🍥｜anime

[quick hill]

Best way is PBRemTools currently.

https://github.com/mattyamonaca/PBRemTools

[unborn elbow]

thanks, i will check that too

[quick hill]

You can check comparsion for PBRemTools in their Github

Uh formatting

[supple raptor]

If the background is a solid color, that isn't used somewhere else in the image you can throw it in Krita or Photoshop, select the background color and apply Color to Alpha. That's the cleanest way to do it. Otherwise those tools the others linked are good.

[quick hill]

test

Wonder if it's discord issue or server can set option to prevent phising.

[unborn elbow]

ok one more thing to try, thanks

[untold glacier]

Kurisuuuuuuuuu

Boy, I feel like we're getting off on the wrong foot.

[supple raptor]

whoa, that's really cool

[wispy canopy]

yeah, i really liked that one too

was running some old lyricsprompts that failed hard on alfahentai, seeing if darkalfa did better and well, it pretty much does

[wispy canopy]

fixing SD limbs, extreme edition!

[merry tendon]

what version is best?

[wispy canopy]

i'd say top one

[merry tendon]

here is another one

[wispy canopy]

again top one, it captures the yandere the best

eye shadow effect

she's pretty hotheaded...

[merry tendon]

thx, everyone(couple of friends, here) says top is best, so ill upload the top one

[merry tendon]

you want the lora when its uploaded?

[wispy canopy]

still only using nearly no lora's at all 😄

the only lora i use permanently nowadays is the goodhands

[wispy canopy]

what would happen if you'd use yandere + wheeze lora xD

[merry tendon]

idk

[wispy canopy]

AND, civitai is down lol

[merry tendon]

yep, just noticed

[wispy canopy]

hmm, i though the prompt was a dud when i got it from the LLM, but it kinda gives cool output?

[cyan burrow]

cat vs soldier of evil

[wispy canopy]

kot will win

[cyan burrow]

found this in my "unchecked" folder (images I haven't looked at)

and this

[wispy canopy]

gotta go, cya laters ^^

[stable nimbus]

Anyone been able to generate anything on par with finetuned models (sd1.4) on SDXL?

[nova remnant]

Is this ai? https://www.fiverr.com/s/2DrjX4

[cyan burrow]

@nova remnant yes

look at how off some perspectives are and how the background doesn't line up with the foreground

and some misshapen stuff

[merry tendon]

yameroyandere v2 out if someone is interested https://civitai.com/models/48579/yameroyandere

[cyan burrow]

that looks more like yanyan than yandere

[merry tendon]

what?

[cyan burrow]

well deredere is the word for cute, lovable, kind etc. Then they add Tsun or Yan to it. Tsun meaning cold, rude, coarse etc. Yan coming from Yanderu, meaning mentally ill 😛 So your Yandere looks Yanderu (or YanYan)

with that smile

scary

I guess she's in her Yan phase

looks good though

[lyric zealot]

[random valley]

So... not sure of the quality when I combined Aurora, Synergix, and dreamshaper 5. Can you tell?
Generated at 1028 x 1028

May try to mix in dreamshaper 6 instead, and maybe replace synergix with deliberate/realistic vision.

note: This is dreamshaper 0.5 mix on top of the already mixed 0.5 aurora and synergix models.

And yes, I used pruned like an idiot

[zinc flax]

Pruned stuff are fine no

[merry tendon]

how

tf

did

you

do

that

[cold summit]

random go yeet

some flowers, beautiful landscape, fire, insane visuals, trending, 8k, absurdres, sun shower, very detailed beautiful girl, yellow dress, (crying:1.5), (half body:1.5), smiling, dark

these are the prompts

you can try

[final vault]

How does she glows?

[cold summit]

the model I guess does that Xd

or the prompts

[final vault]

I see

[merry tendon]

can someone try out the new version of my yameroyandere lora? https://civitai.com/models/48579/yameroyandere

[wispy canopy]

this prompt has a lot going on xD

[unborn elbow]

can anyone recommend me a guide that teaches how to create Loras?

I've made some that were acceptable but not even close to the quality of the best Loras

[sharp bluff]

would like this too ^

[supple raptor]

@unborn elbow @sharp bluff here are a couple of LoRA training guides that are generally considered to be good.
https://rentry.co/dadaptation-for-lazy-people
https://civitai.com/models/22530

[cold summit]

Yeah XD

[unborn elbow]

ty, i will check it

[supple raptor]

Be warned that any Colabs mentioned in guides that you find may or may not currently be working as google is attempting to stop the use of colab for Stable Diffusion.

[unborn elbow]

I will do the training locally

[cold summit]

I wanted though a bit of a different type of scenario, gonna check tomorrow

[sharp bluff]

ty!

[supple raptor]

Testing now on Aurora. Ignore the 3rd one. I messed it up and ran 1 strength again instead of 0.6.

[merry tendon]

thx man, finally someone who doesnt ignore me.. when i asked the same thing on the unstable diffusion server everyone just roasted the name of the lora 😢

[supple raptor]

I'm surprised how well it handles 1 strength. Most loras I have look like shit above 0.8. It seems to do decently at hands too. It actually fixed the hands compared to 0 strength on this one.

alright... 1.5 might be a little too much strength XD

[merry tendon]

bruh

[supple raptor]

If the colors weren't fucked it would be really good though

[merry tendon]

try 1.2

[supple raptor]

I'm gonna try a batch with 1.4, 1.3, 1.2, 1.1 when the current batch finishes. I assume 1.4 will still be too much, but might aw well test it.

here's the grid with that cursed one

1.4 is just a little too much. 1.35 might be doable. I think 0.8 is probably the best at least on this prompt but everywhere from 0.4-1.2 seems work pretty well.

[merry tendon]

can you try using the keywords provided on civit

[supple raptor]

These have all been using: yamero yandere, consticted pupils, shaded face, crazy eyes, crazy smile, mksks style, detailed background, superb, 1girl, long hair light purple hair, curly hair, cute, eyelashes, white dress, pink ribbon around her waist, pink flats, white thighhighs, beautiful 8k wallpaper, outdoors, nature, tree, bush, flower, rustic, extremely detailed, intricate

[merry tendon]

try:

yameroyandere, consticted pupils, (shaded face:1.2), (crazy eyes:1.1), (crazy smile:1.2), mksks style, detailed background, superb, 1girl, long hair light purple hair, curly hair, cute, eyelashes, white dress, pink ribbon around her waist, pink flats, white thighhighs, beautiful 8k wallpaper, outdoors, nature, tree, bush, flower, rustic, extremely detailed, intricate

[supple raptor]

For this one I just added the rest of the tags from civit. I'm trying the prompt you posted next.

