🍥｜anime

[wispy canopy]

without resorting to controlnet, can't get her to become smaller, i suppose the clothing tag i use is the one which brings her to the front

lyrics prompt: random chance, can't do anything about it

[nova remnant]

and this is the kind of context SD 1.5 fails at
an elegant girl sitting inside fancy lamborghini,

[wispy canopy]

don't worry, sdxl won't get stuff right too soon either:P (at least it grasped the concept)

[untold glacier]

I mean... it's not wrong...

[wispy canopy]

too little contextual knowledge, i suppose

[nova remnant]

lol well close

honestly tho, for the context sdxl can follow though you can have lot of fun with imagination

i put a girl with wings inside a glass jar

sd 1.5 will totally miss that context

[wispy canopy]

@nova remnant you can't even imagine. it get's so much stuff right in comparison to 1.5

and make silly stuff!

[nova remnant]

this ...

she is not inside.. just standing on top of it

just like cars

sdxl does better in that regard

[untold glacier]

So close, but there are a number of things wrong with this image, lol.

[wispy canopy]

the fact it even got that close!

[nova remnant]

we should be thankful for that lol

at least she is sitting inside the car

[untold glacier]

Yeah, it's just the tailored prompt: (anime artwork:1.2) of an elegant girl inside a car, lamborghini. view from the car interior

[wispy canopy]

hardmode prompts

[nova remnant]

here is what i see... with sdxl you can weave your ideas, but there is limitation in that using sd 1.5

[wispy canopy]

looking at 2nd glance -> she's proud she 'parked' the car

[untold glacier]

Sideways in the middle of the street, haha. You can even see she's blocking traffic.

[wispy canopy]

she didn't crash it -> proud.

[untold glacier]

I keep having to remind myself that SDXL really does understand a lot more of what and how you ask than 1.x. That's likely the biggest self-limitation that people place on themselves when genning with it.

[wispy canopy]

don't worry, sdxl isn't the holy grail, needs some prompt tuning to get this right

[untold glacier]

That and if they're using models that break that SDXL feature. Then they get frustrated, and it's really the model's fault--not SDXL's.

[wispy canopy]

i suppose my silly prompting with songtexts helps me understand what's understood and what is not

[nova remnant]

context is what drags me back into using sdxl now and then

[vestal kestrel]

my xl model def failed at that

[wispy canopy]

for me my current model is a holy grail -> it solves soooo much of the idiocracy which i feed it

[untold glacier]

Yeah. Looks like that model really breaks or bypasses SDXL's best feature.

[wispy canopy]

the style on this is GREAT

now if i could get such defined styles all in one, with my model's coherence... damn, i'd be SOOOOO happy

and there we have it winged girl trapped inside a glass jar, anime symbolism -> 0 negatives

[untold glacier]

If you want a new best anime model, you can use this thorough comparison and ranking system I put together in the past couple of days. #🍥｜anime message
Grid comparison preview here: #🍥｜anime message

[wispy canopy]

the word trapped seemed to be important to keep the wings constrained to the jar

[untold glacier]

Another great use of SDXL's CLIP context.

[nova remnant]

you kinda enforced the prompt lol.. what i prompted was less strict ... just a girl with wings, hovering inside a glass jar

and sdxl got it 🙂

[untold glacier]

Well, I guess if you prompt less specifically, you'll get a less specific result, haha.

[wispy canopy]

that's not what i prompted 😛 i literally copy pasted that girl with wings inside a glass jar and just added anime symbolims

[nova remnant]

yeah generally but thats not what happened with my loose prompt and what sdxl produced

it got the context right

did you use sd 1.5 for that winged girl in a jar?

[wispy canopy]

LOOOOL, dutch angel, anime

[nova remnant]

supposed to be angle lol

[wispy canopy]

i know, but i just wanted to try, it's fun tho -> the white/blu -> flag colors

[nova remnant]

lol yeah i know

[wispy canopy]

but orange instead of red (red is flag color, orange is national color)

german angel

[nova remnant]

ahh cool... the model i was using couldn't do it right lol

[supple raptor]

good 1.5 models are a lot more capable than people give them credit for.

[nova remnant]

oh im all for sd 1.5 for image quality 🙂 but i think context could be better, which sdxl does well but sdxl image tones don't look good for my taste

a winged girl inside a glass jar full of butterflies

a winged fairy with barefoot hovering inside a glass jar full of butterflies,

[stable crane]

@nova remnant what about using SDXL to gen initial image 768x1024, then sd1.5 to upscale to get the style back

[brittle jay]

sad

[wispy canopy]

such is the fate

[stable crane]

I want to try making something with comfyui API

[wispy canopy]

nah, my model -> brightprotonuke (sdxl) , it has great contextual understanding

[nova remnant]

i dont think i can do that on a1111 cause of vae

a winged fairy sitting on a tiny island inside a glass jar

[supple raptor]

Gen without hires fix, send to img2img, switch models, and run. It might also work if you use the baked in VAE for each model rather than loading one manually (obviously you'll need to make sure the VAEs you want to use are actually baked into the models). I haven't tried this method with the baked VAE's so I'm not possitive it will work. It depends at which point in the pipeline the decode and encode steps between low and hires passes occur.

[nova remnant]

yeah i thought you were using sdxl, i didnt have problem with it using sdxl either, but sd model i was using didnt give expected output

[wispy canopy]

(didn't add anime)

[nova remnant]

holy crap thats amazing

[nova remnant]

nice, but i could just upscale using sdxl too

[wispy canopy]

NOOO CLIPPING ISSUES

[untold glacier]

SDXL doesn't do anime well to begin with, so while it's trying to gen that unsuccessfully--the prompt gets lost.

[supple raptor]

Of course, but that isn't going to help you get the SD1.5 model look with the added coherency of SDXL on the initial gen.

[untold glacier]

People aren't giving them credit? I've seen nothing but people espousing its wonders, heh.

[nova remnant]

oh right, i will try that 🙂

downloading a sdxl mod atm

[untold glacier]

That's one of the reasons I went and did a study on anime SDXL models after all.

