#🍥|anime
1 messages · Page 285 of 1
The crab lady is kinda fun haha
looking great 
Thanks! Tried to find a couple more sfw examples but had none on my phone lmao
Considering you have a 4090, you might be able to run it at full 720p instead of 480p
@fickle canopy my supermerger only works in a1111 but a1111 broke, do you know of any other extensions available for forge or reforge capabale of merging loras specifically
I've been mostly using ComfyUI for the past 8 months now. And I've never used SuperMerger. But you can try https://github.com/wkpark/sd-webui-model-mixer
sweeeetttt thank you 
it was taken down 🙏
yeah, but why take 2-5x so long when most people just see the small preview here anyway 🙂
btw, much better result now!
That does look a lot better. Did you change any settings?
same settings, just another prompt. still feels quite random results-wise, not like with text2img where you at least have some level of control over the output
I was going to upgrade my 3060 12GB card, but honestly, this 5000-series launch has been so abysmal.
I'm tempted to look at those new AMD cards coming out instead. But my recollection is that AMD support for AI is awful. 😦
pretty quiet here these days... sometimes wonder if I'm posting to an empty chasm.. 😛
animated music video of a song i made within udio https://www.youtube.com/watch?v=2DetIbpK64Y
youd be right
AMD is pretty bad for AI
well if compared to nvidia
but its okay for image gen i suppose
also yeah but the peeps i know that post are probably at work or sleeping rn lol
i still lurk sometimes. just busy with other stuffs & servers.
Discord itself feels like that these days. I thought I had some active servers, but most are ghost towns.
naiv4 full releases today though apparently. will probably end up testing that all day long & have things to post again xD
NAIv4 apparently just uses the FLUX VAE (according to their own posts), so why wait for NAI?
ooo let's see what you come up with
this is fun.. I like the pixel stuff
sad. 😦 nvidia not being very pro-consumer 😦 .... not like I could get one anyways
been playing around with it for a bit to see wut it can do
the style control feels a lot better
inpainting also seems way bettr
once i think up some stuff
servers feel a bit slammed rn since they just launched xD
Kirbz
According to rumors prices will drop in a month or 2 and supply will stabilize
there's even a "there will be overstock and oversupply and thing will get really cheap" thread out there since their server GPUs seem to be doing not that well... let's see
Well i managed to snag a 5080 for msrp ish prices
Wanted a 5090 but with all the issues? No way
that's cool - for AI more VRAM would be better but the 5090 is really ... special. happy with the 4090 for the time being 🙂
In upgrading from a 3070ti so its a big upgrade for me
And if i really want to run the heavy stuff i can rent a cloud gpu lol
Oh yeah but if i rent a cloud gpu i just use their gpu/comfy ui in the cloud that gets scrubbed after a few hours
And since im not making anything wrong theres no problems lmao
Can even use my own models/loras
I like your optimism. 🙂
well i managed to get a 5080 at msrp (not FE) things are looking a bit positive
have you done this? online rent?
not yet but Crystalwizard has (and adviced others)
im going to for work tho
I was tempted on the 5090, but that price, and with all the issues, I just.. can't. Power management over the cable is all jacked up, melting is worse, the PhysX drop, the missing ROPS. What a complete cluster.
Nice, I have really considered this for ease of use. Maintaining the local setup is kind of a pain.
You could rent a lot of GPU and still be a lot cheaper than a 5090 😛
So yeah, more likely to go maybe 5070ti/5080.
5080 is also.. interesting to install tbh