[wraith meteor]

what is the best model to specified in characters ? i was trying to do some specific clothes in rev model but its always gives another clothes 😦

ninja or burning or diamond clothes not working

[supple raptor]

I doubt any models have gotten enough training on burning clothes or diamond clothes to do them reliably. I would think ninja should work pretty well, you may also want to try shinobi if you're trying to make a female ninja, kunoichi may also be helpful. I haven't used rev much. Aurora listens to most clothing prompts pretty well in my experience.

[wraith meteor]

I don't know about Aurora yet but let me try and update you

shinobi works well in rev model

[supple raptor]

Here's the link to Aurora. I'd recommend getting the two negative embeddings that are in the files as well and using those in your negative prompt. I use them on pretty much every model, but she made them specifically for Aurora. https://huggingface.co/SweetLuna/Aurora

[wraith meteor]

gonna test this out , i had a hardtime doing some clothing with other models

[supple raptor]

Here's the full negative prompt that I use:
negative_hand, (KHFB, AuroraNegative), (Worst Quality, Low Quality:1.4), border, nsfw, skimpy, grayscale, multiple_girls, (watermark:1.3)
Civitai just went down so I just attached the negative_hand embedding from Nerfgun3 that I use.

Edit: Civ is back up https://civitai.com/models/56519/negativehand-negative-embedding

[merry tendon]

im back

[merry tendon]

works pretty good

[wraith meteor]

model almost downloaded , i have a very good negative prompt that i always use in anime too here it is :

((detailed background)) , duplicate, blurry, abstract, bad anatomy, bad eyes, crossed eyes, disfigured, poorly drawn face, mutation, mutated, ((extra limb)), ugly, two trunks, missing limb, floating limbs, disconnected limbs, malformed hands, blur, out of focus, long neck, ((((mutated hands and fingers)))), (((out of frame))), floating limbs, disconnected limbs, blur, out of focus, mutated foot, long neck, ((((mutated hands and fingers)))), bad eyes, crossed eyes, disfigured, poorly drawn, mutation, mutated, ((extra limb)), ugly, poorly drawn hands, blur, out of focus, long neck, long body, two face, out of frame, text, error, cropped, 3 shoes, 3 legs,2 ears, low quality, jpeg artifacts,morbid, double image, ugly, disfigured, cut off, ugly, grain, low res, deformed, blurry, bad anatomy, disfigured, poorly drawn face, disconnected limbs, mangled, extra fingers, duplicate artifacts, missing arms, mutilated hands, cloned face,logo,(((extra object)))

[supple raptor]

not sure if yandere or just had too much coffee 😅

[wraith meteor]

@supple raptor i guess rev is better than Auro in quality but Auro is better than rev in drawing hands

[supple raptor]

Try my negative prompt on Aurora. I have had better luck with the minimal negative than my old reliable long negative prompt on that model. I also run it at 7.5 cfg. It also has a few activation tokens that can be helpful, mksks style, detailed backgrounds comes from dpep4 which can help with nicer looking backgrounds and SLE comes from the eyes lora tha Luna mixed in which can help with better looking eyes. Generally you'll want to use the activation tokens at the very beginning of your prompt for best results.

[wraith meteor]

will try i am downloading atm more models to compare and choose the best one civitai has alot of good models to check 🙂

[supple raptor]

Here's a few more that I like:
https://civitai.com/models/24779?modelVersionId=56071

https://civitai.com/models/68061?modelVersionId=72732

https://civitai.com/models/70692/lemontea-mix-painterly-25d

[wispy canopy]

why u no like darkalfa 😄

nah, it's a quirky model for sure, and doesn't work at all well with most danbooru type prompts

[supple raptor]

I booru prompt so it just doesn't work well for me.

[wispy canopy]

knew it was coming 😄 yeah, lots of booru tags don't work well on it

(WHICH I LIKE!)

DEATH TO MASTERPIECE

[supple raptor]

Yeah you put out nice images with it. Just doesn't work for my style of prompting.

[wispy canopy]

dumb example: you prompt hands in pockets it goes 'whuh' and fails horribly, you prompt girl, jacket, nonchalant, you get hands in pockets 9/10 times

[supple raptor]

[wispy canopy]

https://tenor.com/view/perfection-michael-fassbender-perfectionist-perfect-gif-7910211

[supple raptor]

@merry tendon I regret nothing

[unreal ridge]

As an average yandere enjoyer. I'mma test this when I get home 👍

[wispy canopy]

.yay. rainbows. so happy. see my joy in all it's colours

[wraith meteor]

why when i make prompt like cute beast , wearing shinobi clothes it turns into human 😄

[wispy canopy]

cute, and wearing are also associated with humanoids

you're probably going to have specify a specific beast to get the beast you want

[wraith meteor]

oh didn't know that , do u have a beast name or a list i can try ?

[wispy canopy]

it's going to be different per model... most anime models, don't even bother with cats -> you'll just end up with cat-girls

lol

[wraith meteor]

yea i tried cat now and i got a girl with a cat beside her 😄

[wispy canopy]

well waddaya know, my model got it right? (it's generating atm, but preview is certainly a beast)

[storm herald]

309

[wispy canopy]

#1 not cute, but #2 would suffice, i suppose

ah, got it working

[wraith meteor]

yes i am trying to learn how this thing works

[wispy canopy]

basically, every word is not only evaluated on a local level, but also on a global level

it's more like 'how does this word remind the model of the things how the word was used when the model was trained'

so if the words cute and wearing were used with lots of girls -> the model will give you girls

cute beast , wearing shinobi clothes, humanoid animal, solo, forest```

adding humanoid animal, solo, forest, was enough to get it give me consistent animals

[wraith meteor]

for me it works like 20% of the times , i am using rev model ^^

[wispy canopy]

100% of the time 😄

[slate night]

SHNAOIT

SHIADIT

[wispy canopy]

ugh, civitai dead again

else i'd have linked to my model

[wispy canopy]

shiadit?

[wraith meteor]

the website is working for me

XD 🙂

[slate night]

Yeah it's working for me too

[wispy canopy]

must be my vpn being wonky

https://civitai.com/models/64843

this the model i'm using

easy to use, except if you're very used to using booru prompts

[wraith meteor]

yes i am using booru prompts , i guess they are working good for any models

[wispy canopy]

not all, you'd find they're not working at times on my model

ketchup love

my model is just best at piecing stuff together, making it pretty damn great for throw shit at the wall, and see what sticks

[wraith meteor]

yup some clothes are not working properly on all models

[wispy canopy]

i hate i lost the ability to do sari's

got this in return tho

[wraith meteor]

wow this is sick

[wraith meteor]

sari clothes is a good one too its working on rev model

[wispy canopy]

i should go to bed, been up too long 😦 byebye ❤️

[wraith meteor]

https://tenor.com/view/sleep-time-gif-25989017

[nova remnant]

And he is selling it as if he drew them xD

[unreal ridge]

nvm

civitai fucking died

[slow needle]

testing multires rn

[keen hollow]

evenin'

[keen hollow]

Been away a few days, any new fun developments?

