Img2img will lose the object completely, n likewise the similarity
#💬｜general-chat
low rain
fervent thunder
anyone using Controlnet in txt2img? i feel like its not really working for me which makes me wonder if its actually working in img2img or its just img2img doing its thing lol
plush forge
does anybody know who the creator of the google colab kohya training is?
warm junco
anyone using Controlnet in txt2img? i feel like its not really working for me wh...
Used it only in txt2img xD
Only scribble i cant get to work
fervent thunder
Scribble seems pretty cool
i guess it must be working for me I see some better results in faces in things. this tech still blows me away so Im not sure if its just good img2img or I actually have controlnet working lol
warm junco
Make sure to check the enable button xD
fervent thunder
ha yes
warm junco
Try openpose and you will see the Difference
fervent thunder
is openpose hand only for images with hands or is that its intent i assume?
warm junco
Yea if the char has good hands use it
fervent thunder
you can mix preprocessor with model righ? i saw someone on reddit post something with i think scribble and dosmix
wide juniper
Just did some testing. It seems that changing the file name of the LoRA file also changes the trigger word. Can anyone confirm that? I tested with the donlee lora on civitai
warm junco
you can mix preprocessor with model righ? i saw someone on reddit post something...
Okay but what will that change?
fervent thunder
no idea
warm junco
Just did some testing. It seems that changing the file name of the LoRA file als...
Idk but it works so for embeddings
fervent thunder
I just saw a post with lots of upvotes that had a mix
I'm also not sure if in img2img you need the image in both places
fervent thunder
Yes. But there is no scribble model I see. just preprocessor
warm junco
Im not at pc rn but i tought there is one
fervent thunder
maybe i missed it last night when i was dl'd so many new things for this and trying to figure it out lol
warm junco
Yea so much new stuff, if someone was away for 2 months he would need a book for all new stuff to learn
fervent thunder
it seems pretty impressive though. like better than a new model maybe lol
haha yes
i also have no idea if its best to check envelope (outer fit) , scale to fit, or just resize
warm junco
If the source image you load for the Pose is higher as your output you will need to check "just resize"
Oh 4% battery, cys later xD
fervent thunder
haha later
plush forge
does anybody know who the creator of the google colab kohya training
smoky holly
is there a way to do image variations with SD? (a la dalle-variations style)
found this -> https://www.reddit.com/r/StableDiffusion/comments/wuu6gf/stable_diffusion_image_variations_similar_to/
but can't make it run inside automatic1111
storm shard
does anybody know who the creator of the google colab kohya training
have you checked his github? https://github.com/bmaltais/kohya_ss
plush forge
have you checked his github? https://github.com/bmaltais/kohya_ss
Yeah but I don't know their ds
storm shard
Yeah but I don't know their ds
you could leave a message in discussions over there
lament steeple
the models for blip2 are huuuuuge
plush forge
you could leave a message in discussions over there
huh, haven't thought of that, thanks!
storm shard
huh, haven't thought of that, thanks!
np yw
sterile blade
Anyone knows about the potential relaunch of Facebook's Galactica (equivalent of chatgpt for medical research) by Stability AI?
fervent thunder
where do lora files go again?
theres so many folders and not just one named lora lol
brisk bloom
models\lora
fervent thunder
looks like i had them in additional networks
does it work in the models/lora too i take it. odd theres two places for them or maybe i just screwed something up on an install
unborn echo
is there any way to make variations of one character
like if i get a result and i like it
i want to have it in diffferent poses for examples
odd stag
severe sonnet
Hi has anyone ever installed this model on invoke ai:
https://huggingface.co/StarwingDigital/Oldjourney
I downloaded it, but no matter what I did, I could not activate it
If you have installed it, I would appreciate it if you could help me
odd stag
https://youtu.be/YephV6ptxeQ
This saves a huge size on control net
unkempt hatch
And things like "make this south park episode live action"
fervent thunder
anyone else playing with Controlnet?
lament steeple
Just finished training another TI
unkempt hatch
Oh my god, The Magic School Bus live action
earnest fjord
Is there an "official" stable diffusion rest API?
lament steeple
Is there an "official" stable diffusion rest API?
if you are using automatic1111, there is an api you can enable
earnest fjord
if you are using automatic1111, there is an api you can enable
I mean as in API as a service like the OpenAI models. i.e. create an account, get a token, and pay per request or something
rigid vapor
Hi All. Noob question here. Would anyone share their favorite website for SD web application using their smartphone? I miss the bots on this page.
worn marlin
guys, how to use LoRA
Can it only be on a laptop or computer?
Because I use mobile
brave matrix
quaint hamlet
hi everyone, is this a good channel to post tutorials? http://the.strange.agency/projects/diffusion-collusion/ 🤔 here's a way to create endless (or just long) scrolling collages
worn marlin
guys, how to use LoRA
only this, can it really not be on mobile?
winter pike
guys, how to use LoRA
lora isnt available in mobile and requires minimum 6gb of vram last time i checked
worn marlin
lora isnt available in mobile and requires minimum 6gb of vram last time i check...
I was disappointed
velvet plover
Hi! Does anyone have a Colab notebook with ControlNet integrated?
finite narwhal
hi everyone, is this a good channel to post tutorials? http://the.strange.agency...
Yeah please share
lost flicker
Hi! Does anyone have a Colab notebook with ControlNet integrated?
https://colab.research.google.com/github/TheLastBen/fast-stable-diffusion/blob/main/fast_stable_diffusion_AUTOMATIC1111.ipynb Get this running and use the controlnet extension on the extensions tab.
velvet plover
Thanks so much. 👍
stable dust
Quick question, how fast are folks generating images using the 768 model? and what GPU are people using? Just a rough guide so I can get an idea
subtle rain
Hello, it is possible to use the power of my main PC but installing stable diffusion on my laptop ?
stable dust
Hello, it is possible to use the power of my main PC but installing stable diffu...
Why not just install it on your PC?
You could install it on your PC but then use your laptop to type prompts into via the webUI?
subtle rain
Why not just install it on your PC?
Because I use my pc for other things
subtle rain
How
subtle rain
ty
hollow apex
Hi
lost notch
hey guys do you have any idea how to deleted multiple characters from the scene ? I want to have 1 person in the image but its constantly putting multiple
*delete
deft bronze
hey guys do you have any idea how to deleted multiple characters from the scene ...
inpaint them out
fervent thunder
is there a simple breakdown of how to get stable diffusion up and running? this all seems very complicated.
trail ledge
I assume you're asking about running it on your local PC?
fervent thunder
I assume you're asking about running it on your local PC?
yes
trail ledge
fervent thunder
thank you
trail ledge
keep in mind that any guide gets outdated quickly as the community is moving fast
you can always go to https://github.com/AUTOMATIC1111/stable-diffusion-webui and follow the README there
which should be reasonably up-to-date, but doesn't really handhold you
steep timber
is there anything that I can use to train models on a GTX 1650? I really want to train some images
trail ledge
how much vram?
steep timber
4gb 💀
autumn portal
DLs from civitai are super slow 😤
celest phoenix
Has anyone used SD to create ad campaigns?
