🍥｜anime

[vital raptor]

hmm

the github implies it should work with add difference, but it doesnt

there's a "fixed bug" from a while back making this work, so it's intended too

[supple raptor]

technically there's no reason why it couldn't work. there's just not really a logical reason to use it that way that I can think of.

[vital raptor]

i've seen most people recommending to use add difference with the base SD model as the "C" model, it just reduces overall added weight

1 + 1 - 1 * block weight is different than 1 + 1 * block weight

but, i don't know how much it matters

[supple raptor]

The idea behind add difference merging is to try to sort of extract the difference that was trained into model b from it's base model, particularly when the models a and b are very closely related. Using MBW allows you to overcome the problems that not doing that can cause though so that add difference step is not really necessary in that case IMO.

If I really wanted to do that, then I would most likely extract a lora from model b and merge that into model a instead.

[vital raptor]

that would probably be easier to test and control as well

[supple raptor]

Yep

[wooden coral]

Throw this into SVD

[nova remnant]

a witch brewing potion, dark night, full moon, mystical.

a lot of the sd1.5 model would miss the witch unless you prompt for 1girl

[native halo]

skill issue

[terse ore]

Any manga models for SDXL? There's LineAniRedmond but I don't know of any other.

[autumn violet]

Pretty cool style, I like it. Is it a Lora or model?

Nice

[steady grail]

just found out that seems dshrink is not working with pipe ksampler

[steady grail]

same result. i was wrong

[nova remnant]

That is no tight suit gen

What are you doing

[native halo]

no more latex 😔

[nova remnant]

[steady grail]

wow, beautiful

how did you do this?

[brittle rampart]

I just use the prompt line.

[zinc flax]

Cuute

[nova remnant]

Grabby horns

[brittle jay]

https://media.discordapp.net/attachments/1019446913268973689/1086767816662843442/bad_gen.gif

[bitter latch]

haha

The power rangers

The gay ranger ,hmmm?

And the villan, The Jester!

ok off to work be back in like umm 5

hrs

[steady grail]

charming

[toxic dragon]

Its been a while since I posted but I've been making a lot of progress.

Tried to get the same pose for 1&3

[supple raptor]

[brittle jay]

@wispy canopy can you try out stable zero123 on your a100 and show results, pretty pls lol

[wispy canopy]

unfortunately, access to those doesn't work that freely 😦

[brittle jay]

lol

when you get access

let's say your access to 24GB+ of VRAM is more accessible than mine

[steady grail]

during the next full moon, after a sacrifice in blood

[brittle jay]

shaikorth be prepared for the sacrifice

https://tenor.com/view/demoman-reference-demoman-tf2-dabois-gif-26377605

[steady grail]

😱

i don't have mailbox, i only communicate with pigeons

even in this exact moment, i'm sending a pigeon to someone with a phone, to comunicate with you

[bitter latch]

not sure where the clock came from

spicey

eh those eyes. wtf

ooh

[supple raptor]

https://en.wikipedia.org/wiki/Project_Pigeon

[steady grail]

i'm utterly fuc*ed.

anyway, i know the truth https://en.wikipedia.org/wiki/Birds_Aren't_Real

[autumn violet]

It's true

[supple raptor]

Thank you, that one required a bit of editing to fix some problems in the original gen.

[stable crane]

nice

[terse ore]

I used controlnet openpose SDXL and it works but the face and body and hand isn't as good, it sort of breaks up

For example, I input this as the controlnet

then it turns into this

[vital raptor]

seems about right

looks like a bit of artifacting in the image, so it's not rendering very good to begin with, probably need to lower your CFG or use more steps (or both), might get a better result

but even then, hands and faces are pretty hit or miss with controlnet and her left foot is probably more of a depth problem, complicated poses are hard to get good

[terse ore]

I see. I tried MultiControlnet but got an error, but didn't try too often. Do you know if SDXL MultiControlnet works well?

[vital raptor]

i havent done any controlnet with sdxl yet

I try to make things work with just prompting if possible, if i can't get it to work it's probably too complicated for controlnet to help much tbh

if you have an actual image of that pose it might work better than a wire model, so you could try that

[terse ore]

I see, yes. I could try using depth controlnet or something

[steady grail]

just started experimenting with depth controlnet.
damn, it's like magic

[vital raptor]

I can't seem to do anything with SD these days without ending up writing a bug report for something on github 😒

It's like no one actually uses this stuff or something

[silk narwhal]

what kind of issues are you having?