[slow needle]

nope

[keen hollow]

right on

[cyan burrow]

what upscale model do you all tend to use?

[keen hollow]

usually lollypop

[slow needle]

latent for me due to most of the models i use handle it well

[cyan burrow]

thanks, I'm trying to decide which works best for multidiffusion and the various styles. I know some work well for anime like animesharp, borugan, lollypop and rsrgan anime whatever.

But what about more western style art that's 2.5d or pseudorealistic ?

[keen hollow]

sounds like you gotta run a grid 🙂

[cyan burrow]

a grid? I do tiles anyway but I need an upscaler that won't leave me with artifacts or remnants or simulacrum

[keen hollow]

yeah run a bunch and compare 🙂

[cyan burrow]

I have like 50 -.-

if not more

might actually be a lot more

If only there was an extension that ran multiple upscalers on different unets

[keen hollow]

have you seen this 1-click face animation swap thing? crazy stuff

[compact warren]

Only seen a lil bit

[cyan burrow]

I wonder if this bot wrote me actual code or gibberish

[supple raptor]

ask gpt

[cyan burrow]

thats who wrote it

here it is

wonder if it does anything

I asked for a script that would use different upscale models for different unet blocks

@supple raptor do you think that script does anything at all?

[supple raptor]

I'm far below novice level in Python, but I think the stable_diffusion_upscale argument would need the WebUi API to be imported first. I'm also unsure of the "as np from scipy...." part of the first line.

[cyan burrow]

it might be poorly formatted since I copy pasted from irc

[supple raptor]

There certainly could also be formatting problems. From a quick search I found from scipy.ndimage import gaussian_filter to be an alternative to the first line. To be honest, my python knowledge is limited to being semi capable of fixing or modifying functional code, and that's largely thanks to python generally being fairly logical. I wouldn't stand a chance at writing even an extremely basic python progam on my own.

I think that Python also expects a blank line between imports and everything else so there shoulc probably be a blank line between line 2 and 3.

[cyan burrow]

Python does seem to follow a more script friendly approach starting out, but to make very complex stuff it gets convoluted

[supple raptor]

I keep telling myself that I should take the time to try to actually learn python, but I haven't done it besides just poking at other people's code to see what happens.

[quick hill]

I used python but I went for node.js since what I worked on python doesn't seem to be optimized for it

[final vault]

damn the good old prompt

[supple raptor]

Coffee Girl is best girl and her prompt will never be forgotten

[final vault]

definitely she's the queen 👑

[quick hill]

I'm still using that prompt regularly for testing.. it's quickest way for me to test holding objects generations

[final vault]

No, it isn't necessary, however PEP8 suggests to have a blank line for better code readability

[supple raptor]

Yep that's why I keep using it. I made an updated version of it that was a little better, but it's burried in my gens. I'll try to find it and send the improved version to you for testing.

[quick hill]

Hopefully I should get back to SD soon, not much happens on SD so I'll have no problem catching up.

[nova remnant]

gimme gimme gimme

[near elbow]

some b&w anime portraits for you

back to kitsch land 🙂

[magic wave]

What model is that? is that wd? :o

[magic wave]

Damn that looks really nice :O

[slow needle]

also made a fellatrix lora for wd 1.5 today although its not that great

[magic wave]

Fair, thats a pretty tough style to do, what were you using for that pic? :o

[slow needle]

fellatrix and mika pikazo lora at half weight and my noise offset at full weight

and plus the base model is wd radiance

[magic wave]

I'll have to try dat, where you upload em? :o

[merry tendon]

Holy! What lora is that

[merry tendon]

Based

[slow needle]

i gonna upload it to hugging rn i guess

[frozen olive]

none it's nijijourney

[merry tendon]

Looks sick!

[magic wave]

Alrighty :o

[merry tendon]

Fr? Looks really good!

[frozen olive]

ye^^

[slow needle]

niji always looks good

[frozen olive]

jeah niji is awesome^^

[supple raptor]

The main reason that I would want to learn python is that it's the language at basically all of the AI tools are written in.

[slow needle]

huh works better than i remember

[supple raptor]

yeh Niji makes nice images. it just isn't nearly as flexible as SD, especially when you consider all of the LoRas and TIs we have available.

[quick hill]

For my usage, I was working on fetching data for my website. I used B4S and a bit of canvas but ended up have to use Node.js, it was very very slow with python.

[slow needle]

tbh the seer quality of niji make up for it

[quick hill]

true

[supple raptor]

TBH I know almost nothing about coding in general. Just enough that I can sometimes stumble my way through existing code just enough to make it do what I want.

[supple raptor]

Fair, it's quality is very good.

[slow needle]

tbh kinda wish sdxl release rn just so we have a way to defeat niji on pure quality

[nova remnant]

i dunno if sdxl could keep up tbh

niji just does eyes, hands, perspective and composition so much better

[supple raptor]

With the amount of data that's been removed from the dataset, I really don't think we'll ever see another release from SD that rivals the powerhouses in the industry like Niji/MJ. I think at this point the best we can hope for is a good architecture from Stability that can be built upon by other groups.

[slow needle]

yup

[quick hill]

I'm more interested in what users will make out of SDXL.

[slow needle]

we will have to wait for WD team to finetune SDXL before anything can happen tbh

[cyan burrow]

heloooo there

[quick hill]

That means forever

[cyan burrow]

SDXL will also demand much more vram

[slow needle]

almost a full year for wd 1.5

[cyan burrow]

it feels like they're moving SD further and further away from consumer and prosumer to pro

[supple raptor]

I think SDXL will suffer the same issue as SD1.4 did. My asumption is that the training requirements will be high enogh that it won't be achievable by the vast majority of users, which wil leave it as a dead platform

[cyan burrow]

damn I forgot what that regional prompt extension that allows you to draw or create boxes was called

especially the one where you draw a color onto a canvas and then prompt that area

[slow needle]

hopefully its not gonna be like that

[cyan burrow]

latent couple?

[supple raptor]

what about it?

