🍥｜anime

[trim current]

i found something on touhou may help you

sec

[scarlet kelp]

ohh thats a nice graphic, which model do you use?

[wispy canopy]

yeah, i kinda figured that out too, but they often work 'together'

some models have stuff in other blocks

[unreal ridge]

I need to try the supermerger thingy

[trim current]

ahh dang was hoping this would help

yeah supermerging is fun

i also like merge board

cos sigmoid

[zinc flax]

must've been a rough day

[wispy canopy]

IN07 is where the most part for my model fingers is

however, IN07 is also where it's compositional knowledge resides a lot

[scarlet kelp]

ohh how you created the speech bubble?

[trim current]

i need to jump into that, been throwing myself at learning lora/lyco training sicne getting the 3060

[zinc flax]

spoken ellipsis

[trim current]

hell took me three days to gbet a good diverse dataset well trained with several concepts and charatcer from Personal Ami

[scarlet kelp]

i do that since a month and my results still not i want

[keen hollow]

The tfm models for this style. They're some of my favorites to use, a lot of fun. This is the tfm-cutesy one.

[trim current]

give me an exmaple, i got a few good lycos out, only like 2 published tho

dm if easier

[keen hollow]

And it can do Tifas, so I'm happy.

[unreal ridge]

Got something interesting, but bonkable

probably because of the nsfw xD

[wispy canopy]

nsfw is applied for 10% of the steps, so it's just to get a scantily clad female outline

after that it's nto applied anymore

[unreal ridge]

nvm

was judging based on the preview

it's dogshit

[wispy canopy]

pure anime models will struggle if not outright fail that prompt

that was my gripe with alfahentai (my first model merge)

[unreal ridge]

weird one

[wispy canopy]

better armor!

[wispy canopy]

can't get headshotted if head = pure armor

[unreal ridge]

[zinc flax]

never tried prompting stuff like this but it looks cool

[pure oyster]

I posted all my good stuff, now I have nothing to spam chat with 😦

[wispy canopy]

can't generate something new?

[zinc flax]

oh jeez now i didn't expect this

[unreal ridge]

Can't escape the 🍑 with this one

[pure oyster]

I can, but I really like to make large and processed set pieces and they take some hours to make.

[gloomy scroll]

Proof that hair has bones

[zinc flax]

that's huge

[unreal ridge]

guys

i think clyde is gone

[wispy canopy]

this is pretty much on the edge...

not because of clyde, but the implicit nsfw

[magic wave]

O, coot :O

[cyan burrow]

If it weren't for the people here that I'd miss if I ended up banned. I'd take a few more risks on posting art that may be too bloody or violent. But since it seems to be at the discretion of whichever mod or staffer is present and they each have their own limit. It's just impossible to gauge what can and cannot be posted in terms of that content :<

[wispy canopy]

and then there's this guy just posting pussy without any worries!

how lewd!

[gloomy scroll]

https://tenor.com/view/cat-pussy-gif-5284960

[magic wave]

[scarlet kelp]

muahahaha also 2 of them

[unreal ridge]

It's always a good looking hand with an extra finger

[wispy canopy]

pretty clean!

[wispy canopy]

friday night, imma play some games with mates ^^

[keen hollow]

dang, love this but the hand is jacked. 😦

[keen hollow]

ooo what game?

[keen hollow]

I remember when I had friends that played games.

or , well, friends.

sigh...

[wispy canopy]

not decided yet, imma start with battlebit, i suppose

[gloomy scroll]

@supple raptor

[keen hollow]

I heard about that. Sorta like a low-poly shoot 'em up, right?

[supple raptor]

two left feet

[pure oyster]

Oh hey, good catch. I'm not going to ask about your familiarity with feet.

[supple raptor]

https://cdn.discordapp.com/emojis/1094779330099433562.webp?size=48&name=uhhhhh&quality=lossless

[gloomy scroll]

Fingers

[wispy canopy]

and it has 127x127 😄

[unreal ridge]

made by 4 people or something

[wispy canopy]

and honestly, graphics never made a game last long

[keen hollow]

lotta folks

[wispy canopy]

gameplay makes games last long, and this is just waht battlefield SHOULD be

[unreal ridge]

karma'd

[magic wave]

OH AY

I PLAY THAT TOO

:D

[magic wave]

For sure, Battlebit is so good

[upper wadi]

*Hmm

Maybe 2.1 on it's own isn't smart LOL

[supple raptor]

https://cdn.discordapp.com/emojis/1025459379287376023.webp?size=48&name=feetpics&quality=lossless

[magic wave]

Its momming hours

[dawn ingot]

Not my type tbh

[dawn ingot]

Too muscular

[pure oyster]

No love for muscle mommies? 0/10

[dawn ingot]

I do like muscle but not too much

[dawn ingot]

This is my type

The girl on the left is who is my type

The last one is good too for me

[keen kayak]

Experimenting with more dynamic looking compositions

[dawn ingot]

Old photo I made it's kinda alright

[zinc flax]

Forgot to add the vae

[dawn ingot]

Vae?

Some different old photos

[dawn ingot]

What's vae and how do I use it???

[dawn ingot]

Sheeeesh

[zinc flax]

I guess the oversimplified explanation would be something to add more color + detail to your generation

https://civitai.com/models/22354?modelVersionId=26689

put in on /models/VAE/ and then go to settings and apply it

[dawn ingot]

I cant use it

I'm on mobile

[supple raptor]

VAE is the part of a model that translates the numbers from all of the calculations into pixels that you can see as an image. There are different VAEs available that can have different effects on things such as small details like fingers or hair as well as effects on brightness and contrast.

[dawn ingot]

Yeah ok but I'm on mobile so I can't use the download

😭😭😭

[pure oyster]

ahh, I take it you are using a service for the images?

[dawn ingot]

What?

Wdym

Like wifi

[pure oyster]

Like an AI website that you pay for with tokens

[dawn ingot]

No I don't I'm using the app

[zinc flax]

Huh, which app

[dawn ingot]

The SD app

[zinc flax]

Gonna assume you're using your own colab then

[dawn ingot]

The top one

[zinc flax]

First time seeing that app

[supple raptor]

There is no "SD app" afaik. What you're using is an app that some other service created that very likely uses SD as a backend for their service.

[dawn ingot]

Ah

Ok

That makes sense

But it's useful

[supple raptor]

If it works, it works and there's nothing wrong with that. But if you're paying for it, then I would recommend looking into free options such as google colab.

[dawn ingot]

It's free

All of it

[zinc flax]

Kinda sucks that it doesn't have baked in vae for the model though

[dawn ingot]

Yeah but otherwise it looks great

[zinc flax]

Color looking kinda bland

[dawn ingot]

That's another photo I made with it

[supple raptor]

Small correction. It does have a baked in VAE, just not a good one. Without a VAE you can't get an image.

