#🍥|anime
1 messages · Page 289 of 1
dmd lmao
hmmm
hmm is automatic post processing cheating in image gen?
when i manually borked my comfy rig i could use automated LUTS but the new nightly one broke it

better colors
yeah, sometimes a depth map
^^^
i would say anything that happens pre gen or during, like controlnet etc isnt cheating, but if you automated like...photoshop or somethin, then maybe thats a grey area
XD
oh like certain steps in comfy to do like background removal, put on different background etc right?
evenin'
oohhh
ill never trust ai to do my photoshop stuff
currently
im still better than ai
i mean if you built a workflow that automatically used AI to detect artifacts or somethin, had it inpaint and correct, with no interaction from you, that would be cheating me thinks
ehe
if i need to photoshop an ai image
working on it still
i plug in my drawing tablet
hope it fucking works
then manually draw/or remove backgrounds/artifacts
buuuuuutttt who cares right, its AI making the image from our prompt, so shouldnt it also fix its fuck up lol
krita SD plugin > everything
ONG
ive tried that but
i always go back to forge
Or have an AI do it for you. Work smarter not harder
oh yeah i got ai to do the contents
local llm
now your cookin
just need to actually make / design it
yeah AI sucks at the execution of that step. always outta order and shit...
le gasp
#exposed
she shunned me for using A1111
oh ill shun you for that too
be still my beating heart
i do it again
a1111 is ass
obsolete
trash
im so laaaazzyyyyy tho...i dont wanna swap to forge...unless i can just copy my stable diffusion folders into it
unoptimized
the models and loras?
you can
yes?
oh
its the exact same as a1111
oh my
forge is even a fork of a1111 lol
what forge?
I'm stoned and just walking back in
hrmmm
yeah forge is a1111 but way faster
im smoking a spliff and cig rn
dual wielding
speed running lung cancer
im trying to get them to use swarm instead but zenia is preventing it
I can't smoke anything, wife is preggers!
o congrats
Thank you!
@sharp imp this ^
so I'm using tincture because I also am avoiding sugar so it's interesting
no clue what that is 🙏
hmm idk if we can get thc oil here
*under the tongue
hmmm
oh goooddd
reminds me of that free sample i got at a dispensary
100 mg liquid thc
guess ill find out once i actually walk into the "coffee shop"
i OD'd ngl
💀
wildest trip of my LIFE
10/10 would do and regret again
i slept so gooooooodddd
Back to SwarmUI tonight.
hell yeah
There's some stuff that's a little bit more difficult than it needs to be in Invoke.
but it's definitely got the polish
the img2img and inpainting is certainly easier
the lack of auto segmentation right in the prompting was annoying
and the file management is kinda weird
and either will do images of burger consumption
in the process of getting forge....gonna see how much of a pain it is to get all my checkpoints etc. should just have to set a path for it to share right?
should just be as simple as copy and pasting your files into forge ngl
the important ones
like extensions, outputs, models
ahhhh....i see
but then its gonna take twice as much space correct?
unless i delete A1111
you can configure them to share a model folder
are we talking forge or reforge?
forge has sketchy obfuscated code, so good luck with that
@echo off
set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=
@REM Uncomment following code to reference an existing A1111 checkout.
set A1111_HOME="D:\Downloads\A1111 Web UI NEW\stable-diffusion-webui"
@REM
@REM set VENV_DIR=%A1111_HOME%/venv
set COMMANDLINE_ARGS=%COMMANDLINE_ARGS% ^
--ckpt-dir C:/Users/User/OneDrive/Desktop/Stable_Diffusion_Models ^
--hypernetwork-dir %A1111_HOME%/models/hypernetworks ^
--embeddings-dir %A1111_HOME%/embeddings ^
--lora-dir %A1111_HOME%/models/Lora^
--vae-dir %A1111_HOME%/models/VAE^
--realesrgan-models-path %A1111_HOME%/models/RealESRGAN^
--esrgan-models-path %A1111_HOME%/models/ESRGAN^
--no-gradio-queue^
--cuda-stream^
--pin-shared-memory^
--cuda-malloc^
--api^
--listen^
--enable-insecure-extension-access
call webui.bat
you can do something like this for your webui-user.bat
use your own flag tho
Still here, just doin fuck shit to get everything copied over, moving old webui stuff to the 4tb external, and setting up output path.
I may pick up another 4TB NVMe
I need to, but right now im just storing outputs and checkpoints im not using regularly on an external SSD.
Yeah, 3:43am for me
And i show up when summoned usually
stored 
Hehe yeah and I've been busy so i haven't been genning a lot on my own

