🍥｜anime

[dawn barn]

Oh nvm ye you fucked then

I recommend these modern bladeless fans

[native halo]

i have never been in germany i dont know germany i dont know where is germany but they dont have one (real)

[dawn barn]

AMD GPU better use SDNext which has standard AMD GPU support

10% is not nothing

I think like 10% have an AC

A bit less

[scarlet zenith]

what did you use for this?

[wet elm]

can someone help? My loras are not working anymore for some reason
i put the lora and the activation text but it does not work, nor does it show it in the generated image lora hashes

[final vault]

i see

[storm ivy]

boch

[analog dagger]

Every anime model should work but I use counterfeit with rage unleashed and detail tweaker LoRa all set In 0,7

[gusty palm]

is sdnet a 111 alternative?

[dawn barn]

It's an A1111 fork

I mean yeah you can view it like that

It should have integrated AMD support

[gusty palm]

ah ok

https://www.instagram.com/p/CwFWNkeOj4y/ heres one but it also has some loras on it

Instagram

[drowsy sail]

https://media0.giphy.com/media/v1.Y2lkPTc5MGI3NjExcnYyZWNmZGZpc3Nxc3k5NXpieHhub3E5YjdmMHNjdGRmYjUxOWh3OCZlcD12MV9pbnRlcm5hbF9naWZfYnlfaWQmY3Q9Zw/S4QbsO9Is8G0WSQswa/giphy.gif

[wet elm]

ayoo how did u do that?

[drowsy sail]

The art I made by AI, the animation I made manually by another program, photoshop and after

[wet elm]

ooh that makes sense

[wet elm]

can someone help? My loras are not working anymore for some reason
i put the lora and the activation text but it does not work, nor does it show it in the generated image lora hashes

[regal pivot]

@wet elm if you are using A1111. there is a possibility that you are missing an update that changes the dependencies of LoRA, Lycons, or Python... if that is the case, consult the support channel https://discord.com/channels/1002292111942635562/1002602742667280404

[wet elm]

oh damn it maybe that

[quiet bramble]

Any good models for making light novel illustration-style covers?

[native halo]

show an example maybe we know some

[gusty palm]

im trying to do stuff like that but in reverse with blender

https://imgur.com/dagY9cP i wanna put like an ai style over text animations or something

prob gonna try with like a single frame first cuz idk how to to gif2gif yet

[pale monolith]

i think it would be easier to do a deforum animation, then put a black cover mask in adobe to shape the text then like a glow layer on top of that / blur the black mask

[gusty palm]

i looked into deforum a tiny bit but i wasnt sure if my hardware was sufficient to run it

do u think 6gb vram would be enough

[bitter latch]

well i'm not sure what he's eating but he's into it

[bitter latch]

yeah i'm just going random 😄

[bitter latch]

😄

3d Anime?

hha

thats better. well ok more anime 😄

tht hand wwtf?

[zinc flax]

Oh wow cute style

Love the eyes

[nova remnant]

how can i get this type of mouth

[young fjord]

expressionless, or :|

[lucid surge]

How to create

[near elbow]

emotionless 🙂

[steady grail]

this image needs a banana phone

[pale monolith]

#1072220168534642768 ❤️

[frozen olive]

inpaint in nijijourney is awwesome

[gusty palm]

https://i.imgur.com/qWsfya0

unrelated but theres a steam thing called vpet

its like bonzibuddy but not a virus

its like a tamagicho

[native halo]

https://tenor.com/view/monkey-deal-with-it-gif-7717548

[gusty palm]

she looks like either a girl from kill la kill or magicval girl acadamia whatever its called i forget

[gusty palm]

lmao

[gusty palm]

i fell into a rabbit hole of weird free games on itchio

[summer shuttle]

miku or miku

[nova remnant]

Hi very nice Pokémon pictures.

[primal wharf]

beautiful style and hanfu

[drowsy sail]

https://media4.giphy.com/media/ng8bihkFt0dDAc8aOb/giphy.gif

[gritty sundial]

model pls

[frozen olive]

niji

[gritty sundial]

model or Lora?

[frozen olive]

model

[tidal phoenix]

Anyone want a new PFP?

[gritty sundial]

is there niji model? I thought it was just Lora, send me please?

[tidal phoenix]

I think they may have just done it on Niji as opposed to StableDiffusion

[gritty sundial]

it doesn't make sense for the guy to send something he did on Midjourney here on the Stabble server

[tidal phoenix]

Yes, well, welcome to people

[native halo]

since we sharin non SD content

[frozen olive]

its where does it say you cant drop images from any ai?

this doesnt make sense either

[gritty sundial]

Sorry man, it was just a guess

[brittle jay]

Happy 1 year anniversary

[native halo]

well its an anime chat,so u can post anything related to anime even non AI

[quiet bramble]

Looks great! Could I please get the model or lora name?