Gotta wait on the 5080 to ensure the power issues get sorted out, and the missing ROPS supply gets exhausted.
3 to 1
... not sure what I am looking at here lol
the GPU
reflection?
didnt pop off my side window
but theres THREE pci cables to 1 gpu slot
my 3070 had 2 connectors
5080 only has 1
I have a new ATI 3.1 PSU, so just the 1 cable for me. It has the new 12v2x6 connector.
No adapter needed.
with the PCIE cables and adapter?
yeah
should be fine 😉
lmao
their are not very many reports of 5080s having the meltdown
im not worried tbh. triple checked that everything is solidly connected
if it was a 5090, i'd tell you to re-wire it and use the 12v2x6,
That was the included adapter/cables, right?
Never mix PSU cables from different PSUs. That's a bad idea. 🙂
it was included
cool
yeah its a newer cable and i useually work on older hardware so i didnt recognise itl ol
Well congrats on the 5080 at msrp, that's awesome.
Yeah, they started with the newer cable "12VHPWR" and had all those melting issues. Most people were fine, but... yeah. Then they altered the standard a bit and have that new 12v2x6 cable where it makes better contact.
THIS. This was most important.
yeah i checked for that first thing lmao
That is something that cannot be fixed. If you're missing ROPS, you have to basically RMA or return. 😐
seriously.... it's ... it makes me want to buy an AMD card and just rent GPU for AI.
Can't stand this bullcrap from nvidia.
also new speeds for SDXL illustrious if anyone was wondering
OLD RIG 3070TI:
prep_time: 12.05 sec, generation_time: 24.91 sec (includes face detailer)
NEW RIG RTX 5080:
rep_time: 0.02 sec, generation_time: 12.11 sec (sometimes 5.5s and 2s prep if i have a lot open) (includes face detailer).
Oh yes, I'd love to know speeds.
quick random gen to test
if i decrease the steps a little it will be faster. 20s ish
ah mb
try a large gen native, no refiner
bet
runnin rn
looks cursed asf but its similar to my old card using normal resoultutions
prep_time: 2.09 sec, generation_time: 29.92 sec
yeah it'll be cursed without specific prompting/model types, that's fine... lol
30 seconds .. that's pretty crazy
I'm on a brand new rig so I don't have any of my AI stuff even installed right now. Otherwise I'd tell ya how long my 3060 takes. 😐
WOW! with 20 steps + a detailer only 7s
ah ok
its basically the same thing lol
yep, very similar. Uses different models.
You can set it up to use the same "yolo" models from adetailer, from what I understand.
yo is that pomni?
ye lol, just a quick test bc shes visually interesting
nice taste
This was the first I heard of PhysX being dropped, how did they mess up this freaking badly, I "own" a bunch of these games on Steam.
https://www.pcguide.com/news/heres-every-game-affected-by-rtx-50-series-dropping-physx-32-bit-support/
they attempted to focus only on ai environment, freeing space from gaming stuff. but what they got was only a very pricy slop that give little or no benefit to ai
what a fail
hello chat
yeah... sorry to bring bad news 😐
good bold lines
It's not your fault. The blame for this bs rests squarely with nVidia.
Was gonna try leonardo but, wow, kinda pricey.
a fast SSD also makes a lot of a difference when loading models, saving etc
hmm true
flux does 2048x1024 native without any issues, just saying 😉
but i feel like if anyone is using any stable diffusion model with a high end card i safely assume they are not using a HDD
and my SSD is like 7-8 years old now 
a recent m.2 makes all the difference
yeah im still on a 2.5incher
nice.. let's see!
ouch, SATA maxes out at around 500mbyte/s... that's 5-20x slower
i should get a nvme honestly. but first eating dry bread and water for the next month
I just got a bunch of new nvmes for the new rig 😛
still loading in is pretty fast for my casas as i dont switch around a lot
990 Pro for the OS drive.