that will definitely get deleted if the server was how it used to be, now i don't know
interesting. its definitely suggestive, dunno if it qualifies as nsfw (im basing this off of what a gen like SeaArt AI would allow, ive seen worse for sure there)
not really talking about your gen, but this instead #🍥|anime message
but well, if it's like before, your gen would also likely gets deleted, but then again, it's a lot more lax now tho
I see. Ive intentionally removed the nsfw wildcards from my outfits, and gave them their own wildcard folder to avoid anything blatant. But if anything i post violates, by all means remove it.
Its s lot more lax because other then #🤝|tech-support this is the only channel with any resemblance of a community
id post in the other channels, but i only really make anime atm XD
Other channels don't get as much traffic ngl
fair enough
and also the fact that not much mods moderate here
Theres mods? I thought it was a 2 man show of fruit & maxfield lmao
yup, just them and community guides i guess. Last time, i don't think the community guide can do anymodding, but now appears that they can
i got wildcard working, but not using what you linked, but using dynamic prompting instead
maybe you linked the wrong one? 
like im using this instead for wildcard
lol that satanic wildcard prompt
nah its not
im gonna let it run a batch of 50 or so images while i sleep, and see what i get in the morning lol
lmaooo
bet bet
@lyric zealot this one?
i got forge up and running, and copied all my stuff over, then sent the old webui to my external to chill so it aint eating space
yup yup, this works for my reforge ui


judge me not 🙏
for i am insane
thats one im working on rn
just give me a few days
itll match my forge one
batman, covered nipples, destroying burger, huge balls

Balls?
Make batman love the burger
the burger returnnnssssss
just give me the weeds pleases
doin the glitch eyes, thats a fun one

oh lawd she got so much neck
This one is a cool style
What could be the good token for this type of mouth ? Pouting ?
Yeah thats pouting
Puffed cheeks could also help
Realistic models don't understand pouting and produce bigger lips instead 😅
Hmm because it's hard to translate cartoon features to real
I try, when i'm bored !
yes this is strange
@fickle oyster enough quality upscale? ^^
:3 Hey how are you!
"fine" -> if having a slight crack in the between vertebrae discs can be ignored (which i can't), dunno the exact term
not full hernia, but not quite ok either
Oh that fucking sucks... You getting any referred nerve pain comorbidities?
Also, I got banned from Civ

yeah, i read about it in your thread
you naughty boy
Yeahhhhh he brought my future fatherhood into it and that was completely uncalled for.
so I flipped and he kept doing DM drivebys and blocks
like 4 times.
I knew him pretty well, well enough to send him money over the net.
i suppose you mean demian?
Zed
ah, zed
Demian was also in that server
yeah zedtheeviltaco.
perhaps exchanged a remark or two? but that's about it
Yeah. Didn't realize he'd be such a turd.
But hey, let's not let anything spill over into yet another server :3.

i was about to mention that

DMs are always open
I'm planning on making a server for you guys
It's in progress
No one came to his cookout
Or hers, pretty androgynous. Curse you illustrioussssss
@toxic lotus i had chatgpt deconstruct,or rather reverse engineer your scary prompt, and create another one that wont veer off into nsfw territory unexpectedly, and included my database of wildcards with yours and some of the outputs are hilarious.
LMAOOOOO let me see that shit 👀
Well the one above was one of them. Im batch generating now. Or did you mean the prompt?
Both
"""Smooth Quality,lazypos,lazynsfw,lora:Eyes_Lora_Pony_Perfect_eyes:0.7,lora:TrendCraft_The_Peoples_Style_Detailer-v1.0A-3_13_2025-illustrious,(masterpiece, absurdres, amazing quality, very awa, best quality, ultra-detailed, very aesthetic),{masterpiece, very aesthetic, absurdres, newest, 1girl, solo, female, {female_hair|hair_styles_long|hair_styles_short}, {hair_color|hair_color_duo}, bangs, {eye_color|eyes_iris_colors}, eyes_pupils, breast_size, { female_clothing_general, {female_clothing_cyberpunk|female_clothing_gothic|female_clothing_fantasy}, {bottomwear|skirts|legwear_main}, pose_casual, pose_dynamic }, {animal_traits|ears_other|tail_types_mammals|tail_types_supernatural}, {accessories|makeup|added_details_character}, {urban_background|fantasy_background|sci-fi_background|interior_background|natural_background|abstract_background}, {angle|anime_camera_techniques}, {lighting|special_effects} | masterpiece, very aesthetic, absurdres, newest, 1boy, solo, male, {male_hair|hair_styles_medium|hair_styles_short}, {hair_color|hair_color_duo}, bangs, {eye_color|eyes_iris_colors}, eyes_pupils, { male_clothing_general, {male_clothing_cyberpunk|male_clothing_gothic|male_clothing_fantasy}, {bottomwear|legwear_main}, pose_casual, pose_dynamic }, {animal_traits|ears_other|tail_types_mammals|tail_types_supernatural}, {accessories|makeup|added_details_character}, {urban_background|fantasy_background|sci-fi_background|interior_background|natural_background|abstract_background}, {angle|anime_camera_techniques}, {lighting|special_effects} | masterpiece, very aesthetic, absurdres, newest, anime_character, solo, {female_clothing_general|female_clothing_cyberpunk|female_clothing_fantasy}, {urban_background|fantasy_background|abstract_background_light|interior_background}, {angle|camera_unique}, {special_effects|added_details_character|mood} },"""
so theres the prompt lol
it still took away the __ : /
i hate discord, but you get the idea. i could probably just share a .txt file of it and that would be easier
aight its gettin weirder
You have to use tilda, or the key beside your 1 key.
Theres never enough ramlethal images ong
~~?
Wait no, it’s grave key
I mixed them up

