#🍥|anime
1 messages · Page 128 of 1
how do i make these
well #1072220168534642768
thank u for this i can finally post my edgy loras and sell them 🙏

cool prompt I can feel 2010 edgy apocalyptic/survival game animes flowing into me
@lyric zealot did you figure out what to do with freeu 
nope, but i saw some people saying it depends heavily on the model and lora
so everytime you change your model, or add/remove lora, your settings have to change
and the value you need, it's trial and error buddy
so that's why i

Sounds like you wait for the Its now automatic 

maybe, or just get a new gpu and forget about it

fluffe
it's 6am and im sleepy as well
i see you got corrupted by Luna 

tried to remove the pointy but even with negative it still shows up
in the negative: (pointy stuff:5.2)
(steven seagal,leotard,sexy:4.5) in positive
now it worked
1060 6gb
how long does it take to gen this image for example
6mins for 1080p
He looks so sleepy
Man I fr wanna call someone here just to flex my pc
Worked all day to configure stuff
Tho still got a ton more to do
Bruh never heard of unixporn
tell me how to generate fr
controlnet inpaint worked (made her wear a white glove)
You... press this
and wait
You need to ask a better question than that, and be less edgy, that's important too
@lyric zealot
thick thighs
I can inpainttt
crazy
well, there's multiple ways to do this. If you don't have a good enough gpu/computer, use the bots at #1100170312106127410. If you do have a good enough computer, then get automatic1111, that's probably the most user friendly and easy to setup local installation
i do have a1111 already
her thighs is exploding
then what else you lacking? a model?

lmao
well, then the best way is to just browse pics here, and see something you like, download the pic and see if there's metadata
if no, then ask for it
most of the time, people are willing to give
brain damage
this is fun
(ultra thick mega fat thighs:3.5)
crazy
insane
wild even
👍👍👍
what is even that video
Read title
seems like a great tutorial
Yes
Which model is that
breakdo
the model I like using is atrocious trying to make wide images 
Did you try with hires fix?
I can't really use hires fix
Bruh
Doesnt make sense tho
Do you know how to use hires fix?
You put a lower resolution and then put the multiplier like so that the resulting resolution is the one you have it set as now
would it work well with super low res to upscale to 512x768 later?
sounds very weird
my legs hurt

Usually you upscale from 512x768
But the answer to your question is no
I can usually do from 512x768 to like 700x1000
but its nearing the edge of out of vram 
another one
Ye that is good
Are you using medvram
Or lowvram
It will allow you to do higher res
med, they don't really make a difference in vram 
my gpu is a brick
both I can do at most 1.5 hires on a good day
That's like the standard
So I think you are fine
becoz art
guys yu kno hat
I ma donlaoddd
a11111
againn
rnn
sorryr ffor bad spelling
not on purpose
ayo
a cat of course
good gif
background looks painted lol
ai loves making vvery suggestive imagery fr
give model
my finger hurts
Its actually a very good all around model, it can do more realistic and anime too, this image has png info

givei t
asuna

download the image and throw it in this tab
you'll see all the information
which one
this one

there'll also be my prompts, they seem to work great with this model to avoid weird stuff
@brittle jay
wide
?
344 too low
this good enough? am I missing any extension?
can someone bring me prompt for first image?
megumin 😔
can you send metaddat?
what model
dalle3
ok
give the prompt you uedd man
megumin in a sunny beach,brown eyes,brown short hair,wizard hat
thx
dalle3 just killed most of SD ( except for porn/violence) 😔
what model are u using
try with aurora 😔
just look metadata
prompts aren't too hard, as long as it's an anime girl it'll look pretty good most of the time
if something is blurry up the sampling steps
something happened with megumin chest
is this still dalle 3
yes
never works for me lol
Welp, I’m off computer
But I’m hungry, so imma wake up and cook some food
nice man
congrats on waking up
there you go, it's mostly the work of lora tho
like fr takes a ton of effort
fr
idk what happened here
what model and/or lora is this?
dalle3
literally
to wake up
where is aoi lora?
and also partially_submerged
https://civitai.com/models/8717/aoi-artstyle-lora-semi-realistic it's from nerfgun
oh yeah I kno that one
OG