1440x768 with flux pro 1.1 on replicate - currently memchecking my 4090 rig 😕
"happy stereotypical anime manga scene" with flux 1.1 pro ultra at native 3136x1344 😉
upscalers be damned
I generate 1920x1080 on SD1.5 (without upscalers) and can generate 3840x2160 on SDXL (without upscalers). What can FLUX really do without upscalers? I'm curious now...
replicate is a paid web service?
like leonardo and civit and such?
https://replicate.com/black-forest-labs/flux-1.1-pro-ultra (and others)
quite user friendly and paid per generation, depending on the compute used, monthly billing... my go-to for "my rendering rig is at home" and other ad hoc stuff
they also have other models than flux
professional service, no strings attached (but of course censored)
with SD 1.x you can generate at that resolution natively but it will be cursed as hell 🙂
not sure how well SDXL does at well above 1024x1024
flux (pro ultra) just goes up to 4k, and with flux on my 4090 i do 2048x1024 in less than 15 seconds. no need to go any higher IMHO, except if you want to print the results
the example above is native, haven't found a way to increase the pixel count
"vibrant friendly wholesome anime manga scene, landscape, peace, furry" for fun and profit
and that took 9.8s on their server, they do have some serious GPUs there 🙂
Have you tried using a dpi count in your prompt? 96dpi, 72dpi, etc Or maybe pixels per inch (like retina display is 200ppi)
it looks hella good but im staying local thogh
though
You have double the VRAM you did before, you should be able to run a lot more stuff now.
Yeah, i just never looked into flux etc before
Somehow the flux ran takes a bit longer but i blame that on the current torch being incompatible so running a hack setup rn
Schell and dev*
That's not even remotely close to anime...
This is the anime channel, why is there half naked 3D men in my favorite channel 
the prompt cannot influence the output resolution, that's a rendering setting - they just offer rations e.g. 21:9 and that results in a certain pixel count with the pro ultra model. in others you can set resolution but just up to 1440, so no 3k with that
if you can run flux locally, just try it. it's a gamechanger prompt-understanding-wise, even though it's not anime focused
With 16gb vram i sure do hope i can run it lmao, though for a 1024*1024 it shouldn't take 40s right
Probably something wrong with my settings or checkpoint but ill look around
i recommend flux fusion, schnell speed with dev detail. and flux is definitely slower than sdxl
Oh that's for sure but depending on the speed i might stick with XL since 5s gen speeds are really nice
20 euler steps with 1024x1024 take >10s on the 4090, but with flux fusion i just need 6-8 steps so that's 5s and totally ok. depening on your hardware, that probably translates to 10s or so
That's pretty good actually. I'll give it a go when i decide to leave the comfort of my bed lmao
try this, they even claim that 4 steps are enough (i tend to add some more)
https://civitai.com/models/630820?modelVersionId=942694
Ive been hitting 40s but thats probably due a checkpoint being falsely labeled as schnell
or using 20 steps when 5 would do 🙂
Haha true
have fun, bbl
You'd be wrong about this if the info was included in the metadata for images. give 6720x4480 (weighted to 1.2) a try sometime at the start of your prompt.
ah tried to run it but my hack torch version is sorta incompatible (due 5000 series)
ill wait till its fully compatible lmao
though in illustrious i can still make pretty images so its allright
Free design, everyone
THE BOULLHATER
early adopter fun, was that way with the 4090 back then as well
Yep lol, oh well maybe soon lmao
I love it! She's is cute! Let's see some more. Maybe though, without the missing arms, that's weird. 😄
They're not missing, she simply has them behind her back to thrust out her chest
if you say so 😛
The arm's not missing. It's typical for character design sheet. To give a clear view of the side
I had to change my in game name in Space Station 14. My name was "Eye Sore" :D
Yeaaaah some love for Teto too. Let's rise against the Miku oppression !
Am I baiting for good content of both sides ? Maybe. You can't prove it.
nice, this a gen?
No, shitpost/meme from one of my other servers.
no more of this character today? I'm disappointed. 😄 😛
no more mrs. nice miku
#artisan-faq or locally
thanks for looking 🙂
i really love this last one
hands need some fixing
i should add adetailer to the workflow
Got WAN working decently 
how long did it take? 
About 5 mins render time and 2 mins of Interpolation and upscaling
But my workflow isnt perfect
that's pretty fast. On a 4090?