[dawn ingot]

Ok

[pure oyster]

also, it is only doing 512x896 and only upscaling it

[dawn ingot]

Rate this

[zinc flax]

Ah, welp something like NAI vae would be nice although still a bit washed

[dawn ingot]

Rate this

[unreal ridge]

🤔

never seen this one before

[dawn ingot]

It's pretty good

And it's free

That made the image above

[supple raptor]

It probably is NAI's VAE, just using a model that it wasn't used for.

[pure oyster]

Isn't NAI's oranagemix.vae?

[zinc flax]

Could've sworn NAI's vae doesn't look like that though

[supple raptor]

That's backwards, but yes. Orangemix and AnythingV3 along with countless other models all stole and renamed NAI's VAE.

[dawn ingot]

It's just titled stable diffusion Ai

My friend star uses it

[supple raptor]

Depends on the model. It's the most washed out VAE out of all of the popular ones though.

[unreal ridge]

Won't even bother searching

[zinc flax]

Speaking of AI services, the only one I've tried was pixai
It's alright

[dawn ingot]

It's just stable diffusion Ai on play store or app store

[unreal ridge]

The app name is basically AI art.

won't find shit

lmao

[pure oyster]

If you do it on your computer, you can turn the first image into the second one.

[dawn ingot]

I see

I like the shading better on the first one

Fsr

[pure oyster]

The first one has no shading lol

[keen kayak]

It's not just about the shading but the added foot, the fixed clothes, the overall level of detail

[dawn ingot]

It kinda does

I'm gone

[cyan burrow]

IMG1
Kid: "Daddy why are the humies so big and run everything?"
Dad: "One day we'll be as large as the humies, you'll see"

IMG2
Kid: "Daddy why are the humies killing us?"
Dad: "Cause we made our nests too big son, one day we'll be smaller and safer. Promise"

IMG3:
dead

[nova remnant]

did anyone try ai animationS?

[trim current]

go add sd_vae in your quicksettings, refresh ui

[untold glacier]

Boy this took a fair amount of work. The text (including behind her head) essentially says "Goodbye! Live your life to the fullest!"

[stable crane]

damn nice!

[stable crane]

special chalk font? how'd you do that

[wet elm]

DAYUM.

[untold glacier]

It's Nagurigaki Crayon, a free Japanese font online. The translation is DeepL, so it should be pretty accurate. It's also been distressed a bit by me, and blended appropriately. One of the exclamation points is actually from a real japanese chalk drawing, and the other is just an edited version of the font's.

[nova remnant]

Hi anyone ever animated ai?

[south sable]

https://tenor.com/view/re-life-money-anime-take-this-gif-20129152

for the font

[untold glacier]

https://tenor.com/view/yy-smoking-counting-money-anime-gif-17852058

[limpid coyote]

Yo anyone know of some good discords that are anime focused? Been searching for some but happy to be here too

[drowsy dove]

It's crazy how "normal" objects can make some generations challenging

To stop inpainting this forever, I'll send it here and call it done

[cyan burrow]

I would have had them sitting on like office chairs or something and then inpainted a car around them

[drowsy dove]

This is the original image, glad I sent it here: #📝｜prompting-help message

I think it ended better than what it started, but there were tortuous lines everywhere. And I got the idea to inpaint certain parts of the image scaled up a bit too late

For my next picture (whatever it's going to be), I will upscale early and then inpaint the upscaled version, hopefully it will be easier then

[cyan burrow]

The Crow God, directing crows to vengeful souls whose scream of agony is powerful enough to reach his ears. They raise from the dead as revenants, gaining nigh-invincibility as long as they're on their quest for vengeance and the crow remains alive and well. After vengeance is done, the Crow God eats the soul of the avenger. As payment

[untold glacier]

This is incredible and fantastic. I have a new addition to my Hall of Fame folder. It fits so well as a wallpaper on my ultrawide--well done sir.

If you had to give the image a name, what would it be?

[cyan burrow]

Is that hair growing from the ceiling?

I know...

"why am I like this"

[shy gyro]

Anyones knows about this model?

I'm looking for a model which can generate good anime landscapes and scenery

[supple raptor]

looks like it could be some version of Counterfeit. It could also be almost any anime model with some LoRAs applied to it. It's impossible to know from just an image with no metadata. You'd be better off asking whoever posted it.

[untold glacier]

Amazing what you can do with CFG 1 nowadays.

[slate night]

Suspicious pink dot placement on the shirt of the second image lmao

[untold glacier]

She just got uh... ketchup there. Pink... ketchup... yeah...

[slate night]

[pure oyster]

I did a terrible job on this one, but I think it is still postable.

ignore the baby hand

[shy gyro]

Any model for anime landscapes and scenery? Aesthetic ones , Lora too

[untold glacier]

A LoRA I have but haven't used yet is https://civitai.com/models/14605/howls-moving-castle-interior-scenery-lora-ghibli-style-v3
As for actual models, I mainly only use a couple now but they both work great with such things.

• Aurora (for booru language-based prompting): https://civitai.com/models/40199?modelVersionId=45601
• DarkAlfa (for natural language-based prompting): https://civitai.com/models/64843?modelVersionId=74900
  With such things, I've found it often comes down to how you prompt. Many models can do fantastically if you're inventive enough with your prompting, though obviously models and other networks that are specifically trained to do such things will do them more easily. Your best path forward is finding a model you like and looking at how other people prompt to get results in the direction you're looking for. An easy way to do this is to search on Civitai's images: https://civitai.com/images?tags=8363 To do this, open an image you like and you can see what keywords they used, the language, and (usually) the general settings and models they used in the info that appears.

[shy gyro]

will you pls tell the upscaler you used?

[untold glacier]

Upscaling model: 4x_fatal_anime - https://de-next.owncube.com/index.php/s/x99pKzS7TNaErrC
As for upscaling method, I use Multidiffusion/TiledVAE which is a fantastic extension that works with hires fix and i2i in Automatic1111's build of SD. It's a lifesaver for VRAM and maintaining detail with ease without generating things you don't want on the upscale.

[shy gyro]

everything about this

[cyan burrow]

I think I'm done merging and training models to release. There's not much point with so many popular ones doing similar stuff, albeit not as versatile

[next thorn]

thanks for this!!!

[cyan burrow]

I might merge things in for myself if I need something but...I think I'd rather put my focus on lora's

the only thing I can actually train

[shy gyro]

prompts and the negative??

[cyan burrow]

You can also do more upscales, like upscale 2x, then 1.5 and then 1.25 I guess if you wanna upscale 4x

[untold glacier]

You can find such info by downloading the image and then plopping it in the PNG Info in your Auto1111 interface. The menu option will look a bit different than mine as I'm using a UI extension, but the option will be there.

[shy gyro]

it tells everything??

[untold glacier]

Usually it'll say most things, yep! Some extension-based info it may not have but if the original user wanted to keep the SD info in the image when it was generated, it should show up.

