🍥｜anime

[broken osprey]

How come you have so many tabs there? At most I have two for the up scalers and that’s it

[bitter latch]

extensions 😄

[magic wave]

Fair :O

[broken osprey]

Okay @magic wave may I dm you with a Question?

[bitter latch]

I think hte big help is the miaoshou assistant one that really helps. it releases vram

[broken osprey]

Oh word?!!!!! How can it do that? Is it available in the program directly?

[magic wave]

Yeee no worries :o

You can go ahead

[bitter latch]

hmm ya know I should publish this model mix/merge on civitAI

[broken osprey]

I want that quality

[bitter latch]

stomp stomp , come here silly human

ponders a name for this merge. .hmmms

[broken osprey]

StompyFoxes

[bitter latch]

wait she looks like girl in a hat in time

kinda

[bitter latch]

no no it's a furry and anime model merge. hmms (ponders)

[broken osprey]

@unreal ridge what is the upscale you are using ? Also how do you prompt for such control on the hands and legs?

[bitter latch]

grins evilly

[unreal ridge]

• 4x ultrasharp
• I don't 😔

[bitter latch]

haz idea's

[unreal ridge]

I can send the prompt tho

[broken osprey]

But like only that?

You use hires fix ? Or extra tab? :0

[unreal ridge]

I just highres fix 0.45 denoise/15 steps on 4x ultrasharp and upscale it on extra 2x using ultrasharp too

[broken osprey]

Okay let me give it a try with an anime model

[weak prairie]

Wouldn’t you want to get close to these three beautiful #niji flower brides? #bride #flower #kawaii #waifus #aiwaifu #aiwaifus #sd #sdxl #stablediffusion #hcup #dcup

[unreal ridge]

I like the eyes on this one... Not that anime tho

[broken osprey]

How did you get all of them on the same image? Did you use control net + latent couple or just latent couple?

[bitter latch]

hmm lets see what that other prompt does with my merge

[magic wave]

Reminded me I gotta train another Nijijourney style

[unreal ridge]

it was from a XL model

I really don't like XL models

[magic wave]

Me neither

xD

XL is scuffed

[unreal ridge]

I hate the fact that it's censored too

most of the times the face looks like crap unless you're doing realistic

[magic wave]

O I thought it wasnt

[silk narwhal]

generate with SDXL and use SD 1.5 in Ydetailer 😄 for the face etc..high denoise

[unreal ridge]

Maybe I worded it badly, but I feel like it's extremely hard to make lewd prompts depending on the model, most of the time it came out without nipples or just scuffed

maybe lack of lewd images on the database

no idea

[bitter latch]

Fancy sci-fi background, blue eye, 1girl, mature woman, Medium breasts, heart glasses, holographic hair, twin tails, button gap, short pencil skirt, collared shirt,

[broken osprey]

I really don’t get what you all use for upscslers or even for models :0 I try the smallest size possible and even then the eyes and mouth come out scuffed :(. Adetailer can’t fix it

[unreal ridge]

send print of your settings

or just an image with the metadata included

[bitter latch]

oh and another extension I love is Stylepile

[broken osprey]

I’m sending you the screen capture if you want, but is this okay in here or should I post it on tech support

[unreal ridge]

are you using comfy?

if yes, I have no idea of how this works

it looks so confusing to me

[broken osprey]

I’m using webui now. Somehow it isn’t throwing me out of memory errros every single second

[bitter latch]

Fancy sci-fi background, blue eye, 1girl, mature woman, Medium breasts, heart glasses, holographic hair, twin tails, button gap, short pencil skirt, collared shirt, (by Artist Edvard Munch:1.3), (Cutester Art:1.3)

[silk narwhal]

try my workflow ..simply drag and drop the image into your comfyUI (may have to change the models to the ones you have (currently using SDXL stuff), and install missing nodes

[broken osprey]

I think I will once this generation is done. It truly is weird how it comes…. Like a big meh

@unreal ridge srry for ping there it is

[magic wave]

Ohhhh, fair then, no clue, I just don't like the prompting for it

Its weird

[broken osprey]

and memory errors :c this is why I hate the webui.

[unreal ridge]

I think you're using way too much strength on the (PROMPT:NUMBER.NUMBER)

[broken osprey]

would that affect the general outcome, coloring and composition?