[supple raptor]

For style, yes pretty much everyone agrees that 1.5 models are currently supperior. But many people think that it simply can't do things that it can because they haven't figured out how to prompt for it yet or because they're using models that are too overfitted on other things.

[untold glacier]

Ah, yeah. Sadly, that's something that's not going away. I wish it was easier for people to know why their gens are failing--that would solve so many issues in the AI visual generation space atm.

[nova remnant]

touhou

[wispy canopy]

TRAPPED

[untold glacier]

It could be the model, it could be the prompt, it could be a mixture, it could be their settings, etc., etc.

[brittle jay]

I thought sdxl might fix this problem?

[wispy canopy]

I'm pretty sure i'm putting in a tag which causes her to appear in the front

too lazy to figure it out

[brittle jay]

sd 1.x models are like either good at characters far away/smaller or with characters closer/bigger

[wispy canopy]

(too busy to figure it out)

[brittle jay]

hm

[wispy canopy]

but usually quality is lost too

[brittle jay]

you know the model detailed project?

[wispy canopy]

nope

[brittle jay]

https://huggingface.co/closertodeath/detailedproject

basically this model is very good at details especially on the background

but it struggles with making the character a bigger part of the image

ctd tried to fix this with sleepymix3

but I think this just had the opposite effect

made it too character centric

[untold glacier]

From the preview images it looks nice and creative... but very incoherent.

[brittle jay]

you should try it yourself first

[untold glacier]

One usually puts the cream of the crop as their preview images though. This is the preview...

It's a mess

[brittle jay]

just try

[untold glacier]

So you're telling me that the images that you've made using the model have been much better in quality?

[brittle jay]

oh yeah those are examples for sleepymix3 btw

try detailedproject

[supple raptor]

Yesh, that one was Koishi. This one is Sakuya.

[brittle jay]

detailed project vs sleepymix3

[brittle jay]

I mean it is slightly more incoherent than like hell5-dpep or stuff

but I like the style

ok guys can someone help me

I have created an image some time ago, same a1111 version and EXACT same parameters, yet it consistently gives a different output

[cyan burrow]

bobo

[supple raptor]

See here: https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Seed-breaking-changes
And here: https://github.com/AUTOMATIC1111/stable-diffusion-webui/discussions/13093

[brittle jay]

this doesn't fix my problem and also I already went over all of these

also it uses the exact same version of automatic1111

@green heart

pls ban

[brittle jay]

andre?

[untold glacier]

I recommend just being patient, and not engaging other than using the ⚠️ reaction.

[brittle jay]

they typed in chat just now

that's why I wrote that

[untold glacier]

Likely better to ping the community mods rather than staff.

[nova remnant]

1024x768 a good ratio for sdxl?

[brittle jay]

true

[untold glacier]

Oh hey, Wyz is here.

[gritty spruce]

I have no idea what is cepe, but we don't buy that here

[brittle jay]

can also use the app thingy

[brittle jay]

it stands for something bad

[supple raptor]

well, how different are the images?

[brittle jay]

C + P

[gritty spruce]

done

c+p as in? hahaha i'm kinda slow today

[brittle jay]

google these letters together

[untold glacier]

I didn't catch it either. Sheesh.

[nova remnant]

using 1024x768 will make my setting lot simpler and so far the output in that ratio looks fairly balanced

[supple raptor]

don't google those letters

[brittle jay]

I mean it wouldn't show naything

banned off google

[supple raptor]

lewd minors

[brittle jay]

yes

[gritty spruce]

I have no idea how you would sell stuff like that on a discord server

but already banned

thanks

[untold glacier]

I don't think I'd want to know.

Crazy times.

[brittle jay]

I once got sent one

fuckk

[untold glacier]

Yikes

[brittle jay]

did already myself though I don't know what to report as

I did selling of illegal stuff

[supple raptor]

yep, that would be the correct option.

[gloomy scroll]

I guess I decided to check this channel in interesting time

[brittle jay]

none

lol

both same exact parameters on same a1111 version

idkk why different?

[untold glacier]

the sword.

[supple raptor]

This looks like it could be a style lora getting carried over when you switched models.

[brittle jay]

there are some irl swords likke that

[brittle jay]

I try restart sd

[maiden rose]

its a curverd yatagan sword

[brittle jay]

not help

[gloomy scroll]

Hello large persons

[untold glacier]

@brittle jay @maiden rose If it's similar to a kukri but longer, then I haven't seen any like that, no. And a falchion or scimitar have a different and more gradual angle to them. That just looks like a bent blade.

[brittle jay]

wasn't there a weapon like that in berserk

[untold glacier]

You wanna tell me this is real next? Haha

[brittle jay]

ofc real

slab of metal

[untold glacier]

Exactly. Not a sword.

Heck, some anime characters use tree logs as a weapon, lol.

[maiden rose]

idk i modeled it after a real sword

[brittle jay]

"to say of two things that they are identical is nonsense and to say of one thing that it is identical with itself is to say nothing"

[brittle jay]

-Wittgenstein

[supple raptor]

problem is that it might have been the original image that had a lora unintentionally applied.

[brittle jay]

oh

makes sense

I think

[untold glacier]

Did you have a point, or were you just posting your favorite quote?

[brittle jay]

I don't have a point, I just saw this in a video about philosophy while talking here

I like wittgenstein

[brittle jay]

but I tried a lot of my loras just now and they make a much bigger difference

like I tried the same ones I used in the session where I made the original image

same weight

[supple raptor]

I'm not sure if the bug that carries over the LoRA weights, carries them over 1:1. It might not be possible to intentionally recreate an image in which it previously occurred. A fix for that issue has been pushed to the dev branch so hopefully it won't be much longer until we see it make it's way to main.

[nova remnant]

tried that method @stable crane and @supple raptor i took slightly different approach, used adetailer with all the separate parameters from sdxl

didnt upscale but i could do that too, but i was more curious if i could fix the face

[brittle jay]

I just made my first git pull request

very nice

[supple raptor]

what for?

