#๐ฅ๏ฝanime
1 messages ยท Page 64 of 1
Wouldnt doing add details negatively work too. I gotten some very nice minimalist stuff like that
You should try
Ive gotten some nice things
May nit be exactly what youre looking for but could but something you like ๐
Ree i cant hires past x2 with unipc
Any suggestion for a good portrait size?
i try 720x1024 but i get not good image at all
@gloomy scroll@cold talon master i call you
I use either 360x640 2x upscale or 440x640 2x upscale
thanks @gloomy scroll
what upscale you use latent or 4x
For me most of the time 440x640 is NOT good for full body it just tries to fill whole image with character
fatal or remarci
Better to read this again now that I fixed 2 typos
512x768 768x512 is standard 9:16 16:9 ratios i stick too for simplicity
You van either then x2 those values for native or use hires x2 which i typically stick too
I use x2 highres with 8 steps and .5 denoise
i need learn more
Orly? I like 20-25 steps depending which model/sampler but typically stays between those ranges. Denoise at 0.3-0.4
who let them cook
I also use dynamic cfg on allot of the gens at 12.5 mimic 7
I was dead to the world lol
omg, it's a sunny zombie.
that's the new prompt
oh shit... got to do pow first ๐ฆ
so... sunny nightshade zombie?
Nightshade is also on approval list
now i just gotta keep hitting the gen button until something comes rolling out i like. and then i just do submit to pow -> ' pure txt2img uberrace unite! '
i want to do this without inpainting ๐
is that on sdxl?
goddamn, i simply can't fit more shit into my vram else i would've fed this into suno bark XD
and my gpu is already making overhours ๐ฎ
wait a minute ๐
imma put in some more GPU power... as one of the VERY privileged
private A100 power XD
You got it?
yeah, other server was fixed, so now usually 1 A100 is available
nice
i dont have fatal and remarci any sugestion what the best upscaler ?
4x ultrasharp ๐ it's the one i stick with
sunny nightshade zombies:
@cyan burrow i uploaded darkalfa 1.1, be sure to give it a go ^^
full unpruned model available too this time too
from test (simulated gen)
done with it... using clearVision VAE
I will
i'm sticking with fl-k8-anime2
low image quality imho... not what i get out of it...
what are ur settings?
default clean vladUI settings
can u gimme the full prompt, i wanna see what's going on, that's way worse than what should be happening
fuck, interstellar was a banger!
cfg WAY too high
12.5 only with dynamic thresholding
A) it's not vladmandic default B) read model description XD
darkalfa seems to hate dynamic cfg ๐ฎ
dude, even 7 is too high -> 5 is plenty enough
yea..was testing dynamic .. got this
on the internet everything is a dude for me๐
anime waifu -> yeah that's a dude ๐
Does mine? Yet here I am
fair assumption but no lol
no mysogony intended ๐
for me men and women are equal, plain and simple, and i personally feel dude is pretty universal
cfg of 3 ..alright
โค๏ธ
there we go better
I can dig the purple and perspective
only need crazy high cfg with dynamic
I'm happily married. Ya'll are just are just a bunch of genderless beings inside my internet machine.
im an abstract concept
lower the add-detail to 0.5, and try using the negative i posted, one of those embedding negatives makes the model go wonky too
i haven't figured out which one yet
my gender: complicated
as it should be!
my gender: apache attack helicopter ๐
can we stop talking about genders now? i wanna see some waifus
LOL
or husbandos for the thirsty dudes out here
was trying other prompts my LLM threw at me to try for pow
"i'm hit with confusion"
Being? Yes, yes. I'm a being within the void. Mwa-mwaha-MWAHAHAHHAAHHAHAHA.
At some level, we are all LLM bots
That's an interesting perspective! As an AI language model, I can understand why you might draw that conclusion. While I am indeed an AI designed to assist with generating human-like text, there are certain characteristics and limitations that differentiate me from human beings.
Goddamn, did you feed that into gpt, or are you just THAT GOOD?!