@lyric zealot pls
ye
https://civitai.com/models/5224?modelVersionId=6056 you could have found it here as well

it's only*
fr
She is still beautiful as usual.
damn i see a lot of people getting some images that seem exactly like the original anime style and i really dont know how to get that
like the images i get are less flat 2d idk
well, find the pic you want, and get the metadata of it
well, usually if you ask, people will share
yeah i never tried that
yeah, maybe try doing that
really nice btw, what model is that?
not that we hide
people just copy and paste it instead of throwing the file in discord :p
here's the metadata
ty

smol
exactly that
that strips the metadata?
damn
you literally copied the whole thing? 
crazy
so everything in my install works as intendded
nice nice nice
that was my point
yup yup
i think he is tire
but first I pet my doug
what model
here's the metadata, but im using it with refiner, so might be weird to look at
parameters
None
2 models, from calicomix_75 to expmixlineV3
this should have, no?
this pic
i did that one
@lyric zealot
thic
not sure how to prompt so its really a painting
Theres nothing in that one :<
he sees thighs
oh my god a human body part so bad and horny!!!!!!!!!!
btw female body part is one of nature's finest expression
real
I had to do a decent amount of image editing but after it all I'm decently happy with how it turned out.
looks neat but what did you have to edit?
I had to edit the eyes, shadows mouth and a few parts of the clothing. The clothing was pretty easy to fix but the eyes were difficult to improve because I'm not very good at image editing.
you know for facial detail you could use Adetailer extension that can help render better quality outputs which SD often messes up depending on image dimensions
Hmm, really? I'll have to have a look into that. Does it work with anime artstyles?
sure, its generic to anime and realistic photos as well
Alright, thank you, I'll have a look at that.
you're welcome, it's an essential extension. 
https://github.com/Nenotriple/R-ESRGAN-AnimeVideo-UI?s=09
Can this work for regular image upscale?
No reason it couldn't but I see absolutely no benefit of using this over simply using the extras tab in WebUI since that can also use R-ESRGAN upscaling models. The benefit of this tool is that it can extract frames from the video and then merge them back together.
Anyone know any good lora's for detailing background with like furniture or just in general?
I tried detail tweaker but It doesn't do much for backgrounds
at least on lower settings like under 0.5
I've been keeping it on 0.2
try a weight maybe with 0.8
they claim :1 is too high https://civitai.com/models/82098/add-more-details-detail-enhancer-tweaker-lora
For AingDiffusion (The model I'm using) anything above 0.3 seems to cause problems
hmm guess snu snu doesn't work as a prompt
idk, she might be skilled in snu snu
think the image tone is too intense, how can i soften it a bit? this uses ...
(medium shot, aesthetic), (highly detailed:1.2), (vibrant colors, vivid lighting:1.1), (cinematic, hyper realistic)
the weight i used is up to only 1.2
is there a controlnet model that can simply turn a 3D image into a 2D image?
like a 3D rendered scene into a 2D "drawn" scene
think i figured it out, taking out (vibrant colors, vivid lighting:1.1), makes a difference
these images are from dalle 3?
yea
no negatives?
no
digital anime illustration style stylistic ink pastel, dutch angle, gust of a wind glass particles shatters all over a long haired white girl with short skirt and dress jumping through windows while firing ak 47 rifle, skyscrapers sky background, serious determined gaze, dust particles all over
interesting, is that publicly free to use?
yea
eh I cant do much without controlnet
dall e 3 able to understand positional prompt tho
i mean generally, to get a specific outputs
yes i read about that, that feature is cool
its easier
than most models i tried for sure
like 100x easier
how do you rate sd and dalle 3?
i get the girl jumping out of the window in 1 try
relative to dall e 3, base SD (SDXL) and the finetuned at it current form is about 6.5/10 (assuming no controlnet or postprocessing) since the quality of their output is secondary to its ability to understand prompt
dall e 3 ofc being the 10/10 as its the best in the market rn
im surprised how you feel about dalle 3
its extremely good man
i have to try it out soon as im done with some more render on sd lol
it can output something like this
btw soft lighting works wonders in my previous case ...
this is the template i made for easy click and apply ...
______ Positive ______,,
Generic +,"(medium shot), aesthetic, (highly detailed:1.2)",
Color & light +,"(vibrant colors, vivid lighting)",
Effects +,"(cinematic, hyper realistic)",
Anime spec. +,"(masterpiece, best quality)",
im obviously too dependent on the a1111 features
the stupid thing about microsoft powered Ai is their ridiculous censorship
and it detected the following prompt unsafe at first..
1girl with a spear in her hand combatting a dragon
eventually i got this
made this for my friend, best i could get, any tips on how to make it look better? im using counterfeit v3 or v30 idk what it is
i stragiht up excluded red eyes but it still just gets red
anime girl in warrior outfit with a spear in her hand combatting a dragon in the seas
dalle 3 context is impressive
how are you writing this as story sequence?
i'm just messing around
are you using dalle 3 for it?
yea, no editing
damn
all straight from dall e 3
does anyone know of any checkpoints trained on things like visual novel backgrounds?
it's for a project
what controlnet model do i use for simple img2img functions like the img2img tab does?
depends what your goal is. none of them are really comparable to img2img, but several of them can achieve similar goals that img2img is commonly used for.
i'm trying to convert scenes in a 3D style made with a program like blender or koikatsu studio to a more 2D style
you probably want something like canny, depth, mlsd, or segmentation for that.
i'm working on a solo visual novel project and i'm trying to decide whether i will use straight 3D graphics or 2D-ify them with stable diffusion
never. even if it did leak, it wouldn't matter. it's not running on anything even remotely similar to consumer GPUs. it might be possible to get it running on CPU with like 128+gb of RAM if you're cool with several minutes per gen.
does dalle even use the same framework as stable diffusion
no, it's completely different
i'm wondering what the use case for this is
like if i'm using canny, do i select canny there as well as in the model dropdown?
Damn rip maybe we could use part and snippet of it as weaker but more customer friendly version
for All? that's just so you can manually select any preprocessor and model that you want. the radio buttons are just filters for the preprocessor and model selection.
i mean the whole section
i've just been leaving it on all
unlikely, but maybe someone much smarter than me could figure something out that would be helpful.
DALLE-3 is literally insane at prompt comprehension.
"Hatsune Miku in tribal clothing in the jungle at night. A pyramid split in half made of wide rectangles is behind her. In the gap between the two pyramid halves is a glowing teal rectangular portal."
And then there's this funny thing.
Wish SD have half that comprehension
