#🍥|anime
1 messages · Page 288 of 1
like what most people usually do
yep a lot of forks and stability matrix also works on it
well, if it works, and you have decent speed, then all good 
tho i will say i have no clue what folder i should put that in since there is 3 or so folders which has upscailers in it
hmmm, what ui are you using?
forge UI but stability matrix has models in it that shares will all UIs u download on it so i have this
should be here then
ty
ΣΙΗ.safetensors my current fav
understood
Also my fave 🙏

Wdym 😅
I like how it does darker colors and textures
Plus it has uhh
Some more varied
generate me
a picture of me
you did your research 
i kinda come from Niji from Midjourny so i came to local but im wondering how to get sim aesthetic quality
gotta find out how good my models REALLY are yknow?
blonde, heterochromia green-blue
too lazy to fix fingers

Aight give me a bit
what does sim aesthetic quality mean?

Honestly was curious about that too
similar aesthetic quality to Niji
oh, i guess you will probably have to download some lora based on niji style
175cm
true didnt think of that although i think MJ models add more prompts in the background when u send in urs so if u put something simple like " a girl dancing in the rain" the will add more prompts in the background that u cant see which may also be a issue but im not sure
yup, they almost definitely do that. You will have to experiment with it yourself, by finding a decent model, downloading some lora, and then just copying people's prompt for their generation, then tweak it from there. If you find something you like, most of the quality tags can just stay, and you will just prompt for the subject/background.
is that me?
so far im enjoying local stuff but i need to work on making stuff that looks vibrant and beautiful
no actually. I can make it green blue eyes tho
What's that in american
well poses and backgrounds i need to work on
wildcards? not heard how would i do that?
best way when starting out is to just copy from others. Look at generations from others, copy their prompt, generate with it, then change stuff with it, and generate again. See how it affects your generation
bahahahahaha
wildcards is just like randomization on crack
well
no
no
its actually very simple
unless you do it like me
i don't know what happened with the eyes tbh
oh thats a lot of letters
perfectly 🙏
i just dont know how to use it
true
wildcard is basically like, out of all these options, choose one. so you can like put blue eyes, green eyes, red eyes
then download/make some wildcards (text files with a list of things to be randomized)
then activate it in the prompt
for swarm its <wildcard:(wildcard name)>
for forge its _ _ wildcard name _ _
no spaces on the double underscores tho
ya
dynamic prompting works the same way except instead of using wildcards you just manually type out what you want randomized
yup yup
wildcard is basically outsourcing your prompt writing to others 
ok i need to process it in my head 😂
most of my own are my hairstyles, pupils, clothing, poses, facial expressions
but i have thousands of wildcards
if you want too use them lmk
I’m trying to come close to this output I got from Niji
that looks achievable on local
Yes just need to figure out how due to the fact I’m new to local stuff mainly
anything is achievable on local 🙏
local supremacy
is that me?
you will get there 
Rn it’s forge
not sure if that's true tbh. But well, if it's not, then it's pretty damn close
Seems like ComfyUI like many people said online doesn’t like Mac hardware
So forge it is
do you already have it installed?
I’m using it rn
I think you should look deeper. I've never heard of any issues between comfy and mac that were considered insurmountable
imma show you what settings to use
give you some extensions
anddd we're gonna change some settings for QOL improvements