[unreal ridge]

yes, heavily

[broken osprey]

I've seen example in civit AI that looks amazing and have sliders like 1.8 1.5 >

[silk narwhal]

yes..lots .. note that each () also adds like 0.1 to the value

[unreal ridge]

Have you tried using neg embeddings?

it'll make your life a lot easier imo

don't know if you're using one

[broken osprey]

yeah I used negative hands, safe pos for character generation, negativeV2 the like

How heavy is this thing? downloading it before killed my webui and now its taking long even with ethernet :0

[unreal ridge]

I use this
(low quality, worst quality:1.4), (bad anatomy), extra digit, fewer digits, (extra arms:1.2), bad hands, by (bad-artist:0.6), bad-image-v2-39000, negative_hand-neg, text, watermark,
and already think I'm using too much

I think I stole part of it from @lyric zealot

[broken osprey]

but like if that is too much then what is the point? Should I just say like *embedding1, 2,3 ... and that it?

[unreal ridge]

Everyone has their prefferences
I just use (low quality, worst quality:1.4), (bad anatomy), extra digit, fewer digits, (extra arms:1.2) and leave the rest for the embeddings

text, watermark, I normally use if the model tend to leave watermarks and shit

[broken osprey]

Okay I get that now, but I dont get how you guys generate detailed 4k looking images and keep details in the eyes, I use (in webui) extras tab and even then image comes out scuffed, zoom in and even on 8k * 4096 it suchs.

Before doing that and only upscale it also sucks, zoom in on any part of the image and it looks like crayoned in, even adetailer struggles to fix stuff, and once I use it upscaling kills the eyes it creates and we back to square one sigh

Comfy on the other hand somewhat generates images, at least they are acceptable

[magic wave]

Default style anime pic

[silk narwhal]

lots of passes and latent blending 😄 , for me at least

[broken osprey]

what even is that :0

[unreal ridge]

I really don't do anything special
I feel like you're using way too many prompts and negatives with :number.number which is essentially fucking things up

[silk narwhal]

in artist terms ... putting two version of a image on top of one another and then draw it again with the stuff that differ in more detail

[broken osprey]

okay.... so how do you achieve that?

I'll go try removing negatives and weights then, see if that gives something tolerable

[arctic nova]

JESUS

[silk narwhal]

with a node setup like this 😄

[arctic nova]

try trimming down that amount of negatives

[broken osprey]

yeah that was an old prompt that is.....sfw..... didnt want to send in something non acceptable for the demo. I'll come up with something normal and see if it works

[silk narwhal]

or nodes setup like this 😄

[unreal ridge]

try just prompting with
(low quality, worst quality:1.4), (bad anatomy), extra digit, fewer digits, (extra arms:1.2), bad hands

• embeddings

see if it improves or not

you really don't need that many negatives

[unreal ridge]

My brain trying to understand this:

https://tenor.com/view/azur-lane-mad-angry-fight-gif-16280033

[silk narwhal]

lol one of my blends was setup wrong

[broken osprey]

how much vram does that use?

[silk narwhal]

🙂 all of the VRAM

[broken osprey]

oh yea you have 12 gb, then it probably wont work for me since I have 8. I am guessing it will throw an error immediately or half way through on the k samplers

[silk narwhal]

well this is doing 2x 1536x768 per sampler 😄

[magic wave]

BlushedArts Style

[broken osprey]

Okay I produced this, not good and meh tbh, it took literally 2:48 min /secs

[silk narwhal]

not sure if my command line settings work with AMD

[broken osprey]

that too, goD i HATE AMD.... or for lack of a better candidate nvidia for creating exclusive techonology for their cards

also that extension didnt release any of the vram weird

[silk narwhal]

it prob does but it just reorganizes it so it won't hog your pc

[rich tartan]

You can get oom on comfy?

[broken osprey]

nop i tried the force or normal release but i got an error with it. It is expecting CUDA libraries in my pc and well, I cant get those haha

[broken osprey]

what/s that?

[unreal ridge]

what did you prompt tho?

[broken osprey]

A close up of a young teenage girl smiling towards the viewer with green eyes and visible perfect teeth, (green eyes:1.1), long white hair, simple background, blurry background, green blurry background

Literally the simples I could make it, also from previous tests I have found that natural language is far better in my use cases.

Most models produce great images through the use of it

(low quality, worst quality:1.4), (bad anatomy), extra digit, fewer digits, (extra arms:1.2), bad-hands-5 easynegative negative_hand-neg (negative; copied yours)

[arctic nova]

What model you using?

