#🏙│ai-images
1 messages · Page 2 of 1
it do be burning my meaty heart... or something like that 
did u use a lora?
nope this is early days of my AI/SD journey, 30th July
thats a whiiiiile back
i know nothing about this lol
i also have no idea what i'm doing
Danny helped me set it up and now i'm just winging it
Felt also helped me
green box is for what i want, red box is what i don't want
i use automatic1111 webui
yeah, i have that too
yeah ur negative prompt basically
i just wanna leard to use this
yes
dont think it's any better tbbh
just wanted to show the differemce, this is my workflow nowadays 
install controlnet on the webui and u should be good
also i can customize it more i think
that's scary
how do u even read this. how is it any optimal from the webui?
wtf are you even making??
meth
create this
damn..
and turn it into this
holy shit
a detail enchancer lora would help
I used one actually, more-detail-wow-art something something
and Faetastic also a little bit
yup thats why u got a much cleaner output later on
but all that work....
webui seems more chiil
chil
chill
sorry my dyslexia kicked in
of course it's not for everybody, but the big work is over now, the WF is finished and now I can create what I want 😉
PORN
hehe im so silly
tf is even that

ComfyUI Workflow, left one is for creation, the right one is for taking an image and divide it into background, body, face and detail it seperately
i will be picking your brain someday for tips
ohh it essentially seperates stuff to get a cleaner output instead of yeating everything together
can I friend* you? @shadow bane
there is def some porn to watch for the plot ngl
I didn't make the left one from scratch btw, it's a huge setup from civtAi that I customized, the right one is basically made 90% by me got some help for the masks(finding person,face,etc.) since I never used that before 

can def see how it would help now
has ultimate upscale and aDetailer built in, as well as img2img, inpaint, controlnets, loras etc. all is possible and you can mix and match however your want
👻
okay hey i wanna ask for a little tip
i like what it's making, but now i wanna make it better
what do i change for the quality to get better?
how far?
more detail? more resolution/sharper? what you want 
@shadow bane has reached level 9.
YES
xD
HF: https://huggingface.co/OedoSoldier/detail-tweaker-lora This is a LoRA for enhancing/diminishing detail while keeping the overall style/characte...

it works with sd 1.5
ultimate upscaler does all of that in one node!
what is that?
i'm downloading it
do you have a link?
have u used a lora b4?
yes
ok set it to 0.8 or 0.6
do you have git?
get this into your custom_nodes folder, with that you can install almost all available ComfyUI tools automatically
u can even try 1 but sometimes it does too much
Yall will need the ultrasharp upscaler
If you want sharper images
yeah how do u get that?
that's a lora?
ok and where do i yeet it cuz im on paperspace not local lol
yes they will need that too for the ultimateupscaler
For SD
im completely in the dark now
Exactly
... lol
i'm gonna start with this and the settings i've changed and see what happens
yup play around with it but i can guarantee good results
You know those upscalers you can select? Latent, latent (nearest), 4x esgran etc etc
its an amzing lora
it adds more detail in the first step, but the first step can only do so much to add detail without losing quality
These are just better upscalers than those
which folder do i yeet it in?
uhmm no not really
I cant check atm
that's why I put the detail lora on a low strength when creating the image and then high strength wen upscaling with ultimateupscaler
hm i am not using comfyui but u gave me the idea
what should i set these settings to?
should be much different
set strength ot 0.8
sdwebui\webui\models\ESRGAN
and leave clip as is
for A1111
yes i just found it
ait lets see what happens

u got ssome cute emojis give me some
fr Yume has adorable emojis
X3
should i donwload all the upscallers 😋
leak the server u are in NOW
i'm going to add that upsccaller later
i just
it looks complicated
oh
THIS IS STARTING TO LOOK SO CUTE
ohh the normal one. got it
oaky okay i will
Its my personal server
it's fine we can take you step by step
you would do that for me?
i asked for this
what have we been doing this whole time here now 
so cute

lemme see how this turns out first
working on this atm(see left), was created and upscaled with the ultimate workflow and now I'm ringing it into the detailer Workflow that I made(see right)
Not yet there...
Maybe ill allow some more people into my personal server
But its for very close friends, so please be kind to the others in there if you end up joining when invited
@shadow bane i was wondering how i would wire the upscaler thing to my setup?