ive been betrayed
You can do 1 at the start, and another at the end, or 3 of them at the start and end
still dont know what the grave key is, none of em look like tombstones to me
or maybe all of them do
"""Smooth Quality,lazypos,lazynsfw,<lora:Eyes_Lora_Pony_Perfect_eyes:0.7>,<lora:TrendCraft_The_Peoples_Style_Detailer-v1.0A-3_13_2025-illustrious>,(masterpiece, absurdres, amazing quality, very awa, best quality, ultra-detailed, very aesthetic),{masterpiece, very aesthetic, absurdres, newest, 1girl, solo, female, {__female_hair__|__hair_styles_long__|__hair_styles_short__}, {__hair_color__|__hair_color_duo__}, __bangs__, {__eye_color__|__eyes_iris_colors__}, __eyes_pupils__, __breast_size__, { __female_clothing_general__, {__female_clothing_cyberpunk__|__female_clothing_gothic__|__female_clothing_fantasy__}, {__bottomwear__|__skirts__|__legwear_main__}, __pose_casual__, __pose_dynamic__ }, {__animal_traits__|__ears_other__|__tail_types_mammals__|__tail_types_supernatural__}, {__accessories__|__makeup__|__added_details_character__}, {__urban_background__|__fantasy_background__|__sci-fi_background__|__interior_background__|__natural_background__|__abstract_background__}, {__angle__|__anime_camera_techniques__}, {__lighting__|__special_effects__} | masterpiece, very aesthetic, absurdres, newest, 1boy, solo, male, {__male_hair__|__hair_styles_medium__|__hair_styles_short__}, {__hair_color__|__hair_color_duo__}, __bangs__, {__eye_color__|__eyes_iris_colors__}, __eyes_pupils__, { __male_clothing_general__, {__male_clothing_cyberpunk__|__male_clothing_gothic__|__male_clothing_fantasy__}, {__bottomwear__|__legwear_main__}, __pose_casual__, __pose_dynamic__ }, {__animal_traits__|__ears_other__|__tail_types_mammals__|__tail_types_supernatural__}, {__accessories__|__makeup__|__added_details_character__}, {__urban_background__|__fantasy_background__|__sci-fi_background__|__interior_background__|__natural_background__|__abstract_background__}, {__angle__|__anime_camera_techniques__}, {__lighting__|__special_effects__} | masterpiece, very aesthetic, absurdres, newest, __anime_character__, solo, {__female_clothing_general__|__female_clothing_cyberpunk__|__female_clothing_fantasy__}, {__urban_background__|__fantasy_background__|__abstract_background_light__|__interior_background__}, {__angle__|__camera_unique__}, {__special_effects__|__added_details_character__|__mood__} },"""
Nailed it

