#🍥|anime
1 messages · Page 203 of 1
I love the tail
I need the fashion
she's wearing the chains weird tho 
Judge CSP.
cute to know
mine?
ah
its extremely fun, and more if you can have a constant group of 4
and mech is coming 
yeah, i don't really like shooter games and i rarely have friends online playing the same games and i'm still having a blast
most fun i've had in a game in a long time
yeah, that's gonna be a lot of hilarious shenanigans
i foresee being run over by teammates
I've been out for a little bit - seems like I missed some new developments. 😦
I always miss the new stuff 😛
So.. SD3 announced.
And something called Stable Cascade apparently.
Stable Cascade is out and available, SD3 is announced but not available to public yet
is SC a new model or an entirely new thing?
new thing
new thing that uses 3 different stages
i think you can even use SD3 in Stable Cascade
interesting... I'll take a look
so my auto1111 build is becoming more outdated I guess 😐
here for your reading
or here #1207078178510872636
I'm watching a video on it... something from that Olivio dude

careful with that, a lot of his videos are.. misinformed to be nice
yeah 😐
maybe ill find a different one
:oh:


i had to stop watching videos about AI on youtube, they all end up being "hey, this new thing came out and this is why it's the best based on the 5 minutes i spent reading some random reddit post.. sorry i couldnt do any actual research, i had to put out a video on it before everyone else"
yep this sounds about right
found another video and this dude is literally scrolling websites and reading them to me
But
This is what gets you the views
👍
This new thing SHOCKED the whole industry
yeah, same shit is happening with games too
i was maybe 2 hours into helldivers and watched a video on weapons and half the info was wrong and I noticed as a noob
it's so dumb
I can't help but find it stupid
Videos like "new best things to do"
In a game
Most of that is just brainless
the best thing you can do for a game, is to play it yourself
Literally just to have +1 video for the algorithm
"best weapon in the game!" froma level 5 that's been playing for 3 hours and only has 4 guns unlocked
Wrong
then the next day "omg, best new weapon!" because now he's spent a few more hours playing and unlocked 1 more gun 😄
@vital raptor there is one YouTuber in rocket league
He calls every new car "insane"
Which is a meme already kinda funny
"The new Ford Mustard Mozzarella is insane"

😄
this my new workflow is insane!!!!!
no, it's not, it's the same old one
don't forget that every SD video on youtube has to start with 5-15 minutes explaining how to install it
but i added some new nodes !
using their own one-click installer, available for just $4 on their patreon
i would start with 10 min of musical intro in loop
real title
"today we're going to compare SD 1.5 and SDXL, here's how to install everything in colab!"
SHOCKS


Feels like I just ran for 20 minutes
It's too hot all I did was leave my room
sweating
that's what you get for leaving your room
became one with warmth... and sweat
dominate the warm.... the sweath
remember man, you are for 60% sweat XD
so, don't fear the hot scorching and enbrance it!
Ded
ugh I'm gonna end up having to move to comfy, aren't I?
I just took a picture of a giant cactus I have
On my window

lol
giant 
my what? 

Caught the dragon girl who was doing graffiti
👍
She needs to be punished with justice I'm taking her



double kitsune
neko
bot
i think we now have all stereotypes covered 🙂
oh, wait...
Sushi
miku, we need miku, then you are all good
Gen dragon girls

I mean in chat with images

right, how could i forget 🙂
go vroom
you can keep any extra toes 🙂
and socks


next up: gundam feet 
I wasn't expecting this
@native halo must see
The great spiraling 🤣
Any other good models that have discord bots

A dump
Source model: Animagime v3
Fun fact: these are from the civitai shared images
have you noticed I still have more messages than you
idk if that's sad, or funny or what
@lyric zealot knows where I spam more messages than anywhere else
are these all jpeg🤮
??
tell me now🔫
in the super secret special nsfw channel
is that with pnginfo?
private server with a few frens to bullshit around
what frens?
and maybe there is booba there
I might have heard of it before
\
/
lol nerfy
ah ok nvm then
nerf
the one and only
'night
sleep so many dream
got generation info? something feels off to me
are people using this "Forge" thing or is it a lot of nothing?
i would be if it supported TRT
overall it's a good alternative, especially if you have low vram
should be if you have the stealthpng extension