Nice, what prompt did you use or is it still mostly rng to get a decent result
Sometimes they move at lightning speed, sometimes smooth
Its a good mix of both, this is also img2vid, I had a decent base to go from
Yeah right now for me its 5-10 tries for a cool outcome
Yeah anything illustrated/painted is a pain in the butt
It really depends on the style but yeah it really hallucinates sometimes. Cant wait for a proper lora/finetune
cute hair
Ooohhh famous painting collection
Woman in blue by vermeer
The scream
I think the girl sitting at the Cafe is van gogh?
Cant remember
So nice though

Which model and lora you use? Want this pixel art style. Arcade style
ha yeah... was just messing around.
Not sure about Woman in blue by vermeer... The first one was literally just a facepalm. 😄
In the set:
The Scream (Munch)
Mona Lisa (da Vinci)
Generic Escher style (Escher)
Starry Night (Van Gogh)
Starry Night again because I liked both (Van Gogh)
The Cafe Terrace on the Place du Forum (Cafe Terrace at Night) (Van Gogh) - this one is hard because it takes the words rather than the work. Thought this output was close.
Napoleon Crossing the Alps (Jacques Louis David)
The Girl with a Pearl Earring (Vermeer)
maybe I'll do some more
dall3, naturally.. gotta keep some alternate options going 'round here. 😉
Try a Rembrandt van Rijn
Cirno & peach combo
More WAN
I see. Wonder about prompt...🤔
nothing special, probably just put 'pixel art' in there. Is there one in particular?
Maybe have to try with Pixel 2d, armored, 16bit or 8bit or something. Need for some medieval art. What do you mean by one in particular?
The Man with the Golden Helmet - Rembrandt
The Girl with a Pearl Earring, again - Vermeer
The Wanderer Above a Sea of Mist - Friedrich
like one image in particular, ill see if I can dig out the full prompt
I'm actually looking for this type of pixel art I shared now. Only Arcade stuff.
In dalle-3?
Sure. As long as the output is produced, any model is fine.
The Swing - Fragonard (1767)
well it doesn't look like i kept these around so I don't have the prompts... but again I'm sure it was pretty simple terms like 'pixel art' and '1990s' type stuff
let me see what I can recreate
these are awesome
Yeah, nice. pretty old school art. The backward quality of 90s and 80s are very detailed.
if you want the prompt for any let me know
Yeah, I want it.
The first post 2nd image. Spider one seems scary. And the blue color tone dragon image. Also the last post prompt.
Small anime woman centered on the floor underneath large thing. In 1990s pixel art LucasArts game, Loom. Dark medieval style. Small graceful woman warrior, centered on the floor underneath large obelisk. In 1990s pixel art Sierra LucasArts game, Loom. Dark style. Small graceful woman warrior, underneath large dragon, cliff near a cave. In 1990s pixel art Sierra LucasArts game, Loom. Dark style. A wide zoomed-out shot of a small graceful woman warrior with a red sash and short gown, underneath a large crystal. In 1990s pixel art Sierra LucasArts game, Loom. Dark style.
They all literally use the exact same descriptor for the style. 🙂
Nice! Thanks. Is that from a arcade game?
Sierra LucasArts game?
Sierra and LucasArts were two companies that created many games in the 80s and 90s.
I like the art from those games - a lot of p&c adventure games.
Pretty old school. Consoles and PC platform I think. Any games you remember?
I remember a lot of 'em. 😛
Is these from one of the games?
... am I being trolled?
No. I mean are these scenes from any of these game?
No, they are not from anything. They are generative AI.
I see. There is health and mana bars. So, I asked.
Last ones looks cool. Can you share the room prompt? I wanna gen some now.
literally just the style portion - 1990s pixel art Sierra LucasArts game, Loom.
Cool.
What tags are you using for the retrowave sun and the Pirate/Venician bay backgrounds? I like those a lot
something like "retro, road, driving a car like miami vice." and french/italian riviera
That one's cool mania. Angry chicken and the girl combo. 
a fancy woman and the anrgy chicken shes emotionally dependent on
what more does one need 
What are some characters/how to generate characters that have shoulder length hair but the hair is equal length all around ie the hair bangs are just as long and the same style as the hair on the back?
Oh
A car and a woman is also cool. Fresh reverse vibe that got from previous retro pixel art.
Like I mean also like stringy hair, not parted like a woman
or "mediium length hair, straight cut hair, whateverforthebangs"
give an image for example
i dont think ive ever made a woman with a car yet before now that i think about it 
I'll experiment later, so tired 😴
fair
Look for the pixel ones @keen hollow shared. Pretty cool.
its only 10 pm rn, ill be up for a while
i saw them before i started posting
i havent been posting much but i do check the images here from time to time

