#🍥|anime
1 messages · Page 234 of 1
fantastic results! what is your workflow like? comfyui?
I only use Fooocus. It's basically, prompt + upscale and a lot of Pony tuning with Loras/models.
foooooooooooocus

does regional prompter work on sdxl and pony models?
yes
i forgor how to use it
i think this is how you would use it to have 2 distinct characters in the same image right?
[background info, general image layout]
AND [subject, details about subject]
AND [subject 2, details about subject]
I dunno
In this chat nubby and eface have explained how to use it, might be easy to Ctrl f and find it
hi hi here!
Yes, I was away for a long time....
is a1111 still the go-to or is there something better now (other than that weird flow chart looking ui (i think it was called comfy))?
Forge is a copy of a1111 with better performance
By illyasviel
is it worth switching to if i'm already well settled into a1111
It's like close to 2x speed on that I think
Or was that on 1.5
I dun remember, friend has a 3070 too
i forgot if you have to use medvram in order to generate using SDXL and pony on an 8gb card, so i played it safe and enabled it
Yee I think you have to use it for SDXL
the warrior is sleeping!
to the land of dreams
I'll take you too
what loras work with pony?
do they have to be specifically trained for pony or will SDXL loras work too?
are we sailing on a ship?
Specific to pony, there's a pony category in Civitai
Wow! Will we be like the crew of the famous Enterprise?
yes it looks amazing
@wispy canopy scam link
Yeah, obviously, but why notify me? i'm not a mod
oh they're at the top of the list, i'm blind
the best unsung heroes of any server
for some reason i thought stable society was mods
is it just like prolific people on the server or smth?
Hello blind, I am dad
hi dad!
i wish novelAI was more affordable, i really enjoy using it the few times i can afford to drop a bit of money on it
I just dont get why anyones throws money at them to generate waifus 
because their v3 model is really really good
I mean yeah, but what for 

to be fair, it's not like there's thousands of perfectly fine local models for free...
idk, people just be doin shit
it's all booba, AI wouldn't have come this far otherwise
i can understand people paying for a service if they have a garbage PC, but even waiting 5 minutes per image for free is better than spending money on image gambling
i understand where the hate for nai comes from, they take from the community and give nothing back while making a profit off of it
some people spend more on generating images than it would cost to just get a decent GPU, some people just don't want to bother learning how to use any of it
What're we talkin about
whenever i drop some money on nai it's usually for the LLM side, i like to use the api on silltytavern
waiting 5 minutes for text is a lot worse than waiting 5 minutes on an image, plus their models seem a lot better than other 13b models
but the subs always come with a bit of the image gen currency too
and realistically, the developers of the UIs aren't really helping new people get into it, most of it is overly complicated for no functional reason
I also like it yet cant affort it
Question how did are they capable of making near perfect images
Just use ollama 
money
it's easy to make a really good model if you have the people, time, and hardware to train it a few thousand times until you get the right settings
Can in generate images near perfect like nai☺️
No but its a free LLM
Oh i that's nice
i'm honestly surprised no one has put together some kind of crowd funding to make a good community model, especially for anime
I mean at least you got pony now
yeah, which is a good start
if only it wasn't broken
hopefully sd3 stuff can be done better
Stable diffusion plss make a fund raiser for this 🙏🏻😭
SD3 already seems really good with anime as a base model
yeah
Who has SD on rn ?😭
~5 more days and I'll finally get to play with the gimped version
i just wish nai would go open source once they broke even
"If you have the hardware to run it, you can do so. if not, continue to use our paid service."
@vital raptor ?
Yk whats surprising??? A mod finally talking in the server😭 Most of the mods dont talk ngl
Or i dont see it
I talk a lot 
i hope i can run at least sd3 2M
I have 7.5k messages on the server
Its good
Dont think its bad
Atleast you're not introverted
if it's like the textual model, 800 is terrible
Its the internet, I can be whoever I want
I can be Micheal jordan
you can be a chopper, a tomato, or even the color blue!
i need help rn🙏🏻😭
what's wrong? not enough potassium in your diet?
No im not home and i really want to make this roblox character into an anime character
I'm desperate can someone do it for me?
( if y'all dont mind)
Pretty plsss
Pretty pls with a red chherry on top