I did a tiffany x Bose mock campaign
steep timber
DLs from civitai are super slow 😤
I use the wget command to download things
idk why but with wget it is faster
trail ledge
4gb 💀
I'm pretty sure you're out of luck. I don't think you can even to textual inversion on 4GB of VRAM
steep timber
at least I can generate images
it uses almost all of my vram
but it works
so I thought I could train something
trail ledge
you can try training a textual inversion embedding, it's the least memory intensive option, as far as I know
steep timber
I dont even know what is textual inversion
steep timber
thanks
trail ledge
I dont even know what is textual inversion
In case you want some background on the various methods: https://www.youtube.com/watch?v=dVjMiJsuR5o
pastel fern
is there anything that I can use to train models on a GTX 1650? I really want to...
you can try using google colab to train
pulsar carbon
Hi, i'm really struggling with my lora training and can't seem to find a lot of good details or any in depth tutorials about how to do it well. There are people out there making killer models but none of them seem to want to explain how they're doing it. Can anybody please help? Thanks
pastel fern
also there are some free lora trainers online
pastel fern
Hi, i'm really struggling with my lora training and can't seem to find a lot of ...
just experiment and experiment i think
pulsar carbon
just experiment and experiment i think
But i'd just be pissing in the wind, it's not an economical way to go about it, i can barely find good explanations of the settings i'm experimenting with. So i don't even really know what changes to make.
pastel fern
Have u tried this?
It seems free to use
Nvm it has limits after a while
But til then
trail ledge
a lot of times it's not the settings (though they can have an effect), but instead it's the captioning that you feed along with your image
sometimes words that seem innocuous can throw a training off the track
I can't link the civitai LoRA model because it's nsfw, but look up "shirtlift-a-lora-for-...". In the intro he gives a guide on his workflow for training something that's not a face.
autumn portal
This whole process feels like trying to crack a new video game lol downloading stuff not knowing if it’s going to work or not 😂
fervent thunder
can anyone make some AI art for me ? i just ordered some parts to run the software on my PC
trail ledge
showing my age here, but when I introduced SD to my friend, his response was "OMG this is like the early days of Linux when you don't know if s*** is going to work!" 😆
pulsar carbon
a lot of times it's not the settings (though they can have an effect), but inste...
Yes i think that is where i am slipping up probably, it really confuses me.
trail ledge
good luck, the one time I tried something other than faces it failed miserably 😆 . Haven't given it another try yet.
gun shy, as they say
wispy jasper
Hmm anyone figure out controlnet?
I keep trying to use my line art and getting back images that just look identical to my b/w line art
potent spire
Tencent has its own controlnet apparently
trail ledge
Hmm anyone figure out controlnet?
I keep trying to use my line art and getting b...
I haven't had much success with line art to images either, the handful of times I tried.
wispy jasper
Doh
pulsar carbon
gun shy, as they say
I managed to train a video game character fairly well, but was a bit of a fluke, i just want to finetune it a bit more accuracy/consistency, but maybe i'm overshooting idk
trail ledge
yeah, I had one model that if I said "with black hair" in the training captions, it just produced horrifically distorted renders with that model. But if I take "with black hair" out of all the input captions and retrain, then the model worked
so 🤷♂️
pulsar carbon
Did the person you were training it on (presuming that's what you were doing) have black hair? That would be why.
trail ledge
yes, I really just wanted the facial features and not lock in the black hair
pulsar carbon
oh
trail ledge
every guide I've read says "only list things you want to exclude"
pulsar carbon
yes exactly
trail ledge
so it seemed obvious to exclude the black hair if I just want the face
but yeah, that sent the training down some weird spiral
pulsar carbon
very strange indeed
i wish there were some authoritative source on this that was easy to understand
trail ledge
I mean, it could also be I had a low number of images and it just couldn't get a good handle on the face by being that restrictive, or something
but I'm no training expert
pulsar carbon
i'm sure stability ai or google has some mad 300 page long white paper about dreambooth and lora, but like hell am i getting anything out of that lol
fervent thunder
I did a tiffany x Bose mock campaign
Did you get good results? How did you do it? Thanks in advance 😄
lost notch
Soooo pix2pix is now built-in but do I need to download pix2pix model in order to get it working ?
pulsar carbon
i believe you do need the base IP2P model loaded, yes
lost notch
okay, thanks
sleek osprey
trail wing
Soooo pix2pix is now built-in but do I need to download pix2pix model in order t...
is built in ?
ripe crest
how do i add the bot
teal anvil
hey guys quick question, SD wont work locally on MAC with M2 chip?
trail ledge
worn marlin
are there any other datasets other than LoRa that are similar to the Korean doll datasets?
dusky blaze
where is the invite link for this server
winter pike
controlnet is awesome
winter pike
guys how to understande what version stable i have
if you can generate good images with 512x512 resolution then its probably 1.5
thorn elk
ControlNet, what a great update. But cant do batch, so video is out of the quest
winter pike
2.1 should be used at 768x768 minimum and anything below that resolution turn horrible and very compressed
thorn elk
I am having this problem when doing batch with openpose model, any idea?-> AttributeError: 'numpy.ndarray' object has no attribute 'save'
trail ledge
ControlNet, what a great update. But cant do batch, so video is out of the quest
It'll be a challenge for it to extract the skeleton consistently from video frames anyway
I did about 30 frames by hand and it had some challenges with resolving the person from the background
better (for now) to generate the animation you want via some other program, then import a control set of inputs
thorn elk
It'll be a challenge for it to extract the skeleton consistently from video fram...
Yea, i agree... the idea it can find the skeleton is amazing already. Maybe too new, i wait for a while until i can implement it in my videos
trail ledge
there was a tweet earlier from someone with a different workflow, let me find it...
that's probably the current state-of-the-art, at least of today
by tomorrow, who knows? 😆
astral goblet
ControlNet, what a great update. But cant do batch, so video is out of the quest
It can do batches but not with the usual tools
trail wing
It can do batches but not with the usual tools
what's the process, is it another extension ?
astral goblet
Anyone knows the minium vram requirements to run controlnet?
Extensions readme says 4gb Nvidia card
astral goblet
what's the process, is it another extension ?
Manual scripting it. Other command line tools. Maybe autoit or autohotkey.
trail wing
Manual scripting it. Other command line tools. Maybe autoit or autohotkey.
good ol'autohotkey to the rescue
astral goblet
Lol you know it
trail wing
autohotkey feels like a hammer in ways. when you pick it up you can turn most things into nails
chatgpt can do autohotkey scripts
lament steeple
52 Downloads in 8 hours. I don't think that's too bad for an unknown
astral goblet
Ohnooo
hidden pivot
i discovered controlNet today, it is super cool 🙂
trail ledge
there are channels for images, answering your question that you deleted 😄
hidden pivot
it didn't work, so i had my answer 😛
trail ledge
like #🏞｜general-with-images , or one of the Dreamer Communities threads
hidden pivot
ok thank you
trail ledge
np
zenith thunder
@outer crescent you dont say >.>
fervent thunder
hey
guys
i need a lil help
is there anyone?
who knows how to use stable diff?
outer crescent
<@304655299631906816> you dont say >.>
run
zenith thunder
zenith thunder
who knows how to use stable diff?
i just follow youtube videos
fervent thunder
yo guys is stable diff 2 out yet?
warm junco
yo guys is stable diff 2 out yet?
2.1 is the latest model of SD
neat badger
Anyone know a site that has sound clips of Actors/Actresses? Not YouTube videos lol. Direct line sounds etc. I rather not sit through YouTube videos if they already exist.
Talking about their movies roles.
stray fern
willow coyote
Anybody know of ways to transform simple audio chords into MIDI? Ie. https://www.youtube.com/watch?v=8_taCHoY24E
I tried https://basicpitch.spotify.com/ but it was so messed up I don't understand why Spotify even has it online.
shy prism
Has anyone gotten the pixilization extension to work
fervent thunder
Anybody know of ways to transform simple audio chords into MIDI? Ie. https://www...