but some people just want to jump right into making pictures so 🤷
Might be because I have 16gb ram and a older chip (m2 pro) which is why it’s slower than the rest not sure
Wonderful can’t wait
oof
i use both
m2 pro chip is fine
I could try Zluda or dual boot into Linux for ROCm but then again I have a 6600 xt 😭
I don't see where you'll run into issues with comfy that you won't generally run into with other UIs
double oof
im under the impression that if i can run it anyone can run it
ai that is
unless you have amd
I am saving for used 3090 or if I’m a lucky and save enough used 4090
not sure about mac tbh but
you said you were already using forge so
should be fine rs
Nice I got a 4090 for $1830 last year
Yh Hi res on Mac takes quite a bit but the gen speed without is pretty decent enough that I can wait
I live in the UK not sure about the prices XD
set your ui up like this
approved
Done
dimensions do matter btw so id look up some you can use but i prefer: 832x1216, 1216x832, and 1024x1024
okay cool go to settings tab for me?
On it
Yes
hit apply
anytime you want to change your default settings hit that
that way you dont have to set everything up again everytime you close your ui and open it
nice right?
staying in the settings tab
lets fix the fact that interrupting gens takes forever
So when I scroll down I have UI defaults but 3 of them
type 'live' into that search bar
XL, SD and Flux
xl
do XL i think
i have one
but
should be fine 💀
Ah ok so just hit apply
yep
I have massive overlapping text
uhh close and open your ui again tho
then we'll fix the next thing
then ill give some basic extensions and uhh
you should be good to go
just need to learn prompting/loras after that
Ok so done that
@woven gust
Done
Done
install from url
copy and paste this into it
install
then do the next one i send
do them one at a time pls
Uhh extension access disabled because of command line flags
oh seeing adetailer, my one is fucked. Still need to fix it 

im unsure about how to fix that as ive never run into that problem
you can manually download them though
i think
unless you cant use any extensions at all 💀
then thats a #🤝|tech-support question
copy and paste the error message from your cmd
or terminal since it's mac
lmao
try adding this to your cmd flag --enable-insecure-extension-access
ok thx
if you know what i mean by that
I do
let us know if that works 
ong
Nope
💀


Ofc it’s sum like that causing the issue
welcome to ai IG 💀
Yh thx XD


Done the 3 u sent
the extensions?
Wrong person sorry
sweet, restart ui
Ye
you should have some new tabs now

uhh
lets see for wildcards tho youll actually need wildcards
if you can navigate to the stable diffusion folder
It’s reloading UI still
and find the dynamic prompting/wildcards folder in the extensions folder
Installing stuff on terminal
I’ll find the Mac version of that XD
Let Python-dotenv
And av download
these are all my wildcards
ya ya, np
I love when some downloads on thermal has a download bar and others don’t XD
🙏
I’m just sitting looking at my terminal to change
fair
Oh lmao nvm the forge UI stopped im starting it up again lmao
That’s why I was just looking at it
💀

So styles has now a million things in it
And now I have
Infinite image browser
At the top bar
yep
infinite image browser just logs all your images you make
itll save the prompt, loras, seed
everything
now you can view past images right in your ui
find extensions
then sd dynamic prompts
then wildcards
replace the wildcards folder with the one i gave you
Oh yes ty

I did it
Uhhh the tabs at the top?
yes
I dont have wild card
💀
I have the image browser only
did it not install 
It shows in extensions it’s checked
And a whole bunch of forge extensions that’s also checked
On extensions my wildcard shows installed but doesn’t show on the top tab
Done
is there any error message on your terminal?
No errors I can see

some extensions have done this to me before but
i could only ever fix it with a fresh reinstall of forge-
typically have to stop and relaunch but i use A1111 so i dunno about others
Also ADetailer and Regional Prompting are godsends, and i highly recommend learning them.
Sadge I tried that doenst show
but it seems like it bugs more often for yall forgies
thats just me tbh