[cyan burrow]

the duality of elf kind

[shy gyro]

its not working

[shy gyro]

tell me atleast the negatives you used for this one

[untold glacier]

Open the image in a browser. Then download it. Then drop it into the Auto1111 interface WHEN you're on the PNG Info tab. You'll then have the info you need.

[shy gyro]

damn there is lot of stuff in that, what is that A deatailer??

detailer??

[untold glacier]

I can't love the image enough help

[south sable]

png info intact, try it out! :p

[pure oyster]

He has more mods for his stable diffusion than most have for skyrim. I wouldn't worry too much about what is what.

[untold glacier]

ʙᴜᴛ ɪ ʜᴀᴠᴇ ᴏᴠᴇʀ 1000 ᴍᴏᴅs ғᴏʀ ᴍʏ sᴋʏʀɪᴍ

[shy gyro]

I create these kind images

[shy gyro]

but , I get wires like stuff, so post processing in photoshop takes a lot of time, that's the reason I'm searching for an alternate model

[untold glacier]

Adetailer is a fantastic extension that I use in literally EVERYTHING I gen. It automatically detects faces and inpaints that section at a higher resolution so you never get things like this:

But instead, things like this.

[shy gyro]

how to use that

[pure oyster]

Post processing is the easy part

[untold glacier]

You can turn it on and never have to worry about it again.

[untold glacier]

Get the extension, turn it on and set the ADetailer model to face_yolov8n.pt
The end.

[untold glacier]

Available tab

[shy gyro]

i have de detailer

[untold glacier]

ADetailer

[shy gyro]

mine available section does not work

[shy gyro]

takes infinity

[untold glacier]

Then put this in the install from URL tab: https://github.com/Bing-su/adetailer

[pure oyster]

Keep in mind it increases gen time significantly and usually there is no reason to inpaint useless seeds.

[untold glacier]

I disagree with your assessment. But to each his own.

[shy gyro]

my gpu can handle that

[pure oyster]

It wasn't an opinion lol, it's math

[south sable]

https://tenor.com/view/chips-eating-anime-lunch-break-gif-11881516

[untold glacier]

You say that it increases gen time significantly--it doesn't for me. It's not double the time. It only inpaints the section of face, and only if it detects one. Most of the images I generate are potential contenders for usage--If I'm worried about a potential terribad gen--then I can click skip.

[shy gyro]

thank you for all the help

[untold glacier]

Glad to help! Hope you find great usage out of this.

[shy gyro]

I prefer landscapes without any character

[untold glacier]

There are some beautiful landscapes and scenery you can make. SD really is amazing in its versatility.

[pure oyster]

You must have a really dank GPU

[untold glacier]

3080. I used to think it was the bomb--though I always thought it could use more memory. Now I'm running LLMs waiting 'till eternity for tokens to generate.

[shy gyro]

currently I use meina mix, i gives tyhe anime feel and some 3d touch at the same time

[south sable]

AnimePunkV10 lately for me

[untold glacier]

Nice! I just wish that I could find an upscaling model that lent a 'painted' look to the image without it significantly increasing the contrast and/or darkening the image like fatal does.

I've tried a few recommended by others, but nearly everything seems to blur the resulting image.

[shy gyro]

i use anime+4x sahrp

[south sable]

that one is good. I also use fatal

[untold glacier]

I use it from time to time, but it smoothes things out a bit too much compared to ultrasharp or fatal if I'm going for a stylistic image.

[shy gyro]

scripts?

and then regular upscaling

[south sable]

sometimes I'll use animesharp in the base gen then in upscale i will switch to fatal

[supple raptor]

Here's a landscape prompt I was playing with yesterday.

[shy gyro]

not my style

[untold glacier]

Love the look of the water.

[supple raptor]

The pensketch style loras are fun

[untold glacier]

Hmm. Haven't tried any--I've always been brute forcing it with composable diffusion. What do you recommend is best?

[supple raptor]

I use all 3 versions of this for slightly different looks. https://civitai.com/models/28687/pen-sketch-style

[south sable]

when you finally remember that thing you forgot

[untold glacier]

Top notch stuff!

[shy gyro]

upscaling is the problem

[pure oyster]

If you want a very balanced 2.5d checkpoint I use "Simply Beautiful" https://civitai.com/models/49037?modelVersionId=53632

[shy gyro]

any suggetions , how I can improve them further

[supple raptor]

Upscaling. Composition is nice, but they're all a bit blurry due to being low res.

[pure oyster]

Don't upscale without inpainting, because if you zoom in, you'll just break the illusion.

[shy gyro]

AND HOW TO DO THAT?

[untold glacier]

How to Upscale with the Most Detail and Least Problems:

• First off, hires fix with Multidiffusion+Tiled VAE at a relatively low denoise (0.4-0.5ish depending on the model and style you're going for. Anime=less usually) with something like fatal_anime as the upscaling model.
• Then take that into your i2i and then upscale again with something like ultrasharp or your preferred upscaling model at no more than 0.3 denoise. Usually in both processes I set the upscale factor for 2x. So 2x for hires fix and 2x for i2i (though you can often go higher at this step.)
• Try to simplify your i2i prompt, keeping quality or generic tags such as "cinematic, detailed, vibrant lighting, fantasy" etc, so it's less likely to generate people or objects in areas they shouldn't be.
• If you are looking for specific details in certain areas then use multidiffusion's region control to prompt/neg prompt specific details for that area, see image. Just be sure to Enable Control for the region prompt itself, as well as enable a region and import the original image to see where you want the section and adjust the size+position as needed.
• To be clear, you'll be using multidiffusion** in place of** any other upscaling method. This means in conjunction with hires fix in t2i and in-place of sd upscale in i2i.

[shy gyro]

i have the tutorial for multi diffusion

[untold glacier]

It's good to know the extension backwards and forwards! But if you want the multi-part process that I use to gen with as much detail as possible with little chance for things you don't want, my steps are up there.

[supple raptor]

good tutorial. I would add some tips for modifying the prompt when running multi-diffusion upscaling in img2img. Removing all of the descriptions of your character is a very good idea to prevent it from trying to change things in the background into a person.

[untold glacier]

That's in there I thought

[supple raptor]

It sure is. My brain just decided I can't read today XD

[untold glacier]

Yeah, 3rd paragraph: Try to simplify your i2i prompt, keeping quality or generic tags such as "cinematic, detailed, vibrant lighting, fantasy" etc, so it's less likely to generate people or objects in areas they shouldn't be.

[shy gyro]

i have a screeenshot of it

[untold glacier]

Thought I forgot something for a sec

[supple raptor]

when in doubt. just make the image bigger. moar big = moar gud

[south sable]

big art is good art

[untold glacier]

https://tenor.com/view/vince-mcmahon-shocked-open-mouth-wwe-wrestling-gif-17762188

[pure oyster]

Yep, I just brute force it with resolution myself.

[shy gyro]

4x fatal 500000G right??