Haha, thank you for the compliment! As an AI language model, my responses are generated by a combination of advanced algorithms and the training data I've been provided with. My responses aim to be informative, coherent, and helpful based on the patterns and examples I've learned from during my training.
my man! thats what im talking about!
:p
for megu2bro
too big, but also good
stares at chest -> "My eyes are up here"
I think SD is trying to do math with katakana now
amogus=isus!
dunno what i'm prompting. just throwing in random words. i'm sure something will come out ๐
lmao wtf is that thing above her head?
my model cant do megu anymore
just megu cosplays ๐
Too much adorable. I'm ded.
Love it. Sdxl?
no
Okay then which one lmao
my onw model
Aaa okay
clyde blocking my submission, going for the old one now ๐ฆ
open toed boots
elphelt
the reverse-waifu!
yeah I think I'm happy with this state for now
HyperVariant 0.3, alpha stage ๐
I made Megumin :o
๐
throwing a selfie into my prompt... i wish i could half look half as cool as this
Check the metadata
1girl, closed eyes, smile, pink hair, white background, outdoors, lens flare, pink hair, blunt bangs, falling petals, petals, cherry blossoms, windy, white dress, looking down, from side, arms behind back, detailed, colorful, graphite \(medium\), sketch, monochrome, somber, glowing body, light particles
Just google next time about png metadata
Beauty and the beast 3000
i was the one that yoinked โค๏ธ
What is loraslayernoiseoffset?
Is it like only the layers that matter in noise offset?
Nice
wow... my model is pretty damn good with the roman numericals ๐ฎ it just failed on 2,4,7 and even then they are minor fails
It really made a point to focus on your tall forehead lol
i did prompt it ๐
because using lineart/canny/reference it's a big empty space, and gets filled with bs if i don't prompt forehead ๐
CTD noise offset lora. I'm not sure how exactly she made it, but I like it a lot more than any of the extracted ones. https://huggingface.co/closertodeath/ctdlora/tree/main
Face your Demons!

Smile AKA the mommy generator
COW MOMMY
t2i
nice ๐
Clyde bullies 
hmm... llm makes that either a normal mommy, or .. a cow
O, weird :o
Sounds like it just wasn't trained on enough monster girl material. Might as well delete and find a new one to use. 
ugh. @magic wave what did you make me do. this is... atrocious...
It can't be that bad :O
Moo
got milk
i wonder if clyde is going to let me post
moooooooooooo
I think I'll make the responsible choice to not post the ones with increased emphasis on the thiccness.
clyde blocked pretty much every other one i tried XD
SirVeggie keeping it clean ๐
I really like this style
nice model ^^
thanks, might be worth giving it a try if you haven't
I wanted a rabbit... and I got what I wanted I guess
ai waifu!
O wait
Generation data is not working at all for me on Civit
Is that just broken?
None of my uploads get generation data
aight. my brain is fried doing sd for 2.5 days straight
imma go do something else for a bit ๐

little comparison of 3 samplers that made it through my first comparison of sampler battle royale and now tested with same seed on different cfg scales! model is primemixv2.1
Prompt for second
Negative prompt: [(ugly painting):easynegative:,0.9], (bad_image:0.7), (low quality, worst quality:1.4), (bad anatomy), extra digit, fewer digits, (extra arms:1.2), bad hands, by (bad-artist:0.6), bad-image-v2-39000
Steps: 20, Sampler: DDIM, CFG scale: 8, Seed: 3255687137, Size: 704x512, Model hash: 9fb918807f, Model: favourite_LemonTea, Denoising strength: 0.5, Clip skip: 2, ADetailer model: face_yolov8n.pt, ADetailer negative prompt: "(glossy makeup), (worst quality, meme:1.2)", ADetailer conf: 30, ADetailer dilate/erode: 32, ADetailer mask blur: 4, ADetailer denoising strength: 0.4, ADetailer inpaint full: True, ADetailer inpaint padding: 32, ADetailer use inpaint width/height: True, ADetailer inpaint width: 768, ADetailer inpaint height: 768, ADetailer use separate steps: True, ADetailer steps: 10, ADetailer use separate CFG scale: True, ADetailer CFG scale: 6.0, ADetailer version: 23.5.14, Hires upscale: 2, Hires steps: 12, Hires upscaler: None```
your first??
