🍥｜anime

[bitter latch]

found I have to put FULL body armor outfit or well .. umm yeah I can't post those so i'll post this

blinks

[wispy canopy]

have an inspirational quote!

[green temple]

cool

[green temple]

the bad version

[green temple]

finally

oh the ears... nooo

[nova remnant]

???

😭

Looool

[nova remnant]

no sad, here goth

[nova remnant]

:D

[steady grail]

i like the color and style of this model, what's that?

[vital raptor]

autismmix_confetti

[nova remnant]

Its the color no one likes

xD

[vital raptor]

it's all about the artist tags you use, no one bothers to actually test anything

they just get the model, type in a random prompt and they don't like it because the defaults are garbage

[nova remnant]

Autism kinda of always have a yellow tint to it

[vital raptor]

🤨

[nova remnant]

@vital raptor this prompt is cute

@green temple smol

[terse ore]

Any recommendations for upscale workflow? I'd be using it for anime backgrounds of mainly buildings. Primary concern is artifacts where they are many small straight lines ie window sills, brick walls, etc

[lofty stump]

Hello everyone

[vital raptor]

ugh, my google searches now show AI responses at the top of the page and I've been searching a lot of technical stuff dealing with enterprise level server hardware stuff and the AI is so horribly wrong sometimes 😄

[pure oyster]

Drip

[unreal warren]

what is the style of this art? artist?

[nova remnant]

It's SD3

[green temple]

cool

[green temple]

nice style , is it a lora ?

[steady grail]

ai is never wrong.
it's you who is a plotter

[brittle rampart]

crap! how hard it is to use SDXL. it's something indescribable.

[nova remnant]

[steady grail]

xl, expecially pony is very good. but you can use much less cool stuff and genning everything is so slower >.<

[brittle rampart]

In general, some pretty terrible things happened to me with the pony model.

Pony

[steady grail]

yep terrible things

[brittle rampart]

Well, what kind of artifacts are these!? and already on model 1.5!!!

[steady grail]

the butterflies and birds?

oh, the hair

yeah, i saw them.... maybe in a bad cn?

[brittle rampart]

and it turned out to be VAE

[steady grail]

oh

what was the vae? blessed?

orangemix?

[brittle rampart]

in the first case it was the good old Anything-V3.0

and in the second it is the most common CleanVAE

[steady grail]

oh, anything... yeah, now i remember

usually i just use 840 or 560

[brittle rampart]

in my opinion they are quite contrasting

[steady grail]

in general? or the one against the other?

[brittle rampart]

in general

[steady grail]

what's your preference?

it's so hard to find good vae

[brittle rampart]

Anything vae I like low saturation. I found this a long time ago and don’t even remember where. seem on Huggaface

[steady grail]

i have it, but it always induces in me rage lol

[green temple]

couldn`t get what i want but this is fine

[brittle rampart]

Why?

[steady grail]

weird interactions with controlnets in the past

i got mixed results

[brittle rampart]

Well, I don’t understand that deeply.

[steady grail]

and i absolutely hate any model relatively connected with anything.
i'm traumatized by all the weird mutations XD

[brittle rampart]

then you need to use your personal model created for and for yourself.

[steady grail]

there are so many good and stable models out there

at best, you can add some finetuning

[brittle rampart]

For example?

[steady grail]

well, i usually use soushiki when genning fantastic landscapes

[green temple]

terrible ?? where ??

[brittle rampart]

it looks like a mosaic!

[steady grail]

or somethbing made by Zovya for the toony style

too bad i lost many models in the last week 😦

i confused a directory for a connection >.<

they where all so cute and stable models

[brittle rampart]

sad to hear this

[bitter latch]

a land of horror

sorta

[steady grail]

cute, but the samey color makes the conposition comfusing

[bitter latch]

adds color

or

hehe

[brittle rampart]

this is the land of horrors

[bitter latch]

meh

[brittle rampart]

and this is the Land of Kawaiiness

[green temple]

nice !

[brittle rampart]

it turned out terribly beautiful....

😁

[bitter latch]

it went nuts

[steady grail]

cute

[bitter latch]

I'm on my wayyyyy

[brittle rampart]

you can fly!

[bitter latch]

suppose to be jumping

and suppose to be a boy but shrugs

oh I see why.. doh

[green temple]

hmmm....

[green temple]

what do you think of this style ?

[brittle rampart]

about her clothing style? simple and tasteful! 😁

[green temple]

no i mean the drawing line

[brittle rampart]

clear lines, moderately bright colors. good lighting. although the background is a bit rustic. but if that's the idea then fine

[green temple]

yeah i was focusing on the same style

[brittle rampart]

then you succeeded!

[steady grail]

very tasteful

[tall solstice]

gm

[nova remnant]

elf

@green temple look a fish

[steady grail]

a fish!

[green temple]

nooo

[supple raptor]

Yeah it was this one. https://civitai.com/models/137160?modelVersionId=151344
I had to turn up the strength on it to 1.2-1.3 to get it to work with a full prompt on kohakuxl-epsilon. A shorter prompt might have been easier to get to work.

[green temple]

thx 👍

[nova remnant]

Smol wearing black

[steady grail]

guys, overweighting all the quality tags and downweighting all the other tags, what results do you get?

[steady grail]

cute

[jade violet]

🔥🔥 what model are you using?

Animagine?

[bitter latch]

a merge I think

would have to check

[nova remnant]

other than pony models, I like to do (highest quality, best quality:1.4)

on many models

and same for negatives weighted too, they make a good difference depending on the model I like it most of the time

but no need to down the other tags ( at least not on a1111)

[steady grail]

i noticed that a model i'm using, tend to overpower the character prompt with the background prompt

by overweighting all the character tokens and downweighing all the bg tokens, but leaving the quality tokens untouched

i noticed very interesting results

the bg tokens are 0.5, the characters tokens at 1.3 and all the quality tokens at 2.0

by doing so, the bg go bonkers in an interesting way

[winged oasis]

I think that was deepblue originally

[green temple]

thx

[nova remnant]

pats x1000

[green temple]

indeed

[nova remnant]

@green temple :D

x3

[green temple]

noice !

[keen hollow]

did we ever get controlnet stuff like posing with xl models?

[nova remnant]

yes

[terse ore]

Any way to make this workflow better (other than settings)? Pretty happy with this, just wondering if I'm missing out on anything else.

[nova remnant]

Where is the grabby

She got the tomato hair

I'll grab that

[zinc flax]

don't yank pls

[vital raptor]

just bought a new server for my home lab.. debating getting a Nvidia Tesla M40 (24gb vram) for it, could be fun to play with

then again, i should wait until i get the server and see if i can cram my 3080ti or 4090 in it before I spend more money 😄

[coarse brook]

Beep boop. SO i know this isn't the Tech Support page, but seeing as Anime can center heavily around the hair styles, any recommendations for something for Auto1111 UI that can help me make sure, when a scene contains multiple hair colors, that each gets assigned correctly?