img2img are a pain
the model can interpreter the image in any way possible, making a gen even more of a gamble
If ya can't do it I'ma find someone elsa
so, you need the right model, the right settings and the right prompt
Im in a different place where home is not near
I'll give it a try tonight when I get back from work. No guarantees it'll look good but we'll see what it looks like
Dm it much appreciated 
Inpainting over hand drawn sketch
Hats off if you painted this
It's Kōhei Horikoshi's sketch
But if you want to see my drawings just check out my x/twitter in the profile 😉
Your stuff isnt half bad either, I just like the style of this one specific
What are you drawing on? iPad or drawing tablet? 
iPad and Huion Kamvas 16 tablet
Well, this exists:
https://civitai.com/images/1896680
Got an iPad Pro as well with the pen but I suck at drawing unfortunately 
Pretty good, try adding "motion lines" and see what you get, depending on your model it might not look the best but still worth a shot
Just grind it 12 hours a week and after 2 short years you'll have a chance applying for junior illustrator position 
Nah I dont wanna make a living with it, just getting better would be great 
Was all dall3. 🙂 I think motion lines was already in there.
well.. hmm.. I can try. Those ships are just sort of gray blobs, so not sure how interesting it will be.
first one
Gonna assume they probably want something like a mecha girl lol
this is the example they PM'd me... https://www.reddit.com/r/Eve/s/TEdYWUZv7Y
second one something like this....
feel like I can do something a little more representative though... lemme see
Oo interesting
ty
its ok these were done on novelai
idk how this stuff works to comment on what can be changed or improved
actually that gen reminds me of the trollface girl thingy lol
could you do this too?
yes that reddish/brownish one is better... more to work with
ooh
trying that first one again but I dunno.. too blobous.. we'll see
its ok these are already way better than I expected
You and Sigma got it perfectly. I had difficulties getting the hood in the picture. 👍 
I like the wings, the huge moon, and the scenery. ❤️
wdym?
lol
we already know the ram requirementfor every sd3 model?
its ok man im happy with what you made , also what software do you use?
Those are all stable diffusion
ok ty
/help
how exactly do i do artist styles with pony models? i've tried using just the artist's tag from danbooru and artist:[artist tag] but it doesn't seem to have any effect
if it doesn't work, then it may not be in the model, otherwise you might need more weight
other than that, just putting it in the prompt should be enough
is there a list of what styles and characters pony includes so i don't have to trial and error?
not officially, @wispy canopy made a list of artists, i can't find the post though
probably not a full list, but it's the best i know of
ah i found the issue
nyantcha isn't in it i guess
i was trying to use their style because it's kind of a shoo-in to try with ai nowadays
plus they have a lot of images on danbooru
How can I use them ?
put them in your prompt
or, if you want to do wildcards, use can use the file for that too
👍
I did a massive ~640 image X/Y/Z run with the list for search and replace (S/R) to pick out the ones i liked
for pony models do i just use the normal SDXL vae?
yes, pony is sdxl
Want me to link you the zip files I made with artist tags?
it haves more results than the file linked before?
it's with previews ^^
hence the zip file 😄
these are the ones i have
one with full artist names, one with artist hashes
1gb? O.O 😮
i didn't render them small XD
lol
those are detailed previews
so you know what you get 😄
instead of some 2 step blurry bullshit in which you'd have to guess what you'd get
what's the difference between the full name and the hash?
Prompt from NAI, I didn't check nor that I have the capabilities of idk reading
Alot more sus than i thought
mr bits ah pose
How has the art commissions scene changed from earlier last year to now?
any time
Hei, I'm pretty new to everything. I want to try and generate a basic Gif. Something like a face slightly moving while breathing and blinking. I have no Idea how to get there though. Someone patient enough to walk me through the process?
never playe with animations
try generating normal one, open eyes, open mouth lol ur desired breathing 1st pose
then inpaint eyes and mouth. to close eyes and close mouth
have those 2 upload on image to gif. u can have more steps on your gif, more images to gen
I already don't know what that means :c
comfy or automatic1111 ?
A1111 I believe
theres image to image gen
and another inpaint, u can mask area that u want to change
most motion gens still suck atm, not really worth doing imo
Even for something as easy as breathing/blinking?
for anime, yes
I've seen good results from things like Sora or runwayml but that's all realistic
I haven't seen a single truly good animated anime yet
one strat I'd consider is generating a good still pic, and hiring a live2d artist
I'm trying a lora but it doesn't work for me, the character doesn't get me, that thing that says training where should it be put? Maybe that's what's wrong with me, any help please?
@vital raptor
@steady grail
better if you don't ask me help for training, all my finetuning and merges are a fail
i can't grasp it too at all XD
but that is on the lora website, it will be for use with the lora right?
the file should contain the images and the tokens associatede with immage if i'm not wrong
typ, as expected
But why doesn't Lora make the image for me like the samples? It's like if you don't get the character, I get a different character.
I'm not saying to replicate them, I'm saying that the character comes out
you load the model, the lora and insert the tokens used to call the data from the lora, but you get a different character?
and that happens loading your lora and not the lora downloaded from that link
right?
Exactly, I use the prompt and I get a character that has absolutely nothing to do with
what about the lora of that link
let's see the difference with lora loaded and without it
you used the same data set from this lora but you get a different result
https://civitai.com/models/391054?modelVersionId=436288
yes
let me see what happened, gen without lora, gen with the lora from the site and than gen with your lora
i want to see what happens
all with the same seed
one moment
with lora
when the other one is generated I pass it
I think I know what the problem is, the lora prompt that appears in the image is wrong, it calls the lora wrong, if I click on the lora directly it puts the lora correctly, on the website it was wrong
it may be true
thanks
why do they keep making the lora menu worse in a1111
it was at its peak with the pop-down menu where you clicked the 🎴 button to open it
the new menu with the subfolder buttons at the top was fine, but i absolutely abhore this file tree layout
Is niji journey still have free trial
Kinda want to abuse the fucking crap outta it ngl
V6 is out holy molly do i want to use it
wife
ignore the floating glass i couldn't figure out how to inpaint it out
i have no idea how to work inpainting in webui compared to nai
she has a hair accessory there and i guess that's hard for the ai to do
no
this thing
that's just how her hair is
ugly
make it better
indeed fox girl better 
@green temple
only smols are cute
lol
@nova remnant do you use forge
no, my pc bad
Do you have questions
yuh
Jab it in chat
shouldn't you use forge then? it lets you use stuff like SDXL on 4 GB cards
There's no helping when your card is a RX 580