plop the end result image and all the other stuff the ksampler gets into it, then add the upscale model loader... done
then it's the same as the ksampler
of course with more values, but you can use the default ones, or mine for now
that's all fine tweaking
ngl i don't understand.. 
upscale to whatever you want, 1.0 is basically sharpen and add detail
bigger than 1.0 is upscaling and adding detail
all the fields that have the same name as the ksampler you can clone the ksampler
denoise will decide how much of the image is changed when upscaling/sharpening, 1.0 means the image will be changed completely and 0.0 will not change it at all
the rest you can just copy from me or leave the default values there
hmm
Nice, i use controlnet in addition to this
yes you can even use controlnet to make it "better", but we do that later when crzlm understands the first few steps 😄
plop the end result image and all the other stuff the ksampler gets into it, then add the upscale model loader... done
then it's the same as the ksampler
^3^
okay i'm trying to dumb this down
exactly

like this?
that's all you need to add to the ultimate SD upscaler, the rest is just as the ksampler node
i'm too dumb for this
never saw that before xD normally the upscale happens after the ksampler is finished, or you feed it an already finished image



Lol
yeah
as i said
i have no idea what i'm doing
i feel like a child
looking at a screen with boxes of text i do not understand and why are there wires
xD
this is how I do it:
oh xD
it takes time to learn n master
put the upscale between ksampler and vae decode
yes, now just need to make it bigger, or will the the same as your old image
2.0 or 1.5 are the default for many
so 1024, or 768
pixels
yis
no I meant either double or 1.5x your beginning values
so that's it?
i can run it?
yep, let's gooo

i did an oopsie

sorry never used latent upscale before, so I don't know the reason x)
you should have the load upscale model too

i'm gonna try something
wait
no this is fun
i think i'm starting to understand
AAAAAAAAAA
how do i make it make the same image tho
it keeps* coming out differently
seed controls the randomness, control_after_generate controls what happens to the number after you hit the queue
btw sd creator now give perms?
OH IT'S AFTER
okay okay
no that was manually done 
No, trusted role
do you know if that role is for anyone who wants to make SD or if it's for PROs ?
Also who was it all that wanted to join my server for the emotes and stickers? XD
Litsa gives it to people that post frequently and post high quality results
So it isnt just given to everybody
ah how can i get it ?
Be active in the community basically
ok ok
Uh oh
Btw pixie, did you want to join the server for the emotes etc?
is there a better setting here i can click on to get better results? the images come out so pixelated
Invite sent
try it
we use upscale models so upscale image (with model)
so then you have more options
all come out the same
i might need to try that
@shadow bane so that would be like a LORA?
ehh I guess 😄
did you download some?
nope
put it into the upscaler_models folder and refresh your browser
ComfyUI_windows_portable\ComfyUI\models\upscale_models
done
don't i need to connect something?
yeah of course, take the image from the vae decode and plop it and the load upscale model node into it
Khazar, you are also welcome to join the srver for emotes and stickers if you want x'3
what
load VAE?
load upscale model
lol kinda floped
throw upscale image away and use the new ones, the little names tell you where goes what 😉
i'm so sorry... i feel so dumb
you still need the upscale image (with model) 
like that?
okokokokok
OMG
yesss
omg that's awesome
congratz!
thank you friend!
you tryna to be gangster? 
@shadow bane has reached level 10.
Looks amazing!
eh sure, sorry, doing 3 things at the same time xD
No worries