If I understand your question right, depending on what DAW you're using there should be a function which converts audio files to midi. I know logic does it.
fleet dock
How much memory would I need for my GPU to generate high resolution images (higher the better)? Wondering before I make a purchase.
willow coyote
If I understand your question right, depending on what DAW you're using there sh...
Sadly I don't think DAWs can do that, however they can ofcourse convert MIDI to audio.
brisk bloom
How much memory would I need for my GPU to generate high resolution images (high...
10gb or higher. 10 can gen images easily, the more vram the larger you can scale to. Dont buy AMD or you will be in a rough position. they arent good for this stuff
fleet dock
10gb or higher. 10 can gen images easily, the more vram the larger you can scale...
Ok thanks a lot, Ill try to get my hands on a 12 or settle for a 10. Appreciate it
winged vine
signal dome
Whats controlnet ?
astral goblet
ive heard some ppl say yes
people are quick to spread misinformation about the new models. It's really easy to because everyone is ready to believe anything anyone says since things move so fast in this field.
most of the misconceptions surrounding 2 are around the disinformation campaign launcehd by a scammer part of this community.
the idea that 2.1 is somehow incapable of "anatomy" is just disinformation meant to create a need they could fundraise towards.
lament steeple
I kind of want to make a VR interface to automatic1111's api, but seems like a lot of work for a toy only I would play with 😛
astral goblet
yay typing with infinitely long sticks
VR interfaces kind of suck imo. and i've got the index controllers
astral goblet
i want my messy desk and physicality for creative work
full lark
wtf with --lowvram I can generate 1920x1080p on 3gb vram
It´s slow af but it´s working, and only 1.5gb of vram is being used
lament steeple
I prefer to have a larger batch size to a larger render. I can always upscale
full lark
x2 but before updating sd I could only generate 512x512, idk why it works with higher resolution now 😁
full lark
I´ll try with another model to see if it works 🤗
720x720 with 2.2 vram and a good model, not bad
lament steeple
Nice!
pulsar carbon
Hey folks, is there any way to do a reverse model search based on the hash? So if you have the hash ID you can look up the corresponding model? If not, someone should really set up a method or database for that, since so many are coming out now.
lusty mason
Is there consistency in animation now or not yet
pulsar carbon
well if not, controlnet ought to help with that a lot quite soon
lusty mason
First anime created with AI gonna blow up the industry
stark apex
astral goblet
First anime created with AI gonna blow up the industry
maybe not. they're already melting down. i respect that so much volunteering is being done , people deciding not to release something because it could displace so many artists so quickly
they're taking the looming job crisis so seriously before it ever happened. really being proactive
biggest example i've seen recently is controlnet aren't releaseing the anime colorizing model
lusty mason
maybe not. they're already melting down. i respect that so much volunteering is ...
It will eventually happen
fervent thunder
Sadly I don't think DAWs can do that, however they can ofcourse convert MIDI to ...
Sure they can. Here's FL studio doing it.
https://youtu.be/pd-pCh8UjCk
blazing ocean
I'm training a Lora with dreambooth, can I give prompts for the class images generated?
fervent thunder
shouldnt they wrap up this general chat and sd-chat into the same thing? like you could say general is for more ot chat but then theres an ot channel. anyway lol
rich lintel
how to control camera angle?
astral goblet
how to control camera angle?
controlnet
brisk bloom
shouldnt they wrap up this general chat and sd-chat into the same thing? like yo...
agree. too many chat channels
fervent thunder
I think when they redid things they just didnt realize they made two of the same channel with this and sd-chat lol
is there not a openpose_hand model?
or do you just use openpose model with it
fervent thunder
how to control camera angle?
I've seen someone using high angle POV , idk how good is works tho, haven't tried - that might be another option, but yea...
Controlnet gives you better control.
fervent thunder
A
fervent thunder
thin heath
Superb tutorail here. I am getting awesome images using these techniques.
https://www.youtube.com/watch?v=7hGd_FjjyBU&ab_channel=ValarHedonist
round harness
fervent thunder
I was wondering why no one posting anime, it seemed like anime audience is biggest, based on amount of popular models 🐧
dawn verge
yup, it's over here now #1072013871730131004
steep dragon
If i activate this setting in A1111, does (text:0.8) still work?
"Emphasis: use (text) to make model pay more attention to text and [text] to make it pay less attention"
dawn verge
If i activate this setting in A1111, does (text:0.8) still work?
"Emphasis: use ...
it should
steep dragon
then again, if i write like this anyway, no reason to keep the settig activated anyway, right?
dawn verge
then again, if i write like this anyway, no reason to keep the settig activated ...
not sure about you, but that setting should be on be default. No reason to switch it off
steep dragon
yeah it is, was just thinking if it might interfere somehow. well if it's not, then just keep it. thanks
dawn verge
yeah it is, was just thinking if it might interfere somehow. well if it's not, t...
it's not gonna interfere if you manually put weight like (something:1.3), but if you put (something), then it will be the same as 1.1
near silo
so just to clarify, high res fix works by:
You select the base res, it generates it, then it uses that image as a reference to lead the composition of the higher res image (which fixes the wonky generations above lower resolutions), right?
lament steeple
That's how I understand it
near silo
Thats how it visually works
I have somebody trying to tell me it works way different, where it lowers the resolution of the original image below what you say, then it applies a hacky upscale to it, which leads to bad results
Which is... not how it works
dawn verge
if im not wrong, it's basically the same as passing the image thru img2img
dawn verge
yup yup
near silo
he called me "dumb fucking idiot" for saying it worked that way
near silo
hes an asshole, he never learns
ok, so that IS how it works
thats how its explained in a1111 documentation as well
near silo
he acts like a little kid, like seriously lmao
sending an ss in #🏞｜general-with-images
marsh pewter
Halo guys, sorry to bother ya all but. can I ask a quick question?
lament steeple
Oh no!! someone's asking a question in an open chat 😛
ask away 🙂
We're all here to help each other
marsh pewter
xD is about my result's on Stable I'm trying some Anime Models and on preview they seem colorful until but when done they get very grany and saturated
marsh pewter
My Cfg Scale is at 7 but about General Setting I'm a newbie
lament steeple
cfg is good
steep dragon
man, it's hard to use anime tenors for anything WITHOUT a person. they always want to add a girl/woman/man
lament steeple
My Cfg Scale is at 7 but about General Setting I'm a newbie
drop one of your images in general-with images so I can get a look at it
lucid vigil
hello everybody !
lucid vigil
thank you @lament steeple
lament steeple
Welcome
lucid vigil
I did it xD
thank you to you too then !
hollow verge
😡
lament steeple
I didn't do it
fervent thunder
is stable diff the only ai program?
ive heard someone saying nai is better
is nai another program?
young ivy
anyone here ever try the art software "black ink" ?
NovelAI is a website frontend/backend using stable diffusion and their own personal models and other functions that they keep even more secret now than before when they had a leak
merry vector
How to upscale images to 4X,8X ?
young ivy
depends what UI youre using, in automatic the easiest method is to go into "extras" tab, select an upscaler to match and then how many times you want to upscale
merry vector
Suggest some free tools
young ivy
what do you mean free tools?
https://upscale.wiki/wiki/Model_Database these are for stable diffusion UIs
you place them in realesrgan or esrgan folder in /models/ folder
reload your UI and use extras and select whichever
this is if you use automatic
this is the best I can help you
merry vector
Is there any YouTube link how to do?
young ivy
I dont know
At this point you know as much as I do and you can probably find your answers googling or youtube searching
maybe somewhere here https://www.youtube.com/@Aitrepreneur/videos
youre welcome
warm junco
Is there any website?
here is a tool:
https://github.com/n00mkrad/cupscale
lilac pine
Hello, guys! I need help.