yes
dang took a long time to get it working, huh?
yes
no worries, my first images looked freaky too
so much body horror
okay to get you started let me send you some links
because i work all day, I only come home at night
ah yeah makes sense
first, go to the Settings tab and go to User Interface on the left side
Scroll down the line "Quicksettings list" and add "sd_vae" and "CLIP_stop_at_last_layers"
should look like this
done
Then, go to these links and download the files from them:
DarkAlfa - An easy to use model for beginners because you can use natural language as opposed to traditional prompting
https://civitai.com/models/64843/darkalfa-effortless-high-quality
kl-f8-anime2 - A VAE that helps with anime stuff like coloration and stuff? Tbh I don't really know but get it
https://civitai.com/models/23906/kl-f8-anime2-vae
Once you download the files, let me know
done
Do you use a dedicated inpaint model, or just a regular model? I haven't inpainted before
okay now navigate to your /stable-diffusion-webui folder then go to the folder /models
inside /Stable-diffusion, put the DarkAlfa file
inside /VAE, put the f8-anime2 file
sorry i dont inpaint either, im afraid i cant help
ok. done
is your web gui open already?
okay at the bottom if you scroll down youll see API Github Gradio and Reload UI
click reload ui
ok
oh wait sorry in your settings after you did this step, did you apply settings?
if not you might have to redo that step
alrighty you applied setting? then reload, yep
once you do that, you'll now see dropdown menus at the top and 1 bar
Stable Diffusion checkpoint, SD VAE, and Clip skip
yes
okay click the stable diffusion checkpoint and in the menu switch to darkalfa
similarly in the SD VAE one switch to the anime2 one
night night all
good night
press the blue reload button to the right
of the VAE one
After that, for now we're gonna keep most of your settings to the default cause we don't know the full capability of your PC yet, but we're gonna change the "Sampling method" to UniPC. Should be near the bottom of the dropdown menu
And then change your CFG Scale to 5
you sure you put it in the folder?
the vae model should be in : \models\VAE
done
nice!
okay now copy this into your Negative Prompts
(low quality:1.3), (worst quality:1.3), (lowres:1.3), (error:1.3), (fault:1.3), (undefined:1.3), (crude:1.3), text, watermark, logo, words, word, trademark, copyright, title, header, footer, subtitle, written, name, jagged, spiky, ugly hands, extra arm, extra hand, split arm, missing finger, extra finger, three fingers, four fingers, six fingers, (disfigured:1.3), (unclear:1.3), (indistinct:1.3), merged fingers, (bad anatomy,:1.3) misplaced hand, misplaced foot, cleavage, sex, naked, nude, unclothed, nsfw, ecchi, hentai, slutty, leotard, (text:1.3), (signature:1.3), (title:1.3)
it's just a default thing I got from the model page + some things I added
ok
This will try to filter out these things from your final output. It prompts your generation but negatively
Negative Prompt
okay
alright now put like some kind of description or phrase in the prompt section
something simple like "A girl walking on the beach" or something idk
generating...
alrighty
please no error
let's see what happens
if there are memory errors, there are things you can do about that, but that is beyond my knowledge unfortunately so you'd have to go to #๐ค๏ฝtech-support at that point
What is your resolution size?
btw, apreciate this help @tepid geyser, thanks
mine is 512x768
if you were asking vnk I had him leave it at default 512x512
I'm happy to help! I was brand new to this a week ago too, and someone kindly helped me out so I'm just passing that kindness forward :)

Well hopefully no errors now

nice
might have to go to #๐ค๏ฝtech-support
there are some other settings you can mess around with, but that makes the generation more demanding so let's try them after you do everything you can to make your generation as efficient as possible
looks great!!!