But if you are intending to do multiple lines stuff, it’s better for 3 of them instead of 1
Its got variety, im rather enjoying this wildcard prompt. I might fiddle with it and make more lol
oh i love this
Ya that'd def be easier lmao
Sorry I was at work
Oh sweett
🙏
I shared some of the best gens so far. A lot of stuff that got a little wild, some that simply didnt make sense and very few that just broke the model lol so all in all id say a successful test run
Yeah I might need to retweak it tbh
I think the mast time I used it I was working on it and may have broken it
Black Magic is doin its thang
Love me some gritty, darker images. Most models do vibrant or dreamy too readily, sometimes i just wanna suffer
Black magic os the model?
Nah Rams model Black Magic Illustrious i believe its called. Its got great colors, im enjoying it
Ooooo bet
Hey people I love the photos in here I’ve just been scrolling up
A thing I’ve seen some “aesthetic” loras around on Civit AI but I have no clue if they are good or not is aesthetic mainly prompt based putting in what u want or can Loras like that help?
Got to try it also
So aesthetic or style loras influence the overall image more so than just a specific character or concept/pose lora. Things like color palette, linework, shading and background will be impacted. You can also use aesthetic as a quality modifier (for example; "very aesthetic") in the prompt without a lora. It will subtly nudge the model toward a more aesthetically pleasing composition. I recommend using the recommended tags for checkpoints or models that the creator suggests, but also familiarizing yourself with others used in different art styles (you can check danbooru for a tag list) and testing them out in your generations.
@fickle oyster made it, and a few others. I recommend checking them all out. They've been really fun to use so far for me, and it's a nice breakaway from my usual checkpoints.
Ah ok thank u appreciate it also DM u a aspect ratio question that i kinda need help with since i was tryna figure it out last night
Appreciate
I replied. I used custom settings to get to a custom ratio that @keen hollow suggested a couple days back, and it turned out decent. But I think that was a resolution upscale not a aspect ratio one. If you need to, you can math out the pixels at a lower resolution to fit your ratio, then upscale by 2x or 4x in the extras tab til you hit your desired resolution while keeping your ratio in tact.
Thanks dude!
No problem! Love the work. I've wanted to get into training and merging and stuff so it's nice to see someone who's actually done it. I'll be posting some of the stuff I make to civitai and crediting your checkpoints so hopefully it drives some more traffic and appreciation your way.
Cheers my friend
Was that the 3840x1600?
I like this one 😄
Yeah, i had to upscale to that resolution lol figured the same strategy would work here.
work where?
The above instance, sorry forgot half that conversation was in DMs lol. She was asking about getting her aspect ratio to match like cinematic wide ratios (21:9 i believe).
21:9 = 1536x640 just a super quick note, then upscale after
Ah ok and upscale u mean hi-res fix or a different form
ah your on forge right?
on the extra's tab theres an upscale option iirc
hi res fix is refining and upscaling
HI-RES is close to what I do
But I do an iterative latent upscale in comfy... it's a hell of a flow
Yes
I would but I’ll get a Nvidia GPU before I have to deal with 20 nodes
Oh I see it it’s cool but I kinda don’t know what to do lmao
hmm id wait on Zenia to come back
o im fine thanks for asking!
shes working rn so were chilling
im a swarm user so i cant really explain how i do it
Ah ok
Can I use swarm on Mac with decent performance or u think it’s best to stick to one Ui that works and learn it since I’m also “new” to image gen
“cfd” and “runity” is crazy
😭
hmmm i recommend sticking with what you know for now
once you get that windows build though 👀
ill help you set it up
hey does well
woah thats really cool how did u get that machine hybrid?
mechabare, see-trough body, exposed wired internal machinery
So prompts no Lora that’s cool asf
I’ve only seen this when people use Loras
eww guts 😛
using modern models means much less need for loras 🙂
"anime style, see-trough body transparent skin, exposed wired internal machinery" with flux
i guess this is SFW - no naughty bits, machinery...
I see
Prolly why I’m always download need Wai updates
Since it’s my fav model
And it gets better each time
Text is quite nice
Was this prompt or u did some inpainting
Thanks auto correct
yep no loras, just illustrious prompting
Ah I see thanks
Steps: 7, Sampler: Euler, Schedule type: Simple, CFG scale: 1, Seed: 3460836797, Size: 896x1152, Model hash: a40a0a48fb, Model: fluxFusionV24StepsGGUFNF4_V2NF4AIO, Version: f2.0.1v1.10.1-previous-659-gc055f2d4```
using flux (or any other model) is like magic if you're used to SD 1.x or even SDXL
Flux fusion is anime flux fine tune?
I use flux a lot for realism
just flux fusion, nothing anime specialized, very generic, does it all
So I’m used to flux cuz prompting is a like MJ which is what I come from but other models that’s not flux seems to like tags more so getting used to that