i break everything
the only feature a1111 has that i like is the regional prompter but
Fair. If you aint breaking it you aint using it hard enough
i dont need that anymore
I have a lot of tabs can forge only load up certain amount of tabs that it can show
i can simply prompt for 40 characters in the prompt box
and itll work
illustrious models are great <3
they can do separate characters even with loras enabled
I primarily use illustrious base models, or checkpoints trained from them. So clean. So consistent.
and having multiple characters with loras enabled is something even regional prompters suck with
yessssss
he gets it <3
i dont use trained/merged models tho
except uhh
the one smol uses
this one
I have a few, but thats because some have a certain aesthetic that i like.
im pretty sure its a merge or smth similar
i pick a clean model and use loras to achieve aesthetics
Oh I would like to do that
and so is the other one but it handles textures better imo
Thats fair, and i have loras for certain outfits im far too lazy to iterate for. Like most of the monster hunter armor sets
i use wildcards lmao
i just make tons of clothing wildcards and mix and match then
them
@toxic lotus wild cards show in settings
or dynamic prompt with them for crazy outfits
But not at the top tab
Oh i have a ton of those as well, but their for random generations for fun
i use them religiously ngl
i liek spam posting on twitter so i make extremely large, convoluted, prompts with wildcards
so i can make, men, women, nsfw, sfw, stuff all in one prompt
ofc it includes different poses, clothing, and backgrounds too
A mystical female_character with eye_color eyes emerging from abstract_background_liquid, hair_color hair flowing like liquid, wearing translucent female_cyberpunk_outfit, special_effects surrounding them. camera_unique, lighting, mood. Art by artist-anime, quality. is an example wildcard wallpaper prompt i have for creating 16:9 pc wallpapers randomly. I just swap the outfit or background tags to change the theme.
Someone said when I searched it up use dynamic prompts instead of wildcard cuz that shows
awww it got rid of the double __
nah see
see
this is mine
it uhh
goes down further
ngl
i have to scroll
😭
yeah discord does that 😅
thats wild lmao
Using wildcards and the random selection between wildcards in the syntax is a neat idea.
Would save me having to save multiple prompts
i do it all the time
you want the prompt?
i need to fix it a bit
its still a work in progress
but it works as is rn

you can include in three "`" and ending it with three of that as well.
I dont think my poor checkpoint could survive that prompt. It might explode
oh shit good looking out
that prompt is actually 3 prompts in one
finally fixed my adetailer, by ||uninstalling and reinstalling
||
you can send it, ill have to swap the wildcard file names in it lol
fair fair
i also posted my wildcard above in chat
so you can just download those and use those
so you dont have to change anything
just like add them to your wildcard collection lmao
i created an extension to hide my preview image.
in case i'm doing some "researching" and people walked in
you can uhh swap some stuff out, like the favorite loras wildcards
done, ill test it out lol
those are actually just personal preferences
bewarned, if you change something it might break 💀
you add an extra } or forget one and the whole thing breaks
😅
its their fault for walking in
if they see sus stuff they have to live with it now
at least in my case
🙏

@toxic lotus it doesn’t show I’ve tried looking at user interface hidden
In settings nothing