[cyan burrow]

was that the one where you draw in color and then decide prompt in color?

and which was it where you drew out boxes?

[supple raptor]

latent couple and regional prompter both do that in slightly different ways.

[cyan burrow]

latent couple tends to fail me miserably

and cuda out of memory on it of course

I'm becoming more and more tempted to buy a second hand 3090 for 800 dollars, then sell my 3070 ti for 300-400

then I will have even better gaming on my 4k tv, instead of playing 2k on it

and I can do a lot of ai

[near elbow]

thanks - haven't been using WD for quite some time now, the newer models are so much better 🙂
this is

Negative prompt: loli,underage,teen, nsfw,nude,naked
Steps: 30, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 2955604206, Size: 512x640, Model hash: d41196e112, Model: notTellin_v17, Denoising strength: 0.65, Hires upscale: 1.75, Hires upscaler: Latent, Version: v1.3.0```

one of the more obscure models, revanimated and lyriel are really good and universally useful

forget about WD for the time being, anything 5 is ok, and there's a ton of other good anime stuff out there

[magic wave]

O, sis :O

[magic wave]

Ohh, waw I haven't heard of that model :o

[cyan burrow]

moving to 2.0 or 2.1 or sdxl means leaving behind every model there currently is, and there's only so many for them

on 2.1 I mean

Not much to do

[slow needle]

moving on to newer models also mean you have to crave your own path too

[cyan burrow]

yeah but then you need the vram to do that

[slow needle]

you guys are sisters?

[slow needle]

yup

[near elbow]

haven't used any 2.x based stuff for ages, the newer 1.5 based models are just incredibly good

all the experts agree 🙂

[magic wave]

Sis of the soul

[supple raptor]

The issue is that SD's training is getting worse in every iteration, most likely due to concerns about copyright, while the technical capabilities and therefore the training requirements continue to increase with every iteration. SD2.x is already basically at the limit of most user's hardware limitations for training and SDXL will only be worse in that regard.

[near elbow]

look, SHE doesn't recommend SD 2.1 either

[steady grail]

hello guys

i was working on this

[nova remnant]

the one n only

[steady grail]

i was upscaling and redetailing it, but it became like this.

[cyan burrow]

That's why I think soon enough we'll have someone fork the actual base model of SD to make their own XL

[steady grail]

😦

[slow needle]

SD 2.X need alot of locons to look good but when it do its fucking good

[nova remnant]

https://tenor.com/view/anime-amazed-eyes-sparkle-gif-11618062

[steady grail]

the model don't support such resolution?

[cyan burrow]

and adobe is really moving ahead with their AI too

hopefully they won't be that stingy with what it can do

[supple raptor]

Yeh, OpenCLIP is very good once it's properly trained.

[near elbow]

maybe... haven't seen anything in 2.1 that 1.5 can't do, though. the CLIP part is better as well, but not a quantum leap either

with hires fix and controlnet you can get to any resolution you want

[steady grail]

i feel like there is not enough chaos in 2.1 😦

[near elbow]

🙂

[steady grail]

in art, chaos is life

[magic wave]

Yes xD

[supple raptor]

Firefly is garbage. It's just reliable garbage. Adobe won't share shit from their project, but we wil benefit from a wider public acceptance as traditional artists begin to implement AI into their workflow thanks to Firelfy.

[near elbow]

e.g. this is at 1024x1536 with hires fix, no lack of details 😉
"((anime style)) a beautiful spring landscape, tokyo, mount fuji in the background, friendly clouds, exquisite textures and details"

no lack of fujis either

[cyan burrow]

dont suppose anyone here is from the EU and looking to sell a 3090 or know anyone who does?

[near elbow]

"((anime style)) ((watercolor)) a beautiful spring landscape, tokyo, mount fuji in the background, friendly clouds"

[cyan burrow]

guess I will look around Sweden. They usually go for 800 dollars

[supple raptor]

OpenCLIP is way better than CLIP, we just don't have much for examples of a properly trained OpenCLIP yet and we're comparing to models based on CLIP that are very well optimized at this point.

[steady grail]

i don't know... it's a feeling

[slow needle]

and also the fact that wd team forgor to train the clip doesnt help too

[cyan burrow]

maybe I should sell my 3070 ti first, that would mean I get stuck to my 1060 for awhile but I will maximize how much I make from it 😮

[near elbow]

more images!

[cyan burrow]

hmmm reading about multi-gpu training though

[near elbow]

how many GPUs do you have lying around? 🙂

better one 4090 than a zoo of smaller things i'd say

[supple raptor]

Yeah but look at the improvement that beta 2 to beta 3 had with the addition of training TENC. I'm cautiously optomistic for the final release of WD and especially for finetunes that will be made from it.

[slow needle]

yup

[slow needle]

anyway for now i gonna train quite a boat load of style lora for wd 1.5

[steady grail]

love it

[quick hill]

Patiently waiting

[slow needle]

maybe 2 more months

[quick hill]

I hope

[slow needle]

also by then most of my lora have gone out of date

[supple raptor]

https://tenor.com/view/kagome-higurashi-waiting-inuyasha-anime-gif-17862090

[quick hill]

Yup, have to train a new one

I'm glad I still have dataset for my loras

[supple raptor]

I'm hoping that all of the people that have been training loras have been holding onto their datasets to retain when/if WD gits gud.

[slow needle]

i always kept mine for later use

[cyan burrow]

sucks that 3070 ti doesnt support nvlink

[slow needle]

most consumer cards dont nowadays

[cyan burrow]

yah, starting at 3090

I would've loved to keep my 3070 ti to add more vram but gotta sell it

[cyan burrow]

hmmm pip install ninja? wot that?

[slow needle]

basically allow compilers to use multiple cores instead of a single core

[cyan burrow]

oh cool

hmmmm gotta update a few other things now that torch is at 2.01

but no idea how

[supple raptor]

SLI/nvlink doesn't combine vram for consumer cards anyways AFAIK. It just distributes the compute to both cards. If that's the case it wouldn't be very useful for SD anyways.

[cyan burrow]

@supple raptor there's a few guides on how to get them to work with SD

I think

gotta add some new scripts and stuff

I think it mostly then comes down to like...one card can load a model and the other generate from it

or train from it

so uh torchaudio 2.0.1+cu118 requires torch==2.0.0, but you have torch 2.0.1 which is incompatible. torchvision 0.15.1+cu118 requires torch==2.0.0, but you have torch 2.0.1 which is incompatible. any cause for concern here?

