
I'm trying to think of the logo, but all I can think of is the HTML5 logo
also, the investors were making moves to try and force MTX to be cramed into the game. Imagine cosmetic loot boxes. CP2077 has none at all. Thye new if they released it without any support for MTX at all, it couldn't easily be shoe horned in later without really upsetting fans. They burned that bridge that the investors wanted to cross. Shortly after, they sued CDPR. I think they lost too, since CP2077 was actually a huge financial success
oh no i looked into it. They settled and paid out 50 cents for each share. couple million
small price to pay to avoid MTX in your product really. glad they released early and fucked that investor group over
They even cancelled the online version of the game entirely iirc. Probably to avoid any form of MTX being forced into their product
fuckin typo
i love SD2.1's weird alt-reality versions of people more than the real thing
high coherence, low accuracy
walter shite, lmao
walter my beloved
Alter White
fast food restaurant chef
that's not how monocles work
jeez probably the most accurate one so far
peter i'm curious if you can get ironman on stage singing with The Beach Boys in concert
no reason to do so other than "can it happen?"
steampunk iron man
reminds me Wild Wild West
I was just thinking that haha
It had some cool steampunk stuff in kt
@wispy nest then we cant be wrong 🙂
Went from 2 it/s down to 1.85 it/s? Huh, that seems backwards and I would have thought a bigger difference than that.
What model/vae
Realistic vision
less vram for the same context window. you can use larger context window with gradient checkpointing

its not something they developed to make it possible to train on consumer GPUs, but rather, to make better use out of their massive 80G GPUs
Time to cook, Jessie!
Yo
when the starting image is kinda bad, the result will be bad lmfao
what the hell is happening to his head

That's at twice the batch size, so about 85% faster
So I went from getting 2 images done 2 times a second, to getting 4 images done just under 2 times a second @dense tapir
I see we are in a breaking bad / Saul Goodman mood lol
Dragon ball
Goku 4k, hiper-realistic,
anyone know how to use open pose?
no
BS2 it said for both though. You are saying BS2 2it/s vs BS4 1.8x/s?
LOL
Yeah, it's 2 at BS 2 to 1.85 at BS 4 lol
That is what had me confused
I could have sworn I typed that properly
that is pretty sweet, shit
I still can't find a PC case for me. 😦 I spent all day yesterday and most of today. I lose so damn much with the modern cases.
I need a case with a reset button, can handle 400mm cards, 180mm deep cooler, and a drive activity light would be nice.
My case has all of that, however it does not have a drive activity light
Come on Jesse
Yeah, 400 mm is not that uncommon lol
walter ehrmantraut
My new 3080 is 338mm
I already gave up my external drive bays
Basically, the new test is, as Jayz 2 cents said, will it handle a Strix.
Next gen are going to be even longer
I have a Lian Li Lan cool 215, if you wanna look at it.
It has two massive 200 mm fans on the front for excellent airflow and very low noise.
Tempered glass side, hard reset button, fan filters on all sides, lots of storage on the inside
Yeah, I am heavily leaning towards the 216
even have the extra fan filter in the cart for 13 bucks
I believe that's how many new steps it learns before it ads that new information to the total information it's learned.
So like if it looks across five images, it should only update the information every five images
Insert basic engineering anime here
I saw a nice case that the power led was a ring around the power button and when red it is drive activity. Seriously, they should all have that.
Cool
This is why I mentioned my perplex of the 250/240mm GPU cases.
Matched everything I needed to a t except max gpu is 240ish mm
My Lian Li Lan cool 215 held my new 3080 with no problem could probably hold a 400mm GPU
sounds chinese
What is that supposed to mean
He cookin
Lian Li is an old name in cases. Used to be for the upper crust
🔊 👵🏽
Lian Li still makes fantastic luxury cases, but they have also brought great quality down to the lower price ranges
Get a hernia picking up a pre 2010 Lian Li they were so big.
yea ok Lian Li salesman
Loved working in them though as a tech
I mainly love my cause cause of the two massive fans in the front
cause cause
They make cases with 200mm x 2 front fans
I don't know of any other cause on the market with the airflow this one has
That's what mine has
Thermaltake X9 Core.
216 is 160
👀 <6gb for 24 frames, 30s (depending on gpu). also try sleeping with 2 a6000 going at 80% fan full blast in your bedroom. like a baby 😴
wtf
I don't even use any fans on the back or top because the front air flow is so strong it forces it out just fine lol
Odd size at 160 😦
I didn't know they made a 216
Yep
god i hate stable diffusion animations these days
superceeds the 215
get good
how do i do that tho
Inter sting, but it has smaller fans?