i had to removed the loras that were naturally in it since i dont have those lol. but its batch generating right now
understandable 🙏
im not sure you can trying asking cs1o in #🤝|tech-support
Should I reinstall or skip?
i usually go crying to him when i break something
My A1111 positive prompt box had a stroke though when i copied that in lmao

its a crazy prompt
uhh just skip it for now
ill ask around-
in a bit
after 5 cigarettes
but
for now you should be able to make great images with little to no issues
so give it a try
lowres, (worst quality, bad quality:1.2), bad anatomy, sketch, jpeg artifacts, signature, watermark, old, oldest, censored, bar_censor, simple background, extra head, conjoined
throw this into your negs
its best to keep them simple
add to it as needed
throw this into your pos prompt
masterpiece,very aesthetic,absurdres,newest
these are quality tags, they help
then just uhh prompt very simply
1girl, solo, blonde hair, short hair, red eyes, narrowed eyes, red shirt, long sleeves, blue jeans-
smth like that
important stuff up front
What should be the hi-res settings again
10 hires steps
denoise of 0.4-0.45
here
Ty
This is a good template for prompt formatting to follow, when starting out at least.
masterpiece, best quality, ultra-detailed, [subject], [character description], [pose/action], [clothing/style], [camera/lighting], [anime studio + artist style], [special keyword: Pixiv, Fanbox, Comiket], [production keyword], [contextual fiction phrase], [visual flavor]
Ty
For models what u recommend for the base stuff
I remember u giving me 3D render style models but I was using a different interface at the time I was using one that people use a lot on mac “Draw Things”
wai and the one smol sus uses
sec
ill link them
Depends on what type of art you want, or rather, the aesthetic. Ilustrious delivers high quality anime, PONY is uh...PONY, i dont use it much. SDXL and SD1.5 are like general use base models, but the SD1.5 leans more into the naturally drawn or watercolor aesthetic. But ultimately it depends, cause theres base models then theres checkpoints that are trained base models aimed towards a certain aesthetic or concept
i do more anime style stuff so i recommended anime models
tbh i wouldnt use 1.5 models
theyre ass
I mean remember the image I send above a bit back that’s kinda what I like
i do not remember
i have the worst memory
ever
💀
Yeah me too. I do have a couple NoobAI merges that deliver really good CGI/2.5D stuff. Ive had mostly nightmare results from the realism models so i just decided that aint for me
Here’s another example image
ooohh then yeah these two models will be bgreat for you
Understood
WaiNSFWIllustrious v130 and Hassaku XL Illustrious are really good models
For general use anime
If you want more abstract stuff, ive been playing around with a checkpoint called PlantMilk that is pretty neat. A little unpredictable though
howd that prompt work for you btw
so far this im downloading sih rn
i need to do it now
the last one was a touch nsfw, but not bad
well
i did mention
it creates nsfw as well
💀
I don’t mind that I’ll mainly do SFW but I’ll dabble into NSFW range after I get prompting down
thats a smart move, cause otherwise youll get some nightmare fuel
XD Yh
Ive seen stuff that would fit into lovecraftian horror, with positions that would make the ring girl contort in horror. The model can certainly create some really good stuff, but if your prompt is ass, your gonna get some ass lol maybe three....that share the same asshole. It gets wild when it breaks
It’s reached 50% I’ll assume from here it does hi-res cuz it goes slower now
yeah