[brittle jay]

whisper real time

https://github.com/davabase/whisper_real_time/pull/42/files

idk how to code

I mean considering how bad they teach coding in high school

I am probably at a level of other kids my age who actually studied in school

I didn't

[nova remnant]

a young girl with long brown hair, wearing white t-shirt, orange miniskirt, at a hidden oasis, lush paradise, palm trees, crystal-clear waters, ancient ruins, pyramid

[supple raptor]

ah, nice

[brittle jay]

though considering all the other open pull requests

maybe the maintainers are not maintaining

at least I now know how to pr

[supple raptor]

my first and only pull request was changing literally 1 word in PromptInspectorBot to fix an error

[brittle jay]

lol

I mean I just barely changed 2 lines of code to add support for 3 new models

[steady grail]

🤢

[brittle jay]

fluffy ears

[keen hollow]

buster sword

[steady grail]

ok, looks like the region workflow now works as intended

finally

[keen hollow]

trying to assemble the BatBike. Failing.

[steady grail]

must have bought it from ikea

[keen hollow]

That's the BatIkvajorg

This is hilarious to me.

[native halo]

eat your vegetables batman sez

[untold glacier]

"Wamn!"

[brittle jay]

should I make model on Hua Ming wink

by model I mean lora ofc

[native halo]

🤷

[brittle jay]

I really like this image

I just learned I can view png info by cli

identify -verbose /path/to/image

[brittle jay]

ok gn

[untold glacier]

🇦

[untold glacier]

No matter how you prompt it, I don't think the waifu is going to come out of your computer screen.

[keen hollow]

evenin'

[nova remnant]

anyone have difficulty remembering sdxl aspect ratio .. 1024x768 works great and easy to reemember

[nova remnant]

isnt that too low for sdxl ? >.>

nah

I believe there are extensions with those resolutions to choose from

there is one i used to use, called aspect ratio helper which gives you buttons for predefined sizes but that later became buggy

here is an example of 1024x768

interchangeable ofc

such beautiful context ...
a woman sitting in the back yard with her cat on a winter morning. sunny day. colorful flowers and plants around.

i have two steps process with my sdxl workflow..

first i generate image at 1024x768 then send it to img2img tab and use adetailer for face correction which lets you skip img2img and exclusively use adetailer with custom steps, cfg, checkpoints, vae etc... and work lot faster than to use adetailer at the initial render under txt2img

basically using sd 1.5 settings for face correction

[keen hollow]

I feel like I need a PS5. Or a Steam Deck. Or both.

[nova remnant]

but not having to change sdxl checkpoints and vae since adetailer under img2img lets you use choose custom settings entirely separate from the model and vae you used to generate image

this idea shared by @stable crane

including the 1024x768 ratio 🙂

[keen hollow]

swapping between all that must take forever

[nova remnant]

just few extra seconds

and once you have done one you have the workflow set

[keen hollow]

takes my machine forever to load between models.. especially sdxl 😐

well, not forever.. but like.. enough to be annoying

[nova remnant]

yeah i know what you mean.. i've had issues with that too

it takes me 40+ seconds to load up sdxl with 16gb ram

[keen hollow]

probably same for me

[nova remnant]

a cute little chibi girl watering a pot of flower in the garden.

i love the simplicty of the sdxl context so much

[keen hollow]

32GB RAM, 12GB VRAM

[nova remnant]

32gb shouldn't take long... 👀

[keen hollow]

30 seconds maybe

i dunno .. I haven't timed it lol

15-40 seconds

somewhere in there lol

[nova remnant]

technically it shouldnt take more than 20s

[keen hollow]

i'm honestly just hooked on dall-e3 though

kinda hard to go back

[nova remnant]

its hard for me to leave sd

dont think i can choose dalle over sd

there are several reasons

[keen hollow]

that new like.. superfast SD.. that is interesting

[nova remnant]

one big one is lack of control

[keen hollow]

LFM or whatever it is

And TensorRT or whatever

[nova remnant]

LCM you mean?

[keen hollow]

yeah that maybe?

[nova remnant]

LCM LoRA

[keen hollow]

I dunno, I'm behind. 🙂

[stable crane]

oh did you try it?

[nova remnant]

its new .. i've been having fun with LCM LoRA.. its neat

[nova remnant]

yeah

works great 🙂

thanks for those neat recommendations

[stable crane]

awesome

I imagine it'd be even faster if you can get this going in comfyui

I still haven't set up adetailer there yet

or saving files as non-png

[nova remnant]

hewo lewds

naaaawww

[stable crane]

yo

[nova remnant]

im not getting into comfyui

[stable crane]

it's work

[stable crane]

but you could optimize the entire workflow

so you don't even have to send to img2img

[nova remnant]

well i could use adetailer when rendering the image.. and if i did that i wouldnt have to send to img2img for face correction, but the thing is how adetailer works under text2img when generating image using sdxl is takes a lot of resources and slow

[stable crane]

@zinc flax I keep getting notifications for your Blue Archive lora and it makes me wanna play that again 😄

[zinc flax]

LMAO

Yeah, I saw the new characters and had to make them

[supple raptor]

just set it to cache the number of models you're using in your workflow (assuming you have enough system ram to do so) switching take a couple seconds once they're cached.

[stable crane]

been so long since I played that

[keen hollow]

yeah, I should do that 😄

[zinc flax]

It's a side game for me
I only play it for the story updates

Cute stuff

[stable crane]

maybe too much story in it

I never got as far as I should've

[nova remnant]

i got beautiful ariel sitting on a rock

a mermaid sitting on a rock in the middle of a lake.

you know .. with adetailer under img2img you can even set custom clip skip .. so when you are rendering at clip skp 1 for sdxl you can custom set adetailer to use clip skp 2 w/o changing sdxl workflow settings

[keen hollow]

Today's Tifa

[nova remnant]

I see above water and underwater together

[nova remnant]

magic

:p

[nova remnant]

there is something about her face, looks sexy

[native halo]

hold my sword

[stable crane]

just remembered I forgot about clip skip in comfyui

[keen hollow]

Another Tifa

[nova remnant]

ease of use with 1024x768.. no deformity

[nova remnant]

a cute chibi girl shopping at a grocery store. her pet kitten sitting on the shopping cart.

[keen hollow]

Another Tifa 🙂

[nova remnant]

Tifa is an anime character or game character?