omnia has indeed been popping off
Some posts are pretty cool to add in music vid or games. Pretty good talent.🤔
Music like?
I see.
I try for some retro, jazz, synthwave music. Old school.
Timecop1983, Wolf Club like.
too slow ngl
i need harder beats and faster tempo than what jazz gives
retro and synth are meh too because id rather sing along, so i prefer music with melodies in it
thanks for the rec though 😅
If you're looking for rock.
Linkin Park, Evanescence are good
ah i know them
theyre in a few of my playlists
i started OUT on them LOL
thats the stuff i was listening to in highschool 😅
crawling and faint by linkin park, and who doesnt know bring me back to life lmao
A new tracks of Linkin Park are there. I like some new ones like Overflow
eh i dont really listen to them anymore tbh
like i said, i was in highschool when i listened to that stuff lmao
Still, it's a good nostalgic feel.
true
i have 3 playlists with nostalgic bands for me, like linkin park, and flyleaf, and Evanescence
you would not believe your eyes 😭
nah aint no way
youre giving me ptsd flashbacks of that song 
Fireflies or some other track?
yeah lmao
Cool
Oooo that scythe one is PEAK
Why don't you post here more often 
Your images are so clleeeaan
idk xD
i dont use style loras
sometimes add detail but most of the time just prompting
I hear you don't need style loras much with illustrious anymore but it still doesn't have like
Cartoon styles and characters among other things so I keep using loras
Even for styles that are uhh trained in models? Ig?
yea illust is nice and for cartoon there are some models aswell like toonify or mistoon
for pixelart i used a lora
because it makes it cleaner
i love the illustrious models they are a godsend for anime pictures
Yeee I used mistoon when I was using pony
Haven't tried the illustrations version yet
Illustrious*
I like using Wai and uhhh SIH?
It has a weird ass symbol in front
Peak models
🔥
i love the nova anime illustrious models
do you use a vae or not?
but the hair in the bed pic and from frieren is grainy
or is it the model?
it was grainy cause i save it as a jpeg instead of png cause of the file size
ohh
but i figured out there was a quality slider for jpg
Rs?
for instance these pictures are the exact same seed but one saved as a png and the other jpg
Pretty
i like it more than pony, because i dont need the score tags xD
with the booru tag autocomplete extension its easy to use
I like illustrious because it's clean
Cleanest models I've ever seen have been illustrious
Hey smol 
I like the twintails
not intended, but a good additional 
LOL
i guess using hair tie in mouth makes the hair tied
It's always so great when it does stuff like that but also a little sad cuz damn I usually like that detail that gets added
Then I have to add it to the prompt
Or worse
Find out what to add to the prompt 

I hate finding out that oddly specific thing I want doesn't work/doesn't work specifically with a lora I want to use

Yeah 
I'm always using a lora of some kind, usually for style, but I often find if I'm having issues with a prompt or concept
Remove the lora and try
It's usually the issue
So I either have to lower the weight until it works, sacrificing style
is that lora on super high weight? It shouldn't remove like concept from the model tho
Or just make a new prompt/find a different style
Nah I use mine on a range of 0.5-0.8 at most

if it's lora for style, then you can try finding a few photos of those style, then using them as ip-adapter
I almost never use higher lora weights unless specified
might solve the issue with prompt with lora
Ip adapter?
Sounds familiar but 
I forgot...
do you use nai?
That'd be dope actually
No I use SD
I bet img2img or inpaint would help but 
basically a module for controlnet where you can "infer" from other images and "copy" their style
Quite frankly I'd rather shoot myself in the foot than use those

Ah controlnet thing
All the things that make my laptop crash out 
ahhh
I'll try it
Yeah I gotta grab the models lmao
It has controlnet integrated I just never bother with it
I stopped using controlnet after I switched from SD 1.5 to pony/sdxl
It jumped from usable to "holy shit even I'm getting impatient waiting"
I wonder if it's worse on illustrious
i don't use much controlnet as well, but sometimes i use the ip-adapter to mimic some stuff