going semireal, same seed 🙂
Steps: 7, Sampler: Euler, Schedule type: Simple, CFG scale: 1, Seed: 2116267029, Size: 896x1152, Model hash: a40a0a48fb, Model: fluxFusionV24StepsGGUFNF4_V2NF4AIO, Version: f2.0.1v1.10.1-previous-659-gc055f2d4```
I’ll download FP16 version and use draw things on Mac for 8 bit conversion and give it a try
last one... same seed still flat 2d cel shaded anime studio ghibli style, speech bubble text "No need for LoRAs or tunes", neko female giving thumbs up, futuristic data center
who needs loras... who needs specialized models... right? 🙂
True
I also keep forgetting fp16 flux is 22gb usually on models
😭
Ah see DT don’t support GGUF
Only standard
DT ?
right... fluxFusionV24StepsGGUFNF4_V2NF4AIO is my favorite
A: 13.66 GB, R: 14.03 GB, Sys: 15.7/23.9883 GB (65.4%)```
I wish they did but the dev said they have no need cuz it doesn’t give any advantage they don’t have on the speed on app
Draw Things
ahhh
why not use Auto1111 or forge then?
Also recently added TeaCache which helps me a lot
forge does support gguf
Idk why but some reason flux takes longer on Forge than it does on DT
Maybe cuz Draw Things is apple hardware full optimised application
Not sure
I don’t know the actual tech behind that
probably
Lmao
I would assume so since u can’t use it on windows or Linux only Apple devices
What was this pose call and the model this style was related to as I think u mentioned it before
the model is IllustriousBlackMagic v3 CLEAN, and i didnt check the metadata for the pose cause im running a wildcard nested prompt in batches lol
but i do have files for every image that contains the prompt, so i can probably find it
its either top-down bottom-up or jacko challenge
Smooth Quality,lazypos,lazynsfw,<lora:Eyes_Lora_Pony_Perfect_eyes:0.7>,<lora:TrendCraft_The_Peoples_Style_Detailer-v1.0A-3_13_2025-illustrious>,(masterpiece, absurdres, amazing quality, very awa, best quality, ultra-detailed, very aesthetic),masterpiece, very aesthetic, absurdres, newest, 1girl, solo, female, crew cut, Wine Red, hair between eyes,face framing hair, x3, Yin-Yang pupils, flat chest, Sweater Dress, Cardigan (Open), ankle boots, Tribal shaman attire with feathers/beads, Woolen tunic and skirt, Knee-high leather boots, Steel gauntlets (feminine design), kneehighs, Lying down relaxed on back, Yoga Warrior Pose , Dragon tail, Earbuds (Wireless), Subtle Noise/Static Pattern Background, Dramatic low angle hero shot, crackling orange energy field around the character ,
this is the prompt
well it did not follow that prompt well at all lmao
must have been an interpretation of Yoga Warrior Pose and Dramatic Low Angle Hero Shot
yeah, ive seen that one used more for nsfw posing than anything
like if i didn't prompt for clothing, it will straight up gen without clothing
im learning that even if i do prompt for clothing, some models are just like...nah not this time lol
that's true 
if you find that a problem, try to put your clothing prompt higher up
or should i say, nearer to the start, or at least at the first 75 token
god forbid you only specify one article of clothing, its a toss up whether it crops the shot to show that piece of clothing or if it goes full body and shows only that piece of clothing lol
this is actually good advice. the way i structure my prompts its usually quality modifiers > subject and subject details > clothing > pose > etc
yup yup, i usually put quality modifiers first as well, but that depends on the model.
and you should also know that it processes your prompt 75 tokens, then the next 75 tokens. So meaning, if your clothing is on the 75th token and the 76 token, example: blue sweater, it will not understand blue sweater, but actually blue, then sweater
i didnt know this, thats actually insanely helpful
so you might find people using BREAK to intentionally fill the remaining tokens with padding, to skip to the next 75 tokens
that makes sense. i was wondering if they were using regional prompting, since thats often where i use BREAK, but their prompts didnt seem to align with that. the 75 tokens rule makes a lot more sense.
just also know that since the first 75 tokens get process first, they are higher in weight as well (or should i say, more likely to appear than those after that 75 tokens)
cheers
i think one of the names for that pose
is jacko sum challenge pose
i remember sum a while back on tiktok a challenge was with that pose
yeah, i got a few friends that told me about it lol the other one that youll see from time to time is the "yoruichi pose"
Yeah when that challenge went to instagram since uhhh IG has “bad” NSFW style auto censor It got more provocative with that pose
lol i can imagine. i dont have an instagram, i suppose i should if im gonna do anything social media-wise, but it just sounds like so much work lol
Lol what a clash of styles XD
For social media work Yh IG might be needed but yeah bro the NSFW auto bot is terrible like u can get real close to borderline uhhhh human body part area showing without getting restricted
I’ve unfortunately have seen some on my page instead of memes
Because the meme page hijacking part of instagram a while back messed up a lot of my fav pages and turned it NSFW
Tryna watch my words with what I say in this server