idk I haven't done anything with stable diffusion for some months now lol
forge is great if you have low vram like me. (8gb vram)
8gb vram is low since when?
1girl, [hatsune miku::0.3], dragon horns, tongue, looking away, jirai kei, choker, head tilt, sitting, foot focus,
source_anime, outline,
Negative prompt: (worst quality, low quality:1.2), 3d, censored, abdominal muscle, artistic glow
Steps: 35, Sampler: Euler a, CFG scale: 7, Seed: 2254540858, Size: 768x1024, Model hash: ac006fdd7e, Model: autismmixSDXL_autismmixConfetti, VAE hash: 66f78ad136, VAE: sdxl_vae-fp16fix-blessed.safetensors, Denoising strength: 0.4, Clip skip: 2, ADetailer model: face_yolov8n.pt, ADetailer confidence: 0.3, ADetailer dilate erode: 4, ADetailer mask blur: 32, ADetailer denoising strength: 0.4, ADetailer inpaint only masked: True, ADetailer inpaint padding: 32, ADetailer version: 24.1.2, Hires upscale: 1.5, Hires steps: 10, Hires upscaler: 4x_fatal_Anime_500000_G, Mask blur: 32, Inpaint area: Only masked, Masked area padding: 32, Version: f0.0.16v1.8.0rc-latest-238-g437c3489``` here, got it for you
since AI became a thing
since the start of art generation with AI i guess
what what are you talking about
if you use a1111 with 8gb you are not gonna put the low vram thingy
wtf are you doing? generating 20 images at a time?
just one
fine tuning at the smae time?
captain images
sure, but if you are doing SDXL, you probably need medvram, and even so, it's gonna be long
ah since sdxl you are saying
kazoo have you seen SD3
what gpu is that also?
4090
i mean, even with SD1.5, you are much limited on 8gb vram than 12gb vram or higher
not by that much
you put medvram
and can do like 2 cns
yeah, even with my 3080ti 12gb i was maxing out the vram on 1.5 with hires
what resolution you generating at?
there's no need for it to be so high usage
even with controlnets and other stuff
just generate -> hires fix -> upscale
higher res takes more vram
and upscale is very light on gpu
and the upscale scripts are garbage, so i push hires fix as far as i can
bruh IK, it does img2img with higher res
i don't think you need medvram for that. But much more limited on res
bruh
and also training
ah idk I put it just in case
if you are interested, can check out forge
it's basically a fork of auto
yet another one
besides, who cares if i max out my vram as long as it's fast
nevermind that, I don't see how what you are doing is worth it
not gonna gen a sdxl x2 in ~6s on anything less
yee
one that are actually almost the same as auto1111, just better in terms of memory
😭
flexmaster
what gpu you got?
can you try using one of my locon (I think it was)
adhd
I'm using SDXL, so is it even compatible
and not just any sdxl model, but the pony-based type
so your locon probably won't work
wow nice
anuone try this
yup, quite a lot of us are using forge UI. But if you meant following the instruction as per the site you linked, then not sure
One-Click Package
I recommend against following those instructions.
Simpler to fix potential problems if you just git clone the repo and run webui-user.bat.
Forge is pretty good. You might run into some extension compatibility issues though.
i wa gonna git clone