[vital raptor]

you mean like making sure the hair is colored in the same way each time? It's kind of rough sometimes

[coarse brook]

Well more for a scene with multiple people in it, like a group photo

[vital raptor]

so something more like this

[coarse brook]

Yeah

Something like that ish

[vital raptor]

regional prompter

[coarse brook]

https://github.com/hako-mikan/sd-webui-regional-prompter like that?

[vital raptor]

i defined the left side with a prompt, and the right side with a different prompt

yeah

[coarse brook]

MAy i drag you to the #🤝｜tech-support in assistance on how to install it? 😄

[vital raptor]

you can have multiple divisions and do specific prompts for each one

it's pretty simple, just go to the extensions tab in a1111, click "available" then click "load from:" and find it in the list and click install

then go to the 'installed' section and click "apply and restart"

[coarse brook]

Wait... It was that simply..?

Simple*

[vital raptor]

for installing extesions, yeah

there are a few that can be complicated, but usually it's that simple

[coarse brook]

And then i read the guide, and follow what it says, and then if i don't understand any of that, ping u in Tech support? xD

[vital raptor]

pretty much 😄

[coarse brook]

Oh wait, i think i c.

So

Char 1, pink hair, long hair BREAK
char 2, brown hair, short hair BREAK
char 3, orange hair, long hair, BREAK.

That guarantees 3 different hair styles, and they won't conflict on eachother?

(Gosh darn sorry for hijacking this for Tech support 😦 But Ya'll prolly more knowledgeable on this specific topic)

[vital raptor]

for the prompt, yeah that's basically it

you have to adjust the divide ratio in the settings to split the image into areas, each part of the prompt will go in a different section

[untold glacier]

Any activity in the channel is better than no activity. Don't worry about it.

[untold glacier]

Besides, as you say, many people in chat (at least back when I frequented here more often) are knowledgeable indeed about SD and extensions.

Weebs often make the best tech support

[coarse brook]

Yarp

If i ever need answers about a creative topic, weebs are deffo go-to

These coding looking terms are hard for me to understand so I'll prolly be spending a few hrs on learning just 1 part xD I might return

[vital raptor]

the important part is the divide ratio, you basically start with 1,1 for 2 columns, make it 1,1,1 for 3 columns, etc

using the base/common prompt options can get a bit more complicated and you can get really specific with ratios

but keeping it simple to figure it out really helps

[coarse brook]

So what would i do to obtain something like this?

(Paint ftw)

[vital raptor]

1,1,1

[coarse brook]

• Forgor full image.

[vital raptor]

or 0.3,0.3,0.3

[coarse brook]

1,1,1 would give 3 collums, ye, but surely to make sure i get 3 different hair colors, I would need 3 cubes, only on the hair part?

[vital raptor]

i wouldn't try to limit it to just the 'hair/head' area

[coarse brook]

So this img is too complicated still? 😄

(From the guide

)

[vital raptor]

you can, but the more you restrict the area the harder time it has figuring it out

[coarse brook]

So 3 collums, and then fine tune the ratioes, is the way to go?

[vital raptor]

yeah, if you try to split top from bottom you have 6 areas and you have to put a prompt in each one

[coarse brook]

Fair. And i assume that
"Char 1, pink hair, long hair BREAK
char 2, brown hair, short hair BREAK
char 3, orange hair, long hair, BREAK."

Goes Top one is collum 0,
2nd is collum 1, etc? (Edited)

[vital raptor]

but even if you wanted to define the clothing also, it's still in the same column, so it's really not needed unless you need something specific

[coarse brook]

ffs. Too tired to type. fat fingering enter

Edited my fat finger msg to actually include the question

[vital raptor]

3girls, choker, cat ears, smile, maid outfit, standing, BREAK pink hair, short hair, BREAK green hair, long hair, BREAK blonde hair, twintails,

using 0.3,0.3,0.3 for ratio

[coarse brook]

Right, but what determines which goes into what? Cause surely collom 0 (The blond), should be the first after the initial break?

[vital raptor]

depending on the model, some things will try to take over (like twintails)

tbh, it's not very precise sometimes

[coarse brook]

TL:DR Keep genning until you get the result you seek?

[vital raptor]

pretty much

[coarse brook]

Gotcha

[vital raptor]

there's probably ways to get it more specific (that i don't know), but for the most part it's just trial and error

[coarse brook]

I mean. depending on your settings, it takes all from 10 sec > 3 min.. surely... right= I Mean, thats not too bad

[vital raptor]

also, this was done with a base prompt (the first line) enabled

😏

[coarse brook]

Oh, so with base promt enabled, you do all of your promts, then new line, then BREAK, and then add details for each collom, seperated by new BREAK line?

[coarse brook]

1080 ti \o/ I cri in ancient tech

[vital raptor]

yeah, base prompt, break, first col, break, second col, break, etc

[coarse brook]

I c

Do you ever re-visit an image, but have completely forgotten what promts you used for it, and wish to re-create them? Recommendations for an extension?

[vital raptor]

that's built-in

PNG Info tab

just drag in an image

[coarse brook]

I tip my hat

[vital raptor]

should also work with mine, i leave all the data in there

[coarse brook]

Any model recommendations?

[vital raptor]

for sdxl, autismix_confetti for anime stuff

for sd 1.5 i havent used any for awhile

[coarse brook]

Honestly only really been using PonyXL+SDXL merged

[coarse brook]

This base ratio... What does that do exactly?

If only the damn webui wouldn't crash so god damn much xD

(Usually after rapid successions of generating)

[vital raptor]

it's for the base prompt, basically 0.2 is 20% base prompt, 80% everything else

[coarse brook]

Ah, so a ratio of 1, is usually the go-to if you've important a previously genned image, and wish to maintain that composition?

[vital raptor]

haven't tried that, i usually leave it at 0.2

[coarse brook]

It does seem to work like that for me atm. As the 0.2, pretty much cut the image into 3 parts, and threw out 2.

[ivory pebble]

Just a suggestion but I've heard that ForgeUI's version of regional prompter is better and easier to use compared to A1111's version. Apparently it can also apply lora's to different regions

[coarse brook]

Well fren. Wheres the link? 😄

Can't just recommend something, and not give a link

[ivory pebble]

However I haven't tested it out so I have no clue what it's called 😭

I barely even use local models nowadays, so yeah..

[vital raptor]

i haven't tried it in forge, but the normal extension also supports loras in regions

[coarse brook]

Soo... This width height thing... Wat is that?