Do I need specific cn models for illustrious models 
don't think so
actually i didn't really use controlnet after using illustrious models, so not sure 
Yeah illustrious can even do two people (using loras no less) with just prompting
So I just haven't really found a good reason to Crack open controlnet
it can do up to 5
wildcards snuck something cute into my gens
Those would be cute emojis 
What makes anime so liked is because it's expressive and that's because the heads and eyes are big. So that must mean the power puff girls are the best animesw
pretty sure powerpuff girls z didn't do that well
irl powerpuff girls 💀 spoilers because its cursed
\
dang I guess I should try an illustrious model lol. I like Bluepencil but it refuses to render more than 1 character. Even Ponypencil, so I've been using Autismmix Pony
i love illustrious 
I also get alot of extra/missing limbs when generating. Do you guys have any tips for lessening that? I've spent a lot of hours using inpainting and controlnet to fix them for seeds I like
depends, what are you doing rn
what are you using to generate, what model, what settings, what prompt, what negatives, what loras if any-
i use a complex series of negative prompts that help me alot
tbh it should just be good model = good image
Been trying to make dakimakura style. Using Autismmix based on Pony, no negatives or extra limbs/disfigured limbs, just this LORA - Dakimakura(2 shots in front and behind)_SDXL(Pony)|抱き枕(正面と背中)_SDXL(Pony)
I started last Saturday lol

but I've put ~45 hours into it
since your new, the newer models sometimes dont require a lora, since you're trying to make a dakimakura try adding "dakimakura" in the prompt alone
@native parrot i assume youre going for something like this?
are you trying to uhh take an image and uhh like impose and style over it or something?
or...

yes I've been using img2img because it's faster and I'm trying to generate art based off of 2003 mmo assets like this
this gives me nintendo DS vibes
Trying to get a 50/50 screen similar to this
but yes would like this style. Pony doesn't do a great job of hitting this one. This is from BluePencil
i think you could just go to txt2img and make those images in the style you want very easily
you could probably just prompt for it but im not sure how as ive never tried that before 
try an illustrious model tbh
I've tried (2 shots in front and behind) because that's what the LORA suggests, but I have to roll like 10 times to get it sometimes
and if I don't have the LORA on I don't get it at all. The problem with the LORA I was using was it would force the bed to be made with normal fabric since that's all the model was trained on. So I had to go back and inpaint the background to get a hay bed
Illustrious is definitely giving me better results than pony was
illustrious is peak
Hi, welcome. let's see those negatives!
Nice, this one is -clean-
good ones here, too
oo nice these are good too.. also clean. This looks like a more vibrant pony model.
steamecommunity 
they're coming here too? seems to be a resurgence of them lately.
here too? we get a few daily here lmao
didn't notice, but haven't been super active in this channel lately. they've been invading other servers I'm in though recently more than usual.
they definitely were more active previously
TOPA
bots?
yea
😮💨
i little fun trend i started a couple months ago