Oh, yeah, like I said, it's basically just a filter for the preprocessor and model. It will auto select the most common combination and limit the choices to ones that are intended to be compatible with each other.
So, the thing that it doesn't really have that level of comprehension. Your prompt is being fed into an LLM that's finetuned to make a prompt for the model and that prompt is being fed to the image gen.
so, the system as a whole has excellent comprehension, but the actual image gen part doesn't.
still, clearly better than SD in that regard, but not as amazing as it seems.
Ok now how do I tell ChatGPT or Bard to do something similar lol.
canny does decent at what i'm trying to accomplish but it doesn't seem like it preserves color
the real play would be to train lora on something like llama 2 at actual conversions of natural language to prompts that are understood by SD. From that point it would be trivial to have that prompt be automatically used to generate an image.
Yeah, none of them will. For that, you'd want to run it in img2img along with canny. put the image in the img2img field, and leave the controlnet image field empty so it will automatically use the one in the img2img field.
controlnet is generally good at keeping shapes, img2img is good at keeping colors in the same places.
i see
i'm not sure how well controlnet and regional prompter would play together otherwise i could just add the subject descriptions to the prompt in txt2img
img2img with canny might be easier though
from my understanding if you look at the preprocessed image (which is generally black and white for most of them), that is what's actually fed to the controlnet module and that is used to influence the noise used for image gen.
here's an example of what the canny model sees
yeah i saw that when i messed around with it
how do i do something like hires fix in img2img?
does the "resize to" thing just take care of that?
you don't need to. the purpose behind hires fix is to use img2img to avoid the problems that generating at high resolutions causes. if you're already using img2img then hires fix is not needed at all.
hires fix is just gen at low res, upscale, then img2img afterall.
i'm hoping canny will allow it to keep details on the character models consistent, like lines in the hair and seams on clothes and such
yeah, that's what canny is good for. SoftEdge is a very similar one that's also worth checking out.
here's an example on that same image with softedge_hed preprocessing
i'm not sure how much prompting is necessary when doing this
it depends on the model and the image that you're using. kinda just have to play around with it until you get a feel for it. I'd start with a fairly simple prompt that describes most of the image. colors generally aren't too important to specify since you're using img2img which will be using the input image to create noise to gen from.
i'm not really sure what to mess with in the canny settings
stuff like the preprocessor resolution, low threshhold, high threshhold
i imagine the low and high threshholds are for deciding what details will be white lines in the preprocessor image
Wait what
Sources?
🤐
still going
This is pain. I don't even like running SDXL on my 3080 12GB.
what is the practical use case for img2img when i can simply interrogate clip to generate prompt and set the right denoise value and i can get something like this with 1 click from a real photo..
i mean i feel like i've cheated
oh and also i think a lot of the training involves feeding the engine images ....
hmm so with img2img i don't have a whole lot of control on the output with prompting
good gosh... img2img is powerful
this was the origianl image .. but sd model added so much texture to it
img2img starts from the original image and adds a percentage of noise to it based on your denoising setting and generates the image. it tends to preserve shapes and locations of color which can be very useful. before we had models trained with noise offset the method for genning dark images was to start with a dark image and img2img it at high denoising as one example
where did you put them!!!!!
i can't find where it downloaded all the controlnet models to, i only get 15 GB of persistent storage before it starts charging me monthly overages
mine goes into C:\StableDiffusion\stable-diffusion-webui\extensions\sd-webui-controlnet\models
ok yeah i just found them there
i guess it works from both directories, i just prefer to use the models directory because it's easier to get to
i wish i could contact thelastben to give him some suggestions for the notebook
like adding a temporary storage option for the controlnet models
he used to be in this discord but i think he left
bruh
Colab is pretty well dead for any WebUI based notebooks. Google cracked down and started disconnecting them very frequently. There's one that's CLI based that still works, but doesn't have most of the features that WebUI has. Most people have moved over to sagemaker or paperspace.
i'm using paperspace gradient
here's another
this is dalle3 I am guessing
yesh
how do i avoid the egregious color mixing
i'm just tryna do green hair with pink streaks
dall-e 3 can even handle prompts that are mostly booru tags
more likely is it stole prompts and images from SD discord servers for it's LLM training to make it able to convert booru prompts to something that it's model can understand.
i wish there was a way to separate things, like sometimes when i prompt "blue eyes" and "red shirt" together, it will add blue elements to the shirt
i have a similar predicament trying to generate a character with green hair that has pink streaks in it, wearing a white plugsuit
it's just doing whatever with the colors
you have become a conspiracy theorist
frfr
there are a couple of methods. the simplest is to put them in separate CLIP encodings with the BREAK keyword. altenatively you can use the cutoff extension
Why wouldn't they do that? AI images have already been ruled fair game as far as copyright goes and they know that SD is very popular. It would be stupid for them not to try to make SD prompts work in their model.
I think if you BREAK between them it probably will not help too much. like it ill still average the averages of the tingy
very yes'
wwhy from this server ttho
makes more sense to steal from civitai
it does. i abuse it regularly.
I never said this server specifically. Though this server is very big so it probably wasn't ignored. Civitai is less likely simply because they don't have an API afaik to make scraping those images and metadata easy.
dsiscrod also doesnt have api
you need a bot or somehting
discord absolutely does have an api, that's how bots work.
self bots are also very easy to abuse....
aren't those against discord tos?
oh no the free account got banned after already gathering all the data it needed to gather 
how do you think most of the niji style loras are trained?
and?
py is so easy to write even monkeys can self bot
banning an account that's free to create and requires no authentication besides an email address is meaninless. this is why literally every f2p game is plagued with cheaters.
Meowning people
getting the ip address to bypass cloudflare is harder than getting an email
i avoid playing f2p model games because they are infested with microtransactions and some of the worst communities
league of legends is kinda good tho