i was just wondering if i should set these command flags on forge (i have a 3070)
--cuda-malloc --cuda-stream --pin-shared-memory
That I have no idea about
Are you the 3070 user
@green temple 
my prompt included "a very evil small girl in school"
so she might bite your knee be careful
i dunno what kind of negative prompt or loras/embeddings to use with pony
Are you using ponyxl or some autism merge
autismmix
no she cant no worries
really the only thing i'm having problems with is hands
hmmm it shouldn't need any by default but you can try stuff like I use on pony: source filmmaker, sketch, chromatic aberration, censored, white background, simple background, low quality, worst quality,
mess around with higher weights on low quality, worst quality and or source filmmaker they should make some difference
its not much on autism tho, it has a very fixed style by default
censored in negative 
see if it gets better hands
i've had these source things in the neg but idk if those are needed for autismmix
for positives just do the scores_9, 8, 7, 6, 5 and 4, source anime (theres anime, furry, pony and cartoon I believe), rating safe (theres explicit, questionable and safe)
on positive i have score_9, score_8_up, score_7_up, score_6_up, source_anime,
I don't think source 5 there does anything
try with 5 and 4
its something for homework
i never really understood what the refiner thing was for
in ancient days of sdxl it was needed to make stuff better
but now you can do half a prompt with one model and refine with another model
its neat at least
like you can do autism on half the image and then swap to pony xl for the better colors or something like that
autism > refine 0.5 > pony would swap to pony at half the gen
i'm gonna try enabling those command flags
it'll either speed up the program or break it
think of refiner as basically changing model "mid" generation
@lyric zealot stoopid

hiresfix 2x doesn't seem to play well with pony sometimes
1.5x seems like it usually works out better