[untold glacier]

Yep

[supple raptor]

If more resolution doesn't fix it then it just means WebUI is being dumb

[untold glacier]

Just keep in mind that it **will **deepen the contrast of your image.

I usually use it in conjunction with another upscaler at different parts of the process.

[shy gyro]

i want that to happen

[untold glacier]

Then boy are you in luck

[pure oyster]

I should stop reposting, but zoom in bitch

[untold glacier]

Nice clear focused details. Well done.

[supple raptor]

Depends on the image that one is a great all around upscaler. DigiPaint35000 is great when you want more natural looking brush/pen/pencil strokes.

[untold glacier]

Oh I've been meaning to tell ya. I tried that one out and it just doesn't give me the detail that I'm looking for. It could just be that it's meant to be used in SD with a higher denoise then I've been using. Some upscaler models are like that. But I just wasn't getting the crispness of the details compared to fatal or ultrasharp.

[supple raptor]

https://tenor.com/view/chefs-kiss-french-chef-perfect-dish-excellent-food-perfection-gif-20341505

[limpid coyote]

I've seen people mentioning control net but not sure what it is. Is it a model specifically for in/out painting?

[shy gyro]

i believe its the biggest tool in SD

[supple raptor]

Yeah, it shouldn't give a high level of crispness. I use all of my ESRGAN upscalers at 0.51 denoising in Hires fix and 0.38 in multi-diffusion.

[pure oyster]

It's an addon that uses reference images to pose/outline your generated image.

[untold glacier]

You don't use t2i hires fix with multidiffusion? What size are your upscales usually?

[limpid coyote]

hmm not sure when I'll mess with it. Still pretty new, getting to stuff like this

[supple raptor]

Here's the controlnet extension. The models are linked in the repo. The model for inpainting is just called control_v11p_sd15_inpaint. https://github.com/Mikubill/sd-webui-controlnet.git

[untold glacier]

The denoise here is higher than I usually run at the hires fix step, so that'll still give more detail than I've been getting. I'll have to try it again to see if it makes any difference.

[supple raptor]

I use hires fix on it's own in t2i always at 2x multiplier with my smallest dimension set to at least 512. Then if I want to upscale further I use Mutidiffusion in i2i, again at 2x multiplier. That last one I posted started as 2048x512. Hires fix brought it to 4096x1024, and multidiffusion brought it to 8192x2048

[pure oyster]

I'm going to yoink this guide in case I ever decide If I want to know what the hell the Multidiffusion addon does.

[untold glacier]

I'm surprised I'm not seeing more evident tiling artifacts at such a high base resolution without multidiffusion. Was it the cream of the crop? Either way, it turned out fantastic.

[shy gyro]

amazing bro

[supple raptor]

Hires fix doesn't do tiling at all. It does straight txt2img, then upscales, and finally straight img2img on the upscaled img.

[shy gyro]

anyone has preset for prompts?? , negative included

[untold glacier]

Sorry, you misunderstand me. I meant when SD generates multiple perspective vanishing points in a single image or mountains in the sky as well as on the bottom of the image, tiling artifacts such as that--which generally happen when generating beyond the max trained size of the model.

[supple raptor]

For the actual prompt, I start from scratch on every one I write. This is my negative that gets used on every image I make, only adding to or removing from it when needed for the prompt I'm running.
negative_hand, (KHFB, AuroraNegative), (Worst Quality, Low Quality:1.2), border, nsfw, skimpy, grayscale, multiple_girls, (watermark:1.2)

[supple raptor]

Oh, yeah I know what you mean. That prompt is pretty resistant to that effect. I find that having the background relatively close or very far away tends to reduce the chance of that happening. It's the mid ranges that it really struggles with.

[shy gyro]

what is that aurora negative

[untold glacier]

My default DarkAlfa (anime model) negative (includes negative embed, "negative_hand"):
negative_hand, (low quality, worst quality, lowres, error, fault, undefined, crude:1.3), jagged, spiky
BREAK
(disfigured, unclear, indistinct:1.3), ugly hands, extra arm, extra hand, split arm, missing finger, extra finger, three fingers, four fingers, six fingers, merged fingers, (bad anatomy:1.3) misplaced hand, misplaced foot, (text:1.3), (signature:1.3), (title:1.3)

Light version (recommended to start with):
nsfw, negative_hand
BREAK
3d, worst quality, low quality, bad anatomy, bad hands, text, extra digit, fewer digits, signature, watermark, username

My recommended negative embeds are negative_hand, auroranegative, and easynegative.

[pure oyster]

exact copy paste from my txt doc lol

Prompt: (x:1.0), (x:1.0), (x:1.0), (x:1.0), (x:1.0), (x:1.0), (x:1.0), (x:1.0), (x:1.0),
Negative: bad_prompt:0.8, bad-picture-chill-75v, EasyNegative, (asian:1.9)

[shy gyro]

you guys are so helpful

[supple raptor]

negative_hand KHFB and AuroraNegative are all negative embeddings. The last two are in the embeddings folder of this repo https://huggingface.co/SweetLuna/Aurora

[untold glacier]

Negative hand: #🍥｜anime message
AuroraNegative: https://huggingface.co/SweetLuna/Aurora/tree/main/AuroraEmbeddings
EasyNegative: https://civitai.com/models/7808/easynegative

[shy gyro]

lol I was searching on civit

[untold glacier]

EasyNegative is there, but the others aren't.

[supple raptor]

Here's negative_hand https://civitai.com/models/56519/negativehand-negative-embedding

[pure oyster]

Keep in mind that your images are still gonna look like eldritch abominations with negative embeddings, they just make it a bit more tolerable.

[untold glacier]

I already linked it above, lol. But one should rate it highly on civitai via your link if they use it.

[supple raptor]

Oh I missed it because I was only reading the hyperlinks XD

[shy gyro]

thank you,

[untold glacier]

I just started looking at your prompt, and boyyyyyyy... all those parentheses, lol.

[limpid coyote]

Anyone got useful tags to adding color to an image?

[untold glacier]

Negative prompting monochrome, black and white, and greyscale add color.

[limpid coyote]

Trying to get more color in this through image2image

[limpid coyote]

I'll try that

[untold glacier]

Wow. You using a vae with that?

[shy gyro]

how to add aesthetic colours to the images

[limpid coyote]

This was the first iteration

[untold glacier]

Anime2 should help. That and if you're upscaling, using fatal_anime would help as well.

[limpid coyote]

then fixhighres sapped all the color out

[pure oyster]

The best option is to put monochrome in positive prompt and color it yourself to flex on everyone.

[untold glacier]

Hmm. I've been having that bug happen to me recently, though usually there a visible dark artifact somewhere in the image with it. Dunno if it's a bug in your case though.