[steady grail]

you should install torch 2.0.0

lol

[mental widget]

Hey, can somebody help me?
Im trying to get some purple lightning bolts into the image, but nothing seems to work for me with inpainting... I tried "fill" "original" and everything from denoising strength from 0.4 to 1

[steady grail]

i would try with editing and controlnet

o remove the bg from the image and make from 0 a new one

[steady grail]

beautiful

they are done with a prompt? or it's a model?

[wispy canopy]

prompt and model

[wispy canopy]

watercolor, official art, promotional art, painting, devil, angel, passion, lies, fall, fade, scar

that's the prompt

[steady grail]

oh right, i was talking about the washed up colors

[wispy canopy]

oh, the one above me? 😮

[steady grail]

talking about morrigan's washed up effect, i love it so much

[primal magnet]

heyhey, great art in this channel. This one isn't finished yet, I still have some work to do in PS to clean up artifacts especially in the waist area

[edgy badge]

Via Midjourney

Although I told it to create normal headphones, it proceeded with a burger headphone🙄

[edgy badge]

Although I told it to create normal headphones, it proceeded with a burger headphone🙄

[latent vector]

img2img ftw

why did that send twice

discord is

[lyric zealot]

well, discord is experiencing some issue now

[latent vector]

glitching a lot

[lyric zealot]

testing 1 2 3

[latent vector]

discord is

[magic wave]

Discord is experiencing so many issues, its so bad XD

Really laggy and duping of messages :o

[lyric zealot]

yup yup

[magic wave]

Yeah, and even after like, 5-10 minutes, it'll dupe a message

[lyric zealot]

yup, it's kinda fucked

[magic wave]

Discord is experiencing so many issues, its so bad XD

[lyric zealot]

[magic wave]

Really laggy and duping of messages :o

[lyric zealot]

[slow needle]

A local mofu is going into a mofu crisis due to this issue lol

[magic wave]

Mofu? :O

[slow needle]

A local mofu is going into a mofu crisis due to this issue lol

Oh shit the flu is spreading

[magic wave]

XD

It is

[lyric zealot]

[magic wave]

Is that expmix? :O

[lyric zealot]

yup yup yup

[magic wave]

O wao, any LoRAs? Feels different to how it usually is :o

[lyric zealot]

nope, no lora

except for negative embedding of course

[magic wave]

Oh damn, fair then :O

[lyric zealot]

[magic wave]

Fair :o

[magic wave]

The waifu :o

Also what that style? Like what model or lora

[obsidian willow]

https://civitai.com/models/63784/kurokawa-akane-or-oshi-no-ko-or-lora

：D

ya might want to adjust a bit to have the hair colour showed above

[mental widget]

What are the best prompts for a wide angle shot? so i get more of the background into my picture.

[cyan burrow]

Gotta make a inpaint version of my model

[keen hollow]

This one is cool, gives Disney vibes.

[magic wave]

O I think I got that one, but I meant like whats the model you're using for the arstyle, since its pretty neat, or just the lora itself? ;o

[magic wave]

Damn niji is pretty sweet :o

[obsidian willow]

ah, i use meinamix

[magic wave]

Ohhh thats fair then, I have heard of that one :o

[trim current]

anyone been using diffusers backend?

[magic wave]

Whassat? :o

[trim current]

in vlad

[magic wave]

Oh, I had no clue about that, how that work? :o

I know its supposed to be faster ye? :o

[trim current]

yee marginally

err maybe that wasnt the right word

slightly faster

[obsidian willow]

How do these seem?

[stuck aurora]

nice!

[brittle jay]

ovefitted imo

all face expressions are too similar

[merry tendon]

new sillytavern automode is sick!

[brittle jay]

hello everyone btw

[magic wave]

I think it was right :o

[magic wave]

Pretty nice :o

[magic wave]

Heyo

[brittle jay]

https://styledrop.github.io/

kinda cool

[compact warren]

Reading

Blegh

[brittle jay]

what

it is kinda cool innit

[cyan burrow]

I'm continually surprised at how easily people can make money with AI generated images. Yet I don't know where to even begin

[brittle jay]

@merry tendon I saw you on the stream

hehe

[merry tendon]

kinda sad seeing him like this

[brittle jay]

fr

though if it's what he truly wants I am fine with it

[merry tendon]

yep

[merry tendon]

here is the link btw https://civitai.com/models/48579/yameroyandere?modelVersionId=88330

[brittle jay]

nice preview images

[merry tendon]

thx, credit goes to hatsu( from unstable diffusion ) for some if the images

the main images is mine tho, i am really f proud of it

[brittle jay]

I don't participate in unstable diffusion even though I am in the server for months now

[merry tendon]

yea, i only post there like once a week

[near elbow]

kawaii neko

(more images, less chat 😉 )

right?

[near elbow]

no such thing as too many screens

[near elbow]

ok, bathrobe, swimming pool, check 🙂

[lyric zealot]

[near elbow]

more 2d look, still SWF i think

back to more realistic

and that's enough kawaii swimming waifus for now 🙂

all at 1120x896 resolution, check out the originals, the discord previews are really bad

[sage mountain]

i wonder what happens if mix this with the crazy expressions lora

[merry tendon]

doesent make sense, its already "included"

[near elbow]

all your pools are belong to us!

[merry tendon]

no reason why you should do that tbh

[sage mountain]

fair, yandere lora does seems like a subset of that other one

[merry tendon]

yep, would be really cool if you could try it out for a bit(bc idk how it works on other models)

[compact warren]

Dunno what youre referring to but i would hold off on it since im gonna retrain that

[sage mountain]

by mix i just meant using both same time to have an expression middle ground

[compact warren]

The point of crazy expressions is to simply allow eyes to generate smaller, so u can make any expression look sorta crazy or traumitizing

Megus is specifically for the yandere aesthetic tho

[merry tendon]

nope, it can do the eyes thing too

[sage mountain]

since yandere im seeing gleeful narrowed eyes and mouth of that specific crazed shape

[compact warren]

Besides, i doubt mine combines well anyway

Currently

[compact warren]

Oh?

[merry tendon]

yes

[compact warren]

How well

[merry tendon]

idk, never tested it much on my model, but haru did some testing and it looked pretty good

[compact warren]

Did they post any results here i can look at? Would be nice if i dont have to retrain mine after all

[merry tendon]

https://civitai.com/models/48579/yameroyandere?modelVersionId=88330 maybe you can find something here you think that fits it

[compact warren]

Already visited the page

[merry tendon]

you looked the the pics?

[compact warren]

Yeah, its evident it can do the small pupils but i only see examples of with smiles

[lyric zealot]

this is at weight 1

kinda crazy even at weight 1, it's not burn or anything

[merry tendon]

yea, you can do without smiles and with, bc its a yandere lora i chose to only show the smiles to show off its capabilities

[merry tendon]

yea, came out pretty good

[compact warren]

Can i get some examples of sad faces and such?