what denoise are you using
ye
i've been using this sampler
R-ESRGAN 4x+ Anime6B
is there a way to gracefully close the webui
just killing it by closing the console window doesn't feel right
after enabling the command flags, it/s went from 2.21 to 2.18

i guess it really only is beneficial for cards under 8gb
What's the model btw
autismmixSDXL_autismmixPony
oo clean
i think the blue archive all-in-one character lora i'm using imparts a bit of style onto the gens though
blue archive
Pony has some scary style bleeds when using character loras
Happened to my Izuna lora too
UD mods HATE him
🟥 <- click here to find out more
Whats UD
don't worry about it
I don't play it lmao
it's another AI discord
to maximize my SD speed, i have disabled hardware acceleration on every program i have that uses it
chorme, spotify, discord, ect
this frees up a bit of vram for SD to then go "fuckin' mine"
and no adblock on the page
that's a thing? 
for some reason yes
Time to use comfyui for low vram usage
i'm using forge
i really don't like comfyui, it seems really unintuitive
Fair
i wish inpainting on webui was as intuitive as it is on nai...
so many settings and shit, i just want to get rid of the damn floating cup
gening smol gives u more vram 
actually does having more vram even speed up gens
or does it just let you render larger
i see
smol
grab
I'd like to use nai again
I really wanna mix some styles lol
okay now smol @green temple 
@zinc flax the blue archive characters lora for pony straight up can't do kaede 
i'm guessing it's mainly because she has no presence in the base traning data for the pony model itself
or not enough
oh, i found the issue
this tag is absolutely 100% necessary for it to even get her somewhat right 
smol is life
Cuute
indeed
what does the generating process do right at the end that makes it like hang on 98% for a few seconds before spitting out the image?
vae correction
Kafka (HSR) - Cyberpunk Edition 💜
testing something big
genning the max res possible for my vram without upscaling lol
till now i could push raw big gens without mutations
oom lol
ok, the tiled vae generated mutations 😦
how do i prompt for eyes like this?
thanks
nice
those two were my favorites when I watched evangelion
regional prompter ?
nice could u send your prompt cause i`m trying learn how to use it too
score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, BREAK source_anime, 2girls, ayanami rei, plugsuit, souryuu asuka langley, blue hair, white bodysuit, red bodysuit, red eyes, short hair, blue eyes, interface headset, headphones, long hair, orange hair, smile, pilot suit
@green temple
neg:
score_5, score_4, watermark, signature, artist name, 3d
oh pony ?
yup, I gave in to the furry gods
they make better anime gens
:kekw:
now I only need to decide on which loras to use
👍
Mannnn
Tryna abuse nijijourney to use 4 free
not going too well. im too scared to root my phone
so i used emulators but it knows 😦
ill try to figure this thing out free lunch is on za table
my first animagine image
yeah, same model
yeah, i was testing it
nice
she ate too many burgers in this one
Whaddya mean?
@green temple smol evil alchemist
not smol enough
now this is smol 👍
She looks pretty skinny to me : shrug :
SD3 Ultra workflow quick tests; the harpy prompt from a couple days ago. Using the API at the official Stable Image Ultra colab, SD3 Ultra with 3:2 and 2:3 has no upscaling or eye correction. Prompt modified to use the "anime art" description from Lykon's example prompt (which seems to pull the anime style as it should):
Some quick initial notes:
-
Eyes are really good from the start and from distances, but still not as good as Pony + Upscale.
-
The colors and overall composition are superior to Pony without any Loras.
-
Backgrounds are great for a base image. I like that it changes hair positions, styles, and lighting better than base SDXL.
-
Did not get the prompt as well as Pony (Pony still wins here); did not get the corset as it should in the first image (got it right in the others, though). It did not get the pose with hands between legs. Got it right with Pony/variants. For a base model, it works nice, especially for colors, composition, and eyes. The PonySD3 that everyone is expecting should be really great if paired with the Ultra workflow. I've tried to put anime artist names, and it does not recognize any specific artist. More tests are needed.
If I understand correctly, Ultra is a workflow using the 2B model. If the 8B model ups the quality even higher than this using Ultra, we will probably have an amazing model.
Here is the upscaled version with eyes corrected. Very nice. Finally, decent base SD gens:
Side by side comparison: left is SD 3 Ultra Original, right is the Upscaled + Eye-Corrected version
What is the prompt
anime art, 1girl, kemomimi, harpy, curvy, [white|black] hair, looking at viewer, [yellow|aqua] eyes, hood, fur coat, underbust, corset, night, moonlight, castle, sitting, squatting, Large_wings, hands between legs, from below, rooftop, wide_shot
Side by side comparison: left is SD 3 Ultra Original, right is the Upscaled + Eye-Corrected version
Side by side comparison: left is SD 3 Ultra Original, right is the Upscaled + Eye-Corrected version
does anyone know if there's a danbooru tag for this gesture?
🤟
i found it, it's literally \m/
not sure how that will play with the prompt box since \ is the escape character for things like ()
its some ultra api
the 6b model I think comes out on the 12
I think
lmao
Yeah, but this is different. They've silently released Ultra yesterday, and if I understand it correctly, Ultra is a comfy workflow (or some type of layer) that was supposed to be used on the 8B model, but it improves the quality of the 2B model. The 2B is the one that will be released in 3 days, and it had really bad quality.
The normal 2B had horrible blurry eyes and bad quality overall. Ultra has better quality, and the eyes come out acceptable, sometimes even high quality, at closer distances.
An example of the two raw gens, one with Ultra and the other with pure 2B:
Left: SD3 2B without Ultra (from Clipdrop) Right: 2B with Ultra (from the official API colab)
left looks real rough
so right is just like, adetailer on face?
hands are less messed though
Yeah, the image on the left was what everyone was expecting for the 2B model. And lots of people were really disappointed, as this is worse quality than SDXL base was at the time of release.
No, that's the thing, the right one is a raw generation. No Adetailer, no upscale, nothing. It is sharper, has better eyes, better details, better quality etc.
And the model is still the same, 2B. That's the interesting part.
Yeah the hands are not good, it is not perfect. But it is miles better than the initial 2B
Left, 2B without Ultra (really bad eyes). Right one 2B with Ultra. Prompt: an anthrofurry shark taur
The catch is that Ultra is expensive.
Ok, another quick comparison round. This is pure SD3 2B from Clipdrop. Horrible blurry eyes even at very close distances, distorted eyes, bluriness at illustrations, washed-out colors, etc.
Here is SD 3 2b with Ultra: sharp eyes at close distances, acceptable eyes when mixed with heavy illustrations, vibrant colors, better quality, nicer composition, etc.
The ultra seems to be a more updated version
The one that looks rough has been out for quite a while already
Adetailer is still superior in some cases I think, from what I've been testing
But then
Adetailer with sd3