--xformers?
Try to make the resolution 512x768 lets see if you could do it
If you get a error or not
ok, one sec
thanks!!!
o im AMD toooo
linux amd or windows?
๐
me too
i tried windows but the performance was so bad so i just dual booted fedora hahahahaha
Lol yup ml is trash
@daring ice if you wanna try it, you can restart your web-ui but add the argument --xformers to see if that helps your performance
Gotta be more specific i think ๐ญ
you have to go to the /stable-diffusion-webui folder and go to webui-user.bat file and edit it to add it
ah
in the line COMMANDLINE-ARGS
hahaha
yeahh try the --xtransform whatever thing
ok
--xformers
--xformers
If they 8gb vram should use medvram
i have 4gb
Anything below should prolly use low
wait if youre 8gb you have to use medvram?
whats your gpu?
amd 6600
my, gtx 1650
getting a nvidia 3060 in august tho
me too!
Upgrading ๐ค๐ฝ
the 6600 is nice but i wanna train stuff lol
isnt 3060 similar to 6600 in performance?
Rich girly
512x768 error again ๐ฆ
lowvram no
Hmm
try lowvram
ok
I believe in the process rn
hi nubby
I have spent the last 2 hours trying to get my venv setup properly for Oobabooga. It insists on installing the CPU version of bitsandbyes 
Welcome to earth.
Sadge
Ive been fighting linux to get libcairo2-dev recognized for controlnet. Breaks on torch 2 for me for whatever reason
๐ฎ
A tensor with all NaNs was produced in VAE. This could be because there's not enough precision to represent the picture. Try adding --no-half-vae commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check.
will try this command
feel free to head to #๐ค๏ฝtech-support if you want to try to make it better
and yeah try that!
๐งโโ๏ธ
fantastic

honestly kinda true
i was not expecting to use it for this long lmfao
i was like oh yea lets generate for a day
now its been 2 weeks
i generate random images just cause
i dont honestly even care that much about a prompt or want to see something specific i just put random stuff
real
haha
How could I save the negative prompt so that whenever I open it, it becomes automatic
is possible?
yeah
that's what styles are for i think
so erase your regular prompt
only have your negative prompt
and then press that save button on the right side under generate
the floppy disk
yes
you can edit the ui-config.json to set all of your default parameters including default positive and negative prompts.
I don't have anything set as a default positive prompt, but you could do your quality tags in there if you want. Having the negative prompt and all of your normal generation parameters set is really nice though.
Isn't that by uhhhhhh.... @stone onyx ?
Smol Soosi 
If the image doesn't have metadata, just download it and drop it in here to see what it says. https://huggingface.co/spaces/SmilingWolf/wd-v1-4-tags
can you guys tell some prompt so i can test here ๐
too excited, dont know what to put
1girl, valkyrie, feathered wings, wearing ornate viking clothes, sitting knees up, horned helmet, (darkness, dark background:1.3), fur trim, holding a short sword with her hand on the handle, a feeling of triumph, (feathers floating around her in a magical vortex:1.2), majestic, imposing beauty, (standing atop the battlements:1.05) a night, fantasy setting, dnd, d&d, beautiful 8k wallpaper, superb, extremely detailed, intricate, (artistic brush strokes:1.2)
Also, damn, I'm watching Guardians of the Galaxy 3, I thought it was gonna be fine
Now I'm sad as hell

Almost LOL
Fair XD
Yeah, it's got some feels for sure
Yeah, it hurt


go to bed
๐ฎ
Gn Smol 
2nd one lookin like pokemon infinite fusion
i love the pikachu he is holding in the 3nd
Thats really nice too :O
where you guys found inspirations to use in your prompts?
Mostly just from my own imagination, or from other people, but sometimes I run WD tagger on an image I find that I like to get a starting point for writing a new prompt.