[merry tendon]

nope, my sd is not running currently(+ im doing some RP testing so my gpu is occupied).. sorry but youll need to test that for yourself

[compact warren]

Sadge

[merry tendon]

it can probably do it tho, if not just put "smile" into negative

[merry tendon]

nevermind.. it can do it i guess

[compact warren]

Thats at weight 1?

[lyric zealot]

yup weight 1

[merry tendon]

how tf

[compact warren]

Mmmmm…..

[lyric zealot]

[compact warren]

I dont think the eyes are small enough

[merry tendon]

can i use your images on the civit page btw=

[lyric zealot]

go for it

[merry tendon]

add (constricted pupils:1.1)

[lyric zealot]

if you need it with metadata

[brittle jay]

@supple raptor do you mind telling what you used for cow submission? (like lora or whatever)

[lyric zealot]

it's inside, just at weight 1

[compact warren]

this is what does it for me

Way more white

[merry tendon]

it works?

[compact warren]

No thats from my lora page

Im just using a reference

[merry tendon]

ok f rp testing, im starting sd

[compact warren]

Lol

I keep getting review updates my expressions lora page but i never get notifs for it

Wtf civitai

[compact warren]

Oh u should also check to see if it works well with realistic models

[merry tendon]

can you try (constricted pupils:1.2)?

[merry tendon]

probably not

[lyric zealot]

got it

[compact warren]

Got what

[lyric zealot]

it's not working?

[merry tendon]

what?

[compact warren]

?

[lyric zealot]

the constricted pupils

[merry tendon]

can you try (constricted pupils:1.2), (crazy eyes:1.1)?

[supple raptor]

I used Aurora with Horrid's EmoteMakerLora and the minimalistAnimeStyle_v10 Lora. The girl was the original gen, it was in portrait orientation and the top of her head was cutoff so I manually expanded the image into 16:9 ar, finished drawing the top of her head and some of her hair that was cutoff, did some manual touchups and filled out the background. Then I genned the cherry blossoms separately, removed the background from that image and added it to the first image.

[brittle jay]

I know of the second lora

[compact warren]

Was the dataset built simply using various images of yandere-ish faces?

[brittle jay]

wwhat is emotemaker?

let me search

[merry tendon]

idk man, the dataset is like 500 images big lol i made that dataset in a span of like 4 days(only high quality images, not thos bad yandere faces that you see everywhere)

[compact warren]

I only did like 80 for mine lol

[merry tendon]

lol

[compact warren]

But anyways that might be why it doesnt work

[compact warren]

Depending on the tag arrangements

[merry tendon]

ill start my sd up real fast and check if it works on my side

[brittle jay]

https://civitai.com/models/53294/twitch-emote-maker-lora

this right?

[supple raptor]

yep, that's the one.

[brittle jay]

thanks

[cyan burrow]

So Meta Quest 3 will be 500 dollars for 128gb version and at least twice as fast as quest 2, slimmer, better battery power etc...Might buy later

[brittle jay]

will there be 64gb version?

I am guessing no?

[brittle jay]

I am just waiting for more SadlyItsBradley on it until I decide

[cyan burrow]

^^

@brittle jay no, 128 and 256

quest 2 will be dropped in price to 300 dollars

[merry tendon]

@lyric zealot @compact warren thats how it looks on my side, so super small pupils do not work. but constricted pupils do

[brittle jay]

yeah, to be expected since they did the same for quest 2

[cyan burrow]

but since you more or less need the elite strap for quest 2 to play it with any comfort and battery time, that's 400+ anyway

I hope quest 3 comes with eye tracking (fast one) and a good fov of at least 115

[compact warren]

Were u using super small pupils as a tag?

[cyan burrow]

with eye tracking it can easily set up optimal PPD and then I can order lenses for the rest

[merry tendon]

no

[compact warren]

So ur just referring to what it does do then

[merry tendon]

yes, yours can do super small pupils, and mine can do small pupils

[compact warren]

Noted👍

[cyan burrow]

IPD even

[compact warren]

Looks like i will go on and retrain then

But first im gonna redo my gun2head lora

[merry tendon]

ill put your new lora on my recommended resources list when its done

[supple raptor]

using only the constricted pupils tag from yameroyandere-v2 here.

[merry tendon]

my lora?

[lyric zealot]

nope, no lora at all

[compact warren]

she can step on me

[merry tendon]

how tf

[lyric zealot]

1girl, pink hair, forehead, sidelocks, yellow detailed eyes, sinister smile, evil smile, wide smile, blood on clothes, (blood on face:1.2), (blood from eyes:1.3),

from front,
Negative prompt: (low quality, worst quality:1.4), (bad anatomy), extra digit, fewer digits, (extra arms:1.2), bad hands, by (bad-artist:0.6), bad-image-v2-39000, censored, monochrome, greyscale, nsfw, nude, loli, smile,
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 22, Seed: 2247440111, Size: 512x768, Model hash: 4d651c7638, Model: Fav_ExpMix_Line_v2, Denoising strength: 0.53, Clip skip: 2, Hires upscale: 2, Hires steps: 10, Hires upscaler: 4x_fatal_Anime_500000_G, Dynamic thresholding enabled: True, Mimic scale: 7, Threshold percentile: 96, Mimic mode: Half Cosine Up, Mimic scale minimum: 4, CFG mode: Half Cosine Up, CFG scale minimum: 4, Version: v1.3.1

Used embeddings: bad-artist [2a38], bad-image-v2-39000 [b03e]```

no idea how tbh

[merry tendon]

model?

[compact warren]

She can poke me with that horn of hers

[unreal ridge]

It's my own

[merry tendon]

link?

[unreal ridge]

Don't have one. I really have no plans on releasing it

[unreal ridge]

😳

[merry tendon]

aurora?

[lyric zealot]

nope, same as before

[merry tendon]

ok

[lyric zealot]

#🍥｜anime message yup which is expmix line v2

[merry tendon]

im off👋 👋

[merry tendon]

i tried replicating that lol

ok.. anyways, now i gotta go sleep fr

👋

[supple raptor]

[lyric zealot]

[serene spoke]

Hi, what checkpoint do you often use and what loras or hypernetwork?

[brittle jay]

same

gn

[small pecan]

aurora model pixel art. not without flaws but interesting

[cyan burrow]

arf aurora, now I understand why people put aurora lora on neg ;D

the pixelation, nay! I mean it's nice work

but I'm sort of admiring what you did and not the model

[unreal ridge]

What was your model name again?