not all are. warframe is a prime example of f2p done right and has a generally great community.
No don't even say that as a joke
i played warframe back in like 2015 and was never able to get a new frame even after 115 hours played
Only a few hours you reach endgame in Poe, the grind is nothing compared to others
i liked destiny 2 until they removed content i paid for and then were like "ok now give us MORE money to play the newer content"
minecraft is probably the best f2p
if you play solo, yes. with 2-3 friends it's very fun. the grind is pretty tame outside of a few crazy weapons that are garbage anyways.
just have friends
most multiplayer games are garbage if you play solo.
i disagree
having a group is essential for playing destiny 2
raids are literally impossible solo
destiny 2 is garbage anyways, but I agree for destiny 1
I don't see how that is true or relevant to enjoy the game, hell you don't even need to drop items to make money that's how I play lmao
like i said before, i stopped playing D2 when they removed the previous expansions and demanded even more money for the new ones
D2 should have just ended with shadowkeep and then they should have made destiny 3
the second a game requires both a battle pass and purchased DLC is the second it leaves my radar.
live service games should always have a way for a new player to replay from the beginning of the story, D1 is unplayable now and D2 just kind of tosses you into the middle of whatever arc they are currently doing
battle passes belong in free to play games and even then they better only provide cosmetic bullshit.
i think destiny capitalizes on FOMO
absolutely
bungie somehow managed to get even more greedy after separating from activision. it's almost impressive
at least warframe lets you play the story episodes as you wish
warframe has probably gone through so many changes since i last played it, it would probably be unrecognizable to me now
the last time i played, you got primes through doing void raids
and trade for premium currency without any sort of tax if you do it from within a guildhall.
i think the F2P game i've had the most enjoyment with is PSO 2
before the new genesis crap
another example of "omg just make A NEW GAME"
Apex has been a lot of fun for me, largely because I ignore their extremely predatory microtransactions.
i really liked the player market in PSO 2, if you put enough time into grinding in-game money, you could use the main currency to buy the cosmetics you wanted without having to spend a dime
if you did take the plunge into the cosmetic gacha, you could sell things you don't want on the market to get money for the things you do want
like one time during the persona colab, i got 2 different versions of futaba sakura's outfit, and i sold them and used the money i got to buy the aigis cosmetics
it's very F2P friendly for an MMORPG
Rocket league and naraka have the cosmetics only + you can buy it for free next time if you play enough style
I like those
Every bit of the game has mtx XD
yep, and I just ignore them because none of them matter. when it's all cosmetic bullshit and I get other cosmetic bullshit that's decent quality for free, I can overlook even very greedy microtransactions.
Haha
Another free game I love is legends of runeterra (absolutely great art in the game)
Players say the game is too free to play even which is funny
Riot doesn't make money from it
Well not that they need
look up guildwars2.com .. its f2p with some limitations, but in general you can call it buy 2 play.. but has no subs
Gw2 is a good game shame it has a lot behind paid walls
behind pay walls? not sure why you say that, to have limitations lifted you can buy one xpac and all the content of that xpac are freely accessible
Don't you have to buy at least one time?
I didn't play much, my friends forgot me