[vital raptor]

well, I was doing some merging using the supermerger extension and it wouldn't let me save as safetensors because of some contiguous tensor error

there was an open report on that, so i just added to it

in my search for a way to fix it, i found the 'model mixer' extension that actually saves properly, which fixes that issue, but it has it's own bug where the generated image of a merge is not the same as the image generated after the model is saved

and, the saved model, even with 0 alpha on model B, is different from the original but not the same as the preview, so it's saving something it's not supposed to, and also 'live' generating something even more different

[silk narwhal]

sounds like rounding differences of the seeds

[vital raptor]

it kind of seems that way, but i'd think using a fp16 model and saving it as fp16 (not merging it with anything) should not change the values

but that's likely the problem, even if it shouldnt be happening

[silk narwhal]

Merging models can be a pain .. especially when one or more are converted

[vital raptor]

these are the original model on the left and saved model on the right. the changes are fairly small and I probably wouldn't really care since it's a tiny bit of changes

BUT this is the 'preview' merge image before it's actually saved, it's not only different, but the quality is better in many aspects of the image.. so i kinda want it 😄

but yeah, this whole project has been a pain

[silk narwhal]

reason i am messing with comfyUI now .. 😄 can preview every step 😄

[vital raptor]

wtf..

decided to confirm my testing by using a1111 1.6 instead of the 1.7RC and generate the images with a publicly available checkpoint, first image is before installing the extension, second is after

it's altering images without being enabled

[silk narwhal]

same seed?

[vital raptor]

yeah

same everything

[silk narwhal]

https://media.discordapp.net/stickers/1018654300110589963.png?size=160&name=Monika Shrug

any extensions that could be affecting the image .. control net etc

[vital raptor]

nothing enabled, i generated the first image, installed the extension and restarted the UI, then used the original image to reset the parameters and generated a second image

[silk narwhal]

maybe the model/vae/lora/embed some can be really random when not using specific keywords in the prompt even with the same seed

[vital raptor]

it makes the same image every time

the extension changed something

[vital raptor]

hmm, seems the second problem is unrelated to the extension, for some reason the first time i load the counterfeit model it generates the first image, but if i restart the UI then it will only generate the second image, even on a clean install

no idea what's going on with that

[brittle rampart]

m?!

now https://civitai.com/models/4468?modelVersionId=57618 this is. before it was https://civitai.com/models/79083/bluepencil?modelVersionId=107812

you can see all in PNG Info

nope. this is Automatic1111 - V1.6.1

I think the point is not in the tool used.

[silk narwhal]

direct from model:

after processing in my comfy workflow 😄, 6 passes of hand and face detailer 😄

[brittle rampart]

scary and gloomy

[steady grail]

i literally love this model

pretty sure is this https://civitai.com/models/4468?modelVersionId=57618

[brittle rampart]

nope

now it this is model https://civitai.com/models/228564?modelVersionId=257926

[steady grail]

wow, seriously? the style is the same https://civitai.com/images/4570260

[brittle rampart]

but LOra same

[steady grail]

wow

yep, i'm in love with that lora too

[brittle rampart]

yes... this is look very kawaii!

[silk narwhal]

AWPainting ... using Dreamshaper 8 as the model for the face detailer passes

[brittle rampart]

face detailer passe - in Adetailer "Use separate checkpoint"?

[silk narwhal]

am using comfyUI

[silk narwhal]

🙂

[steady grail]

still rocky numbers

[brittle rampart]

wow workflow!

[steady grail]

now i'm about to add depth preprocessor to the workflow and a pose workflow 🤣

[steady grail]

now i just need to see where to place everything XD

[silk narwhal]

[steady grail]

you broke your workflow lol

this is mine now and i still need to add this

[silk narwhal]

still messing with it 😄

[steady grail]

probably i need to move everything again to make to workflow more handy

[silk narwhal]

am just messing with it to get a similar look to a graphic novel i love the artstyle of

[nova remnant]

this looks like a new art style

tho it kinda looks like it has lot of artifacting going on

im having a slight dilemma with sd1.5 and sdxl

anime faces are so cute with sd1.5 but lacks so much coherence and thats opposite of sdxl

i prompted for her to be playing with a basket of chicks but sd1.5 shows no signs of visual engagement

i think dalle3 does that better

[steady grail]

it's not what i was expecting, but the result is pretty charming lol

i love how i could give the image more depth

but this workflow is totlly unpredictable, exactly how i love it XD

[steady grail]

next thing, i need to add concat prompting

[steady grail]

this workflow is a carnival. i generate such mayem with 2 words as prompt XD

[steady grail]

she reminds me of the character of an old game

or maybe an anime

but i don't remember which one

oh

🤣

this workflow makes no sense XD

[stable crane]

I tried sdxl as base then upscale with 1.5 but it didn't work out too well in my comfy workflow, I think I need controlnet upscaling to raise denoise

also sad that sdxl is missing a lot of concepts I frequently use

[valid turret]

Evolution of an artwork when SDXL img2img applied

[supple raptor]

The image has PNG info for the full details. It was a pure txt2img gen. I used a noise offset lora to help make it darker and the dynamic thresholding extension helped as well.