I thought the model releasing soon would be the medium one
I've been looking at the bot images for some time, it was pretty bad quality even a few days ago, very similar to the Clipdrop examples I've posted in here.
I could be wrong, but I think the large one was scrapped all together because the medium one is good.
I've read that it is the 2B one. The medium one they haven't said a date yet (at least from what I could gather)
Yeah, things are changing very quickly, it is better to wait for more "official" statements
👍
The announcement made in #📣|announcements says the medium version will be out on the 12th
Yeah, what they are calling the medium is the 2B one, but there were other 2. When I use the term medium, I'm referring to the other intermediary one, larger than 2B, and then comes the larger one, 8B. The naming conventions they are adopting are confusing enough.
Yeah there was this one. I've read somewhere about a model larger than 2B, something in between 2B and the larger one. They will probably ditch some of these models, or already have. I don't know.
Maybe it was 800M (small), 2B (medium), and 8B (larger).
I remember they used to say that the smaller one was supposed to be run on smartphones, maybe. Now, nobody really knows.
Can someone please expand this image and remove the click to start watermark?
I need it to be Android wallpaper size

NAI on SD3 will be crazy 💀
Have anyone tried nijijourney like how good is it when prompted correctly
why am i asking anyway i aint getting it to work
seen few people try its like good but not insane
ugh, it's finally done... why did it take so damn lo-
i accidentally left the batch count on 4
i tried to click on the grid and the program crashed due to OOM
Pretty sure they plan to make their own base model, not sure if they'll be using SD3. Either way, whatever they go for, they got the computing power to do whatever they want. I'm just curious to see if they can make a model better than their v3 one
They better not make a free trial or anything cause my 10000 alts gonna come in
thx 👍
I love it. It's been a while since I've added anything to my hall of fame folder, and this is worthy of it. Well done