I can't send images here, so they are at the general images channel.
inland geyser
gm guys
sinful kernel
shell seal
How good is SD with partial edits nowadays? Can you do something like eyes of person A with mouth of B hand of C etc etc?
fervent thunder
guys do i really neeed to rename the 1.5 sd to "model?"
in order to work?
theres one tut on yt
who did it
anyone knows?
warm junco
no you dont
if it wont start, while searching for the model then you need to rename it
fervent thunder
so i just leave it v1.5 pruned emaonly?
warm junco
yea
fervent thunder
alr thx
maaan most of the tuts on yt are so loong for no reason
over 30 mins
but the work is like 5-10 mins
fervent thunder
yooooooooo this is so straight to the point!
quick and clear
this the best one maan
warm junco
yea and its not outdated
fervent thunder
maaan most of the tuts on yt are so loong for no reason
That's just how youtube works and people abuse it.
You get more money for a longer video...well, as long as people watch it.
In most cases just reading through instructions is alot faster then watching video.
Visuals are nice, but eh...not when people milking it all the way.
bleak matrix
Good morning, everyone! How are we all today?
fervent thunder
That's just how youtube works and people abuse it.
You get more money for a long...
exactlyy
keen kiln
Car
keen kiln
/how can ibgenerate images?
bleak matrix
To learn more, check out #1072220168534642768 #1072229020520947753
lament steeple
Good morning, everyone! How are we all today?
Tired, sore, and feeling like I've been run over by a textual inversion
lament steeple
But it's gotten 186 downloads in 24 hourss
bleak matrix
Nice!
lament steeple
Trained 3 versions of it. one for each major model
willow coyote
Nobody know how to turn audio into MIDI? I signed up for Chordify yesterday, it was complete bollocks.
steep dragon
🤔
it's strange, sometimes after doing a few thousand creations my models simply just produce crap. like faces being a total mess all of the sudden although being created mostly flawless for hours. any ideas?
shut yarrow
trail wing
🤔
it's strange, sometimes after doing a few thousand creations my models simpl...
no idea but after hours of changing models back and forth, sometimes it goes bad... like event horizon horror movie bad
some other times we're just cruising
woven cloak
Where can i use stable diffusion?
astral goblet
🤔
it's strange, sometimes after doing a few thousand creations my models simpl...
to confirm that the issue isn't just imagined, heres what to do. save the generation parameters for an image you believe is flawed because of fatigue. come back later and try them again? same generations? hardware was fine.
rigid vapor
🤔
it's strange, sometimes after doing a few thousand creations my models simpl...
I had the same exact problem yesterday. Been using same model for weeks, but yesterday faces were all morphed like AI was on LSD, lmao
astral goblet
if stable diffusion is being corrupted in memory or something, the same parameters will not create the same image
i keep seeing this problem around a lot. i've even thought i suffered it. but i've never seen anybody confirm it with 2 images that both have the same generation informaiton
trail wing
I can provide that example
unless there is something more in the file meta that isn't shown on the ui and that is the cause
split ruin
Hello to everyone. I have some questions about Dream Studio.
What is Clip guidance?
Does it turn on any sampler by default?
Is it better to set more steps with enabled Clip guidance?
On which Stable Diffusion, Clip guidance has the best results?
Do other parameters affect the Clip guidance?
steep dragon
@trail wing@astral goblet@rigid vapor
Good to see i'm not the only one then ;)
Yeah, providing evidence would be a great way. Shame i didn't think about it myself...
But indeed, i'm not even sure i would get different results for the same seed, but it's rather that bad creations tend to get more over time.
This morning for instance there was about 10% of bad faces. over time until this afternoon and about 2000-3000 generations later i have about 30% of bad generations.
Restarting A1111 and reloading everything made it drop again.
So not quite sure why this is but at least it seems easy to fix.
astral goblet
are you saying that over time, you have a tendency to find bad seeds more and more?
trail wing
I find that overtime, no matter the seed, sometimes corruption happens. This is over big stretches of time where everything was going ok, but error messages from loading 2.1 check points that do not combine with TI for other versions, and a lot of model switching, going to extras and doing some big resizes, etc seems to pile up
but hey, control-c and restart puts everything into place again
some generations use so much memory that the second monitor flickers for a frame with garbled parts of the generated image in it, just before it displays it in the web ui viewer
so that's probably why we see this over time, it's literally pushing and squeezing the vram like crazy
astral goblet
Thats the kind of corruption i thought i'd be getting. grey backgrounds. seemed like jpeg artifact generations. i'd reboot and the problem would go away. but of course, since i thoguht of testing the next time i seen it going on, it hasn't happened. Since then i've updated a lot of different things so shrug
trail wing
cosmic ray flips non error checked memory bit, nuclear explosion ensues
waiting to test it the next time seems to be a great contraceptive
astral goblet
i got cuda 11.8 and 12 sdk's installed. newest game ready drivers. python 3.10.9, a few new pulls on the webui. nothing that i can pinpoint changing something, but it has stopped between then an now
a cosmic ray flip on a fp16 register might do something actually. and it was the week the comet was blastin past
bleak dawn
How to start creating?
trail wing
a cosmic ray flip on a fp16 register might do something actually. and it was th...
The culprit.
astral goblet
ooooo and it makes me wonder
trail wing
please, no jimmy page solos
astral goblet
page? ahem. stairway
wait page is stairway. damn this coffee is weak this morning
or its cold out
or both.. fuck hot shower time. gotta get started. bye all
trail wing
another zeppelin fan repelled
fervent thunder
Imagine this
An artificial intelligence that takes a Melody/song abd transforma It intro any instrument. For example you can transform a piano song into orchestal and a trumpet song into guitar
willow coyote
Looking to transform simple audio piano chords into MIDI
Seems impossible right now
trail wing
Someone sees their scunet come and go from the web ui like it has got a mind of their own ?
fringe minnow
doe this program make deep fake videos as well?
wet summit
idk
fervent thunder
Looking to transform simple audio piano chords into MIDI
What program are you using trying to do this?
willow coyote
There's no program that can do it 😦
But once I have the MIDI I'll import it into Ableton
sacred snow
hi there
nova lily
Is using () technically different effect from using :1.2?
I'm also trying to learn of other commands if there are anymore
Thanks
ashen agate
afayk can Gradio do negative prompting?
fervent thunder
Is using () technically different effect from using :1.2?
You do :1.2 inside of (anyway:1.2)
trail ledge
Is using () technically different effect from using :1.2?
no, just different way of stating multipliers
fervent thunder
(x) , ((x)), (((x))) e.t.c works same way as (x:1.0), (x:1.11), (x:1.22) or something like that , don't remember exact numbers
trail ledge
^^
nova lily
Yea ive been stacking them, no idea if they truly did stack though
trail ledge
each paren is x1.1
nova lily
oh
trail ledge
so (()) would be 1.0x1.1x1.1
nova lily
oh 1.3x
trail ledge
well, 1.21, but who's multiplying? 😆
fervent thunder
I just did to check how far I was with my numbers lol
trail ledge
the (prompt:1.2) syntax just lets you have finer control
fervent thunder
and less mess in your prompt
trail ledge
the ((())) is just easier to type for some people
nova lily
but the :1.2 make me feel like an advanced user 🤓
trail ledge
lol
I like using the number because it's easier to tweak the prompt. Just need to change a number
I know, there is a hot key
but I still like editing
fervent thunder
I just don't like too many brackets, it's painful to read or even look at
trail ledge
yeah, just personal preference
idle granite
does anyone wanna help me figure out how to install/use this? lol https://github.com/ykasten/layered-neural-atlases
trail ledge
is something else sitting on the gpu, like a videogame?
try rebooting, maybe there's a lingering process in the background that still has a hold of the gpu
well, it's not the 3060. I have it running on a 3060 just fine.
did you follow the installation direction?