i mean, its an odd territory since different people would say what they consider nsfw is different. since its a sort of spectrum or subjective opinion these days, they should allow you to set your own filter for posts. would allow people to curate their experience better.
True but a lot of IG is just soft core if they could do twitter bit with actual NSFW warnings unlike Twitter sometimes adds it or sometimes just shows u it that would be better if they are gonna let this stuff happen
yeah. other problem is, sadly, nsfw content or nearly nsfw content gets reactions and engagement. its a similar thing to what ive noticed about Civitai. Gooners be goonin, and that affects algorithms and policies.
Yep
I think also just a filter for what rating u wanna see like civit on social media apps
Would be good
I mean knowing me I’m not gonna mention it here obviously but I wouldn’t have filters enabled lmao
they ask your age, but it seems irrelevant on most social media apps as to what content you can find
lol and that should be a choice as well. but letting an algorithm decide for you is kinda lazy
Yeah. Its funny, ive had better success with cyberpunk or futuristic outfits when i didnt intentionally prompt for it than i ever did
Aw man I lost the hand detailer prompts u gave me can u dm both face and hands so I don’t lose them again @sharp imp
sure
Thank you
Oooo cool I like how much detail this has
The glitch effect is my favorite part. Well that and the ripples and color contrast. @lyric zealot is this a model you made/are making ?
Glitch effect is great
Especially using it a bit like motion blur
no, just straight up from civitai 
i think there's metadata for that 
1536x640 I’m loving as my new res cuz of 21:9 aspect is my fav
i mean, i'm also using a lora, but that's mostly for the face
Yk I’ll give the “aesthetic” Lora a look and see if it actually does anything more to my gens
I use 832x1216 for portraits and 2560x1440 for my landscape/wide shots.
Straight up 1440p res fair enough
I’m not gonna push my Mac a lot when it comes to hi-res
Maybe I would’ve if I decided to get a 32gb ram or higher model
well 2560x1440 is my monitor resolution, and since i started doing wide shots to make wallpapers for it i just decided to stick with it lol
Ah that’s fair my pc monitor is 4k 240hz but I play games in 1080p res unfortunately due to the GPU I have
Unlucky me got hit by the high prices during lock down
I had to get the 6600 xt no other option
i snagged my rtx 3080 right before all that covid shenanigans thankfully, but my ps5 still ended up costing me 1300 due to shortages (had to buy from an individual seller lol), not my smartest decision but meh
Ah I got my ps5 when GoW rag came
So I bought it of Sony directly a ps5 disc GoW rag bundle
Loved the game
ill be honest, havent played it XD i bought the ps5 almost exclusively for FFVII Remake haha
Then I bought GT7 and love it
I’m very much a cat person
Well car
But apparently auto correct knows I’m a cat person also
XD
As ive gotten older, ive gotten more into working on my car. But thats cause i dont wanna pay another 20k lol
Ah that’s fair I’m 19 gonna be 20 in June
Hopefully eventually I’ll be able to get a nice car and tune it
But rn I rather enjoy work as much as possible
tried your resolution out and upscaled to 3072x1280 lol
Ooo nice
I just do 1.6 upscale
So 1536x640 to 2456x1024
yeah i shouldnt do 2x at this size, it slows down the gen a bit lol
Yeah I’ve tried it on my Mac
It does quite a bit
I have a question @lyric zealot since u use forge for hi res CFG should I use the same CFG as the normal one or should hi res be lower than normal one
i let it stay as default, so it's using the same cfg is my generation ones
Ok so my gen is 7 CFG so my hi res is also 7 but both started on 5 on default settings before I got help from u guys in my forge setting
So hence I was asking
I have no issues with it just wanted to know
well, you can try it if you want, i don't know if there's a need to change cfg after generations before hires fix tho
I’ve seen but I have stuck with 7 on all models u guys have given me
And it seems fine
I don’t wanna mess things up to much before I know the proper basics of stuff
heh testing it out. some of the gens are a little....terrifying, but i think thats cause the model probably isnt used to the resolution
well, do a plot xy to test it, you won't screw up much if you are just changing one setting at a time
so cfg guides prompt adherence, the higher the number the more it follows the prompt. the lower, the more creatively the model will interpret the prompt. 5-7 is the "typical" recommendation for most anime models. But there are some that suggest 3-5.5, and ive got a couple that are 1.5-3.5.
Cuz hi res on this looks completely different on draw things which is what I use for flux and before forge which looks like this
which does not look nor actually perform the same as forge
in hires, the cfg just does the same, guides the hires stage to interpret the prompt literally or creatively, and combined with the denoising strength can result in it changing the original image more or less.
tho forge in settings has got a repro of DT
what's the difference between this and forge? 
that i can actually understand forge
and get results thats good