I want to try something new

[cyan burrow]

dark gemini 3

@unreal ridge

https://civitai.com/models/6209/dark-gemini-v3 here

I might release 3.5 and then work on an entirely new one

[unreal ridge]

ty

[cyan burrow]

but I really want a 3090 then so I can train it proper

keep in mind it's a fairly natural language model, but uses concepts and qualifiers like 2.5d, digital art etc

but it's not good with danbooru

or not that good

at the very least not as accurate

[unreal ridge]

https://tenor.com/view/daeth-funi-periflight-dies-from-gif-22212023

What vae does it use?

[cyan burrow]

natural language is fine, instead of the whole 1girl this and that. "A young woman in a fancy evening gown visited a fancy restaurant for a date, while sitting at the table she slit the throat of her date and left his corpse face first in a bowl of soup at the table. While carrying a wicked smile"

I either use kl anime2 or SD 1.5 standard forgot its name

vae-ft-mse-840000-ema-pruned or kl-f8-anime2

I tend to use clip skip 2 but don't HAVE to

[unreal ridge]

I'll just use klf8

[cyan burrow]

vae-ft-mse-840000-ema is better for 2.5d, pseudorealism etc

the other is better for anime

[compact warren]

Wheres the fun in using natural language

[worthy oxide]

🤔

Now I notice that one of the shirt pieces is not drawn correctly

[compact warren]

Its not only that, some models are just made that way

[cyan burrow]

Cause I'm a writer, not a programmer

[compact warren]

Does using danbooru tags make us programmers?

[cyan burrow]

works better for me

no but it's closer to scripting and constructing

variable after variable

[compact warren]

Fair enough

[compact warren]

[unreal ridge]

ayo

[compact warren]

I smell mods approaching

[cyan burrow]

discords overly strict rules are annoying. I think even official channels should be allowed adult channels. By that I don't necesserily mean porn. But where you can make sexual, profane jokes or show a generation of a slutty character. Which can only be entered by people somehow "proven" to be 16 or 18

[compact warren]

At the end of day, we have no control

[cyan burrow]

grrr control net is stuck on processing a preprocessors preview

dunno how to cancel it

sheesh'

[lyric zealot]

what might be some ways to prove that people are 16/18?

[cyan burrow]

oh fuck it, guess I'm reinstalling all of auto tomorrow

I don't know, ask me when I'm not asleep

good night

[lyric zealot]

goodnight

[magic wave]

Good nite

[pastel bluff]

ask them what they think of kim possible

[keen hollow]

evenin'

How you doin' @supple raptor ? Good weekend?

[supple raptor]

Yeah it's been pretty good. Spent some time getting things ready to go camping later this week and got to play Pathfinder yesterday. How about you?

[keen hollow]

oo nice have fun. Mine was pretty good. Had some good food. 😄

[pale monolith]

So jelly I haven't played pathfinder in agesss

[supple raptor]

Usually I play 2 games every weekend, but one has been put on hold because the DM has a new baby.

[obsidian willow]

all emotionless lel

[magic wave]

Those are all some pretty dope pics :O

[magic wave]

That WD 1.5? :o

I really need to get radiance

xD

[steel arrow]

no i just used "depth of field, sun flare"type prompts

[magic wave]

Oh, thats fair then, looks kinda like WD 1.5 radiance style :o

[pastel bluff]

amazing

[tawny helm]

Generic, but screw it, it looks decent. lol

[edgy badge]

Damn, radically beautiful

[obsidian willow]

[magic wave]

[cyan burrow]

hello hello

hmmm 3090 are dropping to around 700ish dollars now here

second hand

I'm actually getting tempted

[slow needle]

go for and finetune some models

[cyan burrow]

wtf File "G:\AI_Apps\Automatic1111\venv\lib\site-packages\PIL\Image.py", line 3173, in _decompression_bomb_check raise DecompressionBombError(msg) PIL.Image.DecompressionBombError: Image size (256000000 pixels) exceeds limit of 178956970 pixels, could be decompression bomb DOS attack.

[brittle jay]

wtf

[cyan burrow]

weird error

You know, fuck it...It's time anyway for a full format

[cyan burrow]

zzz getting it onto usb taking forever, then just gonna install and then take forever and a half to get all software

[magic wave]

Wait why full format? :o

[glacial bane]

Guys can someone try to prompt this face right here

i cant express the words in prompt

[magic wave]

Looks like a hard one to make

[glacial bane]

Exactly but i suggest if yall interested into tryna prompt it

Atleast we got alil razzle dazzle when we wanna generate something yk sus

[magic wave]

Fair XD

[slow needle]

i gonna have to find a way to kill off that watermark

[cyan burrow]

it was about time and I've had a few suspicious things happen and its just about that time to clear all the bloat

and now here I am, clear of bloat

[magic wave]

I think the best way is to edit the dataset, or add really strong watermakr negative :O

[cyan burrow]

installing every driver etc again

[magic wave]

Ohhh, thats fair then :o

[slow needle]

i gonna retrain this anyway

[cyan burrow]

trying to remember what I'm forgetting

[magic wave]

O fair :o

[magic wave]

Fair fair, damn thats a lot of work to do to get everything working right again tho :O

[cyan burrow]

nah, not that much work

[slow needle]

also made a wlop lora for wd 1.5 too

[magic wave]

Fair I suppose :o

[magic wave]

Thats pretty sweet too :O

[cyan burrow]

how the hell did I convince you with that?

installing python nau

[magic wave]

I mean, I just assume you got like a really big external drive or something where you sending all the stuff xD

[slow needle]

works pretty well

[cyan burrow]

the OS/Games drive is an m2 nvmde or whatever and I backed up some settings but other than that...

[magic wave]

O yeah it does :o

[cyan burrow]

and the games will take like 10 minutes per game max

to download

[magic wave]

Thats fair then :o

[magic wave]

Oh damn das some speedy internet, fair then xD

I got a bunch of random trash laying around so I wouldn't be able to format xD

Also really slow internet speed :o

[nova remnant]

I am trying to upscale this image but it is not working

the final image comes out wrong

[lilac yew]

lmao fr

[cyan burrow]

I need to sort all paths now cause the system drive letters changed, brrrr

[slate night]

Another 🙂

[cyan burrow]

ah installing all the venv and crap

got the media codecs working

got afterburner down, faststone, steam, bitwarden secured etc

[cyan burrow]

the moment I find a 3090 for sale (2nd hand) from a well rated source and for a price that's acceptable. Imma strike. Then sell my 3070 ti

[wispy shard]

[cyan burrow]

would be so much cooler if her clothes were ragged and torn like she'd been through hellish battles, blood stains everywhere and biting on a babies pacifier like it's a cigarr

[obsidian willow]

Serve for the motherland

[cyan burrow]

Whenever I make merge models I always merge in with a nude model, cause otherwise the model might have a hard time knowing what the body actually looks like and some clothes would create artifacts

vampire chick upset that her meal left

[obsidian willow]

Yet another Akane.