Sushiro
owl is stuck
🤣
evenin'
hi hi here! Python experts have a very interesting question.
How can I clear the data that accumulates?
or rather the site-packages directory
site-packages directory is required to run your python scripts
Assuming you mean the one inside the python directory, and not one in a venv that you're using for something: Run pip freeze in terminal to get a list of all of the packages that are installed then you can run pip uninstall <packagename> (replace the <package name> with the actual names of the packages you want to get rid of.)
If you mean the one inside of a venv, you do the same thing, but make sure to activate the venv first.
No. namely the one that is located directly in the Python directory.
you can also clean up some disk space by running pip cache purge to remove the downloaded install files that aren't being used to run python
Now I have it quite large in size. and I'm sure that most of it is not needed.
How else can I understand what exactly I need and what I don’t?! 🤔
If you're unsure then the simple solution is to copy the list that pip freeze spits out into a txt document so you can pip install <packagename> to reinstall it if something complains that it's missing something. Without knowing what all you're using python for on your system it's tough to know.
I use it exclusively for SD
Then chances are you don't need most of it if your webui was installed correctly since they pretty much all use some form of a self contained virtual environment.
If you ran any pip install commands without activating your venv then you will have accidentally installed your packages outside of the venv.
this happened a couple of times. but it seemed like that when I used the A1111 interface. now I use Forge.
can I just move everything from the site-packages directory to any temporary one and see the result of restarting and running the entire SD?
guys, someone uploaded a trojan as an api custom node on comfy.
stay safe.
https://www.reddit.com/r/comfyui/comments/1dbls5n/psa_if_youve_used_the_comfyui_llmvision_node_from/
Some packages might install things outside of site-packages so you should run let pip uninstall them for you instead of just deleting. The pip freeze command (before uninstalling things) will give you a snapshot of all of the packages along with the versions that are currently installed so that you can reinstall the same packages easily if needed.
i love how the damage control is just shifting the blame on some hacker group (it's literally them)
I understood with these steps. then another question. can I move an already installed python to another partition on my hard drive? By default, as I understand it, it is placed on the system partition.
what do you mean by them?
apple, probably
this guy was too much of an idiot to be an engeneer from apple
ohhh
Yeah I think the malware was already there from the beginning
finally , a bit wierd but ok
any tips for regional prompter sitttings ?
I think on the website in the description everything is described in quite detail. The main thing is to understand the essence of this extension.
will she give me flowers
why not?
Necessarily! but in the prompt it was a basket of fruit.
how original he is
then give the apples back!
Why did she break the glass?
elves love transparent and sparkling stuff and use them as necklaces
they are like crows. carrying everything shiny...
yep, they are!
and hide on trees exactly like them
Why are LLMs so much larger than SD models?
what are you doing here ram

yooooo
It had better or I will GET the people who make SD
x3
@acoustic locust don't worry its just modded
👍
(I asked for sol badguy btw)
lmao
Oh yeah? Where's his bare feet then?
Like half the mods for GG are barefoot mods
wtf
It resembles him ig
Bro just check the GameBanana page for strive
yea it only resembles characters you ask for, it doesn't know much more
like if I do ky kiske it gives me this
dum loras are fun to use
Horny on main
I've never met an I-no main
I was watching gg recently
since slayer released
hes cool
daru is literally the only person I see playing ino lmao
Slayer is the most powerful character in gg as of now I believe
@acoustic locust funny thing I watched 2 recent daru streams
I didn't see him fighting a single slayer
I saw some playing in the tower when he was walking around but not as many as I see compared to on US streams
x3
I've only ever fought one Slayer
I lost so hard I intentionally avoid them now
Same with ABA and Faust
He was probably avoiding them
Probably because he was looking for high level people