[runic vault]

it's based66 and the character lora used is my own. https://civitai.com/models/133098/mikado-hades-izanami-blazblue

[neon linden]

did you animate it?

[hexed ermine]

cool! How you get it?

[runic vault]

do people not check png info anymore?

(outline:1.5), ((heart background)), (pink background), ((colorful)), mikado \(blazblue\), medium breasts, red eyes, (half-closed eyes), long hair, purple hair, ponytail, headpiece, black thighhighs, detached sleeves, hakama short skirt, sideless outfit, smirk, cowboy shot, <lora:mikado (blazblue):0.7>

aid28, thick thighs, wide hips, huge ass, skindentation, holding, holding weapon, holding sword, holding gun, holding food, monochrome, greyscale, wide shot, jpeg artifacts, low quality, worst quality, text, watermark, signature, lips, teeth, censored, mosaic censoring, bar censor,```

[bitter latch]

yes, as you just proved

or not.. shrugs

[summer shuttle]

monika (blood warning)

[karmic swift]

anyone here who uses dreamshaper ??

[drowsy sail]

Yep

[neon linden]

looks awesome! what did you use?

[fierce jolt]

happy birthday SD, guys

[near elbow]

anime? comic? buff!

[untold glacier]

💘

[quiet bramble]

May I have the name of the model, please?

[gusty palm]

i uninstalled genshin gained 60gb of space for models

[summer shuttle]

gura

[slate narwhal]

Vanilla SDXL tests:

[bitter latch]

there we go

[gusty palm]

thsre are nice

[crude portal]

ola

((letter V)), contemporary, classic floral elements emanating from the center, decorative design, stars ornament, motif, glitter, roses, leaves, flowers, l, negative space, highly detailed engraving, logos

[bitter latch]

and last one for a bit, gonna play some games

[slate narwhal]

Thx, SDXL is getting better. There is a lot of trickery in these, especially to get the eyes right without any upscaling. 👍

[gusty palm]

reminds me of the quilt type embroidery my mom used to make

[brittle rampart]

sensei`s who will tell me how to use embeddings correctly in SDXL?

[sick anchor]

That's crazy good. Is that DeepShaper?

[frozen olive]

omg this is so cute and good very nice, may i ask what model is that?

[lyric zealot]

Nope, it’s not dreamshaper. I’m not at pc right now, I’ll get you the model name later!

[lyric zealot]

Thank you! I’m not at pc right now, I’ll get you the model name later!

[frozen olive]

thank you how do you like baldurs gate 3 so far

[zinc flax]

Cuuuute

[lyric zealot]

it's great! Really loving it now. Still on act 1, playing slowly

@frozen olive @sick anchor this is the same pic with metadata

[slender dock]

一个美女

[wet elm]

what are some anime models yall recommend to get flat anime screenshot images?

[lyric zealot]

not sure what model, but i recommend using "anime screencap" as prompt

[zinc flax]

You can use aurora with anime screencap

But using the negative prompt low quality, worst quality kinda ruins the screenshot-like style though

[wet elm]

oh ok ty!

i just came back to SD after like 1 month so i wondered if there where any new best models lol

[frozen olive]

what is metadata for a model is it on civit?

the name is kinda confusing ngl

can you show your settings?

[earnest orbit]

cute lora

[lyric zealot]

Sorry about that. Metadata is basically embedded data in the pic that shows you the settings needed to generate the pic. Im currently outside right now and will not be able to share the details

[arctic nova]

Thicc

[arctic nova]

Just open the pic with notepad or something

Prompt

1girl, black hair, green eyes, braided ponytail, elf, pointy ears, shadowheart, armor,

<lora:aoi_12:0.3> <lora:TAF2022:0.3> <lora:ShadowHeart:0.9>```

Model
`anmitsumimimi_v10`

[frozen olive]

ah thank you 🙂

[drowsy sail]

Do you think it turned out good?

[frozen olive]

ye seems nice

[drowsy sail]

I managed the art by AI, but I redesigned it myself by hand, do you think it turned out good?

[frozen olive]

u can try to run it with low denoising in inpainting again

it might fix the bad line outs

[arctic nova]

The hand on the right 😭

[steady grail]

impaint the strowbarry and place a banana lol

experimenting with a model. cute

[drowsy sail]

the other hand

the left side I designed, the right side is the original painting

[steady grail]

1111 is so slow, but why results are so much better? >.<

[quiet bramble]

That’s great! Could you please let me know what model you used?

[tropic wraith]

FAQ: What is Stable Foundation?

The Stable Foundation server is a hub of community, creativity and co-collaboration that acts as a nexus between the wider Stability community ecosystems! You can find out more about us & our partners by heading over to #1072229057707659404!