@rustic raft if you only want to sharpen, then I can recommend this:
https://openmodeldb.info/models/1x-ArtClarity
pretty happy with it
hey if I wanna take an old image, would i be able to feed that back into SD and then get the SEED info etc. ?
you can take any .png image you saved with comfyUI and just plop it into empty space of the workflow
then you will have the whole workflow and values you had at that point
omfg thank you
that's awesome
i'm pinning that
you can take any .png image you saved with comfyUI and just plop it into empty space of the workflow
then you will have the whole workflow and values you had at that point
i made it one message to be able to pin it for others
I can even steal it 
lmao
that's the cool thing about ComfyUI makes WF sharing super fast and makes the community make better images at a rapid pace
aww that's adorable
yeah
I KNOW
why do they look so stoned???
its melting
red eye syndrome 😄
@rustic raft slapped the ultimate SD upscaler into your WF:
sadly dreaming faces can happen every now and then, especially with Loras
I hadnt seen that other face before😭
@rustic raft left 0.3 denoise, right 0.5 denoise, I said it wrong before, denoise 1.0 means basically a whole new image, 0.0 means it changes nothing
wtf howwww??
install git:
https://git-scm.com/download/win
then go into ComfyUI_windows_portable\ComfyUI\custom_nodes and open a cmd window:
git clone https://github.com/ssitu/ComfyUI_UltimateSDUpscale --recursive
ComfyUI nodes for the Ultimate Stable Diffusion Upscale script by Coyote-A. - GitHub - ssitu/ComfyUI_UltimateSDUpscale: ComfyUI nodes for the Ultimate Stable Diffusion Upscale script by Coyote-A.
@rustic raft ^
is this git?
if you are in ComfyUI_windows_portable\ComfyUI\custom_nodes you can do this:
wrong folder, but yes like that 😄
how do i set it to a different folder?
@rustic raft go in there and write cmd into the folder tree
then do the same thing again
or cd and paste the path to the folder
example
cd C:\Users\Admin\Desktop\stable-diffusion-webui-directml-master\render
for beginners it's more safe to go in the folder in explorer so they know exactly where they are, how they got there and how it looks like
ye maybe i always used cd
ok write into the field "cmd", just like my screenshot
and hit enter of course
can you show me the folder?
can you find the file it says it can't find?
ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI_UltimateSDUpscale\repositories\ultimate_sd_upscale\scripts
the \scripts is missing
Ouo
i'm only able to go this far..
ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI_UltimateSDUpscale\repositories\ultimate_sd_upscale
I think that's because you're missing the --recursive
git clone https://github.com/ssitu/ComfyUI_UltimateSDUpscale --recursive
ComfyUI nodes for the Ultimate Stable Diffusion Upscale script by Coyote-A. - GitHub - ssitu/ComfyUI_UltimateSDUpscale: ComfyUI nodes for the Ultimate Stable Diffusion Upscale script by Coyote-A.
this is the right command
what this --recursive do ?
ah so it is to perform recursive operations on nested repositories inside the main repository
i think i meed to do this manually
can't i just download the folder from the git page?
and paste it?
wait
it's way too nested, can't you put it a little more lower? your ComfyUI is in 3 folders... 😵💫
ohhhh
yeah i can
wait
C:\Users\Notandi\Dropbox\PC\Documents\Stable Diffusion\ComfyUI_windows_portable_nvidia_cu118_or_cpu
that's where I keep it
this one is just empty right? ComfyUI_windows_portable_nvidia_cu118_or_cpu
besides another folder
no
oh
yeah
i might be misunderstanding
it's so hard to do this xD when just in writing
i hope i'm not making you frustrated
I have it pretty straightforward
C:\ComfyUI_windows_portable\ComfyUI
too long folder structure can make problems with some programs, so it's best to avoid that as much as possible
okay i can do that
all good, as long as your brain doesn't overheat 