[bitter latch]

miku cake?

haha

[nova remnant]

i cant suggest on comfyui, but with a1111 i usually generate at 512x768 with sdxl then upscale by 2x using hires fix with 0.4 denoise, also as of late i prefer not to use lcm due to its loss of image quality, so basically rendering with dpm++ 2m sde karras, 30 steps and 8 cfg

btw @stable crane in case you meant base as in sdxl base model, unless you are training a model i'd suggest using a trained or merged community model

[stable crane]

no I mean bluepencil or whatnot

[nova remnant]

bluepencil is quite nice

i hope you are not using any refiner when using bluepencil, and if you want to upscale you can either try 1.5x or 2x ... i've noticed when i set the render resolution to 512x768 its much easier for my system to upscale by 2x for hires

[stable crane]

for bluepencil I min 1024x1024

[brittle jay]

https://www.youtube.com/watch?v=6i417F-g37s

wow this video really showed me how good the comfyui workflow can be

[nova remnant]

you would be fine with 512x768 for most XL models then upscale by 2x

and if you dont want to hires fix you could try 768x1024

@stable crane is there a prompt you're working on now which you could share so i could try it on my end too?

[stable crane]

only lewd prompts in NAI 🙂

[nova remnant]

dm me

i want to try that on bluepencil with 512x758 and compare with your res

[brittle jay]

somehow not surprised

[stable crane]

it's really good

[nova remnant]

[nova remnant]

White

[nova remnant]

@native halo black

[native halo]

she loves black

[nova remnant]

black furniture yes

[native halo]

[zinc flax]

The only use for nai v3 lol

[nova remnant]

portrait of a woman, romantic, calm, sensuous, long hair, silver hair

[stable crane]

NAI -> SD upscale

it's good for non-lewds too

[vital raptor]

These remind me of the art style from puzzles I did as a kid

[brittle rampart]

you had some great puzzles!

[vital raptor]

i would always pick out the ones that looked like they'd be the most difficult 😄

[brittle rampart]

great training in perception

[vital raptor]

most likely the vae you're using is fp32 instead of fp16

so it takes more vram, but then it runs out of memory so it converts to fp16 and tries again, which takes longer

dunno, never tried it

in a1111 having something like --no-half-vae in the settings will make it always use fp16 and bypass getting that error

i know comfy has something similar but i dunno what it is off the top of my head

[steady grail]

wow, a tomato sauce lora

[brittle rampart]

no no no..

[steady grail]

remember that excessive gore is restricted on the server

yeah, just a remind

i too don't know how much is too much

sfw discord rules are a little too much broad

in any case, in this server, we have a rule, if it's a red stain, is tomato sauce lol

yes!❤️‍🔥

the first 3 are fantastic, the 3° the light is a little unnatural and the last, pretty average

[brittle rampart]

fantastic eyes!

intresting.... thanks

[brittle rampart]

very heavy hand pose. and a great result!

fingers and hands for me are something from the realm of fanaticism. I can’t get them under any control

[nova remnant]

how to use

[steady grail]

i was trying to instal destiny 2, but damn, it needs over 200gb hd free space to install

[silk narwhal]

freefallin'

[stable crane]

omg

that requires some serious dedication

[steady grail]

i need to free a lot of space

i freed 150 gb from the main hd and i was only half way

so, it's maybe 300 gb?

what a joke, it's the first time i felt the desire to format my 2° hd to delete the partition

[light moat]

Nooooo! Who hurt her?!?

[white nacelle]

who

for me ur message just says the users id

@strong elk can you stop trying to scam people???

426417090442821643

oh yea

[untold glacier]

@dull jackal @barren lintel This guy left the server (well, he did when I posted this), and based on his messages and the name, I'd recommend deleting his message(s).

Also, hope your day is going well!

And here is your sacrificial Miku, Fruit.

[white nacelle]

sorry about that, i left for a sec cuz the server seemed to glitch out, like no channels were displaying

but i promise im not a scammer or something

[untold glacier]

I have no opinion or ill will toward ya. I'll leave it up to the mods to decide.

[nova remnant]

what are you even talking about detective, you should get a ban for that handle

[silk narwhal]

discord is having a stroke on the servers 70% of servers are glitching out

[untold glacier]

Yeah, there've been problems.

[nova remnant]

what was that all about ...

[nova remnant]

i dunno if discord was down, i dont see any issues on my end yet

[white nacelle]

well im just trying to find people who id wanna hang out with for hours on end. co clearly torqx is out of the question, but since u dont seem to have ill will towards me would u maybe wanna team up

[silk narwhal]

18 minutes to generate in comfyUI but i like the result

[white nacelle]

@silk narwhal on a scale of 1-10 how much do u like me

[nova remnant]

you wanna hang out, you can hang out here with everyone, we post stuff and chat a bit

[white nacelle]

1 being hate, not like only a little bit

[untold glacier]

Love the layout and it's strangely cohesive despite the noise. Very interesting style.