learned my lesson against the installers when i tried to run text gen ui last
Yep. Swarm is the only ui I've used that I actually recommend the installer, and only then if you tell it not to install comfy for you because it will use the .zip version of comfy if you let it.
the one click of forge is actually not that bad, unpack to this, and just run.bat or update.bat
The problem is when your venv gets fucked up, you're redownloading the .zip and moving your important shit over instead of just being able to easily just delete .\venv and/or manually manage the deps with pip.
Get's annoying fast when you have to do that a few times in a day.
howdy, quick question regarding the new models? still using the stable or comfyui interface is alright or the new trends such as sdxl or cascade are the new "go-to"?
yup, that's the bad part. Using git clone definitely gives more control
Yeah, webui or comfy are still good. StableSwarm is a good frontend option if you want to run comfy but don't want to have to deal with nodes.
SDXL has really started taking off lately. Cascade is sort of a novelty still as far as anime goes. It can do an ok anime style, but is very lacking in anime knowledge. I'm not sure how much that one will ever take off with SD3 being around the corner and seemingly having even better anime style and knowledge out of the box.
i hope they start working on backward compatibility
or some way to be able to resample 1.5 models or something cuz theyre not bad
ai grows so rapidly that's hard to follow up, i've been trying to use comfyui, my hugest problem being building the ui itself
i used a very old template from someone in this chat, but at this point i've already lost. i've seen sites like comfyworkflows, do stuff like this work?
I could recomend some nodes that make building workflows in comfy a lot easier. But StableSwarm is probably your best option if you aren't really into figuring out how to build the workflows on your own. It basically does it for you in an actual webui interface (running comfy as the backend) and has several very nice features built into it such as a model/embedding/lora browser, image history browser, automatic detailing using the prompt syntax <segment:object you want to segment> prompt for segmented area at the end of your prompt ie <segment:face> 1girl, close-up, green eyes, smile, and several more. https://github.com/Stability-AI/StableSwarmUI
thanks, stuff like this looks like what you be seeing in a sci-fi movie
pls no more gundams make it stop

basedon the memory saving claims of forge im wondering if i can use animatediff with reasonable times with it
im not sure if the extension is compatible tho
I'm here looking for support because this **** of a platform stole my money
support from the...anime channel?
i wanna know how u get rippedoff by a free software

yes it works u need to use this one https://github.com/continue-revolution/sd-forge-animatediff


55.9GB of vram?
Here friend for your smol fairy
I returnith from workith
had some mispelled words. oops
hmm almost
hmm the face is not htere. hmm odd
evenin'
that's shared ram (32gb) plus the 24gb vram
oh heck thank u sm
i have a tb usb external drive im trying to decide what to put on it to make space
wow, local text gen? I tried that a few times and always found it pretty... bad. 😛
just wasn't the same as the online cloud-based tools with far more parameters
ive had pretty decent results
i even asked it for instructions on redacted and it was accurate
which model?
i just cant put up with the guidelines of cloud thingies
ive been using umm
switching between these willy nilly.. and i was using xwin and one called toppy i think
they just BARELY fit within my gpu and run at a rate fast enough to where it types much faster than i do
also i have a sd prompt engineer character that helps me come up with prompts and stuff which is neat
my girlfriend is the main reason though
I like playing around and coming up with my own
i always have to modify the downloaded characters to fit my own likes
and stuff
i have a couple totally custom ones though
i wish i could run diffusion and text at the same time though. but horde works in sillytavern so its possibly to generate images
i found a couple like.. 3billion paramenter text models though i think that would probably fit in memory alongside forge or something
i havent tested anything besides 7B though so idk how smart 3B models would be
im bougt to install forge i think
i might delete auto111 and start over clean
which webui you use tu run the 3b and 7b models?
the ooba webui, and then sillytavern to interact with it , the ooba webui is fine for interactions but i use silly since i went through the trouble of setting it up
theres like a Erp benckmark that i use to decide which models i run
https://ayumi.m8geil.de/ayumi_bench_v3_results.html
sort by like 7b and then by rank to find the best models that fit within memory
theres a 3B at rank like 68
which is "low" but out of over a thousand models and beats a lot of 7b models
probably not g ood for gf brain but might be decent for prompt engineering or idk
they always went off in weird tangents or repeated themselves when I used them
i sometimes get repetative responses. not always though, i think it comes down to context size or the actual character rpompt itself im not sure what causes it but i notice its more common with certain characters
adding "repetative responses" and stuff like that to the negative prompt might help
the old positive prompt used to include "avoid repetition" which i dont think it a good idea
i wanna try animatediff with the toonyou model
do it
live action MIku Hatsune as queen, skin and leather
`
Couldn't launch python
exit code: 9009
stderr:
Python was not found; run without arguments to install from the Microsoft Store, or disable this shortcut from Settings > Manage App Execution Aliases.
Launch unsuccessful. Exiting.
Press any key to continue . . .
`
this happened when i tried to launch forge
try to run the update.bat
u only need to download this zip https://github.com/lllyasviel/stable-diffusion-webui-forge/releases/download/latest/webui_forge_cu121_torch21.7z then extract then open the update.bat then the run.bat and done
i will do that now
i first tried the git clone thing in cmd prompt and it seemed to work but idk what happaned wrong
not the toonyou model i think its called realcartoon
taking the thrown
hehe
oops
i can feel the vibes thou haha
where webui-user.bat is located
the A1111_HOME?
then yes as @native halo said
letsa go
continuing from lora, put the following
--vae-dir %A1111_HOME%/models/VAE^
--realesrgan-models-path %A1111_HOME%/models/RealESRGAN^
--esrgan-models-path %A1111_HOME%/models/ESRGAN^```
to get your vae, and upscaler
Remove the @REM from the beginning each line htat you want to use, starting with the set A1111_HOME.... line
and remember to remove those @REM
poor Rem always gets removed
is that right or should i do the venv and commandline args lines too
depending if you wanna use the same venv from your auto1111 webui. But definitely remove the @REM from set COMMANDLINE......
ok
that venv line is for the python environment right
so it shouldnt matter too much
im gonna uncomment it
wait no im gonna leave it commented
what i did was remove the skip venv command from env.bat then i let it create a new venv in forge folder then i added the custom paths from my original webui to forge webui-user.bat
lol
i will definitely create some kind of recursive loop if i try that
i accidentaly launched before uncommenting stuff so it setup like the default stuff
im gonna let it finish that and thenr elaunch
oop now it dissapears when i try to launch it