you need to change that back from my folder structure 😄
just click through the model and loras
ahh
we got green
oh
and the eyes thing you did with the face?
that was a bit too fem for my taste but what is that thing you did differently?
was that another complicated thing you had or just a lora?
the thing that made it into
just stronger denoise
because that cleaned everything up
ahh
it looks so much better
is it complicated to set up?
you can use the doubled text boxes I used and change the prompt to more your liking, you can use different prompts for ksampler and upscaler
what do you mean?
so this is what i have now
i'm just missing the final piece
that makes it magic
OH
WAIT
@rustic raft just take the right image and use that, the tiling is set to 1024 to stop the face dreaming and denoise set to 0.5
I didn't do anything more than that 😉
this means how big the tile is that it will upscale step by step
with 512 it will go through your image 4 times to upscale and each tile has the chance to dream a face since it can't see the character
which box is that?
ahh
if you keep tile at 1024, probably a good idea to up the steps to 30 or 40, since that will make sure all details are finished, 20 might be a bit low for that many pixels
I'm trying to downscale the feminity a little with a new negative prompt box:
(eyelashes:1.3), blushed cheeks,
I'm just saving the ksampler bit, it was always there, just not getting saved anymore like your old WF
hmmmmm
i kinda wanna make you a lora of your character once you got enough data x'3
pixie i mean
you know
omg i would love that
the character
how much art pieces you have?
i'll show you what i have later
a few, but i don't really like how they look
i need to work on it a bit more
to give you some idea
@quaint tiger has reached level 20.
tried androgynous
LETS GOOOOO
androgynous for both ksampler and upscaler
i think i'll figure out how to perfect it, so far this looks hella good!!
whats the prompt for this?
the last one is awesome
i wanna try something :3
i love the horns
take it, put it into your comfyUI and work from there 😉
ty!!

if @rustic raft wants to, they can send you the prompt now 
@shadow bane has reached level 11.
hehe

just don't butcher my OC ❤️
obviously i wont purposely butcher it
just wanna try something
:3
purposely...
thats alot of text

it's the whole metadata of the image
comfyui stuffs that in there by default
😭
æoadrunaæprvhnaærgjhædrfja
it just turned out like that
that's fucking hilarious
ima try some more stuff
please do xD
like studio ghibli style
aaaaaaa
oh
i never used it
only wildcards
and there i also have the templates
i use wildcards too
that's fucking wild
can i see the setup you had to add the change?
like a ss of your Comfy?
I need to use controlnet, my models change it too much 😄
ahh isee
use img2img and Denoising strength 0,4
i know lol
i have done this longer than today XD
but thanks for the suggestion
but im specificly trying to get different styles
so what did you do that it looks like this?
that setting would just cut off the style change :)
different base models
wait lemme try to make similar thing
i liek this
i liek this
it actually has like pencil lines
@rustic raft oh btw, if you want the Loras to affect the upscale too, then just use the model line from the Lora loader instead of the checkpoint loader
I totally missed that and just connected the checkpoint loader with it's model 😄
what will that change?
the loras will affect the upscale too
so more true to the comic lora, add more detail and keep the orc more goblin
with loras connected
with loras disconnected
loras connected upscaled
i don't really know how i feel about it
i like this a lot more bc of the thin lines
yo
@shadow bane could i make the lines thinner by adding that to the negative prompt?
:3
sure, you can to some degree alter the style, try with both positive and negative
finally got controlnet working, had some broken cn models for some reason O.o
if you send me your favorite images of your character, ill try training a lora :p
could I send it later?
sure
i love the flowers and the colors
this shit is surreal
she now looks like a game character for some reason
this is what i've meant for them to look like
@quaint tiger do you use controlnet?
that's them
mind if i try training these images with my furry specialized SD first?
i kind of wanna see that results it makes

alright
first im training using just blip captioning
later when i have more time ill describe the images by hand
^w^