[wet elm]

FGO sprite lol

[wet elm]

hmm im pretty sure that is nijijourney but im not sure

[sly rune]

Is A1111 dev branch acting slow for anyone else?

[native halo]

that one sucks if u using only 1.5 stick to an old version and if u using sdxl dont use a1111 use comfy instead

[scenic oyster]

https://www.youtube.com/shorts/Oi7nl2lpM9s

[bitter latch]

anime rangers?

[bitter latch]

well we found her nemesis

[gloomy lodge]

she's harmless she just really likes sticks and snow

[bitter latch]

ah ha!

wow that came out better than expected

ok with that i'm gonna go pass out now

[frozen olive]

nijijourney

[bitter latch]

thuds

[bitter latch]

sneaks on in...

[slender charm]

https://cdn.discordapp.com/attachments/1019446913268973689/1145297889459970048/00137-shux4.png

[bitter latch]

hmm is this anime CGI?

[bitter latch]

finally got it to go male instead of female

power up!

[keen hollow]

what's the latest technique ya'll are using to get hands that don't make you want to scream? 🙂

[arctic nova]

I already gave up on that. Most models nowadays can make hands relatively well

But I'm way too attached to my own model to consider changing to a new one, so I kind of just live with it lol

[brittle rampart]

I am absolutely not sure that the nodes I built are correct, but the result is far from completely terrible.

[bitter latch]

grinz evilly

[quiet bramble]

Ah, well thanks.

[short coral]

uwu

[summer shuttle]

fumo cow

[bitter latch]

leaves this here

things your brain comes up with when ya can't sleep

[tawdry jolt]

https://github.com/deepinsight/insightface.git

1 GİRLS TALL BIG

[bitter latch]

wves and goes back to bed to hopefully actually sleep

[magic wave]

3 new LoRAs that trained kinda mid

[astral drift]

Sweet, what model do you use?

[frozen olive]

nijijourney

[astral drift]

SDXL?

[astral drift]

I checked out your deviantart, I can tell you put a lot more work into the pieces than normal AI art

[frozen olive]

its not its a complete different ai art platform, you can google it, thank you 🙂

[astral drift]

Oh, it's midjourney based

[brittle rampart]

unexpected and harsh truth! 🤣

[brittle rampart]

First kawaii!

[native halo]

speakin of kawaii

[brittle rampart]

how can she not be sad if her skirt is pulled up and even in front of everyone on the street😱

[astral drift]

Not a fan of (mid) journey myself

[brittle rampart]

but you must admit that the result looks very pleasant. light, not sharp shades, contours, etc.

[astral drift]

Yeah the quality is there, the freedom isn't

[brittle rampart]

I agree with the last one 100500%

[magic wave]

Turns 2 out of the 3 styles I trained didn't actually train that bad

Hero Neisan one stands out though, but the other one is really unique for the eyelashes

[brittle rampart]

sensei. who will tell me how to get and save a text file with generation settings in Comfy same like in Auto1111?

[magic wave]

Anyone here uses Counterfeit XL?

[brittle rampart]

i am cant load it in Comfy.... sad...

[magic wave]

Oh oof, I haven't used it, so I wanna know whats the experience of people with it

[wispy canopy]

Hi all

Been a while

[magic wave]

Ayyyy, Eface

[wispy canopy]

auto has proper xl support, i finally get my random prompt back

it didn't work in comfy 😦

This be the reason why i want my random prompt back 😄

[magic wave]

Yeee, I mean, I'm on the v1.6.0 Pre-releasea atm

[wispy canopy]

my prompt: "cute girl", rest is the randomizer going loco

[magic wave]

I haven't tried using SDXL here now, but before it was horribly optimized

[magic wave]

Fair xD

[native halo]

better coherence but 1.5 still better cause of the variety of styles

[magic wave]

Makes sense :o

[wispy canopy]

i feel sdxl shits on 1.5 in terms of style/application

[magic wave]

Talking about style, I just got done training a LoRA

5/6 done

[wispy canopy]

now to find a good generic xl model, which goes well with my prompting style

[native halo]

also most 1.5 anime models still have the novelai leak merged so thats prob why

[magic wave]

I mean, I feel like SDXL is cool for the prompting, for the styles I feel like atm 1.5 is still the best, at least for me

Specially since I can train SD 1.5 LoRAs

And I don't think I can train SDXL LoRAs since its so huge

[wispy canopy]

mate of mine is shitting out sdxl lora's like mad

[magic wave]

Nice :O

[wispy canopy]

but then again, he bought an A6000 just because

[magic wave]

Oh damn

XD

[wispy canopy]

has a 4090, 3090, this guy is AI incarnate

[native halo]

i just checked an anime sdxl model and it says it was finetuned on 30k pictures,now the novelai leak has i think 5million pictures so u will obv get better results on 1.5 but still have that problem of the deformity or incoherence etc but if sdxl can get that kind of finetune it will deff kill 1.5