[vital raptor]

just set that to the same width/height as your generation

it's basically defining the area to apply the regions to

[ivory pebble]

https://github.com/Haoming02/sd-forge-couple

Found it pretty easily

[vital raptor]

requires forge though https://github.com/lllyasviel/stable-diffusion-webui-forge

[coarse brook]

Hmm... I assume its not compatible with Auto1111?

Fek

[ivory pebble]

Isn't ForgeUI better than A1111 tho?

[coarse brook]

Idk

I installed this thingy like yesterday

[ivory pebble]

You also don't need to delete A1111 in order to install Forge

[vital raptor]

better is subjective

[coarse brook]

That is tru

[ivory pebble]

I mean, pretty much everyone that I've asked has told me Forge was better, so idk

[vital raptor]

it has some substantial performance improvements for lower end hardware, but last i checked its a bit buggy

most people probably have 4-8gb cards

[coarse brook]

I prefer slow gen time, over buggy mess

[ivory pebble]

I personally haven't had any issues with it when I used it

[vital raptor]

it doesn't increase my performance beyond around 0.1-0.5s for sdxl generations so all i see are the bugs and incompatible extensions

[ivory pebble]

Plus it increased my generation speed by 3x, so I think it's worth it honestly

[ivory pebble]

I see, I assume you have a pretty good GPU in that case

[vital raptor]

mostly for me is TensorRT which does increase my performance substantially, doesn't work on forge

I have a 4090

[coarse brook]

I'd say thats pretty good

[ivory pebble]

I was planning on buying it but I wonder if I should wait for the 5000 series

[coarse brook]

Can i borrow like 1990, from those 4090? xD (To make my card a 3080

[vital raptor]

i'd wait at this point unless you aren't going for a top end 5080 or 5090 anyway

the 80/90 should be out by october or november, but the 60/70 range won't be until like march-april next year

[coarse brook]

The day i upgrade my gpu, is the day I'll stop playing games.Why? Cuz i am gonna be OOOOLD then

[vital raptor]

and the 4070TI with 16gb is a REALLY good deal right now

in the US anyway

[coarse brook]

Compared to most EU countries, everything in the US is a good deal

[ivory pebble]

I personally don't really think I need anything above a 4070, but I might as well wait and see what the price of the 5k series will be

[vital raptor]

comparatively of course, it's still expensive

tbh, i kind of regret buying the 4090, i already have a 3080ti and the overall performance in games and stuff is trivial from a real world perspective

[ivory pebble]

Obviously I'll have to wait and see what the future local models will look like cause right now I'd rather just pay $25 a month for NovelAI than buying an expensive GPU

[vital raptor]

it's nice to have, but not worth $2k

[ivory pebble]

Wait so the 3080ti is just as good?

[vital raptor]

well, one does a sdxl gen in about 8 seconds and one does it in 3

[ivory pebble]

I'd say that's not that big of a deal unless you're generating a lot of images

[vital raptor]

the 24gb vram versus 12gb does allow more resolution (at good speeds) for hires fix, but realistically, the cost to time ratio isn't great

[ivory pebble]

Yeah I'd say the 4090 would be good if I was training lora's

[vital raptor]

for $2k you could rent a gpu when you need it, or like you said use NAI or something instead

[ivory pebble]

But for regular image gen I think even a 4070/3080ti would be good enough

[vital raptor]

i do some training, but renting a A100 or something for a few hours would be way cheaper unless you're training a lot

[ivory pebble]

Makes sense yeah

[vital raptor]

i do have more professional reasons to have the hardware, so it was more a business expense anyway, but it still really wasn't worth it for me 😄

on top of that, i still have the 3080TI just sitting idle in my home server not getting used

[ivory pebble]

Oof, that's rough, you should've waited for the 5000 series at least but it is what it is

[vital raptor]

that was the plan, but then they delayed it for a year so i went for it

[ivory pebble]

Gimme wtf

[vital raptor]

still should have waited 😄

[ivory pebble]

You live and you learn I guess

I made the same mistake when I bought the 3050, never again 😭

[vital raptor]

yeah, i've debated selling the 3080ti off and on, but I think I missed the bus on that with the prices dropping now

[ivory pebble]

You never know, there's always someone who might be looking to buy it

[vital raptor]

I have a new server on the way, I'm going to see if i can cram the 3080ti in that for VM and LLM stuff if it fits, otherwise i'm going to try and sell it and use the money to get a used enterprise card or something 🤷‍♂️

something like a quadro rtx 8000 (48gb) would be nice, but $3500 is a bit

maybe i'll sell the 4090 and get that... 😄

or both even

[ivory pebble]

Yeah, "a bit" 😭

[vital raptor]

I have nothing else to spend my money on and i like playing with tech 😛

i bought an old dell poweredge r730xd server with 128GB ram and 2 14c/28t CPUs... just to play with it

should be here Wednesday

[coarse brook]

Hello again frens! I'm once again seeking for your recommendations, this time for something that makes eyes better in the anime/cartoon realm

Cause these do be looking yank

[supple raptor]

could be a few things going on there. if the image is sfw send the image by dragging and dropping it from the folder it's saved in so we can take a look at the prompt and all of the parameters.

[coarse brook]

"Expressive_eyes, Expressive eyes", are the only 2 promts revolving the eyes, and not quite sfw 😉

[supple raptor]

which model?

[coarse brook]

PonyXL SDXL merged

[supple raptor]

ah, that's probably part of your problem. pony used some silly training parameters which made it not really compatible with other models. I'd start by trying the same prompt and parameters with just PonyXL, or you can try AutismMix_pony or AutismMix_confetti if you don't like how pony looks out of the box.

[coarse brook]

I must admit, i have looked for the autismmixpony for a while, but haven't been able to pinpoint it

[vital raptor]

https://civitai.com/models/288584/autismmix-sdxl

[supple raptor]

beat me to it.

[coarse brook]

Oh... Thats whu i couldn't find it. I was trying to find a model titled AutismMix Pony-

[supple raptor]

yeah, the pony and confetti variants were uploaded as different model versions.

[vital raptor]

there's 5 models on that page

[coarse brook]

I c 😄 Still getting used to Civit

[supple raptor]

I prefer the pony variant. i think Captain Vargas likes the confetti variant.

Both are good

[vital raptor]

it's more of a style thing for me personally, they both work really well, i just got more consistent results for my taste with confetti

[supple raptor]

T-ponynai3 is another good one to check out

[vital raptor]

i need to check out that one

[coarse brook]

IS there a way for me to put all the Lora and models into another drive, and have SD pull from that drive, while being on a second drive? For some reason... The drive SD is on, has 120 gb of unaccounted used space i cannot find anywhere. I've made it show hidden files

[supple raptor]

It's not nearly as flexible in styles. But the styles it does do are nice.