[broken osprey]

phantommix_Fp16

[magic wave]

I should upload some of my LoRAs

Just out of wanting to flex, cause they're usually better than the ones out there

[arctic nova]

It followed most of the prompt, there's nothing wrong with it. It just looks generic af, probably the model's fault. Also, I'm pretty sure using 2 hand embeddings is bad, the last one is far superior imo.

[magic wave]

That is so cute

What LoRA did you use for that? @gloomy scroll

[gloomy scroll]

It's NAIv3 image and has pnginfo

[magic wave]

Ooo that sucks, lemme inspect that pnginfo tho

[silk narwhal]

https://tenor.com/view/good-morning-do-it-gif-24435329

[broken osprey]

Well I’m not really in the business of creating delicate and intricate faces tbh. I just want to remove the blush so I’m putting it in negative and see if it works.

As for faces I’m sure I can use contorlnet to kinda get the shape of the face and prompt

[magic wave]

These are some obscure ass artist tags

[gloomy scroll]

I just went through 15k artist list

[magic wave]

Fair xD

[magic wave]

These are very obscure tho

I think I can train sandcasks, but not the other two

[arctic nova]

NAIV3 is so good

[gloomy scroll]

Well I was just look at this kind of list I have no actual idea of those artists

[magic wave]

I see, then I'd have to find another artst as a replacement

[unreal ridge]

Probably the only anime XL model that doesn't look like shit

[gloomy scroll]

What other artist were you looking at since I think there was few different in these images

[brisk valley]

little cheater

[silk narwhal]

starts so good ..:(

[gloomy scroll]

[brisk valley]

https://civitai.com/models/124189

[brisk valley]

I mean it's literally a closed source SDXL model that no one except people who pay can use

[gloomy scroll]

[unreal ridge]

Meh, I'd rather pay for NAI instead

[brisk valley]

the heck

[gloomy scroll]

Well I got what I asked

[brisk valley]

fuckin p2w community

[silk narwhal]

if it ain't local who says it isn't just a scraper 😄

[unreal ridge]

The only way I see myself switching to XL is either by a [redacted] incident, or if I get bored from 1.5 and decide to pay for NAI instead

[silk narwhal]

ask NAI something super obscure that never ever is on any art gallery site and see if it still produces a clean result

[unreal ridge]

literally any model ever

[magic wave]

Well, I looked at the tags

And I looked for those

[silk narwhal]

lol, i manage to have comfyUI have a brainf--- 😄 literally unable to load the exact model it used 15 seconds earlier

[magic wave]

You know, a [REDACTED] incident would be tragic, it would be really scummy and really terrible

but

XD

[unreal ridge]

I mean

[unreal ridge]

We wouldn't be here if it wasn't for it lol

it would still be tragic tho

[magic wave]

True too xD

[silk narwhal]

lol... gotta love Ai generation jank 😄

[silk narwhal]

https://pika.art/video/69c21198-a800-4e4b-b4ad-b0ababd97250

[broken osprey]

Behold, the first image I kind of like :))), no upscale since this is batch processing BUT, it kinda looks good for an obscure chinese model i found

[arctic nova]

model diff

[terse ore]

Is there a way to make anime lineart from img2img and controlnet? I make stuff from a blender render but I want to turn it into manga while keeping the composition

[native halo]

yes

[broken osprey]

The Chinese have amazing models tbh sad thing some you have to buy.

[silk narwhal]

there is a animelineart model for controlnet yes

[terse ore]

Anyone know the name of the model(s)?

[terse ore]

I use that but I mean to make lineart without color

[arctic nova]

Glad it worked out for you, I feel like it was mostly a neg prompt/model issue for you

[silk narwhal]

think there are comfy workflow with some passes that makes that possible

[broken osprey]

I’m not sure, I ran these prompts in webui and it wicks

Sucks*

[silk narwhal]

https://comfyworkflows.com/workflows/4c346fb1-dd4a-465c-afe3-c3acc0c89319

[broken osprey]

In my honest opinion comfy works way better for AMD, quick and responsive; no errors with memory, most outcomes even in the same settings come out better idk what optimizations it uses but even anime looks wonderful in it

Only downside is the strep learning curve. Also @silk narwhal would you mind explaining how to get a good upscale routine similar to hi res from webui in comfy. I’m using latent but I’m guessing that also changes the image far too much

[silk narwhal]

iterative latent upscale node seems to work decent in comfy

[broken osprey]

Huh? Can you show me the pack it comes in?