[magic wave]

Damn, well yeah I'd guess he probably has the gear to train SDXL like crazy xD

[magic wave]

Fair fair

I mean

The deformity isn't a issue for me

But I want a model to listen

Which

To be fair

I think I overestimated for SDXL, it does listen better

But not perfectly

[wispy canopy]

how do you want it to listen, it listens a lot better for me

[magic wave]

Well, I would want a model training both on natural prompting and booru prompts, cause booru prompts are easy for me to navigate and stuff

[wispy canopy]

yeah, i don't do booru prompts 😛

[magic wave]

But the ability to like tell the model to for example make like a dress thats made out of red dragon scales

[brittle rampart]

btw how in XL use "embeddings" ?

[magic wave]

And like be able to tell it that, and still understand the rest of a complex/specific prompt

[wispy canopy]

i think some other tech was in the works, iirc

Dynavision, i think we can become friends

[magic wave]

Fair, what kinda stuff? :o

[wispy canopy]

i think i t was neutral prompt

oh right, and thats the other reason i did very little sd: the HEAT...

[magic wave]

The heat? :o

[wispy canopy]

yeah... my house is highest rated in isolation

[magic wave]

Oh damn o.o

[wispy canopy]

so when summer heat is here, i either get baked by heat outside, or heat building up inside

my heating has been off since februari

[magic wave]

Oh damn that sucks

[magic wave]

Daaaamn, why so long? O.o

[wispy canopy]

my gpu is making enough heat to heat the room 😄

i think i've been venting heat as much as possible since april

[magic wave]

Luckily here is cold so it works out xD

[magic wave]

You should get that sorted out tho :o

[wispy canopy]

how? my small airco can handle it if i don't make it silly

which SD kindly does, and fancontrol doing the thing it does best: make all the fans activate when GPU is loaded

[magic wave]

I mean, do you like heat? I personally hate heat so thats why I think I'd be best to get it sorted asap :O

[wispy canopy]

so i've got a beast pc, capable of producing 900w of heat and properly venting it, in a house which is very well isolated, meaning little heat gets lost to the outside

it's only an issue during summer

i keep windows open, which vents enough heat

[magic wave]

Ohhhh, thats fair then

WAIT

Isolation

I'm kinda stupid, nvm

[wispy canopy]

sorry, insulation

the proper term

[magic wave]

OH

I knew something was wrong XD

[wispy canopy]

my house has highest insulation rating atm

[magic wave]

I gaslighted myself into that was how it was written, and I thought I was reading it wrong and thinking of like the word for excessive heat xD

Makes sense tho, thats pretty neat then :o

[wispy canopy]

i, for one, love sdxl's ability for abstracter stuff 😄

[magic wave]

That is pretty dope :o

Kinda like stained glass mixed with like a dark neon style :o

[wispy canopy]

it's ... uhhh checks notes... "abstract [calligraphy alcohol ink lineart:0.1] of a girl in a meadow at night, dark, comic, anime"

so yeah. it's that.

abstract calligraphy alcohol ink lineart

[magic wave]

Also I really dislike how the tabs are in v1.6.0, why did Auto move the tabs for extra stuff and make generation a tab

[magic wave]

Makes sense :o

[wispy canopy]

1.6 isn't out officially right?

only the release candidate, or am i being dumb?

[magic wave]

It isn't, but I mean, I don't know why he would even think of doing that xD

[wispy canopy]

dang

[magic wave]

That is really dope too :O

[wispy canopy]

sdxl based models do really well on my favorite prompts, it seems 😄

[magic wave]

Nice, I imagined they would :O

[wispy canopy]

it's pushing my 3090 to the limit for a single imiage though

[magic wave]

Also, 3 new styles trained, 1st one is kinda experimental, 2nd one is more of a skin thing, and 3rd one is pretty noticeable

[magic wave]

Is it?

[wispy canopy]

all 24 gb's used for a single highres image

[magic wave]

Well, if its Auto1111 I guess that would make sense

Yeah I think Auto11111 for SDXL is just poorly optimized

[wispy canopy]

comfy is a LOT better on the ram imho

[magic wave]

Yeee :o

I just straight up can't use SDXL on Auto1111, so I guess we just hope he optimizes it xD

[wispy canopy]

#3 best style

[magic wave]

Fair, that one is Afrobull :o

[wispy canopy]

should i know him?

or is this... spicy things he makes?

[magic wave]

Uhhh, little bit of both I think

This peachy using that style

[wispy canopy]

https://tenor.com/view/hot-so-hot-summer-sweating-sweating-hard-gif-16045826

[magic wave]

Fair XD

[wispy canopy]

dem hips tho

[magic wave]

Here peachy in the other 2 styles, Masami Chie is like a classic japanese artist style, is pretty tho, pretty sparklish skin

[magic wave]

I always be making them huge hips

This one has really interesting eyelashes

[wispy canopy]

and a glasgow smile..