slayers are all new so I think he avoided them kinda ye
also
he fought a level 22000 axl
I was like wtf xD
22000 on Axl is some wild work
do you play well
floor ranks are very misleading not sure if asking what floor you play in helps cx
All that remains is to repeat the famous illustration of dogs playing cards! only with a cat twist!
I used to play GG a few years back (Like Xrd and XX) but I only recently got back into GG, so I play Strive now. So I'm only like floor 6.
that one is funny, here one of cats playing poker
middle one is definitely scheming something
Yes, take a closer look! They all have cunning, impudent faces! 🤣
the seasoned cat is a lover of casinos and gambling.
cat thief! 😉
stealing all the cute prompts from you
mugen might just be the best game ever made @nova remnant
I only remember funny things when you say mugen
Like extremely powerful characters AI fighting each other lmao
Can you play one online or something?
meh
onion lamp 👌
and who knows what else exists in the fantasy world. Why shouldn’t this be the case?
yup, no one knows
I don't think so
It's goofed
I love mugen
I forgot i had it installed
I installed it years ago and only rediscovered it yesterday
tell us more about your mugen setup. what characters?
I’m not at my pc rn, i’ll show it when i’m back on it.
It’s pretty vanilla though
nice
It's pretty bare right now, no other addons. Just characters as of now.
the only one I know is 'evil homer' 😄
wow, mugen is still a thing? I remember when that was first released
I thought he looked funny lmao
I only recently rediscovered it on my pc, had it installed for years now lmao
surprised i didn't do more with it
HELLO,SD3
nice
5 minutes of Inpainting, at least it looks pretty good despite the gun looking goofy
np
But man getting 2 characters with correct atributes is not easy
indeed
Even with NAI it's still quite luck based, especially when it's something more complex
:(
i tried
#🆕|sd3 is on fire right now. Lykon is answering everybody. I asked him about the API/upscale, and he said this version in the API with Ultra is not as good as the final one they are preparing. Things will improve further. Very nice.

nom nom
How can I use that thing that says embedding? I don't understand where to put the file or how to use it. https://civitai.com/images/14510180 @vital raptor @steady grail
you put it in embeddings directory and call it adding the name of the file on the prompt.
if you use comfy, you need to add embeddings:name_of_the_embedding
Easy Peasy: https://stable-diffusion-art.com/embedding/
@fresh sleet is pony trained using danbooru tags as well as esix tags?
sorry for pinging i just really want to know
All booru tags are roughtly similar so just use any tags that make sense (with space instead of underscore)
i see, thank you for the reply
.dream prompt 1
@brittle rampart 
Where we go?
and where are the goodies?
your favorite
What else do you eat
I thought you were made of strawberries!
I eat meat! 😁
burgeee
kinda want a burger, been a while since I ate one
Well, where is the meat in it!?
its smol burger
you're too hungry
There is no meat in the burger! The meat must be real! and cooked over an open fire. for example Bonfire!
I have a question! who can explain “on fingers” what “Kohya HRFix” is and how exactly it works and how well or not its use affects the results!?
x3
oh nyo
I put your sentence in the prompt
lmao
Burger on fire!?
just try this part The meat must be real! and cooked over an open fire. for example Bonfire!
thats exactly what I tried :p
burgee 
model in trash! She doesn't know what REAL MEAT is!
model doesn't know how to control the fire
better... but still not the same! it looks like marbled beef!
funya too strict
her burger is so mega big that she holds it with her elbows! I wonder if she will feel bad later?
no burger is too big!
She will definitely be torn to pieces! 😂
This chat has become #food
Well, now I want meat!

time for dessert
🥴
this is the sauce!🤔 maybe..
Ultra, no upscale
is there any cute model for logos? chibi cats and stuff ?
any general SDXL model should be able to do it like dreamshaper
but probably better with loras, there is a bunch on civitai
also SD3 just does it, and better with text
will release soon 
olon johnson
sd3 is out
Chibi for ya
aww
amd life
Why can't those files be kb size
so you'll be fine
because life
remember when this was the best we could get?
hard to believe that was only about 2-3 years ago
😱
couldn't get it to work in forge so i guess i have to learn how to use comfy
deleting it as soon as forge implements support though
I'll just wait few weeks for a more optimized version I guess
is novelai's "vibe transfer" feature replicatable on webui?
i assume it's just some controlnet thing
Not the same thing
But controlnet has a thingy yes
It's called... Reference only
There might be something better
it doesn't look like forge's integrated controlnet has it 