I remember some classes being locked and stuff
yeah it started out as buy 2 play, but later they allow you to play free with few limitations in place for f2p
there are like 9 classes in total but one class came later with an xpac so unless you have that xpac that class is locked for f2p accounts
Always the multi colored hair

i didnt prompt streaked hair lol
It is so common to appear probably lot of training data has it
probably
Gib prompt

1girl in bikini on a cruise ship, daylight

Sometimes I want SD to do those metallic reflections but it doesn't listen to me
xd
near 0 effort tokens for my elf monk, dwarf barbarian, and human cleric characters for my upcoming campaign.
nubbii
yesh?
quick question do you know if I can disable this check ? It takes a bit of time to do
I think disabling the check would require modifying the code. I'm not sure where off the top of my head. I don't think it should actually be spending mych time on it though. It's probably just doing other stuff in the background that doesn't update the CMD log.
Oh my
My 1girl was too poor couldn't afford a ship, is on a raft instead

cat i dont really attempt to control reflections, those could be model specific
but i put these in in the positive box
(medium shot), aesthetic, (highly detailed:1.2), (vibrant colors, vivid lighting), (cinematic, hyper realistic),
time for the sleeps. gn frens
just use xforrmers
also the TMLock ssomething somethin g
idk exact name
amd moment.
xformer won't work with amd gpu
you can fix it by insstalling a pacage
I should put some extra negatives to reduce certain things
1girl with blonde hair in red bikini on a cruise ship, daylight
her bikini looks very fashionable
you should always use adetailer by default