im getting better at tag prompting
you are using auto-complete extension right?

well, with that extension, you can stumbleupon other tags, and even go explore it on danbooru, but often result it's nsfw
example, and if you press that ?, you get directed to danbooru
can u send auto complete over?
safebooru is a non nsfw alternative, if you wanted to look for tags/inspiration
go to Extensions > Available, press Load from:, search for auto complete
yup, but for that extension, it auto directs you to danbooru, you can probably change it tho
i dont see it
oh yeah, theres also the dynamic prompt "magic prompt" but it gets a little crazy
tagcomplete instead?
and some of the magic prompt repos get REALLY nsfw lol and not the good kind
nope
do you already have it downloaded ? 
well, yeah, because it's using some llm to do so

i think not i dont remember that download
true
check your Installed tab under Extensions
nope
ill restart after my gen
it's basically this, if you search it
i was about to share the same clip XD
Despite 21:9 taking longer imo it produces my favourite result
i flipped it to 9:21 and its not terrible lol
It’s just extremely long?
Also I think I really like black magic it’s closest I’ve seen to Niji 6 from MJ style native
kinda? i had one image where it looked like it stretched the hair but that could have been because there was a cross in the background lol
@fickle oyster really great model my friend
Ah I see XD
yeah its really good. i like the depth of field with the grass in the foreground and the mountains in the background. makes it feel more dynamic
yes for sure
Thank you! 🙏
researcher
slightly disturbing
very cool

have a big ear neko 🙂
(no loras, flux fusion with "flat 2d cel shaded anime studio ghibli style,creative chaotic unusual epic dramatic scenery, wholesome neko woman with comically large multicolored ears, sultry look, tsundere")
not amused
poor totoro
\
Amazing
@sharp imp thx for the help i kinda got what i liked here