[magic wave]

Kinda ye xD

[wispy canopy]

and i hate the crotch grabbing dresses...

but that's my personal gripe with all descendants of a certain model

[magic wave]

AI do that sometimes, I don't have "skin tight" in negatives

And fair :O

[wispy canopy]

same as the underglow hair

[magic wave]

I usually don't get that issue much :o

Here is Hero Neisan LoRA, I don't know why the shadows got so dark, but I kinda vibe with it

[wispy canopy]

i like dark shadows tho 🙂

contrast = good

[magic wave]

Yeee, I like em too

Should ask 2vXpSwA7 if they gonna make a LoRA for shadows

[wispy canopy]

who?

[magic wave]

https://civitai.com/user/2vXpSwA7/models This person

For me they like one of the best LoRA makers

They are madman tho, their methods seem crazy, BUT it works flawlessly

Like their LoRAs have like a ≤ 5% effect on composition

[wispy canopy]

nice ^^

[magic wave]

i spent 10 minutes looking up stuff to remember how that symbol was used

XD

[magic wave]

Yeee, its really nice to change and tweak stuff, and it doesn't have any compatibility issues with other LoRAs because they're so good

[brittle rampart]

Experts, nevertheless, someone will tell me how to save the generation text (prompts, seeds, etc.) into a separate file?

[wispy canopy]

it's a setting

first page of settings possible...

[brittle rampart]

in Confy?

[wispy canopy]

in comfy, there is no 'saving prompt'

as comfy is a workflow

getting it to export to file? uhh... unless there's a specific node which does that... i'm not sure if it can

[brittle rampart]

hmm...then how do you use WebUI for XL?

[wispy canopy]

i set it up to change to release_candidate

[brittle rampart]

I apologize in advance for the possibly obvious questions.

[wispy canopy]

'git switch release_candidate'

and then start web-ui as normal

but now it has support for sdxl based models

[brittle rampart]

ah. ok.In that case, I'll just be patient and wait for a full release

[tidal phoenix]

My current boredom filler: Plugging song lyrics into a cutesy anime model

[magic wave]

Peachy

[wispy canopy]

song lyrics always a good way for images

[nova remnant]

Hi very nice picture anime.

[tidal phoenix]

Like Chimeras from Cassette Beasts, if there's any curiosity

[wispy canopy]

I've done a LOT of song lyrics 😄

[supple raptor]

Comfy embeds the entire workflow in the image output metadata by default. You can just drag and drop an image generated in ComfyUI back into the workspace and it will load the workflow. Alternatively, you can manually save the workflow with the save button and likewise load those files using the load button.

[wispy canopy]

yeah, i think he asked if there's an automated node that might save prompt text to similar filename

at least, that's how interpreted it

i knew about the drag/drop

[merry tendon]

wow, you guys are still alive?

i rarely use SD anymore, im very active in the LLM community now tho

[kindred egret]

Does anyone about any checkpoint that can generate something similar to this style?

[supple raptor]

I have a mix that is somewhat similar. https://civitai.com/models/100505

[frozen olive]

"here is my ticket"

[steady grail]

"little girl, this tiket is fake, right?"

[frozen olive]

[steady grail]

🤣

her face is so cute, looks like she is trying to cheat the check point lol

[quiet bramble]

Are there any models like moonbeam (PixAI model) that I can download?

[native halo]

post some pics here to see

[wet elm]

if i go back to an old a1111 version and come back will i lose anything?

[native halo]

no

old version is better if u only use 1.5 models

[wet elm]

i only use 1.5 lol

what changed?

[native halo]

when they added support for sdxl it became a buggy mess

[wet elm]

i wanted to reacreate an old image from my files but i cant rn, so i wanted to change versions to see if i could

[wet elm]

it maybe because of that

[native halo]

yea thats why i dont do the auto update thing,sometimes they fk something up and u gotta roll back to an old version

[wet elm]

and idk how to rollback lol

[native halo]

now u gotta go to where you have webui installed and open cmd there and then type

git pull
git reset --hard "enter whatever hash u wanna roll back to here"

[wet elm]

tysm, sorry to sound dumb but where do i know the hash for an older version?

[native halo]

here https://github.com/AUTOMATIC1111/stable-diffusion-webui/commits/master

[wet elm]

ty, so i just copy the hash like this "af081211ee93622473ee575de30fed2fd8263c09" and put it there?

[native halo]

yea but remove the " "

[wet elm]

oh ok

[sage cloud]

Cool Sangonomiya Kokomi

[wet elm]

tysm!!