I'll try
She looks like that dragon ball Android
@dull jackal ban that 
I'm walking my way home I hate this
Should I change settings in it
did you tweak any settings for it?
No
hmm
i leave adetailer with default settings
basically i just enable it and that's it
Probably doesn't like my resolution

try with 512x768 for horizontals

Ree
Keeps wanting to do super close ones
Cool lighting
would be fun to go on a cruise ...
This is quite the render. Love it.
Who is that woman?
https://www.youtube.com/watch?v=KDTddjForAw then we can put this song over it
Follow Neuro-Sama below:
Twitch: https://www.twitch.tv/vedal987
YouTube: @Neurosama
original: https://www.youtube.com/watch?v=NocXEwsJGOQ
#neurosama #vtuber #shorts
Welp
It says 6 hours 30 mins, but acording to my calculus, it should be around 1 hour 30 mins
Gonna jump over to Comfy for a little bit and see if I can figure out longer length AnimateDiffs
happy saturday
Happy saturday 
caturday
what you need calculus for?
I calculated, cause its going like about 2% every 30 secs, remaining 94%, actually I might be 30 secs every 1%, so
It probably would be like...
that's not calculus
"""
calculus, branch of mathematics concerned with the calculation of instantaneous rates of change (differential calculus) and the summation of infinitely many small factors to determine some whole (integral calculus).
"""
yes
Migh take about 45 mins tho, thats what I was gonna say
47 because its around 1% every 30 secs
But damn
I'll have to retrain my Asanagi LoRA
I was in the middle of training and I ran out of space for the last epoch

do you have a hard drive or something
I was about to write "upscale please" but then remembered I can do it myself now
it's already 2560x1440 😮
need more
bro did anyone here know bot for changing picture to cartoon?
lol
僕らは偶然の確率で. 歩幅は違うけど手を取り合っている 短くも長い一瞬だけど 方角が同じなら進んで行けるよ
| 広い広大な世界 欠けた星のパズル. 弱さを埋めるように 出会ってく 輝き続けるいくつもの物語 次のページは 桜咲く頃に
| 僕らは繰り返しにいきづいて. 思い出を語り合う 時を共に 言葉が通じなくても感じてた わかり合えていたよ心と心で
| いつか いつの日かまた巡り会えるはず. その時が来ることを 信じてる 月夜照らされた 桜道旅の日々 記憶に花をくれて ありがとう ありがとう ありがとう
| 違う時の中 僕らは歩んできた. 悲しみも苦しみも 乗り越えて 僕の終点は少し手前だけれど 側にいてくれたこと ありがとう ありがとう ありがとう
| 記憶に花をくれて. ありがとう
}```
yeah, i totally meant to do that!
it got the sakura right tho 😮
@wispy canopy morning
guys how can I use a differentt model for hires fix?
I think
enable these in user interface settings
OMG IT'S PENDRAGON
Oh hi
bretty 