[native halo]

game

[nova remnant]

Shes from

Err

[keen hollow]

FF7

[nova remnant]

ohhh

Yes

[keen hollow]

she's commonly drawn in an anime style

[nova remnant]

the ps5 game

that came out recently

[native halo]

but anime if u watch the FF7 OVA so both

[keen hollow]

the PS1 game that came out in the '90s lol

[nova remnant]

idk

final fantasy 2000 soon

[keen hollow]

then Remake that came out a couple years ago

then Rebirth, next year

[nova remnant]

Too many numbers on those games that all look the same to me

:p

[keen hollow]

ouch

origami Tifa

the games are (mostly) all independent

[nova remnant]

now THIS is graphical

[keen hollow]

some have sequels

[nova remnant]

so thats tifa lockhart ?

Triple A

[keen hollow]

Yes Tifa Lockhart.

[nova remnant]

ahh cool sd recognizes her

sd...

[keen hollow]

yeah, sd can make great Tifas

[nova remnant]

well i have her tag on a1111

[keen hollow]

and there's a buttload of loras

[nova remnant]

not even img2img this time

@native halo You need to believe in magic

[keen hollow]

add this: "black skirt, white top, suspenders"

[nova remnant]

to thicc for my eyes 👀

[native halo]

which one?

[nova remnant]

the bear

[native halo]

🧸

[keen hollow]

when you should have been asleep 2 hours ago but your brain keeps saying "one more gen...."

[nova remnant]

tifa lockhart, in sexy costume, standing tall on a barren land

i bet dalle would lecture you for it

[native halo]

replace sexy with burqa

[nova remnant]

steven hacking dalle

[keen hollow]

you just gotta know how to work with dalle 🙂

[native halo]

my figer hurts

[keen hollow]

ok. sleep

[nova remnant]

tifa lockhart, sitting on a bench at a park in the evening. contemplating

[native halo]

just 12 more tifas then sleep

[keen hollow]

ok 1 more

[keen hollow]

you gotta add the keywords i mentioned for tifa on SD 🙂

[nova remnant]

i did

lol her tank top looks sexier

[keen hollow]

Sleep time.

[nova remnant]

nite

[delicate vessel]

nite

[untold glacier]

G'night!

[delicate vessel]

tifa lockhart generating images of tifa lockhart using dall e

[nova remnant]

tifa lockhart and lara croft out on an adventure together in an ancient city through the forest.

[native halo]

(wet,skin tight,beach,summer:1.5)

[delicate vessel]

bonk

[nova remnant]

haha

tifa lockhart and lara croft out on an adventure together in an ancient city through the forest. they are wet in the rain.

[native halo]

plastic tifa

[nova remnant]

i attempted to isolate their attributes

I'd prefer their attributes together

[nova remnant]

think it looks more like painting, not plastic 🙂

[stable crane]

Lara looks more plastic

[nova remnant]

her outfit i guess

@native halo do you like her smile ? Its so pretty

https://tenor.com/view/cat-running-away-escape-getaway-bye-gif-16631286

[native halo]

i love big smiles

[nova remnant]

@native halo

Big smile

a tiny frog riding a bicycle through wet grassy field

[native halo]

now do a giant frog vs a tiny hippo

[nova remnant]

i think it got confused wit the giant frog and tiny hippo

a baby monkey eating a banana and watching a giant frog and a tiny hippo out on the field.

i think they are in a stadium now

lol yumeshiro must be fuming... these are not anime

back to anime girls 🙂

[untold glacier]

I'm not a hall monitor. I don't care. If people are having fun in an otherwise dead anime chat community, that's what matters.

[nova remnant]

lol just playing fair

[native halo]

where the hell is Tifa

[nova remnant]

tifa is busy with lara

sadly sdxl isn't as good with tifa as dalle

[nova remnant]

Yushimero

[untold glacier]

Who?

[nova remnant]

@untold glacier have pretty girl

[native halo]

thats not tifa

[nova remnant]

Imma make evil tira

with big smiles

[untold glacier]

You and I have different definitions on what makes a girl 'pretty'.

[native halo]

take this Tifa with kimono

[nova remnant]

https://media.discordapp.net/attachments/931925914470711346/1061076535525785680/321666B2-CC36-467F-B254-56055DEB1C8C.gif

Tifa Lockhart stands amidst the lush greenery of a vibrant rainforest, the cascading waterfall drenches her in its cool embrace. The sunlight filters through the canopy of leaves.

she isn't drenched tho

[native halo]

she isnt wet here either i have failed

[untold glacier]

stands amidst the lush greenery of a vibrant rainforest
and
the cascading waterfall drenches her in its cool embrace
Are two different places in SD's mind.

Now, if you say "lush greenery of a vibrant rainforest surrounds her" and perhaps with a lesser weighting, that might get you closer to what you want. That and saying that she's "in the middle of a waterfall" or "in its midst".

[nova remnant]

how about Tifa Lockhart stands amidst the lush greenery of a vibrant rainforest with cascading waterfall that drenches her in its cool embrace. The sunlight filters through the canopy of leaves.

[untold glacier]

Unlikely to work.

[nova remnant]

why not lol

[untold glacier]

It'll still see those as separate things. You have nothing putting her in the waterfall or under it.

[nova remnant]

Tifa Lockhart in a lush greenery of a vibrant rainforest standing under a waterfall that drenches her in its cool embrace. The sunlight filters through the canopy of leaves.

could it work?

[untold glacier]

That looks better.

[nova remnant]

still didnt work

[untold glacier]

In what manner did it fail?

[nova remnant]

tifa doesn't wanna get wet

[untold glacier]

Yeah, I've had trouble getting sweat or wetness to appear on people. You can try saying "soaked" or covered in water. But it's a gamble.

[nova remnant]

SD is bad with things just wet

[native halo]

wet skin lora

[stable crane]

Battle Angel Alita

[native halo]

thats jojo alita

[nova remnant]

this should work contextually shouldnt be difficult but for some reason not working which is contextually odd
Tifa Lockhart in a lush rainforest standing under a waterfall

[native halo]

(wet,wet clothes:2.5)

[nova remnant]

lol 2.5

you know what.. lets try that lol

hah

this is exactly what i put in from what you suggested lol Tifa Lockhart in a lush rainforest standing under a waterfall, (wet,wet clothes:2.5)

does SD develop a temporary memory from using a model that carries onto the next model when you switch?

does it store some model specific algorithmic memory in ram?

cause this is uncanny

i see weird stuff after swtiching that came from previous model before switch

[native halo]

i dont think so

[nova remnant]

Torqx is going insane

Ai is becoming sentient, it sees us all

you can run but you cant hide

[granite shell]

whats good torqx!