then after that more time for adetailer
Fair enough I’m just used to natural language prompting cuz of Midjourney/Niji
but
Depends on the steps in the initial generation, the steps in the hires fix, and the denoising strength
True especially with laptop hardware aka Mac
yeah its kinda funny ngl
im using a laptop

i have a 2060
tho tbh
Ah fair idk how M2 pro compares to that in speed
fair warning, most Illustrious models are trained on images labeled with danbooru style tags, so often times youll have better results if you mimic that. But the newest Wai checkpoint allegedly handles natural language better than previous ones
i also steal at @dusky linden 's pc
unsure
Understood
my result
Hey thats really clean!
sweett
grats
youre a pro now
go forth and make images young ai user
W XD I wanna see the different styles the other model does with same prompt
I shall come back

Lol i do that all the time
Run a prompt through each checkpoint in a batch of 4 to see how each interprets it
this new style mixing im working on
hates hands
fuck
its such a pretty and creative style too but
the hands are unstable
Ive got a solid ADetailer prompt for hands and face.
i bet i have a lora that could overpower it tho
im using swarm rn
So really quick question if I get used 3090 and then replace my 16gb ram into 64gb cuz I’m running ddr4 my 5600 CPU should be ok?
Thats a dope hand concept, but yeah one is reversed it looks like lol and theres some anatomical clipping
ya my last style was more stable but
i like this one better rn
but the haaanndsss
time to add a 6th lora-
uhhhhh im not a hardware guy, so i aint gonna say yes but it LOOKS like itll be fine...again, i dont know much about your rig
Ooo could I borrow that
@dusky linden cmere
hardware guy
it guy
pspspspspspsps
It’s pretty budget ended 6600 xt 5600 cpu and then 16gb ddr4 3200mhz
Ik it will be ok general wise but for AI wise idk
I know my nvidia 3080 and i910900kf cpu are handling this fine, ive got 64gb ddr4 tho. What that means as it pertains to image generation, other than up to 8 images ina batch takes sub 10 mins, but beyond that i couldnt tell you
yeah thats why im trying to ask him 😭
Since I do tend to help people with pc part picker lists all the time
hiii
ADetailer Pos Hands:
highly detailed hands, realistic finger anatomy, defined knuckles and fingernails, natural finger spread, graceful hand pose, visible palm creases, soft skin texture, proper bone structure, well-lit fingers, high-resolution linework, smooth shading, subtle skin highlights
ADetailer Neg Hands:
blurry hands, fused fingers, extra fingers, missing fingers, broken hands, distorted hand shape, melted palm, deformed fingernails, unrealistic hand anatomy, twisted fingers, blob fingers, claw hand, overlapping hands, poorly drawn hands, smudged fingers, double hands
ADetailer Pos Face:
semi-realistic anime style, ultra-detailed face, expressive glowing eyes with depth and sharp reflections, porcelain skin texture, smooth gradients, soft natural lighting, cinematic close-up lighting, clean facial anatomy, high detail eyebrows and eyelashes, glow iris, symmetrical features, fine lineart
ADetailer Neg Face:
blurry, lowres, bad anatomy, bad proportions, deformed face, mutated features, distorted pupils, misaligned eyes, extra eyes, flat shading, washed out skin, ugly, over-smoothed, low quality, cartoonish, smudged details, plastic texture, disfigured, artifacts, glitch, poorly drawn
@dusky linden this ong
try those out
hmmm upgrading your ram should be okay yeah BUT
be sure your cpu can handle the clockspeeds
mine used to be higher then the cpu could support and i had to lower the speeds a little in the bios
no issues ever since
i overclocked my cpu from 3.7Ghz to 5.1Ghz, i could go further but it already runs fuckin hot as is. I dont wanna melt my liquid cooler.
one of the loras i made refuses to make bad hands so i use it with all my style for stability
🙏
i did not intend for it to do that but it does that
Yh I figured I need to look on that cheers most people said just make a new build not noticing I don’t really have the money for that 😅
lets see iiiitttt
sec its genning
@toxic lotus this is sih
banger
hunter cha fan?
a full build change with those specs is ridiculous ngl, i still rock ddr4 ram and a amd ryzen 9 5900x
or did you just run the sample prompt @toxic lotus gave you lol
upgraded my 3070TI to a 5080 and just replaced the psu + gpu
now to play with the weights, and then segmentation
Just need to grab 750/850w from my current NZXT C650 80+ gold (V2)
i think she did lmao
I’ll check 5900x prices and come back real quick
ahhhhh the heart hands. i can never get that to render right
noo cpu is fine the one you have
probably
ive done it a few times is have to look through my twitter for ts tho
Ik im just checking 5900x and 5700x3D prices
ong
5 5600 say it can only support 3200 mhz ram
Plant Milk doin weird shit again
I’ve been using it ever since I built it in ryzen 3000 days and then switched my 3600 to 5600 when that came out
As I heard the 5600 non X is just a 5600x with Lower clock speed and slightly worse bin
hmm honestly? if it works it works 
upgrading bc its slightly faster but you dont experience lag now?
yall speaking cthulhu rn
Have a friend who’s building a 5080 pc I was so jealous but I’m happy for him
It seems to have latched onto the same subject in each generation with this prompt. Different poses and backgrounds, and outfits. I wonder if i remove that part of the prompt will it properly randomize lol
Gaming wise no lag just worse FPS cuz 6600 xt getting weaker for current games
Also would like to try some RT games
me ngl lmao
ahhh
yeah then upgrading could be worth it but my cpu is also like
Which Yh 6600 xt and well AMD GPUs XD
3-4 years old by now
my prompt?
I heard new 9070 XT does RT pretty well for AMD GPUs wise
uhhh did you move/remove anything aside from the styles?
yes your prompt. and all i removed were the pre chosen loras