[lyric zealot]

[lyric zealot]

you literally being here for like a day not even?

came here to test my discord but saw this instead

[untold glacier]

Well, the original join date was on the 9th before exiting and rejoining just now.

[white nacelle]

well im sure they have a first impresion of me

[untold glacier]

But still...

[nova remnant]

hey smol sus, did you get the new card?

[lyric zealot]

shit, then i was wrong, my bad

[lyric zealot]

nope, not yet

[silk narwhal]

-3

[lyric zealot]

Still thinking about waiting for just 5000 series

[untold glacier]

Is it just me or is that a new avatar? It's nice

[lyric zealot]

it's been awhile

[nova remnant]

oooh the wait game on pc tech is forever tho since it changes every 3-6 months

[lyric zealot]

yup, but i guess it's gonna be worth the wait

since it's like the first card out after the boom of AIs

[nova remnant]

5000 series sounds good tho 🙂

[untold glacier]

I just want more VRAM.

[lyric zealot]

as such, vram and such might be higher

[nova remnant]

must be pretty expensive too im guessing

[untold glacier]

But I'm not holding my breath

[lyric zealot]

hopefully not

since they know it's vram that matters now

i mean, for AI aspect

[nova remnant]

i have to get more ram

[nova remnant]

downloadmoreram.com

16gb is bit too tight for sdxl

[silk narwhal]

VRAM for the rendering and RAM for Storage of AI models

[nova remnant]

yeah

[white nacelle]

not exactly, i need find people who i think would wanna team up with me in what im trying to do, and i need 1-3 people, but like thats only gonna work if i like them and they like me. but like so torqx, and kuromi are out of the question. so what about u smol sus, on a scale of 1-10 whats ur opinion of me

[nova remnant]

sdxl models take up a lot of ram to load

[lyric zealot]

[nova remnant]

Judging by the name

I'd say its professional work

troll

[silk narwhal]

Base SDXL Turbo hasn't gone over 6 GB of RAM for me yet

[lyric zealot]

https://tenor.com/view/tldr-gif-25251690

[lyric zealot]

im joking btw

[nova remnant]

i mean i have to wait 40 seonds to load sdxl on my a1111 with 16gb ram

[lyric zealot]

You are up for the job then, buddy

[untold glacier]

Tbf, A1111 is the problem there but your point remains valid.

[nova remnant]

@lyric zealot

[lyric zealot]

[silk narwhal]

DDR6 will be faster at it 😄

[nova remnant]

im hoping the next a1111 update will be sweet 🙂

when sdxl came out a1111 caught up to speed with comfyui in few weeks

@lyric zealot smug is so funny, seemingly normal gens turning into looking down on you

[silk narwhal]

first pass out of 4 🙂

[lyric zealot]

24gb vram is kinda like the min you need to play with multiple AI stuff

[nova remnant]

I'd pass her 4/4

[nova remnant]

is that what you are targetting with 5000 series?

[lyric zealot]

kinda

[nova remnant]

oh wow

[lyric zealot]

like example that Stable-Zero123

you need 24gb vram for that

[nova remnant]

What is that :oh: @lyric zealot

oh man, i can't afford to buy new gpu atm

[lyric zealot]

#📣｜announcements message

basically 3d SD1.5

[nova remnant]

@nova remnant get ready for the bare minimum vram of 24gb

[nova remnant]

my pc stuttered from reading that

[lyric zealot]

[nova remnant]

me too i made a typo typing that

any chance of stability ai optimizing vram usage?

[lyric zealot]

I think the community is doing that

It's kinda like how SDXL needs crazy amount of vram

then now it's getting better

[nova remnant]

but better how

[silk narwhal]

purging junk data from the base models that just results in bad outputs, kinda like cleaning a messy room

[nova remnant]

i see lot of devs that worked on sd1.5 can't manage to train sdxl just as easy

primarily due to resource demand

[lyric zealot]

i really think

you need to stop with that

[nova remnant]

you know lot of ppl would be very happy to be using sd1.5 if hand glitches could be fixed and maybe adding a bit more prompt coherence

[silk narwhal]

most hands on 1.5 can be easily fixed now..with controlnet,inpainting etc

[nova remnant]

can you explain

ahh inpainting

thats extra work required, what i mean is generating images that wont need extra effort to fix them

[gloomy scroll]

And who are you?

[nova remnant]

btw speaking of inpainting if only hands could be just as easily fixed as faces with adetailer

Vil
just ignore, troll or something

[silk narwhal]

just a horny troll yea

just block

[gloomy scroll]

https://tenor.com/view/milk-and-mocha-pat-cuddle-love-heart-gif-15740749

[white nacelle]

a detective

...of sorts

[nova remnant]

trees on top of trees

[lyric zealot]

my discord is still so fucked

[nova remnant]

what do you mean? i dont see anything unusual on discord

is that about running bots or something similar?

oh

[silk narwhal]

70% of all servers were down for at least an hour now.