[kindred egret]

is not mine lmao

thats why i was asking if someone knew a model similar to that

[native halo]

anything can do that style

or mistoon

[sage cloud]

No idea, I wish I know

[kindred egret]

gonna give that one a try later then

[native halo]

theres also kawaiimix nijiv5 that looks kinda similar

u can go to civitai and filter only the models to compare

[wet elm]

i did it and now got this on the cmd "HEAD is now at 15adff3d Merge pull request #11733 from akx/brace-for-impact"

[native halo]

if there are no errors then it worked

[wet elm]

ok, tysm again lol

[native halo]

🙏

[regal pivot]

Hello, does anyone know of a model that simulates effects similar to the particle brushes that some image editing programs have?...
I am interested in something that is recognized/applicable by some anime model
I know there are some LoRA trained in the fire/flame effect... but is there something that handles particles as a general idea?

here is an example for a "fly" ....
any idea how to simulate it in lineart + some good prompt?🤔

[tidal phoenix]

Would you mind terribly sharing the setup you used? This is absolutely adorable and I love it XD

[native halo]

thats niji-journey

[frozen olive]

i used Nijijourney

[bitter latch]

runs in circles

[bitter latch]

well that came out unexpected

[wispy canopy]

**"MA, EFACE IS LYRICSPROMPTING AGAIN" **
(ngl, getting a "cant contol power vibe from #3)

[wispy canopy]

i think i'm in love ❤️

What kind of mythical beast is this?!

[wispy canopy]

adding abstract was the best thing i did...

feels weird using no negatives at all 😮

WOOO, my monster lyrics prompt is amazing in sdxl models!

lol, it actually tried making prince?!

why did that single paw get fcked 😦

Still can't believe this is "default output" without any negatives

it's just my mega collection of lyric prompts with some dynamic prompting elements. Let the model figure out to make something out of it

2000 tokens. i want to make it 5000 😄

that mario song really being dominant with it's princesses and castles tho 😮 needs more lyrics 😄

[brittle rampart]

you are kinda crazy 😂

[wispy canopy]

it's a dynamic prompt, only 2 or 3 verses are selected out of all of them

3000 tokens, lets do a batch again 😄

[brittle rampart]

How does it work ?

[wispy canopy]

in the short version, A1111+dynamic prompts can both grab "random element" and "seperate elements into averaged weighted token groups"

short example for dynamic prompt:
{2-3$$ BREAK $$ dog|car|bird|sea}

this would cause the prompt to select 2 or 3 words, seperated by the word break, for example "car BREAK sea BREAK dog"

well... instead of dog/car/bird/sea, i have 13kb worth of song lyrics loaded in (mind you, properly formatted)

[brittle rampart]

sound really crazy!

😱

[wispy canopy]

well, it if makes pretty images, anything is valid, right? 🙂

[brittle rampart]

I'm wondering if it will be possible to repeat the result after a while, or is it all "magic of random words"?

[wispy canopy]

how do you mean? yeah, if do enough generations, you'll get similar stuff, or even the same prompt

right now it's 103 possible lines, so selecting 2-3 from those, gives about 1103336 possible combinations

although some lines do seem to warp the rest around them a lot, but that might be model specific

[brittle rampart]

enough generations how much?

[wispy canopy]

more than a million possible combinations currently

[brittle rampart]

for example, if I liked the concept of generation and I would like to change something slightly ... then alas?

[wispy canopy]

you get the specific prompt that was used to build the image

a1111 dynamic prompts saves the template (used dynamic prompt) and the actual prompt used

[brittle rampart]

that a really "magic world of random words"

[wispy canopy]

can't help it, i love song lyrics prompting the best, so i just... improved upon it lol

for example, the actual prompt for this image, that was made from my huge random dump was
anime official art, inspired by Gordon Parks BREAK It's hard to hold your head up high But we must try BREAK A wind so strong (born in the storm to carry on) Will take us high and up where we belong Through space and time (like a comet in the night) We will travel along together side by side (side by side)

and then i could just do this single prompt a couple of times if i really like the outcome

or change small things

still, gotta say, sdxl based models are pretty damn good at this shit, waaaay more coherence than ever before

oof, single batch of 8 images takes 10 minutes lol, so many steps on restart sampler really takes a toll

[wispy canopy]

this... is impressive in it's own right... is this even sd?!

[grizzled moat]

Yeahh

SD 1.5

[wispy canopy]

wow 😮 respect

[wispy canopy]

very cool actually 😄

[grizzled moat]

LMAO

yeah

very cool concept

ngl

[wispy canopy]

i could watch a series of that style, looks so lovely actually

[wispy canopy]

i'm addicted to sdxl model output now 😦 can'tgo back to 1.5 i think

[grizzled moat]

is sdxl good ?

been ages since i tried ai gen

[wispy canopy]

i'm using protovision for my images now

it's an sdxl model

[grizzled moat]

Show me some results 😄

[wispy canopy]

the images i posted earlier :), but lemme transform my prompts so i can make anime again

[grizzled moat]

Ohh

Its cool

[brittle rampart]

funny repetition

I'm sure it took a lot of time 🤣

[grizzled moat]

xD it didnt

[wispy canopy]

ooff.. single generation time going up to 2:30 minutes

[grizzled moat]

ohh

on SDxl ?