cause i dont have them, so it wouldnt have had any effect
are you using uhh
XD maybe one day I’ll get a modern high end GPU rn just looking at sub 600 range GPUs seems like for AI and just gaming I do 3090 used seems fine
hmm amd is doing pretty well lately but they are still behind on drivers and compatibility for AI
my wildcards?
ah 3090 is really good still
Drivers for gaming is definitely good but for AI Yh unlucky
download these? @sharp imp
slightly slower but you can even run video models at a slower speed
yeah, their loaded at least, according to the UI
I unzipped em and put them into my wildcards folder alongside my files
@sharp imp with ur face prompt for detailer
uhhhh how do i do that lmao
Better than wan 2.1 1.3b at 55 minutes for like 5 second gen like rn on Mac XD
just actually download the image
dont copy and paste it or anything
then send it here
oh so send it from my output folder
yeah mac is definitely suffering beyond xl lmfaoo
not just copy it
you can download it directly from where it was genned as well
hmm what platform did you use for Wan2.1?
right click
I mean Flux 8 bit 3-5 min for 30 steps isn’t bad
Draw things application as I don’t trust me running Flux or Video models on a non fully Mac optimised software
ohhh have you tried running video ai on that 3090 yet?
or you dont own it yet?
i can send you the metadata without the image lmao, its listed under the generation lol
Well I need to find a decent offer on eBay XD
ooooh
Then I’ll let you know
like theyre fixed now orr?
oohhh
yeah you could do that too
just send that then lmao
😅
It will be a while but hey when I get it I’ll be a happy Kat
once you do hmm
there's a lot of variables, but on a 3090, a 5 seconds video on 720 14b wan might take like 10-15 minutes ish
put a | here
that is why everyone need to have the stealth metadata extension for alpha channel metadata 
hmmm
as you can see, the wildcards are working, but the first line is actually a non wildcard prompt
that is shared across my batch
Way less time than my Mac XD also u don’t mind if I fren req u I need some AI friends
i can link you the platform i use, it has a lot of documentation for stuff like tis
so i think its latching onto that, even though its randomizing the wildcards
oh sure
@toxic lotus @lyric zealot could I also friend request u lot also?
sure
sure thing 
Heck yeah thanks u lot
looks like its not randomizing
let me run on my end
and uhh
see if i get the same issues
i'm not super active here now tho 
I’m gonna do some brain storming on what type of gen while I shadow watch this conversation and see nice images u guys post
where ARE you active smol 
seed is set to -1 in all my prompts
not really active anywhere else tbh. But a little more active on a small anime ai server
might run it through genspark and have it see if theres anything that might cause errors.
I now have total of 4 AI friends
but it's a really really small community so yeah
fair enough fair enough
Thanks people

wait how do you know i'm ai?
trying incomplete prompts gives some funny images at times
oh wait, i probably shouldn't say that 
I played R6 for 9 years I got irl game sense now XD

gotta say, illustrious is not made for old naval ships + animals. wild scene tho
why are you using incomplete prompts tho
How I feel getting champ every new season
are you excited for the new r6?
😅
Yesnt all I do now every season get champ asap then never touch for the rest of the season I do it solo Q mostly
Peaked Champ #32 rn
accidentally left only trigger words iirc for this one
I don't play r6 so... but the new one looks pretty sick ngl
ooo edit that one
make uhh
make her hold a sniper
and give her a gas mask
and make it
pink
themed
instead of red

Was a Tier 2 pro for rank 2nd British esports team then had mental issues and got taken off the team but I won’t go there today XD
well, hope you are doing better than you used to
wait
Now I really play CoD mainly cuz arcade shooters let me respawn
i removed this:
(pale skin:0.8),long white hair,straight hair,long hair between eyes,pink eyes,white pupils,white eyelashes,pink lips,{black leotard,leather leotard,zip-up leotard,chest harness,full-length arm wraps,purple bandage arm wraps,arm wraps,purple hand wraps,leather fingerless gloves,purple leg wraps,leg wrap thigh highs
and it seems to be working properly not, ill let ya know @toxic lotus
ohh yeaah thats my own character
It was fairly recent so kinda better 😅
i forgot about that
join that server 
mbmb
all good
🍿
i got a whole batch of characters i made that i was thinking id slot in lol
6 kemonomimi idols and 5 non kemonomimi, even started making chatgpt write a story/script for em so i can try generating a whole manga for it at some point
lmao, you actually joined that server for the sticker
Tho me not being on the team so I don’t have to play 8 hours a day now and I can play other stuff like Tekken 8 and SF6 and CoD more
fuck yeah i did
That's the only way to live
Time to get nitro again u lot having to much fun with it
Guns ok with AI I’ve heard people have issues with it
im weird ong
fix the hand now
oh normally i have issues too lmfao