[nova remnant]

i didnt notice that since this server is obviously not affected

[silk narwhal]

about 5 minutes it was down

[nova remnant]

i see, i was offline in the last couple of hours

[lyric zealot]

yup it's slowly coming back

[supple raptor]

tbh I have no idea who you are, but repeatedly mass pinging isn't a great way to get people to like you.

[brittle rampart]

@supple raptor many thanks for your job! the new version pays more attention to the legs. Or am I just so lucky today?

[supple raptor]

v3 has definitely been a lot more consistent with anatomy in general for me. I think it solved most of the annoyances that I had with v2.

v2 struggled quite a bit with backgrounds that weren't nature and was quite prone to diamond shaped pupils as well as some general weirdness with anatomy at times.

[brittle rampart]

In terms of anatomy, the new version is indeed more realistic. but at the same time, a certain cartoonish charm was lost. or is it just me?

[supple raptor]

no, it's definitely a different style than both previous versions. perhaps I should have named each of them differently, but I think of it as more of a series of models that are related, but different from each other rather than each one being intended as strictly an upgrade of the previous version.

[brittle rampart]

then that explains everything. In any case, excellent work as always! THANKS!

[stable crane]

cute

[terse ore]

Trying to use Multi-Controlnet (openpose, depth, and canny) but the image keeps getting messed up (tons of lines all over like crease marks all over the image, hands and feet consistently messed up)

[native halo]

least buggy comfy workflow

[brittle rampart]

maybe Canny result not good?

[terse ore]

Hmm, I'll mix and match and see if that makes it better

[brittle rampart]

take the cheese from the mouse...

[terse ore]

The criss-cross lines are gone...mysterious. Hands and feet are still bad lol

[brittle rampart]

Accidents are not accidental (c)

[dawn jasper]

Who is this person?
@sick pond You saw JJK?

[brittle rampart]

all in PNG Info

[dawn jasper]

@sick pond Pleased to meet you then.

My condolences.

[silk narwhal]

???

should be in the png info ..but it is 3 different models in seperate passes in comfyUI

7 times including face and hand fixing

[dawn jasper]

Do these women you rendered have names, by chance?

[silk narwhal]

nope just generic booru tags

1girl, etc etc

[dawn jasper]

So the beautiful women being generated were on accident?

[brittle rampart]

that a Bot?!!

[dawn jasper]

Wait what?

Where?

[silk narwhal]

just luck out of this spaghetti :

can send the json .. just have to load that and install the missing nodes with the manager if you have that installed

yes

most steps are cleaning up the first generation and doing upscaling etc

[supple raptor]

With comfy, the PNG also has the full workflow embedded in it so you can just drag an image from the outputs folder into discord to share the workflow.

since you have a1111 already, comfy doesn't have to take up much space at all. you can reuse the A1111 venv for comfy.

[silk narwhal]

with the models i have .. i am at 60 GB

[supple raptor]

just have to activate the venv that's in your a1111 directory first, then run the main.py in your comfy directory. I made a simple .bat file to do it for me. if you want to use it, you'll have to open the .bat in a txt editor and change the venv path to the correct one on your system. make sure the .bat is in the root of your comfy directory.

download that, place it in your comfy folder (the same one that has the main.py file. open the .bat in a text editor, change the pat as noted on line 6, save your changes and you're good to go. there's a place in there to add launch args if you need them too.

this is the folder where you'll find the activate.bat file that you have to point the one I attached to. it's inside of the scripts folder. the one I uploaded does not go in here though.

so line 7 of the run_comfyui.bat file should be changed to call C:\ai\stable-diffusion-webui\venv\scripts\activate.bat.

the run_comfyui.bat goes wherever you downloaded comfyui to.

[supple raptor]

You should be good now. As long as that file is in the same folder as comfy's main.py it should work.

[silk narwhal]

the bat file should be in the same folder as the main.py .. no need to touch the main.py file

[tulip acorn]

/A cute woman standing in the living room, scenery, detailed background, masterpiece, best quality, high quality, absurdres, xmas lora:xmas_xl_v1:1

[silk narwhal]

Is not fast and requires a whole bunch of models.. all can be found on CivitAI

AWPainting, Snowfall_V20, all controlnet models, add_details Lora

https://github.com/ltdrdata/ComfyUI-Manager ..will make it easier to fix those red boxes

method 2 is the easiest for you

https://github.com/ltdrdata/ComfyUI-Manager?tab=readme-ov-file#installationmethod2-installation-for-portable-comfyui-version-comfyui-manager-only

download the file in step 2 in the ComfyUI_windows_portable folder and run it

[silk narwhal]

right click..save as

[native halo]

https://tenor.com/view/lucky-star-anime-konata-izumi-computer-desktop-gif-14835314