[cyan burrow]

but monochrome still has a variety of shades that each represent a color, identify it and raise those colors with curve on hue control and you're closer

[limpid coyote]

I think the vae I was using didn't work well with it. or the negatime prompts worked

Started from the first iteration again and seems to be giving me better results

Nevermind

😵😵

[cyan burrow]

dont suppose anyone has any brush suggestions for CSP?

that are free in the marketplace

oops brb morning race taking place in my bowels

[untold glacier]

If all else fails, try restarting SD and then restarting your PC.

[limpid coyote]

I'm trying NAI vae with it now and it did better

[cyan burrow]

that looks familiar

[supple raptor]

Throw in some manual edits to flex a little harder.

[pure oyster]

Even gray blobs with no shading is no match for me!

[cyan burrow]

heh

raising color values on shading values got you cruising

or just hitting autofill in CSP

[cyan burrow]

did you by chance post this on my model page?

[untold glacier]

Help I Can't Increase My Contrast when Generating!

• Use kl-f8-anime2 VAE: https://huggingface.co/hakurei/waifu-diffusion-v1-4/blob/main/vae/kl-f8-anime2.ckpt
• Up your CFG scale.
• If you want to keep your CFG low, use Self Attention Guidance. Enable it and start at 1.0 for scale, and 1.0 for SAG Mask Threshold. If you want it to be more powerful, try lower values like 0.5 and 0.5.
• If you're upscaling, use the upscaling model 4x_fatal_anime: https://de-next.owncube.com/index.php/s/x99pKzS7TNaErrC
• You could just be having a bug within SD. Try restarting it, or your computer.

[limpid coyote]

Thanks for putting that together!

[pure oyster]

I'm a huge fan of orangemix.vae, but it gives me "NaN errors" 90% of the time

[cyan burrow]

isn't orangemix just anyv3?

renamed

[pure oyster]

IDK, it has like a morbillion names

kl-f8-anime2 VAE is too saturated and overwhelming for my autistic eyeballs

[zinc flax]

nai is nai

[cyan burrow]

Cause I doubt they made a vae, when they've afaik never trained a model. "Only" done merges. Which isn't to be discounted or I'd be discounting myself. But still

[cyan burrow]

tried "official" sd 1.5 vae? vae-ft-mse-840000-ema-pruned

[zinc flax]

Almost looks the same as anime2 imo

[cyan burrow]

It does 2.5d, pseudorealism etc better

less saturation

better gamma control

which makes it work great with offset noise loras

[pure oyster]

I did testing myself and saw maybe a 5% diffrence, but that was a while ago.

[zinc flax]

I use either nai or clearvae since I mostly use 2d anime models lol

[pure oyster]

Also, putting worst quality in negative prompt didn't do anything either, but that might only work on certain models.

[supple raptor]

I just fixed the contrast problem with kl-f8-anime2 and use that for everything so I don't have to deal with the terrible textures that all of the NAI vaes do on foliage or the NaN errors you get unless you run --no-half-vae with them, or the bad anime eyes that mse-840000 gives.

[cyan burrow]

how'd you fix it?

bake and shake?

[untold glacier]

Fixed via fine-tuning?

[pure oyster]

Putting Greg Rutkowski in prompt fixed it.

[supple raptor]

VAE Blessup script. Set the contrast multiplier to 0.9. https://huggingface.co/nubby/kl-f8-anime2-blessed

[limpid coyote]

Whoever linked that SimplyBeautiful model, I'm loving it. Great model

[untold glacier]

[supple raptor]

Don't for get Ilya kushiavnlaynvov

[cyan burrow]

thanks, so the vae blessup can be used on any? wjats vram req?

Or Kentaro Miura

[pure oyster]

It might not be the best checkpoint, but it is the best for 2.5d

[supple raptor]

Blessup is run on cpu. It takes like 15 seconds to modify the VAE and maybe 1GB of RAM at most, then it saves the VAE and you just load the modified version like normal.

At least I think it's on CPU. In any case, the requirements are basically nothing.

https://github.com/sALTaccount/VAE-BlessUp

Come to think of it, it is run on GPU because I had issues with pip trying to install CPU pytorch at first.

[pure oyster]

oh, that's the artist who's work is on like every single EDM youtube video

[supple raptor]

that nobody can spell the name of, yes. He's also know for creating the character designs for Ghost in the Shell: SAC 2045. He did the music video for Let You Down from Edgerunners as well.

[cyan burrow]

thanks

oh and I've finally started writing again

my brain is returning to better energy levels after cancer and surgery and all

[supple raptor]

Thats good to hear! https://cdn.discordapp.com/emojis/1091081887134515220.webp?size=48&name=verynice&quality=lossless

[cyan burrow]

thank you, yeah I'm really glad. I'm also gonna buy a bicycle and start getting some exercise in

[limpid coyote]

That's some good fucking shit, congrats

[cyan burrow]

start by travelling around town and then outside and then around my entire region/state

[cyan burrow]

Thanks. Second time too. Hopefully no third

[untold glacier]

Kuvshinov. Early on I used it so much I memorized it.

Just like good ol greg

[supple raptor]

https://cdn.discordapp.com/emojis/1093686767808425994.gif?size=48&name=Fran_Nod_Yes&quality=lossless same

[cyan burrow]

Oh I just thought of a challenge for myself

make a lora that responds to lyrical prompts

this will come to image selection and captioning

[supple raptor]

That might be a bit too much to ask of the text encoder training in a LoRA. Might work fine on a model that already responds well to lyric prompting, but I don't think it will help much for models that aren't already good at it and will probably just be incompatible with them.

[silk narwhal]

Asked ChatGPT to write a lovecraftian story based of the lyrics from the Oshi No Ko opening … pretty nice read 😂

[cyan burrow]

I dunno...not if there's like a shitload of keywords

[limpid coyote]

Anyone know how to change the individual looks of two people in the same prompt? Is there something like "(1girl: pink dress, blonde hair)" Or anything

[silk narwhal]

in painting and masking

[zinc flax]

I use latent couple for that

[supple raptor]

the best solution is just prompting them separately with something like latent couple, inpainting, or regional prompting.

[limpid coyote]

hmm I haven't had good luck with inpainting. Not too sure how to use it and I've been too lazy to figure it out but I think I gotta now lol

I want to get this cleaned up

[untold glacier]

Multidiffusion's Region Prompt Control can also help with that at the i2i stage.

[pure oyster]

I think I found the prompt for my next project, Just need to find a useable seed and inpaint in 4x resolution.

[pure oyster]

too weeb 0/10.

[supple raptor]

https://cdn.discordapp.com/emojis/1009619409347809370.webp?size=48&name=OkinaCry&quality=lossless

[vocal elbow]

10 for weeb

[untold glacier]

[supple raptor]

https://cdn.discordapp.com/emojis/1093687411650875462.gif?size=48&name=Pat1&quality=lossless https://cdn.discordapp.com/emojis/1093687432580440114.gif?size=48&name=Pat2&quality=lossless

[pure oyster]

Fun fact, it takes me 4.5 minutes to generate a 1024x1024 image with my 1060.