Ah ok
we dont make bad hands anymore
Long time thing then
bad hands dont exist anymore
we can fix it
5 fingers tho?
At least majority of the time hands are not so much of a issue
Unless ur on some complex dynamic poseor sum
well, it's a discord server for vencord, a popular discord client with addons
OOPS i notice why it didnt segment the hand
people have been trying to get me to use vencord but
i need nitro for the file size shit
I forgot about modded DC for free nitro features lmao
segmentmented the wrong thing
you can use nitro with it tho. Like how i am doing it now

why use both tho
im not risking vencord due tos violations imo
not sure if there's any nitro features that comes for free on it. But there's some pretty cool stuff for it
$10 nitro gives me everything i need
yup, most people don't wanna do that due to tos
I also didn't use it, surely 
check my account 
Ok I copied a prompt I wanna see the outcome on diff models
hella
thumb on the other side 
Text? I thought anything under Flux or SD 3 models struggled with that no?
struggle, yes. Impossible, nahhh
and then there's Harrlogos too
for base sdxl

it does words like "plap" or other short words it works™️?
U think u could do inpainting with Flux or SD3 for text?
i like how 'plap' is what you went with for an example
SD3 grosses me out
💀
its the one word its trained on really well
Well yeah but text wise it does decent enough no?
sd3 and flux gross me out
😔 I do enjoy Flux fine tunes especially RayFlux 1.0

Swarm released support for HiDream i1 so i wanna mess with that once im home tomorrow
O idk if i told you earlier but im home late late
like 1am type
late
I heard that takes lots of vram
yyeahh work
getting drunk for work is crazy dawg 🙏
yup but i think it fits in my 5080 (16gb)
So Flux does 3-4 min with TeaCache on my Mac I think HiDream will just kinda kill it
hmm since you are using the more complex stuff i do recommend SwarmUI once you switch to windows
On stability matrix what I use I just press a button
That's neat
In comfy people are saying they need to move heaven and earth to get it to work. Then comfy came out with base support for hidream today
I use draw things for flux I turn on TeaCache and enable it from 5-28 steps (30 total steps) at 0.20 threshold
oooff matrix breaks some stuff with swarm
hmmm
but I know it's going to break my baseline flow so I'm really reluctant to update
YUH
QUIT
TRYING
TO STEAL
MY FORGE USERS
YOU
MF



flux dev or schnell?
Dev models
love you too Zenia
😭
cuz its ass
I ran FP8 normal and then it just restarts my Mac cuz out of resources
So 8 bit convert it is
but loras
For me I love it for realistic people
:0 ommgg give me your civit <3
I’ll send u a cool DM @toxic lotus
ill follow
RAMTHRUST
i dont make checkpoints sadly
lol
oo the blackmagic one
i actually was gonna download that later
but id love to know the process
thats perfect ong




Followed both
I kinda sorta am the reason civit discord has a nsfw section...
Maxfield can, but will not attest to this.
i dont make anything yet, but its in the pipeline once ive built a collection of images.

omg so real
i should make more lora's
He is a busy man
i make loras as i need them
so i dont have many
but i made a bounty recently to have people suggest what i make
so i can get back into it
because i do have a list thats been staring at me for like a month-
once i get the patience to fix koyha_ss on the rtx5000 series im training locally
https://civitai.com/user/freckledvixon
I'm really good friends with this person, she's made about $10,000 through civit off of her loras.
jesus chrissttt
ive made
no money

can you uhh
do that
rn

Oh shit
i wanna train locally
followed them
I'm talking to her right now in fact
ill look into it when im home & sober lmao, imma nap for 2 hrs
LOL tell her i said hi ong, im her new fan
she's legit the queen of civit
why a re you drunk 😭
its like
not nowwww
idk what time it is for you
shhhhh
it's 4:20 for me so
just set up kohya
its 2:43am
AYEEE IM SMOKING RN
its 7 43 pm for me tho
consider me torqued tho

ong, used to use a lot of her styles in the past
loved the dappled light and harukami ones
i need to setup torch nightly and manually edit. ive been hoping theres a version out there someone forked 👀
ommgggg
someone submitted to my bounty
the food wars one
the lora i was too lazy to make
it doesnt look as good as the previously submitted one but
25 characters
and its available rn

gonna test it out
yee
imma cook

its in the lora folder
added metadata
sweett
this checks out
got some peak images but then its missing like one toe or is peak goon 
show
not the foodwars style
but