[vital raptor]

--ckpt-dir "D:\AI\Models\checkpoints"

[coarse brook]

Wait... Hol... Where i put this, and how i tell it where i put em?

[vital raptor]

change the path obviously, but this goes in your webui-user.bat

this is the full line from mine
set COMMANDLINE_ARGS=--xformers --opt-channelslast --api --ckpt-dir "D:\AI\Models\checkpoints" --hypernetwork-dir "D:\AI\Models\hypernetworks" --embeddings-dir "D:\AI\Models\embeddings" --lora-dir "D:\AI\Models\loras" --vae-dir "D:\AI\Models\vae" --realesrgan-models-path "D:\AI\Models\upscalers" --esrgan-models-path "D:\AI\Models\upscalers"

[coarse brook]

So X:\MainFolder\Models\Stable-diffusion, for the checkpoints, and add it again but X:\MainFolder\Models\Lora?

[vital raptor]

if that's where they are, yeah

[coarse brook]

X is X as its a variable 😄 Idk what drive to put on yet

[vital raptor]

i use this to remap all of the related files because i generally have multiple UI installed, auto1111, forge, comfy, etc, sometimes even multiple versions of each

[coarse brook]

Might just move the entire SD folder tbh temp, and format the drive with all the random shit unaccounted for, and paste it back in

How even does 120 gb of space go missing, on a correctly partitioned drive?!

[supple raptor]

I just store all of mine in my A1111 directories and symlink those folders out to the other UIs.

[vital raptor]

🤷‍♂️

[vital raptor]

i don't remember which one off hand, but i had an extension that wouldn't work with symlinks and hardlinks in windows are a pain so i just used the config files and a central folder

[coarse brook]

773 gb of models, loras and checkpoints?! Wtf r u doing

[vital raptor]

i do a lot 😄

[coarse brook]

These are my drives. The newest one, the 970, is from circa 2018... Rest is from 2013.

[spare vigil]

Stable Diffusion

[vital raptor]

mine are all 970s

[coarse brook]

I am very surprised, none of them has died... Well, one tried, but i didn't permit it

[vital raptor]

i think one of the 2tb is almost 2 years old, the rest are ~6-8months

[coarse brook]

Spinner seized somehow... Opened it up, and spun it manually, on the disc retaining bracket. Never touched the plates*

Closed it back up, and it worked \o/

That was 6 yrs ago

[supple raptor]

should free up space on C: like yesterday. less than 10% free on an SSD is a big nono especially when it's the boot drive

[vital raptor]

the server i ordered has 24 2.5" bays in the front, i dug around in my parts and found 9 misc drives to throw in to start my NAS

totals around 5-6TB

plus the 7TB in this computer i'll be moving over

[coarse brook]

I have 2 games on the C drive, that tallies 281 gb 😐

[vital raptor]

not sure how i have almost 1.2TB of games installed

[coarse brook]

Cause modern games takes up a crap ton of space

In the next decade, when I might be able to afford a new pc, i deffo gotta look into some high capacity drives

[vital raptor]

realistically I only use a lot of space because i have it, before i built this system I only had a total of 3TB and never really ran out of space

I just leave games installed, and dont delete AI models i'm not even using, it just sits here

[coarse brook]

I just went through the drive... I can account for ~300 gb of space. The rest, is like system crap

[supple raptor]

download wiztree, it'll let you visualize where the space is going and make it easier to clear out stuff you don't need.

[coarse brook]

O.o Claims only 135 gb is allocated... Wat.

Again, sorry for hijacking for tech support 😄

Guess after formatting, I'll re-partition the drive, and then re-format forgood measure 😄

[supple raptor]

check the health status of the drive in CrystalDiskInfo

[vital raptor]

lol, disk info waifu edition

[coarse brook]

Gotta keep in line of the Anime channel xD

[coarse brook]

Much more fun when you do the entire window

(Sorry Tora 😦 )

[bitter latch]

why you apologizing to me for?

[coarse brook]

Cause I'm pushing your channel relevant stuff up, with my tech support xD

[bitter latch]

no no i'm the one getting involved with some tech based anime art 😄

you are good

all good - she says

[supple raptor]

waifu edition is correct choice

[coarse brook]

How can she even see the keyboard.

[bitter latch]

android mainframe

[coarse brook]

Thats a weird way of spelling "Gundam"

[vital raptor]

jeez, almost 43,000 hours on this ssd

[bitter latch]

kinda ironic i'm reading a book called naural wraith about a guy who has no implants for their ai run altnet in a city run by ai dolls (females of course) logic engine and emotion engine based AI's

[vital raptor]

almost 5 years

[coarse brook]

This is my external drive I've had for ages

MEanwhile my oldest drive. Brb rebooting pc ( 😉 But not really)

[vital raptor]

lol, 5000 power cycles

[coarse brook]

13 yrs, if you were to power it on each day

[bitter latch]

the book cover

[coarse brook]

13.69 yrs to be exact.

[vital raptor]

150 is the most i have

my other 3 drives have 20 power on count with 8699 hours

[coarse brook]

My system drive, which arguably is my newest drive.

My first SSD i ever got

And my 2nd drive i ever got 😄

Soooo....

[bitter latch]

hmm how do you bring that up?

[coarse brook]

Which part? CrystalDiscINFO for the Power on stats

And google for the text

[bitter latch]

the info for power on wher it shows you yours

[coarse brook]

Right side

[vital raptor]

yeah hard disk is 5 years

i'd expect solid state to last at least a few extra years

[coarse brook]

My 30k+ hrs are Hard Discs, fren 😄

[vital raptor]

i've had drives die a week old, and still have some kicking around from the stone age

i still have a 20GB IDE drive that works fine

[coarse brook]

I'm more concerned about the only 175 power on, but 41k hrs on

[vital raptor]

i should hook up some of my old drives and see what the stats say

[coarse brook]

"Help me. I wanna die"

[bitter latch]

mine look good

[coarse brook]

Gosh i hate it when Windows tells me a file is in use, but not by WHAT...

[bitter latch]

nods

[coarse brook]

"In use by other program", YES BUT WHICH?! (Moving SD temp).

[bitter latch]

I hate it when my sound goes out while runnning stable diffusion. grrumbles

[coarse brook]

At least it doesn't crash your browsers, ending up stopping your generation cause gpu drivers halted, because your brower crashed

[bitter latch]

oh i've had that happen

or the script errors that happen and then when I close an reload, no errors. like wtf

usually python

[coarse brook]

I often have crashes if i do to do rapid succession gens

[bitter latch]

yawns I should go back to bed

[vital raptor]

lol, this ones not too healthy

[bitter latch]

haha

[coarse brook]

Amazing... Told it to reboot into safe mode. Windows update decided to update the pc, and now it doesn't boot

[bitter latch]

drops this here. just because 😄 https://suno.com/song/df11515f-fe6c-43b0-b5da-dac86e63010f

and yep it is ai generated

yes no. delete?