I didn't
I followed the installation direction
the driver is just my normal nvidia driver
I did not install any new driver just for Automatic1111
it's the front page of the project
scroll down for installation directions
On a typical Windows machine, you should only need to download python and git. The default nvidia driver you have will work with A1111
unless there is something really weird with your setup
well, hopefully one of those things you installed isn't sitting on the gpu 🙂
tough peak
anyone know a way in automatic1111 webui in the img2img tab to get it to just grab the resolution of the img that's there
nova lily
dont think so but you just set it once and it stays til you restart it @tough peak
nova lily
i understand thats a pain
Do the Live preivew settings affect performance or use system resources?
trail ledge
I'm on windows 11 too
brisk bloom
Do the Live preivew settings affect performance or use system resources?
live preview will slow things down a bit, yep
nova lily
live preview will slow things down a bit, yep
Actual Generation or just use resources?
brisk bloom
you can time it on your system try on and off
trail ledge
use artist style prompts
brisk bloom
prompt "painted in the style of artist name "
trail ledge
see the image I just generated over in #🏞｜general-with-images
fervent thunder
anyone know a way in automatic1111 webui in the img2img tab to get it to just gr...
use png info and throw info from it to img2img
fervent thunder
Actual Generation or just use resources?
actual generation too
nova lily
Thanks yea I just read from Sebastian that it does
The "progress prievew mode" tooltips are worded as if they affect the actual image, but no right?
tough peak
any recommended settings to mess with for improving image generation for high VRAM video card? a lot of vids/tips i've seen say do this or that if you have low VRAM but i have 24GB, anything i can do to take advantage of that?
wind bramble
hello i have a problem i had stable diffusion 1.5 and i deleted it for 2.1 and i want to go back to 1.5 because 2.1 is waaaayyyyyyy worst and it suck at everything 😦 the quality is way better for photo on 1.5 where i can find it cause everything i tried returned to this ***** version that i don't want...
anyone have a copy or something of it ?
brisk bloom
any recommended settings to mess with for improving image generation for high VR...
use Hires fix
tough peak
would that be better than just generating and then upscaling after?
brisk bloom
much better
wind bramble
anyone have stable diffusion 1.5 ? cause automatic 1111 force to stable diffusion 2.1...
brisk bloom
it uses the original seed and prompt adding detail. upscalers usually just incresae the res
tough peak
any recommendation for which upscaler to use? i see there's latent stuff which isn't an option in the usual upscale tab
wind bramble
please ?!
trail ledge
anyone have stable diffusion 1.5 ? cause automatic 1111 force to stable diffusio...
uh, no. I'm using 1.5 and derivative models just fine under A1111
tough peak
anyone have stable diffusion 1.5 ? cause automatic 1111 force to stable diffusio...
u can use 1.5 with automatic1111
brisk bloom
the upscaler depends on the image style. they all affect different images differently
wind bramble
i tried all the day to run installe automatic unninstal it reinstall it and each time it's that d**** 2.1 version coming back again and again
im really disapointed right now i don't what to do to get my version
brisk bloom
i tried all the day to run installe automatic unninstal it reinstall it and each...
you can upgrade to 2.1 and select 1.5 models and use thsoe just fine
wind bramble
k,ow what*
tough peak
i tried all the day to run installe automatic unninstal it reinstall it and each...
https://huggingface.co/runwayml/stable-diffusion-v1-5/tree/main u can just download 1.5 again and put it in models/stable-diffusion if u deleted it before
wind bramble
i tried this but the result is a mess a real mess
v2 of stable diffusion i don't know why they messed up so hard
static snow
So no matter what I do I can't train Dreambooth 
literally tried everything
I just want to make stupid pictures of myself
wind bramble
@tough peaki was on this page this morning but its not seem to be stable diffusion .... the files are not like mine
trail ledge
I just want to make stupid pictures of myself
probably easier to use textual inversion or LoRA if you just want to do your own face
tough peak
not sure if it's the official download or not
static snow
Any tutorials?
wind bramble
https://huggingface.co/runwayml/stable-diffusion-v1-5/tree/main @tough peak how to install it ?!
trail ledge
many 😆 Like this one: https://www.youtube.com/watch?v=70H03cv57-o
tough peak
put v1-5-pruned.safetensors in models/Stable-diffusion in your automatic1111 folder
trail ledge
or just search youtube
static snow
I'll give that a try 🙂
trail ledge
good luck
wind bramble
@tough peakand it will download 1.5 instead of 2.1 ?!
tough peak
yea that is 1.5
wind bramble
time to call a priest and maybe jesus on this one ...
nova lily
would that be better than just generating and then upscaling after?
Hiresfix better since its native and oringally outputting a highres image + detailes, but afterwords upscaling an already detailed (usually close up) image with an external app is fine
wind bramble
do i need to git clone this link @tough peak git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git
?
tough peak
i thought u had that already
wind bramble
no i deleted everything
tough peak
ya clone that then
short tapir
I need some help
tough peak
then in there there's models/Stable-diffusion, put the 1.5 file in there
wind bramble
cause i was raging on automatic1111 cause it always bring me the SD2.1 instead 1.5
short tapir
Up until a few days ago my stable diffusion was working just fine, but now it keeps telling me 422 unprocessible entry no matter what I do
wind bramble
2.1 is full of error, photo mess up and cuda out of memory
i don't know why but 1.5 issss wayyy better
i dunno why they does such of things
short tapir
How can I fix it.
short tapir
I had 1.5. it updated it without asking me
wind bramble
.....
trail ledge
Up until a few days ago my stable diffusion was working just fine, but now it ke...
Try posting a screenshot of the entire error output in #🤝｜tech-support
brisk bloom
2.1 is a mess don't use it
its a mess only cause you dont know how to use it. check images in #1045349359044280360
wind bramble
@brisk bloomit's not a huge difference with the 1.5.... i made some incredible things with that one
brisk bloom
I use both, they have strengths and weaknesses
wind bramble
@brisk bloomi checked the channel .....and 1.5 can do that and with many better way
@brisk bloom2.1 maybe and i said maybe have better results with portrait but the rest......
brisk bloom
only because 2.1 is very new. it will get better and has in a short time
wind bramble
like trying to avoid ton get some monsters instead of a photorealistic character ?
hmmm...
i'm working on photos. not drawing or kind of things like that 🙂
and 2.1 is like .......the downgraded child of 1.5
brisk bloom
#1045349359044280360 message not good enough?
potent fox
Is there somewhere a channel for erotic art? Asking for a friend.
wind bramble
@brisk bloomas i said portrait is good.
but for a photo like photoshoot it's the apocalypse
brisk bloom
but for a photo like photoshoot it's the apocalypse
#1045349359044280360 message dunno man Im very happy
wind bramble
@brisk bloom-> https://www.deviantart.com/photokarnage power of 1.5.
celest phoenix
but for a photo like photoshoot it's the apocalypse
1.5 seems to get photorealism better than 2.1 I agree. But the lighting and backgrounds tend to be less complex
trail ledge
Is there somewhere a channel for erotic art? Asking for a friend.