[supple raptor]

She heard someone was trying to deny their inner weeb in #🍥｜anime

[limpid coyote]

I have conquered inpainting. A whole new world has been opened to me

[pure oyster]

It takes me 6 hours to craft my 4000x4000 setpiece images.

[shy gyro]

how you guys get seed, or use randomly

[untold glacier]

She's also not happy about the weeb denial here.

Also, nice prompt. @south sable

[supple raptor]

random, there's no science behind any given seed being better than another one aside from on the prompt you tested it on.

[cyan burrow]

I'm having chatgpt write javascripts for tyranobuilder for me

[pure oyster]

Would a weeb color 183 pages of One Punch Man for absolutly no salary? I think not.

[cyan burrow]

maybe I should go with unity or something and have it write c# or c++ whichever it uses

or unreal engine

make an entire game with chatgpt

[pure oyster]

Using chatgpt I tried to make a batch script to extract information from dwarf fortress .txt files, it did not go well at all.

[cyan burrow]

gave me this for unreal engine
```// Include necessary headers`
#include "CameraLockScript.h"
#include "Engine/World.h"
#include "GameFramework/PlayerController.h"

// Constructor
ACameraLockScript::ACameraLockScript()
{
PrimaryActorTick.bCanEverTick = false;
}

// BeginPlay function
void ACameraLockScript::BeginPlay()
{
Super::BeginPlay();

// Lock the camera in the desired direction
LockCamera();

}

// Function to lock the camera
void ACameraLockScript::LockCamera()
{
APlayerController* PlayerController = UGameplayStatics::GetPlayerController(GetWorld(), 0);
if (PlayerController)
{
// Set the camera rotation to the desired direction
FRotator CameraRotation = FRotator(0.0f, DesiredYaw, 0.0f);
PlayerController->SetControlRotation(CameraRotation);

    // Disable player input for rotation
    PlayerController->SetIgnoreLookInput(true);
}

}```

which was to get a visual novel feel

so the player can only use the clicker

I dunno if that's correct or not, I don't speak c++

[supple raptor]

surround code with 3 backticks on each side so you get a box like this

#include "CameraLockScript.h"
#include "Engine/World.h"
#include ```

[cyan burrow]

thanks for the suggestion

[supple raptor]

Huh... I wonder of that crystal just grew out of her head or what

[cyan burrow]

no she placed it there a few years ago and thought it was an egg, said she'd nurture it. Unfortunately it was an invasive crystalline plant that burrowed into her brain and now speaks to her. It calls itself Codusberry

What does it want? Who knows, it's as dumb as Gary

[untold glacier]

Aight. That's all for me tonight. G'night folks.

[supple raptor]

https://cdn.discordapp.com/emojis/998672574345121855.gif?size=48&name=ChenWave&quality=lossless

[magic wave]

[pure oyster]

This took 70 hours of non-weeb activity https://cubari.moe/read/imgur/hYm5au7/1/1/

[cyan burrow]

Im gonna see how good chatgpt 3.5 is with prompts

after being instructed in which manner to write them

no but fucken hell did it make some creepy shit

like holy hell what is this nightmarish place

[supple raptor]

Time to refine your instructions. ChatGPT is more than capable of writing good prompts. Once you find instructions that work well, save them in a .txt document or something because it will lose context over time and you'll need to clear history and start again.

[cyan burrow]

Im using it on openai site

[pure oyster]

I've had great success using it to give prompts for clothing specifically.

not so much for faces

[cyan burrow]

I get nothing for this link, it's only in case anyone was interested in getting Clip Studio Paint but couldn't afford it. Now 60% off and it's for the non-subscription version. The full software
https://www.clipstudio.net/promotion/giveaway/en/?utm_source=clip_ice&utm_medium

[keen kayak]

Robomaid hours

[vocal elbow]

still has trouble on fingers

[limpid coyote]

This one came out ok for being a first generation

[supple raptor]

That's an easy one to fix at least.

[magic wave]

True, really pretty tho :o

[vocal elbow]

use photoshop layer?

[magic wave]

O damn sample pic looking really good

Just halfway through tho, a lil worrying

[zinc flax]

Cooking

[magic wave]

Yeee

[zinc flax]

I should go back to cooking loras
It's fun

Though mostly character loras instead of style

[magic wave]

It do be, been making LoCons with multires noise and its pretty fun

Ooo fair then :o

[zinc flax]

Skill issued in making style loras lol

[magic wave]

I just make styles xD

Weird :O

[magic wave]

Calm LoRA trained kinda mid

[shy gyro]

check them

[cosmic sleet]

black magic with the hand right there

[stable nimbus]

controlnet?

or just luck

[cosmic sleet]

Luck

Don't know on how to use controlnet

[bitter latch]

last one

off to work.. sigh.. to listen to people whine about 'where's my package, make fedex go faster... whine whine" and no I dont' work for fedex

[abstract aspen]

lmao it be like tht

[bitter latch]

anime customer whining about fedex,

[abstract aspen]

what was your prompt?

[brittle jay]

hello

been afk for like a day or two because I have just setup dualbooted kubuntu

in any case, what is the best way to torrent anime in high res?

[lethal laurel]

i made this! ❤️

[brittle jay]

what cfg?

[lethal laurel]

oh

7

[shy gyro]

it doesn't have any metadata

[compact warren]

Why are u asking that here lol

[brittle jay]

Why not

This is anime channel

[wispy canopy]

and it is of questionable legality depending on country

[limpid coyote]

nyaa.si is the best site for it that I'm aware of

[brittle jay]

ye yumeshiro dm'ed me saying the same

[brittle jay]

uh

tell me about 1 person who watches anime unpirated

they don't exist

[limpid coyote]

Loving this model

[limpid coyote]

I'm doing 6x6 grids of the few models I have right now with same prompts/settings to see how different they are, I'll post here if anyones interested

[quiet bramble]

What is KHFB?

[brittle jay]

one of the aurora negative embeds

https://huggingface.co/SweetLuna/Aurora/tree/main/AuroraEmbeddings

[quiet bramble]

Thank you very much. Do you have any issues using them both simultaneously?

[lucid rapids]

anyone know any similar models to create this type of art

[zinc flax]

Try this prompt anime screencap, retro style
Aurora should work iirc

[lucid rapids]

oh ok ill try

[brittle jay]

no

that's what most people do actually

use them together plus some other embeds

[quiet bramble]

That's good to know.

[zinc flax]

man i love this pose, so cute

[limpid coyote]

Would you say that it's ☆: .｡. o(≧▽≦)o .｡.:☆ KAWAIIIII! ?

[compact warren]

[zinc flax]

https://media.discordapp.net/attachments/818995697042325517/951821545569943552/hapee.gif

[limpid coyote]

Any extensions for A1111 you guys really recommend?

[stone onyx]

prompt pls

[zinc flax]

Got the metadata in it

[stone onyx]

O

[stone onyx]

It doesn't work :c