Weird
The whole selling point of the 215 is its massive 200 mm fans
LIAN LI LANCOOL 216 X Black Steel / Tempered Glass ATX Mid Tower Computer Case ,2x 16 cm PWM Fans Included (Non RGB ) ----LANCOOL 216X
X for black
RGB is 10 dollars more but still 160mm
when they fail, and they will fail, I would have to go to 140s
@oak ospreyhow to use animation
bro uses vga
Will not fit my 170mm
I know, that's MY CASE lmao
I wonder why they went down to 160?
I'm not sure
Is why I discounted the 215 as that CPU clearance is too low for me
216
tl;dr?
Close to what I said lol
video graphics accelerator
@oak ospreyhow to make a animation
I feel a bit cheated by Lian with 216 vs 215
I genuinely have no idea why they did that, I know that at 200 mm, with the way that they designed the blades on my 215, the fundamental fan frequency is below 60 Hertz, which is super non invasive for a fan tone
That is far from true lol
Actually, no. I am not kidding most mid towers will not hold a strix, or anything more than 4080/90 FE
Micro towers sure, but we're looking at mid towers, which are considered full desktops, the only thing bigger is a full tower, but those are extremely expensive purpose-built behemoths
My damn 2009 case can do that
I'm doing some measurements with my calipers right now
I used to only do Full tower but now they barely make any that I would use. A bit sad because they are wide, and tall but not deep which is what we need the most.
@dense tapir All right, so I can say with full confidence that my 215 can indeed hold a 400 mm graphics card, however, it can only do that if you're not using a radiator cooler on the front
Because with my calipers I was able to measure an additional 64 mm, which on top of the 338 mm graphics card I have in, is very close to 400 mm
And I'm sure I had some inaccuracy in the way I measured, I'm sure there's actually a little bit more space
Yeah, Lian only gives that small ass number
Notice the 216 is all the way forward in the picture? Yep, looks like I lost there too
I'm pretty sure they do it because the case is marketed as being able to hold a 360 or 420 mm radiator in the front, and you can't fit a 400 mm graphics card and that radiator at the same time
Anyway, 215 is not as wide so beefy Hatsinks need not apply
don't you have a 5600x?
thats like one of the easiest CPU's out there to cool lol
an AMD stock cooler can handle it just fine
I have my nice Noctua
considerably easier to cool than mine, and I use just a basic crappy hyper 212
You know 370mm would be without rad cause most rads are 30mm deep so fans on the outside plus radiator on the inside would give you 370mm 0mm clearance
make sense
That sucks 😦
new one is about 180.6mm
Now make sense of that since the fans are not nearly as wide, lol
I reactivated my account on New Egg from 17 years ago. Last two items I had purchased from them. Flashback.
Apartment complex fire I lost it all later that year (Oct 28, 2006).
Another issue I found with cases is that they only go up to 180 or 200mm for power supplies. Well, ugh
What I bet is happening is that the case manufacturers are retooling as the absurdity of these gpus being longer and requiring big power supplies they have to accommodate them. Some sites I see coming soon, or just released.
my landlord took the batteries out of the smoke alarms because they woke him up at night. that's when i found out the carbon monoxide detector has a 'snooze' button
somehow all my negative prompts had disappeared earlier :/
better now
In this video, we are going to talk about Generative Modeling with Variational Autoencoders (VAEs). The explanation is going to be simple to understand without a math (or even much tech) background. However, I also introduce more technical concepts for you nerds out there while comparing VAEs with Generative Adversarial Networks (GANs).
*Subscr...
Fire alarm is what woke me up to wake the late wife up. Thing is it was due to the Aimco maintenance man being stoned and he blow torched lit up the cob webs, and stuff, in the wall while doing plumbing work above us.
Fire alarm was going off and in the AC vent the smoke was bellowing in.
New Egg I did business with for my First Intel based PC in 1996.
Was a Super Micro 166
As I suspected
They were blindsided and didn't expect such long cards.
LOL
I had the 700 flux in my cart.
Is that the 700 flux?
I have an ancient 2004ish Antec case and man it was so ahead of its time.
what is the length of a 3090?
Yeah
Oh, a 350ish card is gonna sag for sure
It does as it has 4 GPU connections
There is a reason I don't go small as I like shit I can use for a long time. Case is about 15 years now and only due to the length of a gpu do I have to replace it.
Funny thing is the Antec from 2004 had removable drive cages
Shit, RGB keeps me up
Yeah, Antec is making a huge come back
Really gay bussy rgb
Not passing judgement but for me I don't need my room to be lit up like a Friday Night Amsterdam red light district.
@oak osprey room lit up like a Shrek rave
do yall know where to find some good 120mm fans?
how much do those cost?
buy for 6-8 a piece in a pack of 5
wut
They last for years too
Well, that is the 700 Flux
K
Maybe it isn't made now?
well, geez thanks :*(
so should i be on their website or just looking around in general?