I'm trying to replicate it as close as possible, not too bad
Added controlnet support to your WF
https://civitai.com/models/38784?modelVersionId=44879
I'm going grocery shopping now, cya later peeps 
take care khazar!
omfg
? :3
THAT'S FUCKING AMAZING
holy shit
literally
could ia sk you to render one for me with a special scale?
special scale?
could you render one in this scale?
- one that's a 0.5% wider?
if that's not too much?
you know what i mean
i'm loosing it
loosing
hmmm
i probs cant directly generate in that scale
however
i can generate lower and use my specific upscaling method
yes
that looks SO GOOD
holy shittt
can i import this shit into SD and work with it?
lemme generate that bigger one for you
i can give you the lora
but you would need to download the yiffymix base model too
as its trained on that model
i still cant get over the difference
i will download anything to make that shit
why does it look so familiar
go to civitai and download yiffymix v34
which one is that? i see 4x of them
@rustic raft has reached level 19.
i trained it on the safetysensors 5gb model
the one above VAE
downloading
^^
when we done fucking around with this you gotta make that ai-chan LORA
fr
ima create a few previews for you background
and then u can choose

i'm fucking amazed
nljfsdjlknfsdjlknfsd
nah you're the real fucking deal
seen a specific one you'd really like yet?
i'm shitting myself here
all of these are so fucking good
i'm just trynna figure out what'll fit my screen the best cuz i love all of them
? :3
THIIISSS
alsooo how did you end up making it work so well with the water scene? that wasn't even in the original prompt?
and the arm hand on the left
((SOLO:1.2)), (PixieDis), (full body), (detailed face), (((looking at viewer))) (standing in a lake), (Moon Shine), (bright stars), (falling Stars), blue skin, clear eyes, (POV high front angle), curled horns, androgynous,
(POV tilted left low angle),
[by personalami], by smitty g, [[[by Foxovh]]], [[by Ross Tran]]
break
(shaded), (detailed lighting), (cinematic lighting), sharp focus, [detailed ambient light], [gray natural lighting], ((detailed background)), (long shot:1.4)
hi res, ((raw photo, masterpiece)), nature, (high quality detailed)
lora:add_detail:0.3 <lora:PixieDis V1-000005:1>
thats my prompt
:3
also
please don't share the pixie lora on here, i kinda don't want people making nsfw images with it.. since it's my OC
what is that?
after seeing these pictures i'm 100% going to be making a VRchat Avatar
I fucking need to learn how to make that shit
x3
the face is perfect on these two
atleast you see what i mean
like.. that's the face
Recognizes the image in different ways you can try them one by one and see it in the preview (you can make the preview bigger too)
If you change the preprocessor then you also need to change the controlnet model
If you like the preview then you can download the corresponding model from the CivtAI page I posted
i fucking love you
background has been updated
THIS IS SO FUCKING COOL
thank you thank you thank you thank you
all'dat sounds complicated and brain melty