[nova remnant]

Let them comfyers hack the system

[native halo]

https://tenor.com/view/network-confused-john-travolta-wire-gif-5799547

comfyers

[sterile pendant]

sexyy

[vestal crypt]

https://media.discordapp.net/attachments/1027429244478955530/1185456409953976440/pixai-1692426667379941217.png?ex=658fad50&is=657d3850&hm=f414dc10d646988aad666b177e09f6c4c59915bd11be3aa2b8f98332ac06b92d&=&format=webp&quality=lossless&width=532&height=662

[stable crane]

🙂

[stable crane]

an img2img like process

with like 0.5 denoising

[finite herald]

change your username pls #✍🏼｜rules-and-tos

[steady grail]

wow, charming

[native halo]

i cant its for my redditor collection😔

[brisk valley]

Can check the meta or import into StableSwarm
models are: Tete experimental AhnnOne
ti embedding: aidxl 0.5

:)

[brisk valley]

I should upload some more other stuff here

oh yeah keep in mind Tete is an experimental mbw of AnimeIllustXL which is an SDXL model

[stable crane]

cute

[brisk valley]

ok

[native halo]

1060 6gb

[brisk valley]

?

[native halo]

4min for a 1080p img on normal sampler and 1 min if i use lcm

because xl

also the 1080p img i gen is 540x960 > high res fix 2x

u must be doing something wrong maybe u are gen 1024x1024 native?

how many steps? i do 25 normal with dpm 2m karras then 15 high res steps

do u use auto1111 ?

thats odd cause for 1.5 the difference between comfy and auto1111 is minimal it must be maybe the driver version u are using

[dull jackal]

Sorry--I was on a trip today! ❤️

[near elbow]

have some wallpaper for a change 🙂

happy cherry blossoms

and another style yet

[wooden coral]

Can you not mass ping random people, thanks!

[midnight pivot]

Hello everyone, I'm new to SD, I come from using midjourney and would love some guidance, Im a dungeon master running a furry campaign and want to know how to generate similar style of anime-like SFW here, could someone help?

[cyan burrow]

are there any active cyberpunk 2077 discord servers? Anyone know? Not the official one, I'm not handing out a verify phone number and they want one.

[keen hollow]

Did someone help you?

[keen hollow]

not sure, but I get the frustration 😐

[cyan burrow]

yep....

I guess I'll just ask something I was wondering here too. Google isn't very helpful. Did they remove "Oxy Boosters" from the game? (oxygen boosters)

I can't find them

[keen hollow]

Haven't played it. Waiting for <=$20.

[cyan burrow]

phantom liberty is worth every dollar

it's an expansion, not what is commonly cheaped out as dlc

they also make the relative smaller mapsize of the expansion be much larger because of verticality

[keen hollow]

Trying to get her playing Cyberpunk 2077 😛

[cyan burrow]

reminds me of misty except hotter

[near elbow]

blaster saber? 🙂

[keen hollow]

colored charcoal drawing, anime style, alt woman, kneeling, black eye makeup, choker, spikes, black messy hair, skirt, red chunky shoes, playing Cyberpunk 2077 video game, blue streaks in hair

pew pew

[near elbow]

staying with the theme

[keen hollow]

i wanna try han solo

[near elbow]

just do it 🙂

this looks highly impractical

[keen hollow]

pew pew

[near elbow]

pew indeed 🙂

that's with "(((studio ghibli anime style 1980s 2d flat cel shaded))) "

ha, some firepower

[keen hollow]

Chewie merged with Darth Vader and a Porg somehow... 😛

[near elbow]

totoro hasn't been the same since he became a stormtrooper

kiki crossover

[keen hollow]

this is not how you play nintendo 64...

[midnight pivot]

No, I've been browsing all day long XD

[cyan burrow]

There once was a fox named nox with a box full of sox and in those sox he placed his shutting down

damn, they got to him

[near elbow]

another victim of the wombat police

[lyric zealot]

guys guys guys

new auto1111 update

[near elbow]

what does it break this time? 🙂

[lyric zealot]

hopefully not everything

I'm too afraid to try

but well, i think it's also time to have a fresh installation of auto1111

since my auto is getting buggy

[midnight pivot]

xD

[near elbow]

was it ever NOT buggy? 🙂 (hm, sdxl still struggles a bit with text)

[lyric zealot]

it's much more buggy than before in my case

[near elbow]

live from github

[lyric zealot]

Imma wait for people to update first

[keen hollow]

You're just looking to do anime style generations with stable diffusion?

oh geez... a new update? who is brave enough to try it?

[lyric zealot]

@dull jackal are you here

urgent

@wheat falcon are you able to delete post?

it's urgent

@wooden coral @gritty spruce you guys here?

@wooden coral you there?

well, i don't really wanna link it

someone posting gore

[wooden coral]

?