[silk narwhal]

impact pack

[broken osprey]

ah oke I have that I'll look it up thank you,

also do you know about tile upscaling method for comfy?

[silk narwhal]

this maybe?

havent used it yet

[broken osprey]

Oooooh I have seen it used along with Ultimate SD upscale node, those might make my images sharper

[silk narwhal]

two more after this and then GN all

[terse ore]

I haven't tried it yet, but apparently putting "hands" as the negative prompt actually makes good hands a lot of the time. And it needs to be weighted more or less depending on the Lora.

[broken osprey]

Or use control net for hands :)

[native halo]

https://tenor.com/view/shocked-black-guy-shocked-gif-27437284

[broken osprey]

@silk narwhal before you go, do you know any nodes where I can edit control net pose inside of comfy ?

[silk narwhal]

no idea.. have been using comfy for like 1 week now 😄

[broken osprey]

And you developed that workflow? God damn you smart, okay I’ll look somewhere for it dw

[silk narwhal]

mostly iteration on a existing one 😄

what if i do this, what if i add this etc

i am pushing my rtx 3060 pretty hard now 😄

[broken osprey]

damn well that sounds fun, I think I’ll try replicating your workflow for a change orrrrr trying to create my own for upscaling

[silk narwhal]

generating in about the same time it took one 1024x1024 image on the gtx 1070 but it is now 16x more pixels 😄

[broken osprey]

You sounded like tom Howard 16 times the detail

As long as I can generate good images for myself tbh I’m good, I don’t see a clear way to use this to earn money

[silk narwhal]

25% the workflow process

50%

and i overloaded latent space 😄, black output for the rest

[broken osprey]

So it’s not 100% guaranteed image every time?

[silk narwhal]

nah .. messed something up

[broken osprey]

[silk narwhal]

Pika 1.0 is public now

[broken osprey]

Alright im trying your workflow rn

Wait no I’ve still got 29 more images to go on this run

[broken osprey]

What’s that?

[silk narwhal]

text to video, image to video and video to video website (currently still free)

https://pika.art/video/2eef94e0-0ef4-436c-82aa-bbc6f83c1126

[broken osprey]

Well ain’t that cute <3

It looks like watercolor style hehe cute lil image me like me like. I’ll try to replicate it somehow…. Someday

[silk narwhal]

https://civitai.com/models/124189/anime-illust-diffusion-xl is this model

[broken osprey]

Wasn’t sdxl better for realistic than anime?

Also damn I found two editors for poses in comfy but none of them offer connection to control net output. Instead they want you to load in the pose and edit it there .-.

[silk narwhal]

Goodnight all

[nova remnant]

[nova remnant]

[native halo]

too hot out there to be wearing those clothes 🤠

[bitter latch]

always puts a pencil wtf

[nova remnant]

Very big pencils I approve

[bitter latch]

well... ok then

whats up doc?

[nova remnant]

@wispy canopy you seen this?

https://huggingface.co/segmind/Segmind-VegaRT
https://huggingface.co/segmind/Segmind-Vega

[wispy canopy]

nope, i'll take a look

[nova remnant]

https://huggingface.co/madebyollin/sdxl-vae-fp16-fix/tree/main

[nova remnant]

waiting to hear from you what you think of it

images look like they are better than lcm lora based turbo models

[bitter latch]

😄

[nova remnant]

The Segmind-Vega Model is a distilled version of the Stable Diffusion XL (SDXL), offering a remarkable 70% reduction in size and an impressive 100% speedup while retaining high-quality text-to-image generation capabilities. Trained on diverse datasets, including Grit and Midjourney scrape data, it excels at creating a wide range of visual content based on textual prompts.

[wispy canopy]

LCM OG vs LCM segmind, turbo-ized bpn1.3 8 steps

i feel segmind makes a lot more coherent images at low steps

[nova remnant]

sounds good, did you run them locally just now?

[wispy canopy]

yeah

[nova remnant]

oh nice

did you try with the vae they had up?

[wispy canopy]

just add-in replace the segmind lora

[nova remnant]

diffusion_pytorch_model.fp16.safetensors

[wispy canopy]

although with my latest model tryouts, i had another dumb idea ^^

which... kinda worked 😮

[nova remnant]

ok im gonna try out the model

[bitter latch]

ok... wtf is with the hand?!