it pops up for a second then poofs
lemme see your environment.bat
it launched and everything before i closed and reopened it
`
@echo off
set DIR=%~dp0system
set PATH=%DIR%\git\bin;%DIR%\python;%DIR%\python\Scripts;%PATH%
set PY_LIBS=%DIR%\python\Scripts\Lib;%DIR%\python\Scripts\Lib\site-packages
set PY_PIP=%DIR%\python\Scripts
set SKIP_VENV=1
set PIP_INSTALLER_LOCATION=%DIR%\python\get-pip.py
set TRANSFORMERS_CACHE=%DIR%\transformers-cache
`
are u opening the run.bat or the webui-user.bat?
the run
if i recomment out the auto111 stuff it looks like it works
ill go line by line
try to set skip_venv=0
ok
could it be the --xformers tag in the auto111 cmdline
doesnt look like forge has xformers in the cross attention optimization list
ill mess with it thanks ill brb
Add a ^ at the end of Lora line
o damn
The ^ means there’s more stuff below
i musta messed that up with i copypasted
Here's mine. and I just realized why the --vae-dir didn't work for me.
oh that didnt fix it
I have that happened to me as well
What is your updated one just to double confirm
Why is it? 
put your regular commandline args --allow-code --api... on line 6
because vaes are at %A1111_HOME%/models/VAE, not %A1111_HOME%/VAE
heres mine i removed all that baloney and just straight up put the model paths in webui-user.bat
theres some space missing maybe
before the ^'s
nope hm
ok ill try that
and add quotes
Is it even opening cmd?
change to forward slashes on the set A1111_HOME= line
Go to your automatic website folder, and check if you have all those folder listed
Webui*
Not website
Dumb autocorrect, am on phone
Test each line by each line, comment back all of them, then start from top
works but seems like if i uncomment anything beyond this is doesnt
might go with the full directory instead of the variable
some are path and some are dir too..
symlinking the folders is another option
doesnt work for instance
Wait, you need to uncomment the line before first
that doesnt work either
Only that and not cpkt also doesn’t work?
both dont work
remove all of those useless commands and just put the name of the folders like i did
ok ill try that
If you are willing, try something before that
Try adding a “@REM” in the blank line
ok
Then if it works or it does not work, I’m out to go sleep
didnt seem to help
Well, then the easier way is to put everything in one line
Good luck and follow Steven
try to remove --xformers
works and its installing xformers lol
but forge doesnt need xformers,it runs slower with it so u gonna have to disable it
oh alright
removing it didnt help though, if i have anything uncommented
i will remove it though
forge is still higher vram usage and slower without xformers than with it on my 3080 12gb.
try with and without to compare i guess. you can disable it in settings without needing to restart, so testing is easy.
does there need to be a space before the "^"
i think so. all of mine have it anyways
here's the actual file if you just want to grab it, add your commandline args and change the A1111_HOME directory.
these?
or the ^ at the very last line