[lyric zealot]

boto thanks god

go look

[wooden coral]

Where

[lyric zealot]

#🏞｜general-with-images

fast

[wooden coral]

Heyo, should be gone now

[lyric zealot]

yup

[wooden coral]

Lmk if I missed smth

[lyric zealot]

that's why i said urgent

[wooden coral]

Thanks for the report

[lyric zealot]

even if it's ai, should not be allowed here

[wooden coral]

Didn't look very ai but who knows these days

[lyric zealot]

I think that's it

well, not sure. But obviously not stuff people want here

well

we don't really have to go in depth in analyzing this troll

well, it's just kinda sad that, that's how they get validation online

[wooden coral]

No gore please, even if it's not really obvious

Yes, just to be sure

Thanks

[stable crane]

anyone sub to artists? curious if you prefer:

• download their works in zip files and view locally
• view stuff online at patreon/ko-fi/other/etc

[cyan burrow]

I can feel stable diffusion and other AI applications just leaving the hands of any consumers that weren't able to spend 1000-3000 euro on an nvidia gpu with 24gb+ vram

[stable crane]

I'm hoping we get more vram

[cyan burrow]

I'm waiting to see how AMD will respond to nvidias lack of interest in gaming

anymore

hopefully with cheaper stuff

[vestal crypt]

I don't have to care for how many legs does this beast have

[vital raptor]

for what it's worth, i've been using the 1.7.0 RC for a week or so, seems ok so far

maybe some gradio incompatabilities with some extensions

[dull jackal]

I love this so much.

[dull jackal]

This is super adorable.

[vital raptor]

big bad wolf?

[near elbow]

now we know why she's called little red RIDING hood 🙂

and she made friends with the wolf, too

[near elbow]

wolf mecha

[vital raptor]

fox mecha?

😄

[weak prairie]

Can you find my little friend?

[nova remnant]

[nova remnant]

@lyric zealot

[native halo]

this not a bodysuit

[nova remnant]

jumping

👍

[lyric zealot]

[vital raptor]

a gnome 😂

[nova remnant]

[untold glacier]

My dad sent me this. He misspelled something and bing gave him this instead of what he wanted.

A fantastic mistake.

[vital raptor]

nice

[finite herald]

giv

[true jackal]

nice work... could you please tell me what model where you using? thanks.

[silk narwhal]

She is not big .. it is a movieset 😄

[nova remnant]

angy

[native halo]

needs more latex

[zinc flax]

https://pixai.art/model/1692657116006913280

[bitter latch]

like this?

or

or maybe even

😄

oh here ya go

[nova remnant]

the world is ending

:x

[bitter latch]

nah just simple image 2 image

[nova remnant]

google censored my prompt today

i was shocked and horrified by the power of woke culture

[bitter latch]

hmm ok thats different

[nova remnant]

Google what?

[nova remnant]

yeah imagine that, i used bard a lot for prompt ideas

Oh Bard

all i was asking for woman in short attractive clothing

the short part rubbed bard the wrong way and it lectured me

[native halo]

only burqa allowed

[nova remnant]

the world is ending

and wanna know how it lectured me about 'short'?

i was asking for a nice pose in trendy clothing, and it gave me a prompt about personality

i think the world is transitioning towards an end, unless something happens that could surprise everyone

SDXL 4.0

We'll only need to click generate and it will know what we want exactly

Year 2030 coming soon

prolly

cause time feels like it has sped up in the last few years

[bitter latch]

LCM does great with the right settings

[vital raptor]

magical girl making cupcakes not really what i had in mind for magical girl, but the cupcakes are surprisingly consistent

[bitter latch]

adds to his prompt :D

instead of magical circle it's now magical cupcakes 😄

waits

goes back to watching "Kikansha no Mahou wa Tokubetsu desu; A Returner's Magic Should Be Special"

huh. not what I expected...

[bitter latch]

hmms , changes models and puts thru img2img

wait i've been using Restart.. not lcm

doh!

[white zenith]

IlovetiledUpscale

[bitter latch]

haha

ok forgot I had the 3D Animation instead of anime loaded. oops

[bitter latch]

Have some cupcakes (take 2)

magical cupcakes

[bitter latch]

cupcake cake?

[bitter latch]

hha miku cupcake

[bitter latch]

ah ha! cupcake fairy

[vital raptor]

cupcake fairy 😏

[bitter latch]

hm not what i had in mind but WOW

hmm came out fuzzy

[vital raptor]

same prompt on 2 different custom merges

[bitter latch]

nice

[vital raptor]

second one is really nice, but it leans towards realistic a bit too much sometimes

[bitter latch]

hmm 6am.. I should be in bed but it's my day off so .. meh

yawns and falls over.. thud

[twilit portal]

did u just block me ? what did i do ?

[boreal yarrow]

Nami

[near elbow]

some non-human element for a change

RUN!