[nova remnant]

im curious tho .... what im getting is
vae: diffusion_pytorch_model.fp16.safetensors
safetensor: segmind-vega.safetensors

[wispy canopy]

@nova remnant my comfy workflow should be in them, using both lcm segmind+turbo

you want this one https://huggingface.co/segmind/Segmind-VegaRT/resolve/main/pytorch_lora_weights.safetensors?download=true

that's the lora

i have no idea about the model xD

[nova remnant]

why that one?

im grabbing the files from here https://huggingface.co/segmind/Segmind-Vega/tree/main

[wispy canopy]

that way, you can use it with any model 😄

[nova remnant]

oh.. i was gonna try their segmind model itself

not the lora

[wispy canopy]

didn't check out that one

[nova remnant]

see thats compared with SDXL with far better coherence and its size is around 3gb opposed to 6gb sdxl

plus the speed

they suggest cfg 9 for good quality

dunno what the heck happend, it got bunch of error where it failed to load the model

[wispy canopy]

lcm lora, vs segmind lora

[wispy canopy]

it's a diffusors model iirc

[nova remnant]

should i be installing

pip install diffusers

[wispy canopy]

lemme see if i can get it working natively in comfy

[nova remnant]

ok

ran these two commands ...

pip install diffusers
pip install transformers accelerate safetensors

think they installed properly, no error

failing to load segmind checkpoint still tho

[native halo]

a1111 cant use diffuser models

[nova remnant]

yeah i see it

they are yet to add support for a1111

comfy ui has support

👀

what am i missing? lol im confused

[wispy canopy]

@nova remnant got the model working in comfy just like that... it's just... not good enough 😦

lets see what else sticks

[nova remnant]

the a1111 supported one is here btw https://huggingface.co/segmind/SSD-1B/tree/main but that's not the same what huggingface released now

[wispy canopy]

it's not a lie that it's 100% faster

[nova remnant]

yeah im guessing it can't produce the quality of a trained model

i can't expect the same result from it as i can with bpn

also i doubt it can do all kinds of images

[wispy canopy]

bpn 1.3 vs segmind vega,

[nova remnant]

a1111 supported file is 4gb in size, higgingface released a smalller and newer file

[wispy canopy]

with vega being a base model, and double as fast

still, not as bad ^^

now, can i frankenmerge it? 😮

[nova remnant]

lol maybe you could i wonder too

since that works with a1111 and is about 4gb size, perhaps you could also reduce your model size if you could merge them?

i mean if thats possible all sdxl model size could be brought down to 4gb or something, speedier load time

[wispy canopy]

wonder if it could

i mean, it does seem to be able to merge, but not sure how to get the perfect merge

[modest anvil]

any ideas on how to make the yes look foward

i always get a lazy eye idk why

[wispy canopy]

looking at viewer in the negatives

@nova remnant og segmind vs frankenmerge with bpn

[brittle rampart]

or looking away( or looking up)

[nova remnant]

oh damn, so the merge worked and any changes in speed size load time?

[modest anvil]

still nothing dam

i tried looking left right up down idk ill try more things tho

[wispy canopy]

worked-ish. the upper layers from bpn 1.3 imported, seemed to make it a bit better

[brittle rampart]

#🍥｜anime message

[modest anvil]

yea tried that too

[wispy canopy]

looking at viewer in negatives, do it

[nova remnant]

that sounds promising what you could do

[brittle rampart]

what the promt you have?

[modest anvil]

eyes looking right,golden helmet,(girl) with warrior armor close up, looking right, galactic space, blue star glowing eyes,cute lips

rn\

[chrome mango]

Hello everyone, I've been using SuperMerge to create an anime model lately. However, I've noticed that my merged model consistently draws hands poorly. Can anyone recommend a model that excels in drawing hands?

[modest anvil]

who knows ill try tommrow when i wake up

[nova remnant]

if that doesnt work, try experimenting with weights, something like (looking at her side:1.3)

[brittle rampart]

SD draw eyes lookig right...

[final vault]

which model have you used for the 1st image??

[wispy canopy]

brightprotonuke 1.3

https://civitai.com/models/148871

[nova remnant]

I heard about that model

It's interesting

lol its good, cat

[wispy canopy]

xD

[nova remnant]

Whoever made it, must have taken a long time merging and be really old by now

Really old

[final vault]

thanks

[nova remnant]

lol dunno but that model is pretty new and nice 🙂

[native halo]

kakarot and hotarubreed have decent hands

[nova remnant]

cat typing gif

[supple raptor]

The vae you're looking at in that repo is in diffusers format so it won't work normally. I assume the model should have a VAE baked in, otherwise try using the normal SDXL one.