It will make more sense when you see the preview, it will be the aha moment 
i see
yeah looks very trickster like 
let me know if anything is unclear 😉
haha yeah i don't even know, it's my OC
i don't really know what they are
but they're cute
i'll ping you ❤️
literally its so awesome
Did u make a lora out of squishy lol
aye we had a whole journey with a Lora as a result 
Dayum
5 whole hours lol
Efficient work u guys
yeah in the end a lora of a character will always be superior to trial and error x)
Yup so moral of the story make ur own Lora
how the hell do i fix my image getting deep fried on the last bit of steps
if you tell me more we can figure it out
i don't have image perms so i cant send examples
what tool, what process etc.?
use krakenfiles
probably the model you are using needs a vae i had same problem before
but i have the vae that the model needs
we can guess a hundred times really x)
did you setup it correctly?
show of more of the process and what you're trying to do
yes i followed the tutorial/documentation throughly
@left carbon has reached level 12.
ah so no idea
lemme see if this works
that grim face 
hey can someone tell me what this errorcode means?
Error occurred when executing CheckpointLoaderSimple:
'model.diffusion_model.input_blocks.0.0.weight'
File "C:\Stable Diffusion\ComfyUI_windows_portable_nvidia_cu118_or_cpu\ComfyUI_windows_portable\ComfyUI\execution.py", line 152, in recursive_execute
output_data, output_ui = get_output_data(obj, input_data_all)
File "C:\Stable Diffusion\ComfyUI_windows_portable_nvidia_cu118_or_cpu\ComfyUI_windows_portable\ComfyUI\execution.py", line 82, in get_output_data
return_values = map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True)
File "C:\Stable Diffusion\ComfyUI_windows_portable_nvidia_cu118_or_cpu\ComfyUI_windows_portable\ComfyUI\execution.py", line 75, in map_node_over_list
results.append(getattr(obj, func)(**slice_dict(input_data_all, i)))
File "C:\Stable Diffusion\ComfyUI_windows_portable_nvidia_cu118_or_cpu\ComfyUI_windows_portable\ComfyUI\nodes.py", line 476, in load_checkpoint
out = comfy.sd.load_checkpoint_guess_config(ckpt_path, output_vae=True, output_clip=True, embedding_directory=folder_paths.get_folder_paths("embeddings"))
File "C:\Stable Diffusion\ComfyUI_windows_portable_nvidia_cu118_or_cpu\ComfyUI_windows_portable\ComfyUI\comfy\sd.py", line 410, in load_checkpoint_guess_config
model_config = model_detection.model_config_from_unet(sd, "model.diffusion_model.", fp16)
File "C:\Stable Diffusion\ComfyUI_windows_portable_nvidia_cu118_or_cpu\ComfyUI_windows_portable\ComfyUI\comfy\model_detection.py", line 120, in model_config_from_unet
unet_config = detect_unet_config(state_dict, unet_key_prefix, use_fp16)
File "C:\Stable Diffusion\ComfyUI_windows_portable_nvidia_cu118_or_cpu\ComfyUI_windows_portable\ComfyUI\comfy\model_detection.py", line 36, in detect_unet_config
model_channels = state_dict['{}input_blocks.0.0.weight'.format(key_prefix)].shape[0]```
here's an image if that's somehow easier to understand
this confuses me
idk anything about comfy so i have no idea how i can help u
try to use this checkpoint in automatic1111
oh lol i just saw what's wrong

oh i see it now blueprints shit is stupid
wait can you help me with this?
i'm feeling silly
put the VAE in the VAE folder, then make VAE loader node
you can btw double mouse click on an empty space and search for a node
you don't have to manually go through the tree
i don't even know what i'm looking for
look i just threw this image into the thing and i got no clue what i'm doin wrong
it's not the LORA's or the Checkpoints
it's something else that's wrong
comfyui tells you what node has a problem, look at the purple highlighted node
look like guy from avatar movie
pals just blue
yeah but like, this doesn't explain anything
when i drag in the image
naah i still see it
it does that
you need:
I told you 
😭
beginning steps beginning steps, happens to all of us 
something is very very fucking wrong
this image was made with SD, but the new one was made with Comfy..
dud wanted more earrings
lkjdfskdflkjdfg
WHOA
help
what the actual
i dont use comfy ;-;
@shadow bane
i use a1111

i knowwwwww i'm trying to work with the LORA in a1111
add negative prompt
AFAIK, comfy doesn't use:
break
[
and is it normal that the Pixi lora is loaded 2 times?
badhandv4, deformityv6, boring_e621_v4 are embeddings/Textual Inversions(or short TI), did you download them?
@rustic raft
https://civitai.com/models/16993?modelVersionId=20068
https://civitai.com/models/16807?modelVersionId=53718
https://civitai.com/models/87781?modelVersionId=94126
介绍(中文) 基本信息 此文本嵌入为负面文本嵌入。它能够在对画风影响较小的前提下改善AI生成图片的手部细节。如果它让你的模型表现得比以前更糟,请勿使用它。您可与其他负面文本嵌入一同使用。 如果你想使用效果更强的版本,请移步:NegativeEmbedding - AnimeIllustDiff...
介绍(中文) 基本信息 该页面为推荐用于 AnimeIllustDiffusion [1] 模型的所有文本嵌入(Embedding)。您可以从版本描述查看该文本嵌入的信息。 使用方法 您应该将下载到的负面文本嵌入文件置入您 stable diffusion 目录下的 embeddings 文件...
put them into the embeddings folder
then write them again in ComfyUI after !REFRESH! it will fill it out automatically for you
i have now downloaded them and will be refreshing
woah
works
it works!!