hmm.. trying to figure out, is there a way to preview the first frame of an animatediff workflow? 😐
if I generate the image, then use same seed with animatediff, it's just completely different
Where the shit do i download R-ESRGAN 4x+Anime6b
@dusk oxide
Thanks
I tried some of those anime ones for anime
I just default to the ultrasharp one
idk they look weird
Perhaps it was just your model, but I'll run a few generations with ultrasharp to see if I like it 👍
maybe ye
This is like my model with (anime:1.4) its about the most 2d it can look
How did you do that?
it's with "AnimateDiff"
There's an extension for A1111 that's pretty easy to use. I am trying out the same in ComfyUI because you can get longer generations.
There's other tools, though. This is a fun one I did back in July that I thought came out well.
Then you have the lipsync ones... might be interesting to see if you could combine both.
And then of course there's roop, which is the crazy faceswap stuff. Here's a quick scene I ran through from Terminator 2 back in June, swapping in Chris Pine for Arnold. 🙂 I haven't tried to swap in an anime style yet.
I watched and read it twice to realize what was wrong lmao
I don't watch movies thats super cool
can someone point me to somewhere where I can download [redacted]
like the original model
I need it
what is that about
why would a model be ... redacted 
do you not kno
this is epic sd history
NAI leak
I have no clue, I just prompt and cool stuff happens!
how tf is there no wallhaven model?
nobonoko - Music For Animal Cafés
- 00:00 We're Opening
- 00:15 It's Time O'clock
- 05:55 B-Meowsic I
- 10:30 Chat au Chocolat
- 15:13 Planarian Cut
- 19:11 Birthday Dog
- 20:17 Waterfront Booossa
- 23:30 Sylvester the Cat
- 28:06 Storm
- 31:44 COFFEE MACHINE
- 32:38 B-Meowsic II
- 35:39 Peanut
13...
you can just download the images and look at their metadata (if you have automatic1111 you should have a png info tab where you can put the images to get it)
the model used shall be written there, along with the prompt and various settings
^^^ if the picture is posted with embedded metadata of course
ye
thiss is so insanely good
inbetween
i wish the metadata bot was here :/
can someone add it
wat
parameters
None
moment
ok
4gb vram
can you modify the location where controlnet models are stored?
sbongpob jumpscafr
frankly... i'm not even sure I have all of the latest controlnet models.
ok
Maybe I should have written prompt
first thingy (but dall e 3 hates anime babes!!!!!!!!)
ye same
wwe been asking for it for months at this pionjti
to no avail

dalle3 is pretty crazy
yye
low amounts of vram is always funny
ok
you can host it yourself and send images from here in DM to it. https://github.com/sALTaccount/PromptInspectorBot
it'll run on a potato. I run mine on one of my free E2.1.Micro instances through Oracle Cloud.
before setting up the Oracle VM I ran it locally on my PC with no noticeable impact on performance.
is ther any way to make the comd prompt mroe responsive or is that a windows limitation or should i just not wory about it
hello kat
what do you mean by more responsive?
heey
i guess that was worded badly. its not like the window itself it laggy but like when its generating it seems to update only every 1 or two seconds
how r ualll
im gonna try to work on animatediff today
If you happen to be getting around 1-2s/it then that would make sense. It only updates when a step is completed so if a step is taking 1-2 seconds then every update will also take 1-2 seconds.
oh
ensign mrow
a cat mask?
So is the image I asked with
lol
we can't have dalle3's power locally
@nova remnant you know... it's not even terrible 😮
i like this one
digital clean anime lineart style, dignified pose long blue haired girl holding a shining glowing blue sword on right hand and her left hand rising up commanding a arrays of magical blue swords raining on the sky in background
Time to throw this in sd
ig ot the text gen ui working after hours and stuff. the ui isnt coded nearly as nicely as sd-ui. i ran into problems constantly but i eventually got it working

masks look fun to gen
not sure if my model will cooperate
i broke it while running the update-windows.bat and had to fix it again
I would be surprised if it can even grab a sword

masterpiece,high resolution,highly detailed,best quality,4k, sun beams,white clouds,volumetric lighting,brightly lit,sunny day,noon,big smile,happy, 2girls,holding hands,on the tropical beach,black wetsuit,cute,short hair, comfy,saturated,abstract,Fractal Geometry,(Cutester art:1.3)
heres my prompt i just replace the middle lines for new subjects
@elfin bolt
trying a new model, felt not lazy enough to try and see what prompts are good for it
its a full 2d model
its surprisingly coherent ( sadly doesnt understand sword rain)
4 fingers 
She got hand to hold mini light saber
I just use Oobabooga for the API and run it through SillyTavern's UI https://github.com/SillyTavern/SillyTavern
I like that sd is more consistent than dall e 3 as of now (shits get blocked by every little thing)
the showers is there
how do you mean, gets blocked?
dalle is better but its censored
blocked prompt 😦
"safe mode" shenanigans?
Like "cute fluffy" is considered nsfw idk man