Very soft coloring and soothing atmosphere. Can you share the prompt? I'm trying to do something similiar for this picture below.
metadata are still inside
(iuui,ajishio,marushin:1.05),year2023 , general, BREAK teto, vocaloid, 1girl, red long hair, big eyes, dreamy relaxed expression, looking at the cat, smiling, silk_dress, playing with, animal, cute cat, feral, outstretched paw BREAK (chibi:0.7), (pastel gradient, pastel colors, pastel, pastel palette:1.2), clouds colorful palette, colorful, chromatic aberration, bioluminescence, vignetting, reflection, neon palette, solo, vector trace, yellow theme, official_art, masterpiece, best quality, amazing quality, very aesthetic, high resolution, ultra-detailed, absurdres, newest, scenery,depth of field, volumetric lighting, very awa, (symmetrical:0.3)
focus on token before the birst break and after the last
Oh, my bad. There it is. Got it. Thanks
cat
Guess who?
Guess who?
wow
How it started/how it ended
steanmescommunity 
@finite herald 
What is her name?
Ginny?
Difficulty up. Any guesses?
Genderbent Malfoy?
Ok I cheat this time with google, is it the unidentified Slytherin girl? Played by Chloe Rich? (I never watched the movies past #3)
Pansy Parkinson
Always imagined her like this
Oh. I always imagined her as black haired for some reason with an ugly nose. It's been years since I read the series
There was (nose:1.1) in the prompt 
These sisters shouldn't be hard to guess
My first guess would be fleur and gabrielle delacour but these are wearing Slytherin... 🤔
Ok I had to Google it again. Daphne and Astoria?
I think I've seen that second one before, shouldn't there be an iceberg somewhere?
maybe she is sitting on it?
seductive smile, smug
@lyric zealot make 12 continuous cartoon pictures for the story stoncutter
beep boop~ error receiving command
Could you @lyric zealot make 12 continuous cartoon pictures for the story stonecutter
@lyric zealot Episode 1: A poor stone cutter chipped at a rock.His hammer went TACK and his chisel went TOCK.
Episode 2: Then a rich man walked past, in his rich man’s clothes.“I’m just a poor stone cutter,” the stone cutter said.“I’d rather be a rich man instead.”“
And he became a rich man.
Goooooo Chgans!
Felt like a somewhat cohesive story. 😛
And eventually he lived happily ever after.
He offered her stones...
And flowers...
They were married.
And yes, they lived happily ever after.
Hope that helps, Lin, and that it's the best story you have ever seen. 😄
Would you play this game?
Or this one?
I'd play "Tank 'Splosions"
I bet @wooden sail and @woven gust would play this one.
Maybe a sim game.
ok that's enough time wasting.. sheesh...
Is this Gemini?
dalle3
I haven't messed with gemini in a while.. maybe I should give it a go.
I don't even have any local AI stuff installed on my new build. Got annyoing maintaining it. Just feel like using the web services now. 😛
Is shoule probable be able to handle very complex scenes 🤔
Can't have it without fine controls like that
[stadium:gym:0.2] - switching prompt mid generation.
And with Comfy it's even 'grainier'
gemini's filters are too much... can't get very basic prompts. Maybe it doesn't like video game boxes?
ok, enough time wasting... gonna go get something done. byeee
yes, I like smug
Hi! I haven't used image gen ai for a long time and now when I'm checking the prompts some of them have something like "score_9" or 7 and so on. Is that some kind of lora?
It's a feature of a popular model Pony Diffusion. It has been trained on captions of a certain 'style'. Here's an article about that from the creator:
https://civitai.com/articles/4248/what-is-score9-and-how-to-use-it-in-pony-diffusion
Thank you so much
manga cover fun
Very fun indeed
this looks more like a comic than manga but still great gen 
that model looks colorful
A girl with long, flowing white - blonde hair is depicted. She's wearing a black leather jacket over a black top, and a pair of large, over - ear headphones rests on her head. Around her neck are some necklaces, adding a touch of style. She is playing an electric guitar, which is predominantly blue. The background is dark with blue light effects, creating a cool and dynamic atmosphere, suggesting a musical performance or practice scene.