Damn

[wispy canopy]

well it is 1536x1024 x 70 steps x restart sampler x sdxl model

[brittle rampart]

now try add faceDateler node for correct face!

[wispy canopy]

i had beauty in the negatives, so i don't think it would help a lot lol

[brittle rampart]

for distant angles, this will help a lot

[wispy canopy]

correct... but that would even make my generation times worse lol 😄

[wispy canopy]

this good enough for a "random" output?

[brittle rampart]

I think on your hardware - it will not play a big role

[grizzled moat]

heavily upscalled right ?

[wispy canopy]

no, single shot

[grizzled moat]

i mean its good tho

how much time ?

[wispy canopy]

single shot (no upscale), no negative prompt except "error"

1:50

[brittle rampart]

😭

[grizzled moat]

i think sdxl does the upscalling on the background of the process

[wispy canopy]

nah, it's been trained on larger resolution, so it doesn't fall apart on higher resolution

so it is more consistent on resolutions above 1024x1024

[brittle rampart]

my time generation average 3-5 min

without upscale

[wispy canopy]

had a 3080, upgraded to 3090 for llm, now back to sdxl, all 24gb still in use 😄

[brittle rampart]

bourgeois! 😱

[wispy canopy]

not as crazy as a friend of mine

finding a used 3090 is a godsend 🙂

found mine for €750

[grizzled moat]

Which country ?

wow

[wispy canopy]

NL

[brittle rampart]

give me a rope and soap and show me a beautiful tree!

I will sing my sadness 😭

[grizzled moat]

ceiling will work

[brittle rampart]

very low. and no space!

[grizzled moat]

mhm

[wispy canopy]

dang, "cute anime girl" is a LOT of token weight

[brittle rampart]

Miku is also depressed today?

[supple raptor]

She's sitting in front of the TV because you weren't paying attention to her.

[wispy canopy]

"disappointed miku is judging you"

[brittle rampart]

Oh Miku! your little sister. so sorry! She was very busy looking for a rope of soap and... a beautiful tree!

[slate narwhal]

I'm with you on this one. I'm not going back to 1.5

[wispy canopy]

the image coherency is just leagues beyond what 1.5 models can do

[slate narwhal]

Yep, and it keeps getting better and better

[wispy canopy]

it even has flippin' birds!

do that in a a1.5 anime model, and prepare to laugh your ass off lol

[brittle rampart]

open wallpapers site... and sell it!

[wispy canopy]

i'd personally never sell my ai art

can't really claim it as something 'mine'

[brittle rampart]

oki-doki... then try make of it a very very big puzzle

[grizzled moat]

i can safely claim this as mine

🙂

[wispy canopy]

which model are you using?

[slate narwhal]

The newest SDXL_Niji_V5.1 (released a few hours ago), it is awesome

https://civitai.com/models/120765?modelVersionId=151548

[wispy canopy]

well, it doesn't seem like an over-stylized anime model, so lets give it a whirl 🙂

not completely anime (it'll have to do as anime render, i suppose), but goddamn these outputs 😮 (is even skin off considered too nsfw?)

[slate narwhal]

It has several Niji styles in one model

[brittle rampart]

smile, just try smile......

[slate narwhal]

You sure? Apterus, the king of evil smiles is in the prompt ^^

😮

[wispy canopy]

meh, niji doesn't like my prompting style

[wispy canopy]

for now i stay with protovision (#1 tho... )

[brittle rampart]

guys, tell me, as a girl who does not understand all the technical aspects, but only uses it. what you think about AUTOMATIC1111 v.1.5.2

[wispy canopy]

switched over to 1.6 so sdxl works properly

[native halo]

buggy mess

[brittle rampart]

but 1.6 not a release

[slate narwhal]

I don't get it, you are using dynamic prompt right? It should have worked.

[brittle rampart]

stay still on v 1.4.1?

[wispy canopy]

some models don't like it for real

[wispy canopy]

it's the release candidate

[native halo]

i wouldnt know im still on a june commit because i only use 1.5

[brittle rampart]

oh i see v 1.5.0 support XL?

[native halo]

yea but its buggy u have to switch to the dev branch so it works better

[wispy canopy]

yeah, 'git switch release_candidate'

but it's essentically 'not ready', that's why it's RC

[brittle rampart]

tell me how i can try it, without update my v1.4.1?