NEw ilustrious V2 seems pretty good (no upscaling, adetail and some natural prompting support)
Compared to the fine tunes of others does this compete or are the others better?
\
this one is way better
1536x1536 gens
no after detail
refinement or upscale
Ah ok I’ll download it link?
Ty
Might be my new fav model who knows
W
Dataset making it sound like open source Niji
XD
still suffers from sdxl quirks though but thats great
again havent adetailed yet
and bad prompt lol
No hi-res and adetailer?
yup
What steps and cfg u use?
Ok ty
DPM++ 2m
Was gonna ask XD
Compared to what I use now Euler A
I wish mine would gen that fast XD
7s but my hardware is probably better
Me with a MacBook says no
with 5080 n all
but you can use euler a
luigi from the mario brothers standing on a giant green pipe, standing with zero suit samus,
like i said bad prompt
I still wanna know how fast the Mac Studio M3 ultra with 128gb ram would be
Vs Nvidia GPU with DDR6 64gb or sum
euler A natural prompting falls off
Rn the M3 ultra with 512gb ram is £11k of apple XD
Ah i see I’ll use the one u used first and see how it goes
at 1024x1024 is also truggles a little with following it more closely
but eh its probably a seed game
Ok I’m doing mine now
No detailer or hi res 1536x1536
If anything it will be my 16gb ram that will be the issue
If it fails to gen
Unified memory
Yep
Anyways it’s 13% done
hmm it works really well with artist tags too
I don’t gen this high of a res so it will take longer I’ll assume
hmm at 1536x1536 yeah will take longer
1024x1024 also works fine
but its less ai prompts
Btw ur gen would’ve been done 8 seconds ago by time I reached 23% now XD
43% chat
XD
If we used same settings and I had a 3090 roughly how long u think it will take @dusky linden vs ur 5080?
hmm honestly? with SDXL not a big difference i think?
since it fully fits in the vram
hmm maybe a little slower
So if I can say Yh I have 5080 performance kinda (only on SDXL)
Just need to find the perfect 3090 used deal now
i wont be able to guarantee that since i never ran on a 3090
only on a 3070TI
but on a 3070TI i had with like 2 lora's 20s gens
I mean sub £600 for 3090 I’ll take it
Fair enough
hmm honestly id save up a little more for a 5080 then
prices will normalize soon
On 1024x1024 does it do eyes well
Prolly these are current 5080 prices in the Uk
SS taken yesterday
oh yeah thats better since blackwell is newer and will last you longer
This is 5090 😭
hmm it is ai afterall still
Seems like they really like watermark stuff
oh no thats the cutesexyrobutts, tag
hella poisoned with watermarks
but its one i knew out of my head lol
tag based ones work just as well
I’ll assume they have a lot of styles and artists baked in?
yeah its like illustrious
so danbooru
but this time added natural language support
so aka "prompts cooked by llms"
try remaking one you did before
with the other models
I will but would it be better with 1536x1536 without high res or 1024x1024 and then do 1.5x hi res?
up to you fam
I’ll keep 40
CFG scale 5~7, steps 28~35 recommended,
For now
but i dont follow guideliens
Can I try 1536x640 and hi-res 1.6x with 10 steps if u can im gen sum diff as of now
It may take a while
😅
send me a prompt and ill give it a go
Ok
running one Zenia did before (she has great prompts)
(ultra detailed), (masterpiece, best quality, amazing quality, highres, absurdres, very aesthetic), chiaroscuro lighting, heavy shadows, heavy shading, hu tao (genshin impact), bruises on face, blood on teeth, eye scar, subtle wind, particles, dark background, dynamic pose,
If yk what else to add to make it more cinematic u can also try that and tell me what makes it “better”
dropping the res to 1536x1536
ran 35 steps, no lora, no hires, no adetailer, natural language prompt, with the same cfg/sampler/scheduler you suggested and the creators suggested negative prompt.
1536x1536 also
Ik it just came out but I can’t wait for 3.0 XD
Cuz 2.0 is great
She wears a celestial dress made of stardust and moonlight,trailing behind her like a comet tail. Glowing particles swirl in the air around her feet,and her reflection ripples across the star-lake beneath. Her expression is calm,eyes closed,at peace in this impossible place.,
The background is a surreal night sky filled with stars,galaxies,and faint auroras. Soft clouds hang low on the horizon. The color palette leans toward pastel purples,soft silvers,and deep cosmic blues. The scene feels ethereal,like a half-remembered dream,suspended in silence.,```
personally, i removed the ( ) because that doesnt do anything imo
only when its (ultra detailed,1.1)
Ah ok I just copy and pasted from its description
its supposed to softly nudge the model, like a really weak weight so ((( ))) would nudge it harder, but i feel like it doesnt do anything for quality tags
be different if you used it on eyes
hmm maybe its a forge speciic thing
Ooo nice it’s 1:1 res but I kinda asked for 21:9 aspect but it’s fine XD
Yay ty it looks great
also you were asking about gen time for a 3000 series gpu, i have a 3080rtx and it took 30 sec
running again with adetail
I think I’ll save up for 5080 but then will my pc be an issue? @dusky linden
so theres a big difference in the 3k and 5k series lol
hmm youd probably want a new PSU but honestly i havent made a new rig in ages
Ik that one I was gonna do that and ram anyways but the cpu is 5600 on a B450 board
my rig is like the ship of theseus at this point
adetailed
it added another scar, thats all i can tell lol
ahhh little sharper line art
but not terribly noticeable unless your looking at them in a lapse like that lol
gonna try some nsfw ones, see if they used safebooru or not
I’ll assume swarm has this adobe lightroom video maker on edits thing?
yeah i had lazynsfw in my negative prompt so i wouldnt get anything i couldnt share lol
oh no i just used the snipping tool

but i have LUT's but i didnt use them yet
i just upscaled to 3072x3072 with hires and that added soooooo much time haha. might go with a 1.5x instead of 2x
same prompt
1024x1024 and its usual variations work fine
but if you base it on 1536x1536
only ultra wide, simple prompts
otherwise mutation time
like the good ol 1.5 days
gonna test it with my nested wildcard prompt
Width: 1824, Height: 1152 squares or papes only with this high
mutations hella
hahah yeah

good ol two headed mutation
cant post the full examples due them nsfw (good anatomy tests)
good ol detached head too
malformed bodies
etc
BUT what it makes it does GOOOD
lol had an image that was slightly nsfw with an extra random shoe in it.
random artifacts, but thats kinda a fault of the wildcard prompt too. it creates some pretty wild prompts
oop my lora broke it
specifically prompting what it messed up fixed it
the clothing details in that second one are really nice
them hands doe
tactical gear is fun. you using open pose for the holding the rifle or just natty prompting that?
loras were at fault
didn't even prompt for it or use open pose
it's random 
nice!
some people have all the luck, ive not gotten a single decent holding a violin bow or drawing a drawstring in any gen
tried out 4xfoolhardy remacri with the last batch and got these beauties







what do you mean?