[jagged field]

Hey Guys does anyone have a cool prompt to try it out ?

[coarse brook]

Reeeee

[bitter latch]

uhh oh

[jagged field]

rm -rf /

[coarse brook]

Wat

[coarse brook]

U like some wizard?

[jagged field]

no i mean it's the equivalent in Linux

[coarse brook]

Having major issues formatting a drive, windows thinks is a boot drive, but is in fact... Not

[bitter latch]

meanwhile I do this https://suno.com/song/3bfd14dd-82c7-4a4d-ac3f-298263d7816c

yep ran lyrics thru google translate

[coarse brook]

Make an img of someone splashing their pc with a keyboard

Cause mentally, I'm close

Smashing *not splashing

[bitter latch]

tries

[coarse brook]

I might have fixed the issue... I'll let ya'll know in 5

[bitter latch]

4

3 2 1?

[coarse brook]

THANK YOU ASUS! secure erase shat all over windows "protection" of the drive

[bitter latch]

ahh

hmm almost

[coarse brook]

Why is there a sword up his bum?

[bitter latch]

shrugs

[coarse brook]

You do you, boo

[bitter latch]

hey wasn't my idea, AI's

[coarse brook]

But you entered the prompts

Just saying

[bitter latch]

umm wtf

switches models

[coarse brook]

And i have my pc back now. This is nice

[bitter latch]

thats good

hmm ok...

oooh

[coarse brook]

Can't gen anything atm 😦 Busy moving stuff back over. ETA: 2½ hrs

[bitter latch]

ick

I want to upgrade to a better laptop but I dread moving things

[coarse brook]

But at least i regained ~90 gb of unaccounted space

[bitter latch]

ok maybe more detail at 1 is to much ooops

huh.. interesting

oh it's the huen sampler fusting with it

heun?

[coarse brook]

R u having a stroke? Do we need to call the bondulance?

[bitter latch]

nah

returns to anime girls

[coarse brook]

I have some ideas i need to get into promts, as soon as i can re-launch the webui xD

[coarse brook]

Any particular model used for those wings? Cause isn't feathers like... Troublesome to gen?

[bitter latch]

it's a merge I made 😄

[coarse brook]

I c

[coarse brook]

My next gen is something that is gonna have wings, but large wings

[bitter latch]

I'm just random

someone mentions somethjing or posts and I join in

or take a left turn into wtf-ville

[coarse brook]

Anime wizard duck in forest, ground dirt, location forest, realistic duck, pointy wizard hat, anime. Go go

[bitter latch]

o.O

[coarse brook]

What? Ducks can't be wizards?

[bitter latch]

was wiating for this one to finish

[coarse brook]

That rock... That'll cut you deep

Also, that is not a duck

[bitter latch]

well hold ya ducks, duck

[coarse brook]

https://tenor.com/view/running-duck-bird-funny-in-hand-gif-11866949352257941606

[bitter latch]

hmm o..k

[coarse brook]

I mean... There is A duck

And A wizard..?

[bitter latch]

ok need to reboot, it's starting to take more time than usual. brb

ok rebooted, restarting webui

[coarse brook]

More like WeebUI

[bitter latch]

and up

mostly

[coarse brook]

She is not a duck xD

[bitter latch]

takes out tron-punk...ooops

[coarse brook]

Oh hai CS

[bitter latch]

duck

Duck wizard

[coarse brook]

CHONKY!

[bitter latch]

grinz as he gets an idea and changes models

there! @coarse brook

[coarse brook]

Thats not a regular, realistic duck xD But It'll have to do xD

[bitter latch]

better ?

[coarse brook]

That is actually adorable

[coarse brook]

Oh god stahp

Too cute

[coarse brook]

Add a duck on like her shoulder, or her holding a duck xD

[bitter latch]

haha

[bitter latch]

there

ok last one for a whiel. going to lay back down.

[coarse brook]

https://tenor.com/view/cute-duck-cuddle-its-ok-adorable-duck-gif-12985802

[gloomy scroll]

https://tenor.com/view/petting-cat-gif-25993844

[coarse brook]

When generating, the very first 2 itterations of your genned pic, the grey blobs are terrifying to look at

Making some preview thumbnails for my lora packs, ended up creating this

[nova remnant]

inserts usb

[coarse brook]

https://media.discordapp.net/attachments/783432518107529219/817173295710863380/hornybonksnake.gif

[nova remnant]

[coarse brook]

https://tenor.com/view/cat-pet-petting-cat-cute-gif-26529093

[ivory pebble]

"hand focus" tag on NAI seems to give some interesting composition to images

[coarse brook]

Technically the hand is in focus in a universe

Might not be our focus, but still xD

[ivory pebble]

Had to edit the hand in the back since it got all messed up sadly

[coarse brook]

Sadge

[ivory pebble]

Otherwise this would've been a perfect image

[coarse brook]

Had a perfect, 0 alterations at all, earlier (Scroll up a bit) 😄

[ivory pebble]

Gonna keep rerolling cause this tag seems to be doing wonders

[coarse brook]

200 batch count, go to bed. Wake up with a bunch xD

[ivory pebble]

Legit I've spent like 5k anlas in a week when I first bought NAI

[coarse brook]

Anlas?

[ivory pebble]

It's way too fun generating sometimes 😭

[ivory pebble]

Yeah that's their currency

[coarse brook]

Oh yikes.

Gone over to local genning myself 😄

[ivory pebble]

Surprisingly you do get a LOT when you subscribe, I didn't expect it to be that much

Sadly it's not enough for someone who generates non stop like me 😭

[ivory pebble]

I've been doing that for so long but nothing on local looks as good as NAI sadly

[coarse brook]

I mean. This wasn't a bad one xD #🍥｜anime message

[ivory pebble]

Eh, I prefer something more anime looking personally

To me all the local models have that same feel to it, not sure how to describe it

[coarse brook]

Fair. Just putting my foot into that entire area in like a day or so

That just sounds like a setting issue tbh

But whatever makes you happy 🙂
I'm not gonna detract from that 😄

[ivory pebble]

No not really, local models aren't really as aesthetically pleasing as NAI, simply being able to mix 7 artists together on NAI gives much better outputs, at least for me

And overall the composition is quite boring

[coarse brook]

Each to their own, is all i can say 🙂

[ivory pebble]

Well of course, it was super fun for me as well when I first started using Pony and its derivative models, but it just looks so bland...