Erotic/nsfw art is against this server's tos, so you'll have to find a different discord
brisk bloom
but at the end of the day choose what suits your style more
potent fox
Erotic/nsfw art is against this server's tos, so you'll have to find a different...
Thanks.
Can you recommend a server or is this too precious to ask here?
tough peak
i am the friend he is asking for
wind bramble
have you got a theory about why 2.1 suck so hard ? just a question ? cause photography and Dgital art is my job and i dunno why 2.1 is like a downgraded version of 1.5, it's like incredible
celest phoenix
Can you recommend a server or is this too precious to ask here?
Avoid here. Have a look at civitai and check the right filters
wind bramble
lol
tough peak
they decided that 1.5 was too OP so they nerfed it in the 2.1 patch
celest phoenix
have you got a theory about why 2.1 suck so hard ? just a question ? cause photo...
2.1 is heavy on the negative prompts more so than the prompt
wind bramble
i have better and you can create your own if you want @tough peak@potent fox
celest phoenix
So if you have poor negative prompts it won't be as good
trail ledge
have you got a theory about why 2.1 suck so hard ? just a question ? cause photo...
maybe after they stripped out all the data from porn images, there weren't many good photos left of people 🤣
wind bramble
@tough peakit's like the creator of call of duty working on this mess ?! xD
tough peak
yup
wind bramble
@tough peakoh we have some sabotage professional in there wow impressed xD
brisk bloom
a lot more development in 1.5 as its been out longer, a little easier to use. As time goes by more dev will go into 2.1 with better custom models
the quality comes from the models
wind bramble
@brisk bloombut why they changed the prompt negative prompt why to make it harder instead of 1.5 who worked just fine ?!
@trail ledgelol
celest phoenix
Negative prompts are easy though just copy and paste
brisk bloom
I am not sure exactly but they changed how the files are created
wind bramble
i had a bunch of that
but not seems to get better resultats with or without xD
i can understand this version hahahaha it's like trying to understand a fish.
and talk with him
tough peak
fish can be good at conversation
celest phoenix
Took me a while but I'm getting better with 2.1, and with 2.1 there's more depth at the sacrifice of some realism
wind bramble
with 1.5 i coul play chess with it with 2.1 i play paint. GREAT IMPROVEMENT !
tough peak
is there any good anime models for 2.1? i'd like to try them if there are
celest phoenix
No idea
wind bramble
another stuff...the yaml file ...
what the.....
no yaml file on 1.5 !
it's already in there !
leaden mural
<@824919846570819584>but why they changed the prompt negative prompt why to make...
It's mostly that the model broke due to the removed material in said model affecting all of the weights in the neural network. Thus the reason why most people just stick with 1.5 for the time being.
wind bramble
2.1 is like a puzzle of a fish. and when you done the puzzle you can try to talk with it. but he doesn't care cause...it's a fish.
and i understand them.....trust me
trail ledge
see? I told you it was the porn removal 😆
celest phoenix
Eh its all about development. Its all rapidly progressing and developing there will of course he some slip ups along the way
wind bramble
yeah i know that's cause of porn bro..... it was the first theory when 2.1 cames out
tough peak
they removed the porn? 😮
trail ledge
in 2.0/2.1, yes
tough peak
cmon now, u cant do that
wind bramble
but it was hard to do porn on 1.5 too....so why they changed the protocol .....
trail ledge
it's all sfw all the time now in 2.x
wind bramble
1.5 too bro
tough peak
there could still be models derived from 2.0+ for nsfw tho?
wind bramble
buf if you like black rectangle we can do a geometry porn in there......xD
PERFECT!
trail ledge
there could still be models derived from 2.0+ for nsfw tho?
maybe, but probably more work than anyone is willing to do
wind bramble
@tough peakstill at Cloning Taming Transformers into repositories\taming-transformers...
trail ledge
unless you have deep pockets like a porn producer or something
wind bramble
i HOPE THIS VERSION WILL BE THE 1.5 I PREY FOR IT
ly internet connexion cry after so many Gigs downloaded......all the day to get 1.5 back 😢
btw for me automatic1111 is like a kinder egg you know ? i never know what will be in it 1.5 ? 2.0 ? 2.1 ?
maybe i will get the version 160000.7 one day
brisk bloom
using the correct installation instructions you dont get a random version
river maple
maybe, but probably more work than anyone is willing to do
Waifu diffusion 1.5?
trail ledge
that's why people are on 1.5
wind bramble
i does all the instructions on youtube on many video and the result was always the same "hello im stable diffsuion 2.1 do we meet before ?)
trail ledge
just use the instructions on the front page of the github project
wind bramble
and i say... YES and NO give me my mates back !
and he say, i can be ur new best mate for ever ifw and i say no you suck at everything it say ok...but BACK AGAIN AND AGAIN
brisk bloom
thousand of people come here and get it right, some have issues. most get it right
tough peak
yea click that
wind bramble
.......
NOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOO
HE IS BACK AGAIN !!!!!
HELLO DID U REMEMBER ME ???? WHY WHY WHY
2.1.
please......please please help me guys !:!!!!!
and it load v1.5pruned as a default model.....
it laughing at me right now
what the...........
they removed the 1.5 from the internet to force us to use that shitty version....
i bet that
river maple
1.5 is not removed, what are you even talking about?
wind bramble
bro...
trail ledge
are you confusing UI with the model?
tough peak
download the one i linked and put it in models/stable-diffusion
trail ledge
A1111 is the UI, it can use 1.5 models and 2.1 models
tough peak
then load it top left
wind bramble
i want my BFF back and is name is 1.5 ...
brisk bloom
sounds like this guy is trolling to me
wind bramble
ok so let me precise, i want the 1.5 UI with 1.5 model.
no no no trolls i cannot post any screenshot guys
tough peak
the UI doesnt change how the output is gonna be
brisk bloom
the webUI doesnt change with 1.5 or 2.1.. its the same one
wind bramble
no on 2.1 you have sketch and inpaint sketch not on the 1.5 !
and here i got sketch back again then it's 2.1 !
and not 1.5
brisk bloom
and here i got sketch back again then it's 2.1 !
its the latest webui. Its independent of the SD version
they arent different
brisk bloom
you select the model.. thats it
tough peak
ya if it says top left stable diffusion checkpoint is 1.5 ur on 1.5
trail ledge
that's what we've been trying to tell you
wind bramble
a mutant version ?
trail ledge
the UI is independent of the model version
brisk bloom
the webui will get constant updates... it works with both... looks the same
tough peak
the UI can load either, it isn't a hybrid of the two you would have to download 2.1 or 2.0 yourself
leaden mural
but it was hard to do porn on 1.5 too....so why they changed the protocol .....
To avoid lawsuits for copyrighted and suspicious material, though they also added a base model to build upon if you want to put those back in.