Amazon
F andP12
shrek forest rave
P is for going through heatsinks and rads f is for back of cases
huh £12 for a pack of 5
ARCTIC, a leading manufacturer of low-noise PC coolers and components, initiated and continues to shape the trend towards quiet cooling systems. In addition to an extensive range of CPU/GPU coolers and fans, ARCTIC produces high-quality monitor arms and wall brackets as well as practical tech equ...
Well worth it
Rgb gayness overload
why couldnt there just be a pack of 6 💀
RGB jacks their prices to the moon
I think I saw that one but I am like wtf at the cases blocking the air flow.
?
Well, fuck the black version doesnt look like that
LMAO Shrek: Where the Wild Things Are
See obstructed
Well, it has my precious lights
HDD activity and on
btw, this is the 800 flux
has my reset button too
nope, no reset button. SOB
Do turtles eat bugs only
Wait
I am stuck in what I like and I like a separate button
whats the difference between the f and p fans?
You can restart Deez nutz
so p is for stuff like water cooled?
help me get better DJ Shrek at a Rave plz
Worst case I have to buy this
and f is for regular stuff?
rads AND heatsinks.
Jack black as Luffy
Bastard Amazon doesn't carry the 800.
max is 700
I haven't trusted them in eons, and I only just reactivated my account on them after 17 years
Hack Black maybe 😄
Well, the biggest reason I ceased to use them was they used USPS and UPS.
i dont even know if the f12s will fit on my computer
f12 is 120mm
i... i dont know
Sen, no need PST allows daisy chaining
all i know is that i have a b450 with 6 molex rgb fans
yeah...
1996 is calling
Monkey. D Luffy
the case came with those...
and its the reason why i want to replace the fans
dont blame me...
I know, I just ran into that on one.
blame my annoying mother
your b450 has 4 pin
she got me the damn wrong case
My artic fans are only 3 wire and I use it on my 4 pin MOBO b450 all the time. It controls them in bios via voltage
so it should be fine then yeah?
John Travolta
thats a very odd nick cage
teeny face
135ish MB/s of raw speed.
Nick Cage as Roronoa Zoro
haha... funny ._.
Adam Sandler
@bold cave
Adam Sandler as Vegeta
Yes, please
Jack Black as Goku
Oh, nice. So far 800 vs 700 only the front changed? Gotta be more than that for 10 bucks of a difference.
Looks like IT clown
El Ricitas? Is that you?
Yeah, seems last year
oooooh man this britney spears tho ❤️🔥
Yep, July of 2022
Suzanne Sommers, i prolly misspelt her
That's Vegeta
lmao at steve buscemi and his many shoulders
guys how do i get ctu to make photorealistic results
Mick jagger as Goku
Brad Pitt be lookin good
Anyone in here AIO, or like I am scared of AIO and do open loop instead?
Tom cruise as Goku
that's pretty good
I see 0 reason for you to get an AIO
your CPU is stupid easy to cool
Taylor Swift on Laundry Day
TAHOE
Are you blind? This is Britney Spears
lmao
Megan Fox as Bulma
Oh, it stays super cool just I despise how I can hear the fans ramping up when I go to sites. I actually prefer dead silence.
Watercooling has less spikes like that.
what does that shirt say
Dirty thoughts messy
Only reason. I think if I push the shit out of this benchmarking it gets 61-70c depending on my room temp but it does like to jump fast so I go to a site and the fans will rev up instantly
foxed her for you
that rev up and instantly down annoys the f out of me.
Taylor swift as android 18
at about 1k rpm it begins to annoy me but with the new case we shall see
My current case has a sloid top so heat can be felt up there
*solid
for a reset button I may just grab my drill and put one there
Arnold Schwarzenegger as Broly
@stone cipher
Good
What about GTA San Andreas?
hmmm. Carl?
Dr. Phil as Goku, first
i did img2img on the real san andreas and it's fuckin terrible LMAO
Claire Redfield
train to convergence 
zero loss, babeeeyyyyy
haha SD2.1 knows the nightstalker, richard ramirez
lol fml trying to make CJ and Big Smoke photorealistic was a fail
SEN
WAT
Convince your father that he's stuck in a time loop? Hmm, let me think. How about you start by planting little clues around the house that repeat every day, like a broken clock that always shows the same time or a vase that keeps getting knocked over in the same way. Then, begin dropping hints that maybe things aren't quite right, like suggesting you both have déjà vu or that you've had the same conversation multiple times. Eventually, start saying outright things like "I feel like I'm in Groundhog Day" until he starts to question his own reality.

thank you ChatGPT
Anyone know if it's better to use blender's mist pass or SD depth preprocessor to use for ControlNet?