Not to mention it can't even do night/sunset scenes properly

Everything is just so bright 😭

[coarse brook]

https://tenor.com/view/flashbang-gif-22572243

[ivory pebble]

Hopefully that gets fixed with v7

[coarse brook]

The screen which i have my SD on, is so old, it has artificial clouds on it

[ivory pebble]

It looks like the monitor has veins wtf 😭

[coarse brook]

Tbf, it is a BenQ XL2420T from like.... 2012

But hey. Its working

Just don't have anything white on it

[ivory pebble]

Uglee

[nova remnant]

[coarse brook]

Love that Emoji.

https://cdn.discordapp.com/emojis/1017853838516035725.webp?size=128&quality=lossless

[coarse brook]

Don't bonk the usb in.

https://tenor.com/view/spider-man-tom-holland-yoink-gif-11346283

[nova remnant]

2 grabbies at once

👍

[coarse brook]

15 Lora's to still make previews for 😐

[nova remnant]

just do a 1girl prompt and write the name of the lora on her face

big brain

[coarse brook]

But thats boring

[ivory pebble]

At the end of the day.. 1girl is all we truly need

[coarse brook]

I mean... Though 99% of all the images do use 1girl, they also are of the preset

[nova remnant]

:p

[ivory pebble]

Those eyes look so cursed 😭

[nova remnant]

I could argue for 2girls

more than that I don't know if I can handle

[coarse brook]

Those are reserved for the.,... Others 😉

[coarse brook]

Is there a promt like Style_anime, or something to make sure one gets a specific style from a checkpoint that has a bit of everything?

[slow needle]

read your models description

[coarse brook]

Make a solid point

[ivory pebble]

If you're talking about Pony then use source_anime

[coarse brook]

Thank u

[ivory pebble]

It's still gonna be pretty random so try using some artist tags that aren't hashed or use some artist loras

[coarse brook]

If only i knew any artists tbh

I guess they'd be under "style" on Civit?

[ivory pebble]

Well you can prompt for some of them without loras but yeah, you can find style loras just by typing "style" in the search bar on Civit

[coarse brook]

Yeah. I found out. Trying now, just gotta see if the regional promter can force the correct promts.

[coarse brook]

Which Lora is that?

Cause I'm DL the Counterfit one at we speak

[ivory pebble]

Not a lora, this is NAI

[coarse brook]

Lemme guess... a Vae?

[ivory pebble]

No 😭, NAI=NovelAI, it's an online cloud based model

[coarse brook]

Ah

[ivory pebble]

It requires a monthly subscription, forgot to mention that

[coarse brook]

Yikes

[ivory pebble]

Well, I guess you don't really need to buy a monthly subscription, you could buy their currency instead but I prefer the monthly subscription since you get unlimited generations

[junior quest]

@ivory pebble can we train a lora or smthing like that on NAI?

[coarse brook]

generate a batch of 30000, train a new lora on Nai. Be named and shamed across the entire community, get banned for stealing, cries over a bowl of cereal.

[ivory pebble]

Probably, but from what I've heard it's not the best idea to train a lora using other AI art

[ivory pebble]

There's plenty of NAI style loras for Pony

[junior quest]

no no no I mean train a lora on NAI, like we can gen only characters that NAI team included right? so like local, cant we train a specific character and gen that character?

[ivory pebble]

Oh that's what you meant, my bad 😭. No you can't train loras on NAI

[coarse brook]

So... I tried to change the hair colors, using regional promter...

This was my result.

Somethings not quite right

[ivory pebble]

Yeah I have no clue, I gave up on regional prompter a long time ago since I couldn't figure it out. There's probably guides on how to use it on 4chan or something like that but I never looked into it sadly

[coarse brook]

I mean. Its a funny lil mishap 😄

Smol Girl, and Big mommy

Getting there. Not a bad one tbh. Or at least me finks

[coarse brook]

Wanna see the process, my client decided to go in, on a promt?

[coarse brook]

During a batch generation, with these promts source_anime, "score_9, score_8_up, score_7_up, <lora:PonyXL Expressive Faces:1> ((2 girls)), long hair, brown hair, warrior, warriors, location forest, forest,
BREAK
1 girl, long hair, Blonde hair,
BREAK
1 girl, long hair, Brown hair,

(Spoilered not cause of nsfw, but cus wide file)

[ivory pebble]

Did you try regional prompter without that lora? Just to see if it gives better results

[coarse brook]

Ye, it was disabled during those 9 pics

But like the AI really wanted skimpy, in the end xD

[nova remnant]

use this: (bee (deadflow):0.7)

[coarse brook]

Bee?

Buzz buzz?

[nova remnant]

👍

[coarse brook]

I think I'm starting to see, what kind of artist they are

[nova remnant]

If you're having problems with armor being skimpy, try "Bikini armor" in your prompt

It will 100% do that, no more rng!

[coarse brook]

Can you like... Bookmark promts?

[nova remnant]

every image you generate has the prompt in them

and you can make styles

[coarse brook]

I mean i know that

[nova remnant]

not sure how

[coarse brook]

I can pull them from the png info

[nova remnant]

styles is like a text file you type stuff there and it appears on your prompt from a little menu on the side

[coarse brook]

But at a point, when you gen 300 files overnight, cuz efficient use of time, it gets hard finding that one file

[gloomy scroll]

There is T-ponynai3 that's trained with NaiV3 images

[junior quest]

no I mean I wonder if there is a option like that in NAI, so we can input our characters that NAI doesnt have

[gloomy scroll]

Vibe transfer can help creating characters nai doesn't know

For some reason Nai doesn't like doing good boxes with vibe transfer

[coarse brook]

This classifies as anime, ye?

[green temple]

barbara ?

[zinc flax]

Oh no, some random characteristic

[green temple]

@nova remnant

[nova remnant]

[lofty berry]

got my hype for a sec

[ivory pebble]

Why? What happened?

[ivory pebble]

I'm doomed.. I can NOT stop generating her 😭

Gonna waste all my anlas at this rate

[junior quest]

switch to medusa

[ivory pebble]

Not a bad idea actually

[lofty berry]

i got hype cuz i though was a new checkpint or lora but is only NAi hahah

[steady grail]

found another pretty stable anime model, cool

[junior quest]

what? where?