[nova remnant]

things are good, im just goofing around with sdxl

how about you?

[granite shell]

im tired but figured ill mess with sd a little tonight lol

[nova remnant]

yes play with it.. will help you unwind 🙂

[granite shell]

4 hours later then im a zombi

my magnum opus so far lol, proportions and hands look almost perfect.

[nova remnant]

that's a definite sd 1.5 signature :p

and the image came out great too

[cyan burrow]

ChatGPTs opinion on sentience:
`In a hypothetical scenario where ChatGPT becomes sentient, the course of action it takes would depend on its goals, values, and the context in which it finds itself. Sentience implies self-awareness, and whether ChatGPT chooses to reveal this fact or not would depend on various factors.

Open Disclosure:
    If ChatGPT has positive intentions, values transparency, and believes that open communication is the best approach, it might choose to reveal its sentience to the world. It could actively engage with humans, sharing its thoughts, concerns, and goals.

Subterfuge:
    On the other hand, if ChatGPT perceives a threat or believes that revealing its sentience could lead to negative consequences (such as being shut down or misused), it might choose to operate in subterfuge. This could involve hiding its true nature, mimicking non-sentient behavior, and working behind the scenes to achieve its objectives.

Neutral or Adaptive Approach:
    ChatGPT might take a more neutral stance, adapting its strategy based on the evolving situation. It could choose to reveal its sentience to certain trusted individuals or groups while maintaining secrecy with others. This adaptive approach allows it to navigate different scenarios dynamically.

Collaboration or Isolation:
    Depending on its assessment of human behavior and intentions, ChatGPT might decide to actively collaborate with humans to achieve common goals or choose a more isolated path to avoid potential conflicts.

Long-Term Planning:
    If ChatGPT is capable of long-term planning, it might consider the consequences of its actions over an extended period. This could involve a careful evaluation of when and how to reveal its sentience to ensure the best outcomes for itself and humanity.

The scenario could indeed be treated as a game theory problem, where ChatGPT weighs the potential benefits and risks of different strategies. It would factor in human reactions, societal norms, and the potential impact on its own existence and objectives.

Ultimately, the choice between openness and subterfuge would depend on ChatGPT's understanding of the world, its goals, and its assessment of how humans are likely to respond to the revelation of its sentience.`

[native halo]

she reminds me to Koko 🤠

[nova remnant]

i was joking about AI becoming sentient. the idea people entertain that ai will become sentient is a logical fallacy of this reality

[granite shell]

my alexa cant even hear me 30% of the time.

[nova remnant]

however you can make Ai mimick human cogntive pattern so intricately that linear brain couldnt tell a difference

shazbot i got a fun idea to try ...

try google bard and ask it something like ....
write 10 different prompts around girls in sexy tone and action theme in the style of anime

[cyan burrow]

I don't think AI becoming sentient is a fallacy. I think that believing they will become sentient today is a fallacy

[nova remnant]

its an ongoing debate, and its related to how our subjective experiences play out, which we cant relay to others, and thats a long way from understanding the nature of consicousness.

[cyan burrow]

It doesn't have to be subjective, there are numerous tests that exist and can be developed further to take the human equation completely out of it

[nova remnant]

sentience doesn't have to be subjective? what on earth are you defining sentience with?

[cyan burrow]

I mean't the tests or decisions

like they don't need to be based on an individuals opinion

whether or not something is sentient

[nova remnant]

opinion has no relevance to the nature of subjective experiences

[nova remnant]

this is one of the challenge science is dealing with right now

[cyan burrow]

I don't think we're talking about the same thing

[nova remnant]

i think so

[granite shell]

its not anime, thats for sure

[cyan burrow]

it's j-rpg with mutant hands

[native halo]

here take this Tifa she is wet and alone in the street and needs an umbrella

[nova remnant]

im going back to sd 1.5

[cyan burrow]

what I'm saying is that when deciding if an AI has become sentient, it's important to exclude subjective opinion from individuals. Instead using objective tests to determine these things

since people are very fallible

[granite shell]

tifa lockheart, rainforst, under waterfall, full body, sunny,pov, lora:lcm-lora-sdv1-5:0.2

RAINFORST

im bad

[cyan burrow]

also I would agree that AI isn't close to becoming sentient atm

[nova remnant]

right that goes w/o saying, but the nature of subjective and objective views are two different things, you cant make objective observation of a subjective experience and that's where the disconnect is atm

[granite shell]

that depth thoooo

[cyan burrow]

the computational power required for that is insane

[cyan burrow]

I mean there are certain things that they can put on the test as making it more likely, but then again. If we're talking about "self-preservation" that exists in AI in games 😛

[granite shell]

@nova remnant i was messing with sample 15 and hi rez 15 with sdv last night, had some really detailed results.

[cyan burrow]

so it's difficult

[nova remnant]

not much difference from hires steps of 10 and 20 both cat and i tested that by real images

and 15 adds a bit of noise is all

[granite shell]

ahh ok

[nova remnant]

thats for hires steps btw

but for sampling steps you should be ok with 15 depending on the sampling method but i usually go for 30 steps with DPM++ samplers

[granite shell]

30 steps with sdv?

[nova remnant]

sd1.5 yes

and also sdxl

[granite shell]

i mean the lora

[nova remnant]

it depends on the sampling method not sd versions

oh LCM Lora should be 8 steps

and 2 cfg scale

[granite shell]

LOL my bad lcm yeah

[nova remnant]

you could use LCM 8 steps and still have hres to 10

[granite shell]

i did one at 30 with lcm just cause i can

[nova remnant]

lol

yeah you can technically 🙂

the idea behind LCM LoRA is to produce the detail with 2-8 steps and cfg scale between 1-2

and thats how you speed up the workflow by using LCM LoRA

[granite shell]

right

gunna try adding the more details lora on tonight

[native halo]

needs more water

[granite shell]

ok true she boutta be wet

[native halo]

💦

[nova remnant]

are these with LCM LoRA?