Agghhh, damn it you discord and your slow response times.
tough peak
it's just how you interface with the model, the models u download are different to that
wind bramble
there is no "about" section on this Ui ?
brisk bloom
if you watched videos and read instructions you should know this already
wind bramble
but why the result is waaaayyyyyy different than my version with the same models ?
do you know guys how to get back to the old UI ?
you know without skecth inpaint upload etc ????
i don't want this the old one was perfect
tough peak
what even is the old UI? you said you had uninstalled and reinstalled several times maybe u had different settings that are now erased causing the difference?
wind bramble
on 1.5 i had not such of things, but 2.1 have those new variations that i don't want i checked all video of 1.5 on youtube and same they don't have that too
fervent thunder
Just don't use them if you don't need them , why old version?
wind bramble
1.5
tough peak
ur on 1.5 lil bro
fervent thunder
1.5 , 2.0, 2.1 is just a model version, it has nothing to do with UI itself
wind bramble
i want to go back to 1.5 cause 2.1 suck at everything and too many cuda error !
tough peak
fervent thunder
download 1.5 model , put it inside
\stable-diffusion-webui\models\Stable-diffusion
and change model version inside webui
wind bramble
i have like all the 1.5 models on my computer already i put the 1.5pruned and ....im on 2.1 now...(again)
tough peak
you are not
fervent thunder
just change it, you have model list inside webui , I'll send screenshot in other chat
wind bramble
yes. i am. cause when it lauch it talk about yaml files. there are no such of things when you start 1.5
fervent thunder
Pic I sent is your model list. If you'll put model you want inside folder I sent you above - it will be on the list.
tough peak
Creating model from config: C:\Users\bradl\stable-diffusion-webui\configs\v1-inference.yaml that?
tough peak
that is fine
that's only to do with the UI bro
not the model
if in the top left u have 1.5 selected u are using 1.5
wind bramble
but....it load v2 model too....
tough peak
it really doesn't tho
trail ledge
yaml file isn't just a 2.1 thing. People use yaml for a bunch of other stuff
tough peak
bro is on the zaza
trail ledge
hell, it's not even an SD thing
tough peak
paranoid delusions about 2.1 haunting him
trail ledge
lol
wind bramble
ok
tough peak
is 2.1 in the room with us now?
wind bramble
so let me ask you differently the thing
how to get back to the old UI then ?
ok i understand that Version is based on the model you can load or not
right ?
and im not trolling
im just disapointed cause if you add yaml file you can load v1 or v2 models then you don't have any clue on wich version you are on. so why they released 2.0 2.1 etc ?!
and what about the UI then ?
fervent thunder
Technically you can clone old repo version from github...but that's not an easy thing...
I will hint you to direction ...
https://stackoverflow.com/questions/12256137/cloning-an-older-version-of-github-repo
https://github.com/AUTOMATIC1111/stable-diffusion-webui
But finding version you want - on you
fervent thunder
im just disapointed cause if you add yaml file you can load v1 or v2 models then...
You have a clue which model you are using , it's in the list on my screenshot.
On your screen - you can clearly see you are using 1.5 model
wind bramble
yes cause it's a fresh install and i followed what @tough peak says
but i'm still on the wrong UI
fervent thunder
What's the difference in UI?
Are you sure you were using automatic1111 before?
There are other UI's
wind bramble
wich others ? cause i need to confess i don't remember if it was auto or nor
not
but i think you are right ! it was not autmatic
fervent thunder
https://invoke-ai.github.io/InvokeAI/
That's another one somewhat popular
wind bramble
have you got an idea about what it could be ?
no that was not invokeAI
i used that before
fervent thunder
This one?
https://github.com/ddPn08/Lsmith
wind bramble
nop it was like automatic1111 but with no sketch and inpaint skectch
but the rest was exactly the same
trail ledge
how far back was this?
fervent thunder
yea that might be just really old version
wind bramble
december 2022
warm junco
Maybe he used stable Diffusion ui v2
wind bramble
or october
trail ledge
Maybe he used stable Diffusion ui v2
yeah, I'm thinking that's why he keeps confusing UI and model version
trail ledge
warm junco
This one?
https://stable-diffusion-ui.github.io/
wind bramble
thanks for the help guys really thanks a thousand thanks
no i posted in #🏞｜general-with-images
fervent thunder
yea that's just older a1111
wind bramble
how to get it ?
wind bramble
and why my work on this UI and i do the same thing as usual are worst than the version i used then if its the same version ? with the same models ?
fervent thunder
all this new scary buttons just gives you more options you can use or not use...
wind bramble
and more cuda error
fervent thunder
What cuda error?
wind bramble
i was able to do 1600px rez now it's just blow up when i try
warm junco
You need xformers
wind bramble
already installed....
fervent thunder
maybe cause you're missing some arguments?
wind bramble
--xformers....
warm junco
What is your card?
wind bramble
nop the same argument is in it as the old version that i had
trail ledge
so, this whole discussion over the last two hours, was just that you can't do 1600px with a CUDA error?
wind bramble
like with DPM2a you can have a grainy face like real photo and now with the same set it's plastic ....
wind bramble
never use that....cause it cause me clones everywhere it's a mess
fervent thunder
It's opposite, actually.
It is supposed to fix issue with clones
tough peak
if you really want to go back to an older version of the UI pick a commit you want to go back to and do git checkout <commit hash> in the automatic1111 folder
fervent thunder
yea I sent him a link with explanation and to original a1111 github
wind bramble
bro ....you are great in this but not me lol
tough peak
the last one made in december 2022 was 210449b
so u can git checkout 210449b and u will be back to the way it was 31st december 2022
wind bramble
i followed the same tutorial but now i have a weird UI it's not like i had when i followed the same tutorial :/
tough peak
i just told u how u can go back
if u want further back than that scroll through the commits and find one even further back
trail ledge
yes, tutorial teaches how to get the latest version
of course it's going to change over time
wind bramble
https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Seed-breaking-changes someone teach chinese or arbian here ?
cause i don't know what they talking about
wind bramble
i just want my old uIIIIIIIIIIIIIIIIII
tough peak
i tell u how
git checkout 210449b
that command in the automatic1111 folder
if it's not far back enough i can give u one thats further back
wind bramble
im crying and laughing at the same time it's incredible i never exeperienced such of things
oh !!!
let's try it 🙂
fervent thunder
i'm shaking and crying rn
, sorry , couldn't resist for a quick meme
tough peak
*i'm shaking and crying rn*
, sorry , couldn't resist for a quick meme
my dog stepped on a bee 😢
wind bramble
@tough peak kimho@DESKTOP-IDD03SF MINGW64 /d/sd
$ git checkout 210449b
fatal: not a git repository (or any of the parent directories): .git
kimho@DESKTOP-IDD03SF MINGW64 /d/sd
trail ledge
you need to be in the automatic1111 directory
tough peak
ur not in the right directory
tough peak
no exactly what i wrote
wind bramble
#🏞｜general-with-images @tough peak
wind bramble
let's try it ^^
fervent thunder
I think you should be able to
git checkout @{X.days.ago}
or
git checkout @{'x month x weeks x days x hour x second ago'}
if that's easier.
wind bramble
venv "D:\sd\stable-diffusion-webui\venv\Scripts\Python.exe"
Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug 1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
Commit hash: 210449b374d522c94a67fe54289a9eb515933a9f
Installing requirements for Web UI
Launching Web UI with arguments: --xformers
A matching Triton is not available, some optimizations will not be enabled.
Error caught was: No module named 'triton'
No checkpoints found. When searching for checkpoints, looked at:
- file D:\sd\stable-diffusion-webui\model.ckpt
- directory D:\sd\stable-diffusion-webui\models\Stable-diffusion
Can't run without a checkpoint. Find and place a .ckpt file into any of those locations. The program will exit.
Appuyez sur une touche pour continuer...
OH
i know why
fervent thunder
just put model inside a folder
\stable-diffusion-webui\models\Stable-diffusion
yeah that was the probleme..it load now 🙂
omg it download things ?
why ???
1.59G ????
and why it talk about triton ?
trail ledge
probably because you didn't have a model installed?
wind bramble
i have 19models in the folder....
trail ledge
No checkpoints found. When searching for checkpoints, looked at:
- file D:\sd\stable-diffusion-webui\model.ckpt
- directory D:\sd\stable-diffusion-webui\models\Stable-diffusion
Can't run without a checkpoint.
are your 19 models in those locations?
wind bramble
ohhhh there is a thing to do rename a model i guess into model.ckpt
but i dn't remember wich one
grim crown
Hello guys. I have a problem. I using Windows and i use Stable Diffiusion Webui for latest and i have a GPU / 30 GB Ram but its so slow making a picture.
i how to draw fast?
trail ledge
what's your GPU?
wind bramble
xformers @grim crown
grim crown
what's your GPU?