[wispy canopy]

you'd have to create a new install

that's the only way to know for sure

(my cursed lyrics strike again)

omg.

these. are. horrible.

i'll post in the horror channel

[brittle rampart]

but how should I indicate that exactly the release I need is downloaded?

[wispy canopy]

you create a new folder, and git clone there

[brittle rampart]

from dev branch

[wispy canopy]

then you use the command git switch release_candidate

that switches you to the new dev branch

[brittle rampart]

git clone ?

[wispy canopy]

how did you set up a1111 in the first place?

[brittle rampart]

Download sd.webui.zip from v1.0.0-pre and extract its contents.
Run update.bat.
Run run.bat.

[wispy canopy]

uhh...

yeah... about that 😮

[brittle rampart]

yep...

[wispy canopy]

that's the thing i never did

(sysadmin by trade, git clone is what i do)

you'd have extract to a new folder

then from that folder, execute a command prompt window to use that git switch release_candidate command

then run update and run

[native halo]

my method is even worse,i just downloaded the obama leak from last year and i just been slowly updating from time to time

[wispy canopy]

don't mention downloading that here, it's not ok to talk about it here as stability.ai ai doesn't want to promote that

[native halo]

🙇‍♂️

its ok there i changed it

[wispy canopy]

👌

obama leak 😄

[brittle rampart]

i can use that way? clone my v 1.4.1.

[wispy canopy]

i think so, but you'd still have to switch

[brittle rampart]

omg i think it do all copies my models...

[wispy canopy]

yeah... that's another issue you'd have to fight with

symlinking model folder is the 'smartest way'

as you'd have to save and update only one location

the other installation will use the models from the other folder

[brittle rampart]

ok . try clean install v.1.0.

[brittle rampart]

omg it really best idea!

[brittle rampart]

hmmm Version: v1.5.2 now installing.. but it really use dev ?

[slate narwhal]

Anime Joker + Apterus smile

[magic wave]

Tomboy

[slate narwhal]

Raw version, more modern style. Classical SDXL teeth problems on a few

[wispy canopy]

did you run the command to switch branches?

[brittle rampart]

yes. before run update.bat

[slate narwhal]

Aesthetic changes

[wispy canopy]

perhaps update.bat switches it back? 😦 could be

you could always run the command afterwards

so after update.bat run the switch command

[magic wave]

Whats going on with the versions? :o

[brittle rampart]

but in dev branch 1.5.2 is current release or not?

hmmm...

[wispy canopy]

should be release_candidate

at least, that's the one i'm using

[brittle rampart]

oki-doki... in WebUi

[brittle rampart]

that not a dev

[brittle rampart]

try reinstall

[magic wave]

Are you trying to go into the dev version?

[brittle rampart]

i just try install dev version for try it.

[magic wave]

You should be able to just open up cmd in your folder and then do "git checkout --force v1.6.0-RC" and that should work fine

[brittle rampart]

unzip my v1.0.0 in new folder, then do that comand?

or?

[wispy canopy]

yeah, that would work too

[brittle rampart]

oki-doki!

[brittle rampart]

not run update.bat?

[magic wave]

Yeah, I mean if you wanna make a new folder you can clone the 1.5.2 version, then just use the git checkout command on that new folder

[magic wave]

No need

[brittle rampart]

just comand and run.bat?

[magic wave]

Yeap

[brittle rampart]

oki-doki! try it

[magic wave]

For me that worked just fine, and now I'm on the dev branch

With a bunch of new funky samplers

[slate narwhal]

Post apocalyptic evil blond Batman, kawaii style

[brittle rampart]

i am sorry. i must make a symlink on "model" folder or ..?

[wispy canopy]

ever made a symlink?

in windows the type you want is a junction

[brittle rampart]

nope.

[wispy canopy]

mklink /j models <path to actual model folder>

[brittle rampart]

oki

[wispy canopy]

you'll have to run in from an administrative window

[brittle rampart]

try. i can make a symlink on a extentions folder?

[wispy canopy]

yes, you can do the same

howeve,r if you update an extension in a installation, the other install will be affected too

so i'd be a bit more careful doing that

[brittle rampart]

$ checkout --force v1.6.0-RC
bash: checkout: command not found```

[wispy canopy]

git checkout 🙂

[wispy canopy]

are you in the right folder?

[wispy canopy]

bruce wayne lost his parents at childbirth?

lol

before childbirth

batman just magically spawns into the world with lost parents included

[brittle rampart]

yes. e:\3D_Projects\stable-diffusion-webui_dev folder where i am installing v 1.5.2

[slate narwhal]

lol, yeah something like that, and it gets more complex, it has the Darth Vader aesthetic in the prompt, it is supposed to have something from Vader

[wispy canopy]

lol, ultimate anime tragic self-imposed backstories 😄

literally born without parents "the force made me do it"

[slate narwhal]

rofl 🤣