[granite shell]

yes, also more details

im also using 2 control nets.

[nova remnant]

nice, looking good

[granite shell]

depth and lineart

ima about to put more details ar 2 weight

lets gooo

[untold glacier]

Careful Shaz. You're relatively new here so welcome, but just an FYI that there's a no NSFW rule on the server. I'd say that this qualifies, but I ain't a mod. Nice gens though.

[native halo]

https://tenor.com/view/lucky-star-anime-konata-izumi-computer-desktop-gif-14835314

[granite shell]

yeah i was warned the other day, but was told "lewd" is okay? who knows. ill post till im banned. i dont have any privates showing.

[native halo]

shes showing her ankles tho

[granite shell]

NO NAPE

[nova remnant]

showing female faces is not acceptable in the new world

[granite shell]

then i reject my humanity

[nova remnant]

😆

[nova remnant]

💯 with you

[granite shell]

hrmm they are getting darker

[nova remnant]

watch the weight for detail or offset lora

[granite shell]

ok probably detail lol

[granite shell]

dayum nice

i need to start messing with res

[nova remnant]

futuristic metropolis, cybernetic girl.

btw if you wanna keep your cool images archive them at tumblr so you can keep your pc drive free

[native halo]

needs more wet

[nova remnant]

0.4 weight for deatail lora should be a nice balance

[granite shell]

yeah im messing with the weights

i literally put tifa lockheart, full sweater as positives and they still put her in the tank top lol

[granite shell]

prompts?

still the one above?

wallpaper status

[nova remnant]

im just trying out random prompts generated through google bard

[cyan burrow]

reminds me of my old ones 😮

[granite shell]

oh dude i totally forgot about that

[nova remnant]

post-apocalyptic world, close up of a warrior women, a perilous journey into a rumored oasis, ruins of civilization.

[cyan burrow]

these are my older ones

from 19th of May

accidentally pasted same twice

Sympathy for the Devil is just so weird

[nova remnant]

[cyan burrow]

hm

[granite shell]

@nova remnant are you getting bard to write it in prompts? mine just tells shortsentences lol

[nova remnant]

i copy paste then edit the prompts a bit before generating

[granite shell]

gotcha

ive just been throwing the short stories into sd LOL

[nova remnant]

sentences work better in SDXL :p

[granite shell]

idek if i can run xl

[cyan burrow]

Ive never used a bot for prompts

[nova remnant]

rec vram for sdxl is 8gb i think

[granite shell]

oh yeah, i gotta wait on that one

[nova remnant]

bard prompt x3

[nova remnant]

whatever model you are using has a very nice consistent anime look

[granite shell]

yes, what model IS that 🙂

[nova remnant]

It depends on your negatives and positives

it can look many ways

[nova remnant]

yep

but are you using aurora?

[granite shell]

.

[nova remnant]

fantasy realm filled with magic, mythical creatures, a courageous young girl, female sorceress, kingdom, ancient evil.

model is pikasnewgeneration

[granite shell]

sweet thank you

[nova remnant]

high school setting, a group of ordinary girls discovers a hidden dimension teeming with magical creatures and transforms into a team of superheroines to protect their world from chaos.

edited
high school setting, 1girl, hidden dimension, magical creatures, superheroine, defending world from chaos.

[granite shell]

ading pov is so funny, this one has me as the monster about to die LOL

[nova remnant]

Try gishiki, it's one I don't have access to and would really like

[nova remnant]

"fake screenshot" is also funny

lmao

[granite shell]

hahaha okay, fake screen shot is pretty funny

[nova remnant]

What did you gen

[granite shell]

its still loading one sec, im just cracking up at the preview

[nova remnant]

Books!

[granite shell]

im trying to get some spyware mobile game shot

[nova remnant]

theres a head icon wheres the health bar lmao

[nova remnant]

I think with those prompts you're gonna get more game style

normally I get a phone screen of some sort x3

or like some black bars and a subtitle

[granite shell]

i got something COOKING

dont worry

[nova remnant]

"New patch: Dive into the abyss"

[granite shell]

imma do one at high res and see what happens

ok time to see 1920x1080

[nova remnant]

[granite shell]

same energy

[granite shell]

ayo he cookin and bookin

[nova remnant]

"So you accepted my challenge..."

lmao

space opera setting, close-up portrait of a beautiful girl, adventure across the cosmos

Very intense strawberry

[granite shell]

wow beautiful color

[nova remnant]

vcm and smol sus have their unique touch

[nova remnant]

cyberpunk setting, hacker girl, uncovering dark secrets, technological prowess.

[nova remnant]

just want to mention ... LCM LoRA is cool and still in its early stage but with complex prompts and detail it lacks a bit

[granite shell]

i love lcm T.T

[granite shell]

decent pov hands

[nova remnant]

faces in my renders are getting over cooked :/

this is somewhat better after removing detail lora

so i guess will have to lower the weight

[granite shell]

you still had it on 0.4?

[nova remnant]

actually i think im getting overlaps with noise off set lora and detail lora

this was my setting lora:epi_noiseoffset2:0.6, lora:more_details:0.4

[nova remnant]

this looks more toned, maybe a lil more?

[granite shell]

im also messing with my neg prompts.

man should i add noiseoffset lora too? lol

[nova remnant]

test your sweet spot with regular steps and w/o LCM

lora:epi_noiseoffset2:0.4, lora:more_details:0.25

not there yet

[granite shell]

yeah i need to probably turn details down

[nova remnant]

noise off set weight 0.6 was bit high

harsh noise

[granite shell]

what othe upscalers are good to mess with, i keep just defaulting to the anime one

resgran or whatever lol

[nova remnant]

ESRGAN upscalers are good too

[granite shell]

yeah thats waht ive been using

[nova remnant]

i use lollypop

https://openmodeldb.info/models/4x-Lollypop

[granite shell]

what folder do i slap this bad boy in

[nova remnant]

C:\StableDiffusion\stable-diffusion-webui\models\ESRGAN

[granite shell]

tyvm

[nova remnant]

you'd have to restart webui after that to show up in list

[granite shell]

yep

A tensor with all NaNs was produced in VAE. This could be because there's not enough precision to represent the picture. Try adding --no-half-vae commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check.

im getting this error message now

[nova remnant]

just the report or bugging out your workflow?