[zinc flax]

aw

1girl, tank top, claw pose, half-closed eyes, [purple hair:black hair:8], open jacket, long sleeves, brown eyes, x hair ornament, shoulders, blunt bangs, trousers, outdoors, grin

[limpid coyote]

6x6 of all the models I've been using right now. With same seed and prompt

[zinc flax]

But what's that in her mouth though

[unreal ridge]

something

[limpid coyote]

I thought the mouth hold tag would make her cover her mouth lmao so it's probably supposed to be grass or cloth

Didn't bother changing the prompt when I realized

[unreal ridge]

Dark Sushi ones looks nice

[limpid coyote]

Yeah my favs are darksushi, meina, and simpylbeautiful

[limpid coyote]

Is this first generation or inpainted n stuff?

[unreal ridge]

first

but i'm using adetailer

[limpid coyote]

Which model is it? Looks good

[unreal ridge]

It's my own merge

not public

but try prompting art nouveau

[limpid coyote]

Nice, looks good and I'll try it. I need to mess with more tags

[bitter latch]

removed chibi

there we go

hmm ok...

goes more anime

[wary hedge]

hi, how do we prompt for normal classic anime style pls

[bitter latch]

hmm well you can use anime, or some use anime line art style, or for the last couple I just use mistooneanime_V10 model

https://civitai.com/models/24149/mistoonanime

tho I had line art style

with that

[wary hedge]

thanks i'll try, but can it do very classic stuff like sao

[bitter latch]

shrugs

[brittle jay]

Model?

(+loras)

[gloomy scroll]

pnginfo
https://huggingface.co/Surficialruby/LoraDump/tree/main
https://huggingface.co/segole2003/Lora/tree/main

[brittle jay]

I am on phone, but ok

[gloomy scroll]

Aurora

[brittle jay]

What about loras

[gloomy scroll]

lora:LoraMerge_dim128:1
lora:aha128dim-epoch-000010:0.7
lora:shanzhagao64dim-epoch-000010:0.35

[brittle jay]

Oh nice thabks

[gloomy scroll]

Actually I was also using this lora:GoodHands-vanilla:1 since I had some problems with hands lately

[brittle jay]

Oh ok

Never heard of any of these

[gloomy scroll]

Loramerge is one I mixed

[brittle jay]

Also who are surficialruby and segole? You know em?

[gloomy scroll]

surficialruby is mine

[brittle jay]

Oh nice

[limpid coyote]

Guys got any good lora recommendations for better anime gens in general?

I just started using them and gonna add that good hands one

[brittle jay]

for that we usually recommend embeds

[limpid coyote]

For hands?

[brittle jay]

yes

negative embeds

[limpid coyote]

Not sure what embeds are so gonna watch a few videos now lmao

[brittle jay]

they like add words to the dictionary of the model

[limpid coyote]

ahh ok

[brittle jay]

they don't actually add anything "new" to the model

they just help the model associate a word with a concept it already has

you could think of it like that I think

anyways

here are some random ones I have on hand

aurora and khfb embeds are practically fine stand alone

like just put em both and then you don't have to worry too much about negative prompt

[limpid coyote]

Alright, thanks!

[toxic dragon]

Its been a while since I've posted here but I've gotten back to trying to fine-tune stuff after a long break.

[limpid coyote]

Looks good, do you use inpainting?

[toxic dragon]

No.

I probably should learn to make it easier.

[toxic dragon]

There we go.

[supple raptor]

ded chat https://cdn.discordapp.com/emojis/286619073226342400.webp?size=48&name=playdead&quality=lossless

[supple raptor]

coffee girl is eating her cup now

[trim current]

brought extra

[supple raptor]

She's nice like that. She often brings one for you.

[zinc flax]

Gotta save the environment by eating every plastic cups you bought

[supple raptor]

I'm gonna have to gen Coffee Girl with a seaturtle at some point now.

[keen kayak]

Finally done with this one. Not perfect but I'm gonna be happy with it anyway

[south sable]

[south sable]

Not me. But looks like urban samurai lora which I like to use.

[drowsy dove]

Is there a list of hashes of models somewhere? I'm looking at some of the images I created that I placed in the "potential" folder, and I can't find the models I used to create them 🤦

[limpid coyote]

Anyone got tags for a tokyo neighborhood?

[untold glacier]

You mean outside of SD?

[drowsy dove]

Yes, because I'll have to download it again, probably

[untold glacier]

Civitai has them under Hash, though I'm not sure you can search in-site by them. You can try via google, something like: site:civitai.com "hashtag here"

[drowsy dove]

I didn't know it was possible to search by the hash on civitai

Thank you

[untold glacier]

Awesome. Glad it works.

[drowsy dove]

But I did misuse this model back then. Photorealism...

[untold glacier]

Lol. A solid break of the model. I love doing that.

[stable crane]

very cool

[keen kayak]

Thanks, gotta find a good way to get the arms clearer/better at some point but aside from that and the badly scribbled in hand I'm quite happy with the outcome :D

[zinc flax]

dang this style is old

[stable crane]

reminds me of a cool version of a final fantasy dragoon

really enjoy it

[keen kayak]

That makes me quite happy. Thank you ^-^

[zinc flax]

gonna try genning also lol

[stable crane]

ooo

is that parfait mix?

I missed its cuteness

[zinc flax]

Oh no, a random merge I did

[stable crane]

I kinda wish that, every time I prompted, it'd also run it through parfait mix

just to see the cute version

[zinc flax]

Very cute

Been a long time since I've used it though

[stable crane]

parfait version, also real cute

though maybe hands are a bit rough

gonna try mixing it with another lora to see

[zinc flax]

Yeah, hands are pretty bad at that model lol

[bitter latch]

well.. hmm

[stable crane]

a similar cute model is akkaimix

[bitter latch]

hmm yeah mine.. shrugs

[zinc flax]

Ok now this is 100% true

Very nice model

[bitter latch]

oh gawd.. umm.. giggles.. oops

[stable crane]

llol

[zinc flax]

Oh man

[bitter latch]

guess there's a furmeet at disney

[stable crane]

Disney gonna sue you

[bitter latch]

which would not suprise me

nah I have no money

???

[stable crane]

indentured servitude

they'll make you generate wholesome disney content until your time is served

[stable crane]

her shirt is so cute

[bitter latch]

haha

[pastel bluff]

anyone see the tweet from NAI

[bitter latch]

hmmm?

[zinc flax]

The leaked models thingy?

[stable crane]

no, link?

[bitter latch]

switches to the anime model

[stable crane]

you know what I'd like to see

[bitter latch]

yes but we can't post those here

[stable crane]

lmao

I was gonna say "reverse ratatouille"

[bitter latch]

hmm explain

[stable crane]

imagine a giant rat, cooking, and there's a tiny human on his head making him do the right thing

[bitter latch]

copies .. pastes

[pastel bluff]

they say theyll start suing anyone who uses their model or smth

[stable crane]

lol

a bit late

[zinc flax]

They may take legal actions