[valid turret]

Finally I found out about ADetailer

So, every SDXL model is anime-style now 😄

[near elbow]

hehe... TBH i don't use any upscaling anymore since the SDXL models are good enough for 1024x1024 with a wide margin, and faces are much better, too...

[valid turret]

With another model tag "fishnet" created nets and fish in interior 🐟 🥅

[valid turret]

1024p is the lowres now

[near elbow]

all relative... most people i know only click on the image once and then get the large discord preview which is 800x600 pixel only, anyway...

look at her, she approves 🙂

[brittle rampart]

Hi guys. can anyone tell me why when I post an illustration on Civitai I get a message "TOS Violation - Blocked for lolita" ? and how can this be fixed?

[valid turret]

Check metadata of the file

[brittle rampart]

Bot of CivitAI reaally joke! hate it!

[near elbow]

more fun

[nova remnant]

how well would it do exploding cars

[native halo]

dont upload pics to civitai post them in pixiv instead 🌚

[silk narwhal]

that mood when the model does great hands but it put left hands on right arms and vice versa :/

[remote vine]

haven't been using ai for a while, what's the best model rn?

[nova remnant]

Pixiv

Many bodysuits there

[stable crane]

lol

[vital raptor]

The best model is the one you like the most 🤷‍♂️ It's completely subjective

[silk narwhal]

DreamShaper 8 and DeepBlueXL are my goto, next to AWPainting

[steady grail]

i personally dig much more anime models

[native halo]

well u are in an anime channel

[steady grail]

eheheh

[vital raptor]

for anime stuff I'd probably suggest bluePencil/DeepBlue/kawaiiPencil

[native halo]

imagine downloading an anime checkpoint that cant prompt anime characters

[vital raptor]

the definition of chaos in picture form 😄

[silk narwhal]

look it's the cableguy

[silk narwhal]

accidentally moved one of the node links to the wrong node in my workflow :/

just assume she is moving almost at the speed of sound

[vital raptor]

It looks like you're generating 2 different prompts and then blending the latent noise before rendering or something. Pretty interesting affect

[brittle rampart]

Uploading to civitai is simply a way to support the authors Checkpoints and LoRa`s. And on pixiv, of course, I re-upload the illustrations

[vital raptor]

man, CFG is confusing. The SD documentation, and therefor all the videos and reddit posts about, all say the same thing.
Basically, lower CFG values are 'more creative' but better detail while higher values are more strict to the prompt, but worse quality

I'm not convinced this is an accurate description of what it is or how it works

especially with more simple/shorter prompts

I guess it depends on what you think is 'quality' though, one of the videos I just watched the guy found "anything over CFG 16 starts to get unusable" but the images he was showing were badly burning around CFG 9 🤷‍♂️

[brittle rampart]

"Burn-in" of the image at high values allows you to avoid one addition. but his work did not satisfy me much.

[vital raptor]

or maybe I'm reading into it too much and i'm looking at it wrong, this is a very small subset of a grid I did to try and find the best CFG/steps combo for this model

[brittle rampart]

as they said earlier, many models do not support values ​​more than 8-10

[vital raptor]

CFG 1 and 2 basically look the same, just lower contrast on 1, 3-15 all look mostly the same, a few details here and there but nothing drastic

[vital raptor]

I must have missed a conversation at some point, I don't know what you're referring to

[silk narwhal]

the more details in the training data the higher the cfg end up being

really low detail data ... low cfg needed

[brittle rampart]

I asked the same question here earlier. and it was discussed.

[vital raptor]

Ah, I didn't see that

[silk narwhal]

when generating i use 11 as my average point .. for realistic i do 12 or 13

for hires fix i use 4 or 5

[vital raptor]

I can't go over 6 or 7 without burning

[brittle rampart]

for me it 5-8

[vital raptor]

on this model specifically

[brittle rampart]

how do you use different meaning in hire.fix?

[vital raptor]

to me, it seems like the higher CFG is the one that's "more creative" since it just starts adding random details

[brittle rampart]

why not use LoRa to add details?

[silk narwhal]

i use comfyUI .. the face and hands fix nodes i set the cfg to 4 or 5

[vital raptor]

I don't need details, that's not a problem

[brittle rampart]

oh comfy! then that explains everything

[vital raptor]

I'm not trying to fix anything, I'm just trying to figure out why CFG doesn't behave the way it's described

either it's using subjective terms, or it doesn't do what it says

I don't like over saturated high contrast images, so CFG 7 in my above grid is worthless

[silk narwhal]

SDXL based model?

[vital raptor]

Does CFG 11 have more detail? sure, but it's not usable, so the detail is not worth anything

yeah, this is sdxl

[silk narwhal]

yea..SDXL is lower cfg than 1.5

[brittle rampart]

or there is still a dependence on how the model was trained. what if view dependency data using different models?