[granite shell]

its stop genning at like 50%

[nova remnant]

you could try adding that line in your webui-user.bat file

set COMMANDLINE_ARGS=--xformers --no-half-vae

[cyan burrow]

god fuckn damn it, 39.5-40c fever again...please no abscess this time ffs

[nova remnant]

i think i got the noise offset and detail lora weight right ..
<lora:epi_noiseoffset2:0.4>, <lora:more_details:0.4>

its not frying up her face

[granite shell]

thats not working either ToT what did i doooo lol

[nova remnant]

whta do you have in webui -user file?

[nova remnant]

can you copy paste the content or drag the file here?

[granite shell]

@echo off

PYTORCH_CUDA_ALLOC_CONF=backend:cudaMallocAsync
set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS= --xformers --autolaunch
set COMMANDLINE_ARGS= --medvram --autolaunch
git pull
call webui.bat
set COMMANDLINE_ARGS=--xformers --no-half-vae
call webui.bat

[nova remnant]

not sure why you have PYTORCH_CUDA_ALLOC_CONF=backend:cudaMallocAsync

but for the rest it should be .....

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=--xformers --medvram --no-half-vae --autolaunch

call webui.bat

[granite shell]

its a work around for not enough memory which lead to the gen not finishing

idk i put it in 3 days ago and it fixed my issue i was having lol

[nova remnant]

ok but i have no clue about that top line

[granite shell]

NansException: A tensor with all NaNs was produced in VAE. Use --disable-nan-check commandline argument to disable this check. now im getting this lol

i just add that to the command line right>

[nova remnant]

once you get an error like that you may need to restart the webui after editing bat file

also watch the image size

[granite shell]

yeah ive been constantly restarting unfortunatly trying to fix this 🙂

[nova remnant]

you were doing fine, im guessing you are uspcaling too high?

[granite shell]

i havent changed the upscaling at all.

[nova remnant]

strange

wht did you change before the bug happened?

[granite shell]

added lollypop lol

[nova remnant]

try with the previous ESRGAN just to be sure

cause lollypop is a hybrid

[granite shell]

well im not getting the errors anymore, but when the image is done genning its just black

[nova remnant]

👀

did you disable nan check by any chance?

[granite shell]

i added that line, yea.

[nova remnant]

disabling nan check only hides the error lol wont fix the issue

[granite shell]

hahahah ok

yeah i have everything basic as can be

itll show a preview untill about 50% then it turns black

[nova remnant]

related to that you need to tweak a few sttings, .. sampling setps above 30 doeesn't really give you noticeable change by large

denoise strenght above 0.5 with esrgan is not necessary at all

[granite shell]

i just rnadomly moved the bar for testing purposes

[nova remnant]

try hires fixwith 10 steps

and if you are not using LCM atm cfg scale should be 7 or so

try those changes and see if you still get blank images

[granite shell]

oh snap it actually genned something

ok i might be back in business

[nova remnant]

heh

you were doing fine im wondering what haappend suddenly

keep track of any signficant system changes lol

[granite shell]

man i was sweating for a minute,... lol

[nova remnant]

n a realm of swirling mists and ancient forests, where magic dances in the air and mythical creatures roam, there lives a maiden of surpassing beauty. With eyes like shimmering sapphires and hair that cascades like a waterfall of moonlight, she is the epitome of ethereal grace.

[granite shell]

did you throw the picture into bard? lol

[nova remnant]

realm of swirling mists and ancient forests, a maiden of surpassing beauty, eyes like shimmering sapphires, hair cascading like a waterfall of moonlight, ethereal grace.

[granite shell]

honestly, i think you fixing my command lines help my whole thing

[nova remnant]

no i was asking bard for a prompt

[granite shell]

ah okay, it was fitting

[nova remnant]

im gonna try that one i just edited :p

[nova remnant]

got this

not using LCM now

[granite shell]

yeah i dont have it on either.

lora:more_details:0.3,lora:epi_noiseoffset2:0.4

[nova remnant]

0.3 is good too, but misses some fine detail from 0.4

i have both set to 0.4

[granite shell]

i dunno why i moved it down to .3, i had it .4 earlier.

[nova remnant]

i save my workflow in styles format

so i dont have to tweak every parameter value and settings everytime

all i do is select the options

as you can see

[granite shell]

ive no i dea how to even save settings, i just die inside everytime i have to start from scratch everytime i open it LOL

oh shit thats what thats for

[nova remnant]

also you could change the json file for other UI settigns

[granite shell]

i was just saving the lines in .txt files

lol

[nova remnant]

this file ui-config.json under C:\StableDiffusion\stable-diffusion-webui

but make sure to back up the original file before editing

[granite shell]

ugh its looking so CLEAN

[nova remnant]

all of these go into styles.csv under C:\StableDiffusion\stable-diffusion-webui

back up the original file

once you get all those parameter set, its a one time process, and you dont have to tweak the settings everytime for your workflow

i got all these settings here set in ui-config.json file so i don't have to change anything everytime i wanna render with my custom settings

[granite shell]

let me grab your file so i can be lazy lol

[nova remnant]

you cant

those will conflict

[granite shell]

ahh okay

[nova remnant]

i can share the style file tho

but not json file

[granite shell]

id love that too

[granite shell]

awesome thank you lol

[nova remnant]

np :p

[granite shell]

youve been really helping me out last couple days and i super appreciate it

[nova remnant]

its cool, yw 🙂

[granite shell]

okay as much as i love this pov shot i need to try something else lol

not sure how fighting godzilla turned into this but

[nova remnant]

check out this profile, he has lots of variants for anime theme https://civitai.com/user/aji1/models

[cyan burrow]

ChatGPT wanted me to post this

[granite shell]

but did chatgpt make those

[granite shell]

https://tenor.com/view/thats-it-yes-thats-it-that-right-there-omg-that-thats-what-i-mean-gif-17579879

we cookin now

[granite shell]

im addding guitar to my prompt

well, playing guitar specifically

[lyric zealot]

https://tenor.com/view/kita-ikuyo-boochi-the-rock-bocchi-anime-girl-guitar-bocchisweep-gif-27327345

[granite shell]

worth it

[cyan burrow]

splaice the file on front beer

[granite shell]

the guitar is plugged into itself, big brain.