But sdxl is coming so meh

[bitter latch]

hmm no

[stable crane]

I saw sdxl can do text now?

[zinc flax]

No clue about that

Huge

[bitter latch]

dangit

[stable crane]

Clyde blocked my next post RIP

[bitter latch]

oh?

[stable crane]

it was even nicer

[bitter latch]

ok thats all kinds of fail

sorta

[stable crane]

idk how mouse ears got so cute

there's basically no mouse ear girls in anime right

who's even drawing enough of these for the AI to learn

seriously underrepresented

[bitter latch]

furrries?

😄

umm.. haha

[stable crane]

curry for dinner

[bitter latch]

umm.. fail

almost gives up

[swift lava]

This one is hilarious. I love the facial expression. It is just so cursed.

[bitter latch]

double cursed ?

surprise shorty!

[stable crane]

brothers lol. his mom has some explaining to do

[swift lava]

Ok, that one is probably worth cleaning up. It tells such a story.

Fruit bowl angler fish. Or any other lure predator.

[bitter latch]

here ya go

[stable crane]

hm for multi diffusion I find you do need to keep certain aspects of your prompt

[unreal ridge]

💀

[stable crane]

dangerous alchemy potions within

[bitter latch]

accidentally went same route

[stable crane]

99 bottles of beer on the wall

your girl is an alcoholic 😢

[bitter latch]

nah, bar maid

[compact warren]

That one vtuber exists

[pastel bluff]

who knows if theyll actually sue

[bitter latch]

oh wtf.....?

somewhat better..

...

[unreal ridge]

I'm pretty sure that this is about people using the NAI hashtag and posting stuff made with models that aren't NAI's or some shit

[unreal ridge]

leaked models

which is probably the big majority of the current models lmao

[stable crane]

I haven't seen anyone train a non-NAI that was actually good

waifu diffusion or whatever seemed to try and accept money and

did the money just get embezzled? lol

idk there were a few but, I wonder how many models on civit are just NAI lineage

[compact warren]

Waifu diffusion tried but failed imo

[bitter latch]

ummmo... k then

[stable crane]

I think when SD can fix high frequency details like hands then we're golden

[compact warren]

90% of them

[pastel bluff]

all anime models at least

[compact warren]

We’re getting closer
Theres a lora on civitai that really helps give better better hands by default

[bitter latch]

https://giphy.com/gifs/naruto-shippuden-4gsjHZMPXdlGo

[pastel bluff]

unstable diffusion too lol

[stable crane]

is it negative_hand?

[pastel bluff]

its been so long still havent released it

[compact warren]

No, envybetterhands

31K downloads

[pastel bluff]

hey i think my model is pretty okay, okay

[bitter latch]

https://giphy.com/gifs/foxhomeent-RLW9YEaSBfBMt79fm4

[stable crane]

what's your model?

[compact warren]

https://civitai.com/models/47085?modelVersionId=55199

[stable crane]

LoCon
rip I'm still on a really old SD

maybe I update tomorrow or something

[pastel bluff]

not anime though

i trained a photorealistic model called BRA

[stable crane]

ty looks promising

[stable crane]

from scratch? no base model?

[stable crane]

oh I see

[pastel bluff]

yea

wait no

from 1.5

[stable crane]

yeah I see what you mean, it's 1.5 so it can't be NAI

I suppose on the photoreal side, a lot has come from Stable Foundation's work

[unreal ridge]

To anyone who's curious. It translates to something along the lines of:
「Using leak models means we cannot provide customer support, documentation, UX and general quality standards, or encryption, which are the standards of our service. Additionally, these models do not accurately reflect NovelAI, and we strongly recommend against using leak models. Installing them on your own computer or hosting them online may pose security and privacy risks.
In some cases, we may take legal action if leak models are found.」

[stable crane]

though perhaps NAI piggybacked on Stable Foundation?

[unreal ridge]

Additionally, these models do not accurately reflect NovelAI, and we strongly recommend against using leak models. Basically the exact thing any company that get their product leaked says

kek

[bitter latch]

and back to anime

[unreal ridge]

Oh

I'll try this one

[compact warren]

I get really good hands in pretty much 80-90% of my gens, highly recommend

[compact warren]

They are late asf with this tho lmao

[unreal ridge]

^

[compact warren]

Was their ceo in hibernation or something

[pure oyster]

I wonder how long it takes for thier users to learn they can generate images for free.

[bitter latch]

goes random, rogue even

[unreal ridge]

with better quality

[pure oyster]

Speaking of quality I made this today after a bunch of photoshop and inpainting.

[unreal ridge]

https://tenor.com/view/raccoon-steal-stealing-yoink-seeya-gif-10550534

saved

[bitter latch]

haha

[unreal ridge]

500mb?

[compact warren]

Tbh i never notice the size

[unreal ridge]

The biggest I've ever seen was like

200

[compact warren]

Idr care how big something is until i actually need to clear space for it

[bitter latch]

....... cackles

[bitter latch]

oh.. really?

[compact warren]

[stable crane]

I feel like it'd be really fun to work for novelai

[bitter latch]

not sure why the ears are like that wtf

[pure oyster]

I too wish to not have any work to do.

[unreal ridge]

yep

[bitter latch]

would you say it would be a Novel idea?

https://giphy.com/gifs/ben-tattoo-giant-cD7PLGE1KWOhG

[stable crane]

lol

[stable crane]

they gotta be improving the software somehow right?

[stable crane]

I actually never even saw a screenshot of their UI

but my dream job would be working in UI dev for a image gen product

[unreal ridge]

They just recently implemented inpainting

hurray

just took them like

[stable crane]

huh

[unreal ridge]

1502 years

[stable crane]

it sounds surprising

because I would've figured they'd have a lot of $ for hiring

midjourney definitely slurping in the mad $$$$$$$

that'd be a fun one too and it seems absurd to me it's STILL discord based

[unreal ridge]

I kind of want to pay for niji

[stable crane]

niji is so good

even just looking at the images people gen there, I'm always impressed

it's quite a pricey subscription cost though

[stable crane]

so I figure they gotta be printing money

[pure oyster]

Huh, my images in imgur are showing up as 404 if I'm not logged in

[trim current]

ive been recently doing a mix of SD and Niji then back into SD

started with this in SD into

working on seing how far i can take it in img2img

[pure oyster]

What other image hosting sites are better than imgur? I have never been so annoyed by a website in my entire life.

[stable crane]

catbox.moe

[pure oyster]

I'll check it out, trying to find a host for all my good images

[pure oyster]

A little inconvenient to use, but yea, 10x better than imgur and google images https://catbox.moe/c/9q1x8v

Also, a little slow, but that might be just me