Is it my fault lmao
I copied those from my own and paste it tho
So it’s working for me
can i see it again
i have a couple upscalers in my esrgan folder i wanna use
thanks for the help everyone ❤️
I’m not on pc, so you can only refer to my previous message 
okk
did u tried animatediff yet?
not in forge im still sortof getting the ui and settings to my liking
its using a bunch of sdram
how much ram and vram u have?
oh yea 6gb its gonna be rough 😔
Oooh oooh stable diffusion 3 boutta be a game changer
I know I've said shit about sd but I've always had hope for it
may actually be peak who knows
we were told there's models of all sizes for "everyone" so.....
hmm interesting
but cant be so sure until release
TRUST or DOUBT
that's what they claimed, so I guess we can only wait and see
yeah yah
i finally realised what's the issue. There should not be a ^ for the last line
my version works because that's not my last line
like example
the last one should not have ^ and it will work. Tested by adding a ^ at the end of --pin-shared-memory and the cmd closes straight away once opened
do not copy my ckpt dir tho, since it's a symlink folder that i am using
how to do it right 🙂
deym
nice
photo of ginger woman, in space, futuristc space suit, (freckles:0.8) cute face, sci-fi,dystopian, detailed eyes,blue eyes
Style: paparazzi
framing:Head Shot
ligthing:overcast lighting
camera:Canon EOS 5D
tyy for my lunch 
what's that?
a joke
name of this dress?
I'm gonna put this here since ya'll are my typical engagement...
Is it normal for Comfy to have worse performance than Auto on 1.5 models at higher res?
I'm seeing quite a big difference at 1536x1536 generation.
Comfy 1024x1024:
1.5 - 1.35it/s
XL - 1.49it/s
Auto 1024x1024:
1.5 - 1.54it/s
XL - 1.32it/s
Comfy 1536x1536:
1.5 - 3.17s/it
XL - 1.72s/it
Auto 1536x1536:
1.5 - 2.13s/it
XL - 1.83s/it
no, at worst it should be basically the same. if you're using separate environments for them, check to make sure they are using the same torch (and xformers if installed) versions. also make sure the same optimization is being used for both.
just installed comfy, it's like a brand new vanilla install
im no expert but high top dress?
they are separate environments, though...
I don't apply optimizations to either (apart from enabling xformers for Auto)
bing
iirc you have to manually install xformers for comfy. i don't remember for sure though because mine share the same venv
hmm ill take a look
i love bing (dalle3)... it's awesome
yeah @supple raptor looks like comfy doesn't ship with xformers anymore... they claim it is no longer needed for performance. Interesting.
I can try to install it and see what happens
@bitter latch
lots of people claim sdp (which is integrated in torch since like v2.0) is equivalent, which is only partly true in my experience. speed wise it's usually within margin of error, but xformers is more efficient which is likely to matter at high resolutions.
some are saying nightly torch builds might help... ill see
I run torch 2.2.0+cu121 and xformers 0.0.24 fwiw
ok I changed my mind I like it 😄
what python version are you using?
From the little testing I’ve done, xformers and sdp have about the same performance but xformers has better vram management so you don’t run out as fast on larger res which results in a faster gen
3.10.11
any particular reason, or that just is where you landed back at install?
But I have a higher vram gpu so my test results may not apply to lower end cards
If I remember right, torch has issues on newer python versions, pretty sure 3.10 is what auto says to install
It was the latest 3.10.x version when I originally installed and figured if it ain't broke, don't fix it. Torch only recently started to support 3.11.x
been farting around with this for a little while and still have to say, I prefer Auto 😛
random starbucks cup lol
With the exception of having the ability to play with newer model types (cascade,etc) a few days earlier, I’ve never found a need to use comfy. The I face has a cool factor but I find it more time consuming for no real benefit
@bitter latch
wow
gonna try this nightly build
hmm this one looks more like star wars
I get images like this from autismMix randomly. Never tried prompting for it specifically though
ahh finally a use for those stupid MS reward pts
grrrr stupid 'unsafe content detected' wtf?!?!?!?
ever since forge installed xformers from me launching it with it on accident it wont let me launch it without xformers
you need to delete xformers directories
where would those be
i can search the dir
ah that didnt help
oh i think i found it
i don't remember, i used "everything" to find it
hirez fix is doing this weird thing
i normally use img2img for upscaling so im not too worried about it
what res are you doing?
if it's a lot bigger than their trained value, then it's gonna have this problem
ah ty
whats the last 3 lines of ur thing do
ive seen cuda stream in the console but not sure what its for
i think it was the miltudiffusion thing making the tiles
tiled diffusion doesnt cause that on auto111 i guess its different here
huh, turned on FreeU ext and it's making better images
hmm suppose to be on moon looking down
FreeU off
how do i make super high rez work on forge
img2img upscaling dipped into my shared memory and got super slow
tiled diffusion seemed to be the answer for a111
wont let me enable it on forge
froge
frog
i guess the built in multidiffusion thing is basically the tiled vae diffusion extension
this upscaled without going all into the shared ram and only took 5 minutes
i think you can disable that in nvidia control panel
seems the tiled diffusion only produces weird results with hirez fix
i looked for that but couldnt find it ><
i dont have the new nvidia control panel yet
ohh maybe i missed that
im curious what the neveroom vs multidiffusion settings are
i do have that fallback policy setting