I would recommend https://civitai.com/models/56519/negativehand-negative-embedding
instead of Badhandv4
I just get better overall results more of the time than with badhandv4
I just posted the things that were already in the prompt they are trying to replicate 😉
Ah, ok
from SD to comfyUI some things are different x)
@rustic raft do the negative TIs now have embedding:: in front of them?
booba 
idk
i have boobs as negative XD
like this:
embedding:FastNegativeV2,
but, I kinda like the first results better for some reason.. it's the same seed tho. one is with the embeddings, the other isin't
i just feel like the face got ruined
here is the real negative prompt, the TIs are not doing anything for you atm
embedding:badhandv4, embedding:deformityv6, bwu, embedding:boring_e621_v4, (multiple tails:1.5), female, pussy, (pawpads ontop of hand), ugly face, watermark,
i've updated the neg.p.
now it's loading
here is the ComfyUI fixed positive:
((SOLO:1.2)), (PixieDis), (full body), (detailed face), (((looking at viewer))) (standing in a lake), (Moon Shine), (bright stars), (falling Stars), blue skin, clear eyes, (POV high front angle), curled horns, androgynous,
(POV tilted left low angle), aheago,
(by personalami), by smitty g, (((by Foxovh))), ((by Ross Tran)),
(shaded), (detailed lighting), (cinematic lighting), sharp focus, (detailed ambient light), (gray natural lighting), ((detailed background)), (long shot:1.3)
hi res, ((raw photo, masterpiece)), nature, (high quality detailed)
sadly break and [] are not recognized by it
also need to be careful with weights: () and :1.3 etc. in SD/A1111 they are getting balanced, in ComfyUI it takes them as is and it might be too much, so if the image starts looking broken or overblown, then that could be a reason
ohh that's a nice shot!
im testing multiple models now
i have now 26 pixie loras
trained at different strengths

26?
ýes
AAAAAAAAAAAA my boyyyy
PLEASE UPSCALE THIS
tf happend here
@rustic raft how did yours turn out, losing your mind trickster 😄
OIJUSDFLIJHFSDLKJDFS
looks pretty good expect for some yellow melting 
... i just changed the text.. and well.. it didn't turn out as I thought it would
LOL
ik this is very eye catching
Crash Bandi- entered the chat
ahh the furry came through 
lmfao*
IT'S CASH BANOOKA!!!
im idiot i didnt run embeddings
fr
i don't know why but the old Prompt text worked better
which one?
so what eye color has ur character?
bright blue or brown yellowish.. idk? maybe green heh.. whatever looks good i guess
i'll respond in a sec, KSampler is loading
this worked a lot better
when i did the new one, it broke it and made them a furry

yeah sometimes it be like that 😵💫
stop
:3
no Neko-Arc-chan, don't do it, you're PG!!!

please shitpost somewhere else @sleek sequoia
I don't even have image perms
What is PG
Wait a minute I thought that was verified chat
i can't get the face to look right anymore 
I mean you don't even have any hands in that image, try taking out the positive hand Lora and negative badhands
@shadow bane has reached level 12.
OH
there are hands..
it's just
not safe
kjngfdkjnldgf
it's cropped
ait, i'll fuck around
we'll see what happens
ttyl
im trying to fix the hand on it
@rustic raft the point I'm trying to make: everything you add will fight to get it's limelight in the image 😉
8th July, another oldie
love the details and the white ear part makes it more interesting
heeelp😭
what's up?
what did you change?
or no one eye is too big
i add negative prompt xd