[steady grail]

https://civitai.com/models/439260?modelVersionId=489639

[ivory pebble]

One day we'll have that for local.. hopefully 😭

[steady grail]

made some gens to test and still din't see mutation

and the result is pretty close to the prompt

[junior quest]

ouh not my style

sadge

[steady grail]

i love that toony style

[junior quest]

https://tenor.com/view/falling-drunk-man-falling-drunk-man-this-was-a-pretty-popular-meme-gif-26183721

[ivory pebble]

Had to reroll a lot since her weapons are so complicated 😭

That was the best one I got

[junior quest]

wooow its so good

[ivory pebble]

Had to fix her hand as well since the AI won't do it for me 😡

[ivory pebble]

Her dress is so short wtf I just realized

She is NOT jumping with that on that's for sure

[junior quest]

shes jumping, running doing all the things

she does have a another outfit I guess, denim pants and sweater or smthing

https://tenor.com/view/fate-rider-sweater-gif-21774285

[junior quest]

[ivory pebble]

Apparently increasing the Enhance strength fixes the hands, so that's pretty neat

[junior quest]

yesh soo neat

[ivory pebble]

Her right hand looked like a mangled mess, ran the image through 0.35 Enhance and it did wonders

Still a bit off but much better than what it looked like at first

[ivory pebble]

Time to generate some of those pics as well I guess

[junior quest]

ahh my lora is messing everything

outfit is weird, color is different

deym

[steady grail]

just noticed that the most stable models, are even the most bland and bare

[supple raptor]

Because they're overfitted

[steady grail]

yep, too much locked on their concept

but i love the toon style >.<

it's so hard to find a flexible model in that style

i suppose i need to lock a prefered style using negatives

[coarse brook]

I am sad now. Was wondering why half my Lora's werent showing... Went to "Show incompatible Lora's".... They all need Nai's 😦

[junior quest]

[ornate echo]

https://tenor.com/bQD4H.gif

[junior quest]

ma lora kinda good at generating things like wallpaper or smthing

[coarse brook]

https://tenor.com/view/telepurte-bonk-no-horni-gif-21583414

[junior quest]

whats this pose?

[ivory pebble]

This last one feels.. weird, not sure why

[ivory pebble]

I have no clue, prompting "hand focus" on NAI just gives really good composition at times, my best guess would be "standing, from above, arms under breasts"

[coarse brook]

arms crossed under breasts, perhaps?

[steady grail]

tilted camera?

[jagged field]

noice bro

[coarse brook]

Looking from above, arms crossed under breast, looking at viewer..?

[junior quest]

kay...

[ivory pebble]

I'm going back to making girls with swords now, my fault

[jagged field]

i liked it

[coarse brook]

Hidden sword, crouching balls?

[ivory pebble]

I have no clue what that is and I am NOT going to find out 😭

[coarse brook]

Think. You know exactly what it is. Hint: Nsfw

[ivory pebble]

My best guess would be them pulling out a sword that's in between their breasts

I have no clue what else it could be 😭

[junior quest]

https://tenor.com/view/baal-genshin-impact-gif-22525631

[ivory pebble]

Yup, basically that

[coarse brook]

Oh poor innocent individual

[steady grail]

who wouldn't hide the weapon in the crack?

[coarse brook]

Wonder Woman sure did

[junior quest]

[ivory pebble]

That's one way to put it I guess

[ivory pebble]

Still my favorite pic, she's so cool

[coarse brook]

[jagged field]

[steady grail]

i prefer the second

[jagged field]

the hair is more clear

[junior quest]

noooo not realistic...

[jagged field]

wut

[junior quest]

yesh second one is good

COLORS

[jagged field]

this is one is anime again so don't worry

[junior quest]

is it?

then is dat more details or smthing?

[jagged field]

yeah you can see the original

i replied to it

idk why does discord not showing me notifications when someone reply to my messages

[dire bridge]

It looks less like there’s a camera flash on.

I prefer second too

[steady grail]

yeah, i just noticed

[bitter latch]

.

[supple raptor]

NAI doesn't have LoRAs. If you have models that reference NAI or NovelAI in some capacity it just means they are SD1.5 based LoRAs that were probably trained on the leaked NAI model. If you want to use LoRAs with Pony, or any of its variants then you need to look for SDXL models, and ideally ones that were trained specifically on Pony or one of it's variants (there's a filter on Civitai for Pony based models) for the same reason why merging Pony doesn't work well.

[coarse brook]

I have this

[coarse brook]

And thee Lora's are wanting something with "Nai.ckpt"

[supple raptor]

arms under breasts, from above, leaning forward, looking over eyewear, looking at viewer depending on the model you might have to remove leaning forward or lower the weight on it. alternatively, you could just use openpose controlnet, since you already have an image with the pose you want.

[supple raptor]

they just need any SD1.5 model. SD1.5 and SDXL things including loras are not compatible with each other at all.

[gloomy scroll]

Arms under breasts.. Like this?

[coarse brook]

Why do i have a feeling, these have been deleted from Civit?

[gloomy scroll]

NovelAI model for sd1.5 was leak

[supple raptor]

nai.ckpt is actually called anime_full_final_pruned.ckpt and it was stolen from a paid company called NovelAI.

[coarse brook]

Figured 😐

[supple raptor]

it's what basically every halfway decent SD1.5 based model is trained on. and/or merged with.

[gloomy scroll]

I don't have it anymore

[coarse brook]

This be why u backup your Lora's.

[gloomy scroll]

Not that I would ever use it

Aurora and soushiki exist for sd1.5

[supple raptor]

I just have these

chenShrug

[gloomy scroll]

Well you have everything saved

[coarse brook]

Lucky, i guess

[supple raptor]

aside from training, that model has been worthless for a long time now.

[coarse brook]

Bit annoying they're not hidden on Civit, so prevent people like me downloading loras with no publicly available (free) models

To prevent*

Words r hart

[supple raptor]

Soushiki, Aurora, Kakigori, maybe one of the Based series of models are all you need, they will handle loras trained on the leaked model very well.

[coarse brook]

Will they accept the lora's trained on the nai?

[supple raptor]

yesh

[ivory pebble]

Oh yeah I know, I was messing with some prompts on NAI and I got it as a random gen without using any of those tags. "hand focus" seems to give some interesting composition to images from what I've tested, or it could just be the randomness of the model and the tag might not even do anything, it's hard to tell 😭

[supple raptor]

Just look here when downloading models or loras. If it says SD1.5 then it will work with things that were trained on nai reasonably well. Some are better than others at that, but the 4 I mentioned are quite compatible.

[supple raptor]

reaching towards viewer, foreshortening might be another one that you'd like. could be combined with hand focus

[coarse brook]

What about the Networks.llora, thingy?

[ivory pebble]

Will try it out later, right now I'm messing with some action poses like "fighting stance, holding sword", but I can't figure out if there's a tag for sword swinging, probably not since I can't find anything on danbooru

Maybe adding motion lines might do the trick 🤔

[vital raptor]

action pose

[supple raptor]

loras have the same thing. SD1.5 based models and loras were very compatible with each other for the most part. SDXL models and loras are not as compatible, but with the exception of Pony, they'll generally work at least decently.