what did you use to get this outcome?
oh welcome back, it's been a long time 
best quality, amazing quality, very aesthetic, very awa, masterpiece, absurdres,
1girl, solo, red hair, tied hair, red eyes, stud earrings, beanie, round eyewear, korean,
Negative prompt: bad anatomy, bad hands, error, missing fingers, extra digit, fewer digits, (worst quality, low quality:1.1), jpeg artifacts, signature, watermark, username, sketch,
Steps: 24, Sampler: Euler, Schedule type: Automatic, CFG scale: 7, Seed: 795778899, Size: 768x1024, Model hash: 938234ffc1, Model: ΣΙΗ, VAE hash: 66f78ad136, VAE: sdxl_vae-fp16fix-blessed.safetensors, Denoising strength: 0.4, Clip skip: 2, Hires CFG Scale: 7, Hires upscale: 1.5, Hires steps: 10, Hires upscaler: 4x_fatal_Anime_500000_G, Version: f1.3.0-v1.10.1RC-latest-1010-g9277598e```
pewpews
I can't post a thing here, I don't do SFW stuff. =/
What models yall using? ive been messing around with the popular illustrious models (WAI, Hassaku, Prefect) and I just havent been able to make it work the way i want to. I feel like everything in illustrious has this same 2.5d shiny look, or a grainy look to it no matter what i do
What sampler do you use?
Euler A?
Its a matter of prompting nhl
Ngl*
Abs were fine iirc
But if not 🙏 forgive me fruit
I mostly use Euler A or dpmm 2
I can post some examples tomorrow
sure, also i switched from euler a to Res multistep + KL optimal
KL optimal? Ewww. I tried that when it was first thought up and it just can't compare to AYS (with proper n +1 sigmas)
how can i make this better
*"Weird Science" by Oingo Boingo starts playing in the background*
It's a movie that's as old as I am about two teenagers designing a woman in their computer and she becomes real. The theme song is also titled "Weird Science".
So the only way to make her better is to make her real
meant if anyone has pointers on what could be improved
Well, your windows need work but that's probably an issue with your model and will be nearly impossible to fix.
Your vase only has one handle instead of two.
The maid outfit lace disappears before her hair would cover it
Holy shiiittttt
I KNOW THAT MOVIE KOI
You're giving me nam flashbacks
You're making ME feel old koi
I bet you know what it actually smells like old man 🙏
I want to watch weird science now
I may or may not...
How many wars were you in 😂
Zero

can i dm you with some questions?
sure but im probably not the best to ask as i just started SD yesterday
it's fine!
There are modded 4090s with 48gb. Anyone know their performance?
It's modded, I wouldn't take the chance as you will have zero warranty at all in case of issues.
Are they that unreliable? I saw some on Alibaba with good reviews. But, I'm just curious to know their AI gen speed. And old cards have their warranties expired long ago
not really nsfw but just spoling just in case
Hmm dm me the prompt
whats the prompt? (i forgot to reply in my last msg)
masterpiece, ultra-HD, impressionism (1.5), high detail, (blurred background), masterpiece, best quality, very aesthetic,8k,anime, masterpiece, best quality, 8k, (flat colors:1.6)
1girl, Tatsumaki, short hair,navel,jewelry,underwear,green eyes,panties,earrings,green hair,midriff,pants,necklace,bra,side ponytail,dated,see-through,black panties,highleg,black pants,black bra,hoop earrings,pearl necklace,highleg panties,see-through shirt,tatsumaki,green belt,cowboy shot,kneeling, v & 3d, realistic, blurry, low quality, grainy, noisy, distorted, jagged edges, artifacts, rough, worst quality,muscular, hearts, heart symbols, skinny,
I bolded the & so you could see P & N clearly.
ah attatched meta data. sometimes discord strips those
Discord doesn't strip anything unless you download the webp
You're also forgetting there's a command line argument --disable-metadata
Some use it, some don't
If someone doesn't use that, and then creates a checkpoint, You can even drag + drop the checkpoint into ComfyUI to see what their filthy mind was working on.
hmm imo its the negative AND quality tags giving that plasticy look
why no spaces though 😭
so hard to read
No idea, I copied and pasted it verbatim so you wouldn't have any issues.
did it with my usual tags made it a bit more SFW tho (no refine/adetail/upscale)
Still looks semi realistic around the midsection and anime for the face
hmm yeah but less plastic
hmm if i use a artist tag (no artist lora)
quick and dirty style changes. 2 was with a lora the rest with just a artist tag
Yeah, i still feel like the WAIillustrious has this sheen to it that is distinctly AI. an issue i didnt really have in pony
CSR one doesnt have that issue, but it is distinctly CSR
And if you want a highly realistic one, I recommend Crystal Clear XL
oh yeah didnt use a lora for that one
different checkpoint: ΣΙΗ
stomach thing is still slightly plasticy but no grainy
think its just a matter of applying the right style and fitting promtp
prompt
You might have to get rid of some of those quality modifiers, like ultra-hd
oh yeah definitely. i use diferent ones though
i WISH i was
but noooo

One doesn't escape Walmart that easily