basically i just wanna replicate the auto111 ability to upscale as much as u want by tiling queue
neveroom possibly means to just not oom, however with multidiffusion, you will need to do some settings for best speed vs best res
i think that setting above basically does it
so you might oom in multidiffusion
auto111 multidiffusion tile size is like i think a multiple so i needa figure out that
i can't speak for neveroom extension, since i never really use that
i think its like 96 by default instead of 768
blender cycles recommends tile size of i think 512 or 256
Let me dream
Can i have the model name or lora or whatever pls really digging these images
1girl, frog girl,
source_anime
Negative prompt: 3d, censored, abdominal muscle, artistic glow
Steps: 24, Sampler: Euler a, CFG scale: 7, Seed: 3273378199, Size: 768x1024, Model hash: ac006fdd7e, Model: autismmixSDXL_autismmixConfetti, VAE hash: 66f78ad136, VAE: sdxl_vae-fp16fix-blessed.safetensors, Denoising strength: 0.4, Clip skip: 2, ADetailer model: face_yolov8n.pt, ADetailer confidence: 0.3, ADetailer dilate erode: 4, ADetailer mask blur: 32, ADetailer denoising strength: 0.4, ADetailer inpaint only masked: True, ADetailer inpaint padding: 32, ADetailer version: 24.1.2, Hires upscale: 1.5, Hires steps: 10, Hires upscaler: 4x_fatal_Anime_500000_G, Mask blur: 32, Inpaint area: Only masked, Masked area padding: 32, Version: f0.0.16v1.8.0rc-latest-238-g437c3489```
ok thx
Stupid frog
lmao

thoughts on the 'bad dream' or unrealistic dream negative ti
Never heard of that embedding
i use bad dream a lot cuz the picture on the model page looked good lol
Bad Dream + Unrealistic Dream (Negative Embeddings, make sure to grab BOTH) Do you like what I do? Consider supporting me on Patreon 🅿️ or feel free...
Fast Negative Embedding Do you like what I do? Consider supporting me on Patreon 🅿️ or feel free to buy me a coffee ☕ Token mix of my usual negative...
this is the other ti i use on almost evreything
how is the SVD thingy work in froge
i almost don't use SD1.5 now, and am too lazy to test it, so unsure. But I always believe simple negative is best negative
you need to download the SVD model to use it
simplenegative is the name of the ti or just like a simple prompt
ok i might test it later
got this to go in 3 minutes and didnt dip into shared ram yay
no, i meant like keeping your negative prompt very simple
What does that gif mean

Extremely simple negative
Yamraj with a Twitter handle
i dont use negatives except maybe something like 'nsfw'

almost perfect hands

just like me
just like us

dress name?
woooo thats great
I will find png info for you later
thanks
jacket, crop-top, pleated skirt, off-shoulder, thigh highs, sneakers
anyone else have images gen like this with certain prompts?
adding or removing a word from the prompt, or even changing the order of things seems to be fine, but specific combos with generate like this no matter what seed
I have similar problem on forge when using Autismmix
this one in particular is autismMix, but on auto1111
I also have the same problem with my SugoiXL model randomly, which isn't pony based, so it's not that
I think there is extension that can fix it but heard it affects quality so haven't checked it
i'm wondering if it's a bug with the text encoder or something, which would affect auto and forge