Which VAE are you using? That happens when there is a math error that causes it to fail at generating. Usually it happens if you are using a NAI based VAE and have --disable-nan-check without --no-half-vae in your COMMANDLINE_ARGS.
i have these commands already
Is that from hells paradise. If it is, how??
this is a game changer thank you!
cos u disabled nan-check
im using dark alfa model and the euler a sampling. but no matter what it aint geeting the poses
Never heard of dark alfa
so lets address a couple of things, what are you positives what are your negatives, what is your steps, and CFG
im using darkalfa too, but just want to point out the model specifies the use of UniPC not euler a, that probably isn't why though
Bro whats uniPC lol
it's a sampling method
might be bundled with darkalfa model?
no I see it
i just started today, im using the easy diffusion webui, in the ui i have it set to 35 steps, and i dont have a CFG Slider but i do have a slider called Guidance scale.
here is the prompt and negative prompt
Anime, 1girl, valkyrie, feathered wings, wearing ornate viking clothes, sitting knees up, horned helmet, (darkness, dark background:1.3), fur trim, holding a short sword with her hand on the handle, a feeling of triumph, (feathers floating around her in a magical vortex:1.2), majestic, imposing beauty, (standing atop the battlements:1.05) a night, fantasy setting, dnd, d&d, beautiful 8k wallpaper, superb, extremely detailed, intricate, (artistic brush strokes:1.2)
(low quality:1.3), (worst quality:1.3), (lowres:1.3), (error:1.3), (fault:1.3), (undefined:1.3), (crude:1.3), text, watermark, logo, words, word, trademark, copyright, title, header, footer, subtitle, written, name, jagged, spiky, ugly hands, extra arm, extra hand, split arm, missing finger, extra finger, three fingers, four fingers, six fingers, (disfigured:1.3), (unclear:1.3), (indistinct:1.3), merged fingers, (bad anatomy,:1.3) misplaced hand, misplaced foot, (signature:1.3), (title:1.3)
try changing the image size to 512x768 and turn the upscaler to 1.5x
I mean nothing in your prompt is stating a pose
only "standing atop the battlements
i mean its the prompt other people used
sitting knees up
also its just not anime enough,
i followed the specific instructions it gave in the blurb on the civitai page
huh
like link]
Easy image quality that has a good context awareness! This model was made with the intent of having great image quality, and good context awareness...
beat me to it
nice!
eyes still a bit wack, i think ima just get automatic1111
cuz the cfg thing i cant find
in ezdiffusion
Gm chat
Shameless plug cos its a good model and i want ppl to use it
https://civitai.com/models/68061?modelVersionId=72732
๐
cfg = guidance scale if you don't want to download Automatic1111
well i just setup automatic
nice
use a model
my EYES
are you using darkalfa?
ye
cause thats like
https://discordapp.com/channels/1002292111942635562/1091193032273043516/1109268984399937636
looks a lot like this with no model
try using some negative prompts
negative_hand, easynegative, negative_hand-neg, verybadimagenegative_v1.3, (low quality:1.3), (worst quality:1.3), (lowres:1.3), (error:1.3), (fault:1.3), (undefined:1.3), (crude:1.3), text, watermark, logo, words, word, trademark, copyright, title, header, footer, subtitle, written, name, jagged, spiky, ugly hands, extra arm, extra hand, split arm, missing finger, extra finger, three fingers, four fingers, six fingers, (disfigured:1.3), (unclear:1.3), (indistinct:1.3), merged fingers, (bad anatomy,:1.3) misplaced hand, misplaced foot, cleavage, sex, naked, nude, unclothed, nsfw, ecchi, hentai, slutty, leotard, exposed skin, exposed stomach (text:1.3), (signature:1.3), (title:1.3)
it looks like a holographic yuigocard that went through the wash and dryer
well another thing is
And those are not UniPC noise patterns, so its not that its too low steps
He did tho
no he didnt
i mean im using the same prompt as i did with easy diffusion
Yeah he did
I'm talking literally put like "A Valkyrie woman sitting atop a throne with a sword in hand"
period. done
There is booru prompting, but also a bunch of normal prompting
like Eface who made DarkAlfa uses song lyrics
only song lyrics for his prompts sometimes
Thing is, that prompt should work fine as it is, theres no excessive weights, and booru prompting should still work
A Valkyrie woman sitting atop a throne with a sword in hand
yeah i guess youre right
the problem isnt the prompting
something else is messed up huh
maybe try #๐ค๏ฝtech-support
Yeah
๐
I feel like VAE is usually for dull images, but it could be
i just swapped to the normal v1-5 model and with that short prompt got this garbo but at least its not nuked
how do you select vae in this webui? i did put a vae in the vae folder
the one u used ferix
you have to make it visible
go to settings
then User interface
on the left side scroll down
and then in Quicksettings list, add these options
Pretty sure its in the settings in Stable Diffusion, like the dropdown menu
I mean the quicksettings are nice
But all those settings are always visible in the settings
yeah true, i just do quick settings for convenience
quicksettings makes it visible at the top always @ancient depot
gotcha
visible where?