[zinc flax]

Next episode never

[obsidian willow]

epik

[cyan burrow]

--xformers --opt-sub-quad-attention --sub-quad-q-chunk-size 128 --sub-quad-kv-chunk-size 64 --sub-quad-chunk-threshold 90 --api I'm testing out this to see if it'll be fast and stable

[pastel pollen]

share the prompt please

https://tenor.com/view/please-begging-gif-25375904

[cyan burrow]

hey hey eppi

[pale monolith]

heya!

[cyan burrow]

--xformers --opt-sub-quad-attention --sub-quad-q-chunk-size 128 --sub-quad-kv-chunk-size 64 --sub-quad-chunk-threshold 90 --api this has been working pretty well for me so far. Pretty fast generations and little less oom

that sub-quad thingy is confusing so I just winged it

[pale monolith]

[cyan burrow]

my new model is merging well so far. My goal is a 2.5d/pseudorealism model

that might be a lightsaber but its supposed to be a pool cue

[brittle jay]

why so blurry? upscale now

[supple raptor]

Sharpness wouldn't fit the art style at all IMO. I had to apply slight blur to the face because adetailer made it too sharp.

Also, more resolution would just mean that it has to be downscaled for it's intended use which would potentially make it look worse.

[brittle jay]

photoshop

just increased sharpness

though upscaling would be even better

[supple raptor]

I prefer the softer look on this one.

[brittle jay]

yes

same

but the blurriness hurts my eyes for some reason

idk

[cyan burrow]

I tend to upscale like 25-100% at a time. Depending on what resolution I'm on. Using different samplers depending on what flaws I start to see

[brittle jay]

man I swear sometimes my eyes are too sensitive to stuff

[cyan burrow]

Sometimes you also create your own flaws when there aren't any, or at least aren't what you think there are

but you tend to connect it to the worst :p

[brittle jay]

well everybody imagines flaws where there aren't any

[cyan burrow]

yes but you have a partucykar talent for it with your sensitive eyes

this was not my intention

[brittle jay]

I just researched, according to DSM5 sensitivity to sensory stimuli is a symptom of ASD which I have, so it makes sense ig

[cyan burrow]

Yeah, not in my case'

in my case its sounds 😛

my adhd I have harnessed as a tool to come up with ideas, story etc

[brittle jay]

well it is also a symptom of ADHD

which I also have

lol

well practically everyone has adhd these days

to varying severities

[cyan burrow]

meh its overdiagnosed to people who don't have it just to sell medication

I've been dealing with this crap my whole life. Tiring myself out to fall asleep, to the annoyance of my mom

[brittle jay]

I go slep now

[brittle jay]

Asymmetrical

isn't she supposed to have those thingys on the right side as well?

[supple raptor]

Left half of image is Flandre, right half is Remilia. They're sisters.

[brittle jay]

oh

some toho lore

[brittle jay]

GN

hopefully I do well on my important test tomorrow

[cyan burrow]

is there a pix2pix around easy to ihstall etc

[supple raptor]

PIx2Pix has been natively supported in Auto1111 for months.

https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features#instructpix2pix

[cyan burrow]

but I dont have a tab for it

[supple raptor]

it uses img2img tab. You just have to select a pix2pix model.

[cyan burrow]

oh ok

alright downloading it

[teal pivot]

[supple raptor]

@teal pivot I really like v2 of your model btw

[teal pivot]

Thank you, I did my best ^_^

[supple raptor]

It has a huge improvement on the hands and general anatomy over the first version.

[teal pivot]

Yep!

[sharp bluff]

yall know whats happening?

tried SD today and this happens every time 💀

[supple raptor]

Check cmd to make sure it didn't get paused by accident. Aside from that, the most likely thing is that it's out of VRAM and trying to swap to ram or pagefile.

[sharp bluff]

ive restarted it a couple times definitely not paused

i might be out of vram?

not sure how i would be but let me look

[supple raptor]

Could be something else using extra vram.

[sharp bluff]

yep

i am indeed

i forgot i had a 100000x4000 image open in photoshop

[supple raptor]

Ah, that'll do it

[sharp bluff]

oh yeah it actually works now ty

[supple raptor]

[edgy badge]

Damn, how do you make images with such quality?

This is the best image I got from StableDreamer

[slate night]

Ooh she's cute!

[edgy badge]

Yeah, Uraraka Ochako

[slate night]

Model merge + prompt is the secret sauce

[edgy badge]

What's model merge?

[slate night]

In automatic1111 you can get two different models (checkpoints) and combine them

[edgy badge]

Is that an extension?

[slate night]

Nope! Hang on I'll get a screenshot

It's just on a tab at the top

[slate night]

For this image, I made my own model by combining a bunch that had bits that I liked- pastel color palette, softer shading, messier coloring

[edgy badge]

Prompt: photo, Uraraka Ochako (From my hero academia), beautiful face, brown hair, black sundress, sharp focus, hyperrealism, cinematic lighting, photography NIKON D580 lens 35mm F1/100
Style: Anime

photo, Uraraka Ochako (From my hero academia), beautiful face, brown hair, black sundress, sharp focus
Style: Anime

[slate night]

Idk what the thing in the water is beside her lol

Kinda looks like a large mouse trap

[edgy badge]

It looks moe like a raft

[slate night]

Yeah I'll bet it was probably supposed to be a raft lol

[slate night]

Bowsette my beloved

[pale monolith]

b-beloved?!

[slate night]

The Ochako lora plays better well with my other model. Outfit is still a bit off though but whatever

[regal pivot]

...hi ..Do you know a good workflow, to add a background to my character (who has the background removed) in A1111..??

[pale monolith]

if the character has a transparent background, i'd say generate some separate backgrounds then paste together in something else (photopea, clipstudio etc)

[edgy badge]

Definitely

[stone onyx]

I tried, didn't work :(

[slow needle]

wat

[wispy shard]

F

[magic wave]

[lyric zealot]

Oh wow, I got the username update

[lyric zealot]

Well, it’s my waifu

[magic wave]

Tru :o

[magic wave]

Woa

Got lil badge now

[lyric zealot]

Yup, probably cos I’m a 2015 account