NVIDIA Quadro P4000
wind bramble
what is the file i need to rename
grim crown
%PYTHON% launch.py --xformers --force-enable-xformers --disable-safe-unpickle --api --no-half-vae --disable-console-progressbars --enable-console-prompts %*
im using this
but its so slow
wind bramble
hmmm
grim crown
Time taken: 1m 45.68sTorch active/reserved: 5310/6808 MiB, Sys VRAM: 8192/8192 MiB (100.0%)
tough peak
what kind of resolution u tryna do?
tough peak
also 100% vram usage, prob not good
wind bramble
try to keep xformers a retry
grim crown
if i dont use high res
Time taken: 15.22sTorch active/reserved: 3005/3630 MiB, Sys VRAM: 5172/8192 MiB (63.13%)
tough peak
how high were u goin?
fervent thunder
What resolution are you trying to generate?
grim crown
What resolution are you trying to generate?
512x512 to 1024x1024
wind bramble
lol
grim crown
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 6, Seed: 23432658, Size: 512x512, Model hash: 0d27c62ffa, Model: realdosmix_
wind bramble
ok delete all the flags and keep wformers and relauch and try
grim crown
i delete all xformers?
%PYTHON% launch.py --disable-safe-unpickle --api --no-half-vae --disable-console-progressbars --enable-console-prompts %*
like this
wind bramble
no like this
grim crown
oh
wind bramble
set COMMANDLINE_ARGS=--xformers
on the webui
just keep xformers
and retry to generate or upscale
tough peak
Time taken: 20.46s
Torch active/reserved: 6450/10972 MiB, Sys VRAM: 13336/24576 MiB (54.26%)
doing 512 up to 1024 using latent upscale
wind bramble
is it better ?
tough peak
but thats on a 3090 which according to techpowerup is 371% faster than ur gpu
grim crown
set COMMANDLINE_ARGS=--xformers
okay wait
tough peak
so idk if that is that slow for ur gpu
grim crown
so idk if that is that slow for ur gpu
i used huggingface
but its now
adding guard in billing system
(virtual cards is are disabled)
so i using now paperspace
tough peak
i think it's just relatively slower hardware
tough peak
and this is a gaming site so it's prob even more than 371% faster for AI stuff
trail ledge
at least according to benchmarks
grim crown
The virtual card is currently unavailable on huggface. When do they fix it?
grim crown
what you using
wind bramble
i followed that tutorial btw https://www.youtube.com/watch?v=Pyze0seDHzA&t=369s&ab_channel=OlivioSarikas
grim crown
do you have a good GPU u can use on ur own PC?
im using rtx 3050
tough peak
neither is a p4000
wind bramble
2080TI vs 3050 who win ?
tough peak
2080ti
grim crown
i how to use
wind bramble
wind bramble
here is the specification of your card @grim crown
and yeah... that card is a little bit weak
but for 512 even 1024 should be working
trail ledge
it works, and the speed is about inline for the card
he's just not happy with the speed
wind bramble
without the result will be surely near a castrophy and mess
or maybe twice at the same time
grim crown
omg.
wind bramble
...
grim crown
u need to change ur card bro
i using vds
grim crown
it's good buddy !
no its not good when i use A100 its so fast 10 seconds
tough peak
yea an A100 is way better than a p4000 and has 40GB of VRAM
obviously it's going to be faster
wind bramble
yeah sure
then it's not a problem of flags or something just your card limitation
or weakness of the card
you cannot do better bro
or try to generate 140x140 then upscale it
but the result....
i go to sleep have a sweet night guys !
and thanks for your help !
i appreciated that !
fringe lion
hey guys quick question, if im training a face with lora cause my GPU only has 8gb VRAM, will it lower the quality of art i can generate using that training? should i have a friend with a more powerful GPU train for me?
trail ledge
training quality shouldn't be affected if you're using the same parameters
fringe lion
just takes longer then. How about quality of art generated?
trail ledge
same, unless you're generating large images
some methods are vram bound
if you're going super high res
trail ledge
speed is probably the most noticeable
especially when you want to experiment with a bunch of images
nova lily
Does prompt order matter? I understand your main subject being first. But what if you put "blue shirt" "Red shirt"
Will red shirt come out more often or will blue cause its earlier? Is it placebo? random?
empty knoll
I recently saw a local training application that you could export to cloud from the app, but I cant remember what it was, does this ring a bell to anyone?
it would set it up in a google colab notebook
quick narwhal
What does the preprocessor do for controlnet?
empty knoll
it preps photos for the model, so if you use depth it will create a depth image for the depth model to process, if you already have what you need you can choose none
empty knoll
What does the preprocessor do for controlnet?
Sorry that was in reply to this
quick narwhal
Sorry that was in reply to this
So it should match the model?
empty knoll
correct
quick narwhal
Got it, for openpose, does the input have to be a stick figurine, or can it just be a photo?
empty knoll
it can be a photo, it will do its best to interpret the pose and create the little line pose model
quick narwhal
Got it, thanks so much!
empty knoll
yeah for sure, its pretty neat, I was playing around with controlnet and the gif2gif extension to recreate gifs
quick narwhal
I'm just starting to try it out, a few days late to the party
need to figure out what all these models do
empty knoll
somethine ground breaking is coming out every 3 days, so you are prone to be late on a few things
did you look at the github, it has examples of each with the raw preprocessed image and then generations https://github.com/lllyasviel/ControlNet
so the first image in each set is just the pre processor doing its thing
trail ledge
Aitrepreneur also has a summary page here: https://miro.com/app/board/uXjVPnB9L2M=/
quick narwhal
The examples don't really convey what each model is best suited for to me, It feels like the depth model does all of that
empty knoll
yeah I have been leaning on the depth model and sometimes canny if depth does not work too well
empty knoll
Aitrepreneur also has a summary page here: https://miro.com/app/board/uXjVPnB9L2...
Thanks for this
trail ledge
yw
empty knoll
The examples don't really convey what each model is best suited for to me, It fe...
Pnce you get familiar, this might be worth playing around with, I havent yet myself but plan to https://www.reddit.com/r/StableDiffusion/comments/1152ius/mindblowing_controlnet_trick_mixed_composition/
quick narwhal
Pnce you get familiar, this might be worth playing around with, I havent yet mys...
I saw that post but was too lazy to figure it out lolololololol
empty knoll
once you get the hang of controlnet it should be easy
just using img2img instead of txt2img and supplying a photo under img2img and the controlnet section I think
and adjusting denoising
quick narwhal
ok so this means I can blend elements from 2 imputs? 1 in img2img and another in controlnet
empty knoll
I believe so yes
quick narwhal
Got it, I shall give it a try
empty knoll
I am no pro, but if you have any questions hit me up, and I will help if I can, I am pretty far down the rabbit hole at this point
quick narwhal
I am no pro, but if you have any questions hit me up, and I will help if I can, ...
I appreciate it, the circle needs more people like you 🙂
empty knoll
I appreciate that, there are definitely a lot of helpful people in the space though
I made a subreddit, but I have not done much with it yet, if you feel like sharing any of it, feel free to post here, I might try to get the subreddit going a bit more soon https://www.reddit.com/r/aiwallpaper/