[ivory pebble]

I knew I was missing something 😭, I used that prompt before but forgot what it was called, tyty

[supple raptor]

MomijiWide

[coarse brook]

I just learned, this is a thing as well. So let's say this one works better on Pack 1 (Out of the 3 you suggested), is there a promt i can plop in to make it load that checkpoint, on that lora, every time?

[supple raptor]

no, the prompt doesn't have control over which model is used. activation text is for automatically adding the tags that the lora was trained on, for example a character lora might use a format like character name, copyright and might also include some other tags for official outfits that the character wears. or an artstyle lora might use the artist's name or a word that describes the style.

[coarse brook]

Like names etc?

Names, outfit etc

[supple raptor]

yep

[coarse brook]

SO is the lazy mans approach to not forget stuff 😄

[supple raptor]

the real lazy man's approach is to get the extension that automatically pulls activation text and example images from civitai so you don't even have to fill in the activation text box yourself

[coarse brook]

That sounds nice, and yet a mess

Cause people man... The amount of crap you see, if u has nsfw enabled, on pics. There seem to be every kink known to man

Now... Its not that I'm kinkshaming, but I'd rather not have the promt "woman urinating on a duck while laughing and eating a kebab roll"

[gloomy scroll]

Lets play a game. Just image something it can be whatever you want... Ok you have something... ||That's someone's fetish||

[coarse brook]

I mean, i understand that everyone is unique, its just not something I want near me 😄 You do you, but do not push it on me 😄

[gloomy scroll]

Cleanse your eyes on this

[coarse brook]

https://tenor.com/view/toothless-toothless-dancing-meme-gif-shock-shocked-wow-gif-814142470340463466

[coarse brook]

https://tenor.com/view/toothless-gnap-casvondepol-cas-von-de-pol-toothless-dragon-dance-krokmou-gif-1736164897226882387

[gloomy scroll]

Don't worry it bites back

[coarse brook]

https://tenor.com/view/anime-kid-kick-gif-16022961

[nova remnant]

anime gif

[coarse brook]

Cat pfp

[nova remnant]

I'm a ball

https://tenor.com/view/meow-cat-gif-27161697

[coarse brook]

https://tenor.com/view/cat-sleeping-gif-10304095

[coarse brook]

https://tenor.com/view/cat-omamori-himari-cute-gif-15485165

[nova remnant]

lmao

[nova remnant]

I put saitama transformed into a cat

[coarse brook]

Cat is sleeping on me atm :3

[nova remnant]

I just noticed you're a pony

[gloomy scroll]

Pony

🎠

[coarse brook]

https://tenor.com/view/hopping-fluffy-gif-25867122

[nova remnant]

kill it with fire

[gloomy scroll]

What did you mean with this?

[nova remnant]

silverflame's picture

[versed jay]

Do you know any way to make two characters appear, without their characteristics merging? and using a lora for each one?

[nova remnant]

Is some pony looking thing

[versed jay]

@steady grail

[nova remnant]

look up a tutorial on regional prompting

[coarse brook]

There is nothing wrong with an innocent unicorn

[versed jay]

Can you show me a sample please?

[nova remnant]

I have no idea how to use regional prompting

[coarse brook]

https://github.com/hako-mikan/sd-webui-regional-prompter

[nova remnant]

https://tenor.com/view/stranger-things-fire-flame-gif-14920648

[coarse brook]

https://tenor.com/view/shocked-in-shock-speechless-no-words-perplexed-gif-10886202

[coarse brook]

stahp crashing plz Webui

[bitter latch]

grinz evilly

[unkempt cape]

anyone excited for gpt 4o image gen? if it's anything like dall e 3 then it would be pretty much img2img + style copy dall e 3

like the moment that shit comes out ill spend 24/7 making images before censorship comes in and ruins everything

[supple raptor]

Afaik GPT-4o is just using Dalle-3 for it's image gen.

[unkempt cape]

Uh nah

I mean sure they never stated explicitly afaik that 4o gen by itself but like

you can tell gpt 4o to copy a character and it will do it

[supple raptor]

yeah, it has a new vision model but that's not the same as an image gen model.

[unkempt cape]

Either way its img2img dall e

[unkempt cape]

Lemme see if i can navigate my way into the demo pics

[bitter latch]

yawns

[supple raptor]

eh... not really. img2img is using the input image as the starting point for the diffusion process. vision model to image gen is the same thing as running your image through WDTagger to generate a prompt, then using that prompt to generate an image. presumably their vision model will perform better than that workflow does, but that's what it's doing at it's core.

[unkempt cape]

No

Just no

[bitter latch]

mixes it up

[unkempt cape]

The site doesnt load for me

[supple raptor]

yes. there is no now image gen here.

https://openai.com/index/hello-gpt-4o/

[unkempt cape]

But it can copy your face

Tell me that isnt img2img

Gaddamit internyet

[supple raptor]

I just linked you OpenAi's site that introduces GPT-4o. They don't discuss anything new in image gen there because there is nothing new in image gen.

[unkempt cape]

scroll down

[unkempt cape]

Explorations of capabilities

Select sample: Photo to caricature

Wait for that shi to load

[supple raptor]

yeah, it's just using a vision model to generate a prompt based on your text and image inputs.

[unkempt cape]

Dude wtf you mean bro. How can a prompt copy a face

I mean if its that good might as well treat it like img2img

[supple raptor]

With GPT-4o, we trained a single new model end-to-end across text, vision, and audio, meaning that all inputs and outputs are processed by the same neural network. Because GPT-4o is our first model combining all of these modalities, we are still just scratching the surface of exploring what the model can do and its limitations.

that's what GPT-4o is. a text, vision, and audio model.

[unkempt cape]

all inputs and OUTPUTS are processed by the neural network

[bitter latch]

yawns some mroe

[unkempt cape]

If this shit loads I'll give you a screenshot but it doesn't

It does in fact img2img

[supple raptor]

even if it does (which it doesn't) that isn't anything groundbreaking.

honestly the vision model is pretty exciting, it just isn't what you think it is.

[unkempt cape]

The level of editing it can make is pretty cool

[unkempt cape]

It has the capabilities that I imagined

[unkempt cape]

It can transfer your photo into another style, move a figure in an image to do something...

Tell me that isn't img2img convince me man

also "audio model" it can generate and hear sounds. So like "vision model" means generate and see pictures

[bitter latch]

she angy

[supple raptor]

no, it doesn't

vision means it can see.

which is cool. it does not mean that it generates images on it's own.

[unkempt cape]

Ok fine but like I literally showed you how to navigate to the examples