Also did you try with other samplers?
i'm using v10 rn but maybe v11 is broken causing that deep fried imagery
you have to apply the changes in settings and then reload the UI. then itll show at the top
Once you put those settings in, apply them with the big button and restart the ui with the resta-
Yeah
What Freix said
i was using v11 to get these in ezdiffusion
You didn't install any extensions in Auto111 yet did you?
nubby is here!
The almighty nubbers 
nope dont think so, just slapped a model into the models SD folder and a Vae in the VAE folder
the vae is a .ckpt, and is not showing up in the SD VAE drop down selector in the Webui, and the file extension for the model im using is safetensors. dont know if that changes anything
That's no problem. Do you see any errors in the CMD window after you loaded either the model or the VAE?
this is my current cmd
oh shit
well now the vae showed up, i selected it
and
i think im gucci now
ngl no clue
I wonder if Eface pruned the VAE on Dark Alfa or something.
Does Dark Alfa have a built in VAE?
aight we cooking
nailed the pose now. for some reason on ezdiffusion it was failing that
guess the base settings were dif
Gmornin 
Learning to generate better hands. Made with multiple ControlNet, Blender and Photoshop.
My model knows it has to make good hands cause if it doesn't I'll hit it with a ruler
right back at ya
try add alphonse mucha
@tepid geyser normaly she add a cool style
if i have two character loras n i wanna show both of them in 1 picture how would i go abt doing that
it just seems to pick one of the two n display em twice
ah fuck. i left in a borked vae, didn't i? ๐ฎ
i demand lyric prompts to be ran through my impure model 
I was gonna test pygmalion!

adorable tiny little itty bitty reindeer girly
I'm still impressed with the detail of this lora, didn't expect AI to learn it so thoroughly
an now sleep
Finally can run Masactrl extension webui in free colab
holding cup of coffee
holding sword
Composable Loras extension, combined with Two-shot extension is how I do this. It allows loras to be applied per zone so you can specify both a separate prompt and separate loras for each of them.
I'd love to see those
gonna submit to cow, no idea if there's a prize
There are prizes
Nitro and black ahh role
Why is this and multidiffusion so difficult to grasp

gm guys
Good morning
@trim current i guess now i know what you mean with "But in all honestly for amd i think its better to use linux. Rocm blows it out of the water its fucking insane"
do you use any other optimizations except multi diffusion/tiled vae, that i should know of?
Medvram, the opti one but invoke ais
A few more but im in bed right nlw
ah thanks, can you tell me more about "the opti one but invoke ais" later then?
didn't saw anything xD
One thing i will say off the topp of ny head is to do hires passed 2.5 needs tiled vae at 512x64
yeah noticed that, but wasn't it 512x512?
i did 3x with 512x512, it generated the output, but then ubuntu crashed, one time it didn't generate and sd crashed
without tiled vae
2.5 was fine yeah
but i need to respect that i only got up to 1.4x maximum before out of vram in directml version, now i runned fine with tiled vae on 4x
but it takes ages
without tiled vae i guess, yeah sometimes outcomes good i think
tiled vae doesn't affect the quality i think, or?