If only there was something you could do to add in new tokens/parameters without touching the underlying model... Oh wait, thats literally what LoRA's are

My LoRA's work excellent, just have to know what you are doing
I am actually currently trying my first style LoRA
Can the 7950 come out yet
I have yet to see a single example/usecase where uniPC is better than DDIM
I mean, they look like they could have been made with any sampler, and uniPC is considerably slower than DDIM
too bad it looks NOTHING like her lmao
looks JUST like her lmao
yeah I mean, people who need glasses typically wear them
the sheer depths of your depravity never cease to depress
it's not realistic to imagine her without eyeglasses. SHE WOULD DIE
This feature is only available to Premium users for now. Premium 🌟 also gives you many additional benefits; view /premium info for more.
what lmao
probably
its not a bot lmao
This feature is only available to Premium users for now. Premium 🌟 also gives you many additional benefits; view /premium info for more.
raw photo, photorealistic, ultra high res, photorealistic, perfect lighting, lifelike textures: blonde woman standing in a sunlit field, her hair slightly tousled by the gentle breeze and her expression serene --v 5.1 --ar 2:3
what the hell is going on here lmao
pretty sure mj got a v5.1 recently so it prob is
Does anybody know how to get rid of this when prompting or other ways? Getting some texts like from games or posters idk lol
Is it in negative prompts?
don't give in to the journeeyyyy lmao
i did none of it correctly
stretching the image helps
who do you think it is, first impression. Thank you
Bernie Sanders
genius
see, I saw putin first
what an edgy thing to say
hm i must work again, should be st. George 😄
best of luck
@kind quartz Saint Bernard
lmao
such fuzzy boys
yes i like this bread, but more i like basset hounds 🙂
Saint Bernard dogs are imposible to feed. Financial bankrupt 🙂
nobody knows, it has 100's of possible settings
wait until you find out its a sampler that you have to tune for your specific generations, and that it has like 4 sliders you have to mess around with
I almost purchased a case and had all my info/cc etc... in then I just couldn't do it. Seems there are issues with it.
Yeah, I have yet to see anybody find a good way of using it unfortunately
from all of my tests and general consensus, its just slower and worse DDIM
I know for we without tensor cores I was told it isn't for us and why it crashes on me.
I can sometimes get it to work but it sucks
Really, DDIM, Euler both kinds, and DPM++ 2M Karras as the rest I just never use.
I never could get anyone to tell me why it exists except that it can.
As soon as I found out I can't really run it I removed it.
Same
Auto shoved it on us
You don't have to use it
If you were following along you would see
this is how i see sytan in my head
Needs more Shrekifying
Lmao, that is nothing what I look like haha
you can't stop me from seeing it still
Did you see how many open boxes there are on Nvidia and AMD? A lot.
People are returning them in droves
this one is kinda closer
further
further
closer
here, just have an actual picture of me lmao
There lol
thats what I look like
ohh you're the dude
you're annoying lol
🤪
yes lmfao
what, like you guys are gonna get anything from these images? lmao
I was hoping you'd look like this
they aren't anything I haven't shared before
i can't wait to hear this transcript being read out in a youtube true crime video
unrealistic body standards
lmfaooooo
believe me people have done a lot with less than this
I've already been in the public eye before, considering my friend is kinda a celeb
well, by proxy, you are famous
and I am looking to work in the same industry, so
yeah, basically
they aren't like famous famous, but they are pretty huge in their scene
so pass your headshots around and do your job for you? make an AI version that's overfitted, and distribute it widely? ensure it's merged into every model on civitai so that when you type 'male' it just makes you? and then you go to an audition and they're like "this guy feels safe and familiar"? HA. that'll never work. i've tried it
lmaooo
you are too much
that is the 1990s way of saying "extra"
But you look like Shrek and it worked
that was my big break
put your ear to the wind. do you hear that? the sound of my little legs breaking as i became typecast
seen me in ANYTHING else?
Can you do training on RNA2
the fuck is this that
Whatever the second from last gen AMD gpu
RDNA2
most likely unless it's a laptop GPU
I need some beer
that is counter-productive
RDNA 2 is an architecture, its destkop and mobile
I'm worse sober
you don't know if they have always been drunk or sober
you're an incredible friend and human being
for all you know, they are always drunk, and now you just said thats how you like them
SMH my head
I need atleast 3.5 beers to fix my lack of water since beer is mostly water
Basically drinking water

i asked for a bad thing from the prompt and it gave me this look
Mr Bean and toy story
LOL
alright, have to clean the house up for mother's day breakfast tomorrow. got my maid outfit on and ready to get cleaning
How could I generat the image here, what's the steps?
bro