or if it's the models
@supple raptor Do you know the extension that can fix this?
yup
ofc
so many time
Does it repeat that if you change seed
yeah
animagine and autism models
it's different, but same type of thing
What prompt is in that one
like, it's trying to do something but has no idea what to do
2girls, petite, black hair, short hair, purple eyes, choker with (bell:0.6), bear onesie, happy, blush,
cant say its coming randomly for me
actually, that's not the whole thing
marota, eba, score_9, score_6_up, score_5_up, score_4_up, source_anime, 2girls, petite, black hair, short hair, purple eyes, choker with (bell:0.6), bear onesie, happy, blush,
I can fix that with entering more or erasing some
Put bell without weight
i forgot i had a preset adding the pony stuff
yeah i said that
Pony often breaks for me with low weight on something
Might be the cause even tho you're using a merge

ok, but just removing 'blush' makes it work
the weight isn't the problem (although it fixes it)
basically changing the prompt at all can fix it, weights, adding words, removing words, etc
Well from what I tested the weight below 1 causes lots of weird things
It doesn't break 100% of the time but on default pony it seems useless to do low weight
added 'looking at viewer' to the end and it works fine, even with the weight at 0.6
under weighting the bell changes the size of it, otherwise it's massive
more cowbell..
Cow dragon
Here's a writeup on why it happens. https://civitai.com/articles/4044
in short:
We should rename it "A⑨⑨⑨⑨'s WebUI"
actually close
I just said what she was wearing
full pnginfo
1girl, red eyes, big eyes, ponytail, hair extensions, midriff, skirt, thighhighs, sneaker, hair ornament, stud earrings, facial tattoo with heart, off_shoulder jacket, graphic shirt,
source_anime, (bee \(deadflow\):0.6),
Negative prompt: 3d, censored, abdominal muscle, artistic glow
Steps: 24, Sampler: Euler a, CFG scale: 7, Seed: 1522538317, Size: 768x1024, Model hash: ac006fdd7e, Model: autismmixSDXL_autismmixConfetti, VAE hash: 66f78ad136, VAE: sdxl_vae-fp16fix-blessed.safetensors, Denoising strength: 0.4, Clip skip: 2, ADetailer model: face_yolov8n.pt, ADetailer confidence: 0.3, ADetailer dilate erode: 4, ADetailer mask blur: 32, ADetailer denoising strength: 0.4, ADetailer inpaint only masked: True, ADetailer inpaint padding: 32, ADetailer version: 24.1.2, Hires upscale: 1.5, Hires steps: 10, Hires upscaler: 4x_fatal_Anime_500000_G, Mask blur: 32, Inpaint area: Only masked, Masked area padding: 32, Version: f0.0.16v1.8.0rc-latest-238-g437c3489```

@dull jackal Here's gift for you
Haha
i thought the gift was for @wooden coral
it's for @finite herald
ty for the lunch
what model?
autismmixSDXL_autismmixConfetti
I'm assuming it's the same model for this?
yes
what model?
kakigori?
AutismMix_Pony
its a W
Another ping, a ping...what fun did I miss?
free gift of 50$
Free!? Well, by all means, if it's free, it's CLEARLY GOOD, RIGHT!?
Yes ofc it's good but for some reason Fruit didn't want us to have it

Well, I mean, it was only $50! Psssh! Not free enough! We deserve better, on account of our obvious intellectual and aesthetic tastes.
same thing happens to me on forge,it happens on 1.4 , 1.5 and sdxl idk whats the cause but it fixes after i put an img on png info tab and click on "send to txt2img"
The issue is "math" https://civitai.com/articles/4044
Here's the extension to fix it https://github.com/hako-mikan/sd-webui-prevent-artifact
Apparently a fix is already on the dev branch of a1111 and will be in 1.8.0
I’ll just wait until then or update to the dev branch. It’s not a big deal usually
lick

The extension is an easy temp fix while we wait. Who knows how long it'll be until 1.8 is pushed to main.
Hello. Do we make images here ?
There's no bot here. Most of us here uses our own local machine to generate stuff
Oh cool
one of the most popular option is this https://github.com/AUTOMATIC1111/stable-diffusion-webui