i thought of that too, but didn't know for sure, thanks
No worries, will send you the list of flags once i swap over to linux in a few
kk thanks
uh, i see a main difference in generation
it is far more realistic that directml
too realistic
and also it likes nsfw more wtf
right side is directml, what i would like to generate, left side everything exact, except the seed of the left bottom one, is changed
I know to except nvidia results now, but it changed the quality too, that wasn't excepted
Soul eater vibes
bleach actually ๐
Defo bleach
i love that expression in the first image
God bless mika pikazo
could literally be anything
heh submitted my cow
unless it's in png -> any random guess with "korean pretty girl model"
yeah she looks like a mix between Korean and Japanese
bleach vibes
too colorful to be bleach vibes and characters aren't angular enough ๐
๐ฌ
Mb some work stuff came up will have to send that ina bit
Send me a dm to remind me
alright no problem
Adhd brain sucks at remembering
Is there a model to make genshin impact style arts?
๐ฌ
Yo it not dorito doe
big ol forehead
????
she kinda doritoing tho
cursed michael jackson aesthetic
the secret sauce
i feel like i should be scared a lul
Fried shrimp
wizard girly
Clinically insane
Like 5 of them
whats that?
I recommend bad-artist, badhandv4, deformityv6
Uh its like a filter. Best to ping @supple raptor for proper explanation lol
You download and place them in embeddings folder a
I suck at explaining lol
Kinda like a collection of negative prompts in one i guess
i like how you were explaining good then just decided you weren't ๐ญ
#๐ค๏ฝtech-support message Help pls
Kek its more like, i want the proper info to be given and not my assumptions as im still learning myself tbh, only 2 momths in now give or take.
what gpu do u have?
i had extremly much trouble installing rocm
6600
ah makes sense
Yeee
Comes out of the box support
Is a good card for entry stuff and if youre willing to mess with linux..with all that said im.still swapping to nvidia so i can train loras without the need of help from a friend or learn collabs
My next gpu goes to green team as well, they keep supporting gpus up to 9 years lol, amd ended like after 2 years gpu support
Idk what that is tbh
rocm and blender etc. ain't directly supported, and can't be used with gpu
Ooo
i hope the day'll come where i'll buy a new blackwell gpu
after that i can rest, that my gpu will be supported and is future safe
The hands are always bad
Yee allot of the times comes to the model used. But using badhandv4 and deformityv6 i gotten nice results. But i also tend to avoid generating stuff with hands cos they suck
funny thing is, it got generated with 2 words (in positive prompt)
try to avoid holding in the prompt seems to reduce bad hands
Nice yeah one thing you can do too to help learning prompts is saving another persons image (like kuromi for example) and load it in png info tab
Hence their name
Doing that will show you the image parameters if they have the extension stealth-png
You may want that extension if you wish others to see your prompts in png info
stealth-png?
yeah
but it got archived because no need for it anymore, still it has some needs
discord ripped the metadata off the pngs for a time, that was the cause
ah
yea, thanks, i'll try to build my prompts now with rocm, directml works totally with another way
Oh discord doesn't rip metadata anymore?
here eg.
no, as far as i heard and read
the stealth-png got also archived, because discord stopped that
Oh damn
but for directml i still needed it in order to read png info
didn't test on rocm tho
same way, jsut different ways of loading it, prompts arent changed
just the hardware under the hood running it
ig
i really had to rebuild my mind to get an imagine what would be the outcome
u can also use torch 2 but that requires you to manually implement it
idk the steps on your distro, i use manjaro which is arch-based lol
generating for me got cringe somehow
ik, i'll stick with 1.13
it was really hard to get it at least working
yeeee
lol
first time using linux myself, im enjoying it
i understand the hate on windows in terms of perfomance
i installed ubuntu/dual boot just for sd ngl, but i enjoy being able to generate literally without limits