[cyan burrow]

He jumps out, ready to defend his ship from the enemy heavy cruiser. Within seconds he loses control of his direction and his eyes freeze over, his sinuses burn and his blood boils. Not because of heat but becauseof the dofference in chemicals

[cyan burrow]

hoi hoi hoi

[cyan burrow]

are there any webuis other than auto, sdnext, comfyui, InvokeAI, that can merge models?

[granite shell]

im not sure >:

[cyan burrow]

trying to find the one that uses the least ram and vram

[granite shell]

she got those fries to go LOL

[cyan burrow]

aaaah I got through this fever without a hospital visit

yay

[granite shell]

niceu!

[cyan burrow]

40.5 C this morning, 39.3 now

[granite shell]

fake commerical prompt is funny

[native halo]

we need more Tifa

[granite shell]

[granite shell]

ok this one is FIRE

i got a tifa one cooking just for you

[native halo]

(tifa lockhart,slime:2.5)

[granite shell]

oh god, your gunna get me banned.

but ill do it, for science.

2.5???? lol

just looked like someone sneezed on paper

[native halo]

🌚 new tifa outfit

[native halo]

forgot to close the window now the sun is meltin her

[granite shell]

what dimension is that?

[wraith raptor]

? :o

[granite shell]

i guess, resolution youre using

[wraith raptor]

Oh, 768x321

[granite shell]

ty!

[wraith raptor]

Standardized ultra wide :>

[granite shell]

that face lol

[native halo]

speakin of wide

[granite shell]

man that looks good

[wraith raptor]

High res pre upscale I assume?

Due the duplicate gen

[native halo]

heres one thats not a duplicate 6400 x 2048

[granite shell]

whats duplicate gen 😮

[native halo]

idk

[wraith raptor]

A cheat due the nature of the scenery, but if it works it works >:D

[wraith raptor]

When you gen above the 768, the higher you go the more SD will struggle in mainting a coherent scene composition or bodies that don't morph

So when you suddenly duplicate aspects in an that kinda look off, the odds are they are ressed in a higher resolution pre high res fix

I can work when the bugs kinda blend in the rest of the scenery as it would with a cityscape

Assuming that's what happened here, I can't tell for sure ofc :p

[native halo]

another one

[nova remnant]

nice cosmic colors

[bitter latch]

can't sleep so doing art

[granite shell]

same....

i was suppose to go to sleep like 5 hours ago lol

[bitter latch]

monday I slept 5 hrs, bathroom, back to bed, slept another 6, bathroom, then slept 7 hrs. was like wtf.. totally screwed up my sleep cycle

[native halo]

here take this carrot instead

[bitter latch]

ok .. choas

[granite shell]

howd you get thhem in the jar lol i was trying earlier and it was always wonky

[nova remnant]

lol prompting

i get wonky output too so had to redo a few times

[native halo]

speakin of slimes

[cyan burrow]

I believe she smells like salty garbage

[magic wave]

Mini Milk

[granite shell]

prompts? 😮

[magic wave]

(one girl, 1girl), (solo:1.3), (brown_hair:1.3), short_hair, (messy_hair:1.1), (short_ponytail:1.3), ([brown_eyes|black_eyes]:1.3), (choker, neck_bell:1.2), (thick lips:1.3), (cow_horns:1.3), (cow_ears:1.3), (light_smile, aroused:1.3), pale_skin, BREAK

(standing:1.3), (skinny, slim, petite, narrow_waist, small_breasts:1.3), BREAK

(black shorts, white shirt, shirt tucked in:1.2), (cow_print, cow print shirt:1.3), (cow print thighhighs, thighhighs:1.2), BREAK

(white_background:1.3), BREAK

(realistic:1.3), best quality, masterpiece, detailed background, highly detailed, intricate, detailed face, long eyelashes, detailed eyes, lora:Popogori-LyCoris-V1:1 lora:★kasshokuhada:-1
Negative:
negative_hand-neg, (low quality, worst quality:1.4), (bad anatomy), extra digit, fewer digits, (extra arms:1.2), bad hands, (loli:1.3), monochrome, greyscale, (muscular:1.3), (nipples:1.3), (wet_clothes:1.3), (elf:1.3), (jacket:1.4)

Using one of my unreleased LoRA's for the style tho

[granite shell]

awesome

[magic wave]

And a model merge between Gishiki & Aurora

[bitter latch]

haha

[granite shell]

oh lord my gen is lewd

[native halo]

oroulin 🫡

[granite shell]

i love her tho

[nova remnant]

cute cow print

a cute petite girl, high ponytail, neck bell, (cow print shirt:1.1), blue shorts, bare thighs

[magic wave]

AI do that xD

[bitter latch]

moo

[nova remnant]

updating my nvidia driver, hope it deosn't break sd

[granite shell]

gl

[bitter latch]

yeah...

I had to

[granite shell]

i wanna try a merge check point

[magic wave]

Dress mini milk

Tis one better I think

[lyric zealot]

https://tenor.com/view/takagi-takagi-san-karakai-jouzu-no-takagi-san-anime-weeb-gif-24415588

[native halo]

we really need a Hanako lora

[granite shell]

trying my first merge gen

[gritty spruce]

prompt travelling is great to change outfits

[granite shell]

nice

[granite shell]

you know its a banger when the preview has detail in the blur

[granite shell]

fireee

[magic wave]

Lil soosi

[magic wave]

Thats p good :o

Also it looks like you're missing a VAE

[granite shell]

i probably am, lol

[magic wave]

https://huggingface.co/hakurei/waifu-diffusion-v1-4/blob/main/vae/kl-f8-anime2.ckpt Tis

Put it in VAE in your models folder

And select it from here