[chrome mango]

thanks! will check it out

[nova remnant]

ok but the problem i had was it failed to load the model even

this is the model i was trying with segmind-vega

[supple raptor]

Ah, in that case it might not be compatible with A1111.

[wispy canopy]

yeah, i don't think it'll load in a1111 -> that one doesn't have diffusors working

you could use sd.next, i think that one supports diffusors

[nova remnant]

yeah probably, they have a 2 month old model that is a1111 supported but i was curious about the recent release

[wispy canopy]

or comfy

(it does work in comfy, confirmed by me)

[supple raptor]

There's a compvis format .safetensors model that would theoretically work if the config was present in A1111, but i'm not sure if it is.

[nova remnant]

i got to intall the diffuser tho via pip install

not sure if that is enough to run it

[wispy canopy]

but a1111 doesn't support afaik

[supple raptor]

Yeah, we use some diffusers libs, but not the diffusers pipeline.

[nova remnant]

ahh got it

[final vault]

Do you still have the kenshi model?

[supple raptor]

https://huggingface.co/SweetLuna/Kenshi

I haven't used that one in ages.

[nova remnant]

Luna

[final vault]

I know
I saved your message a year ago (7th Jan 2023) where you created a mix with kenshi so decided to give it a go

[final vault]

I miss luna, sirveggie, pendragon

[nova remnant]

that kenshi model looks really good by its cover image

[supple raptor]

Luna's been busy with life. Veggie still posts, just not as much in this server. Pendragon has been mostly quiet for a while but he pops up now and again.

[supple raptor]

It's the precursor to AuroraONE. It was good for it's time, but Aurora is pretty well a straight upgrade.

[nova remnant]

cool, i will have to try them out

Aurora best model

followed luna on civitai for any updates in the future

[brittle rampart]

how to make Adetailer recognize areas containing faces? For some reason, it recognizes faces where there are none at all.

[vital raptor]

i think you change the detection confidence threshold

[brittle rampart]

this happens at the default value of 0.3

[vital raptor]

there's also the 'mask only top k largest' if you set at 1 (or how many faces you have) it should stop at the best matches and not check for other things

[supple raptor]

1. Make sure you're using face_yolov8n. It's the most accurate.
2. If you're only trying to gen a single person then you can set it to only mask the largest face. (Mask only top k largest)
3. Slightly increase the confidence threshold.

[brittle rampart]

@supple raptor @vital raptor Mask only top k largest = number of faces I want to be recognized?

[supple raptor]

Yep

[brittle rampart]

@supple raptor oki-doki... what about "Detection model confidence threshold"?

[supple raptor]

Higher number = model has to be more confident that a thing is a face.
Adjust in very small increments.

[brittle rampart]

@supple raptor @vital raptor thanks. i try.

God knows what! No matter how I twist it, it still recognizes a face! This happens when using the Scaling script

[vital raptor]

you have a typo in your detection confidence, it should be a decimal (.) not a comma (,)

also you might need to raise it further, when the detection script runs you should see boxes appear around the detected area with a number, make the detection confidence higher than the fake face as long as it's not also higher than the actual face in the image

[brittle rampart]

wow... My regional settings imply comma(,) separation of integer and fractional parts. Do you think the problems might be related to this?

[vital raptor]

i'm not sure on that, i don't know how deep the localization code is for that type of thing

[brittle rampart]

right. detection areas are highlighted.

[vital raptor]

but the higher value may be needed at least, even if the comma works. This image for example detects the face at 0.87 if i set the detection confidence to 0.87 it still works, if i set it to 0.88 it won't mask the face

[brittle rampart]

those. next to the squares it shows with what accuracy the face was identified in this area?!?

[vital raptor]

when it highlights the area during gen it should show the number, that's the confidence number for that spot

[brittle rampart]

didn't know about this.thanks

[brittle rampart]

Oh! I figured it out... on top of everything else, when upscaling I overestimated Denoising strength the overall value and the face in that area was generated even before Adeteiler

[brittle rampart]

really?

[native halo]

https://tenor.com/view/ez-noob-skill-issue-simply-a-difference-in-skill-gif-20780203

[brittle rampart]

everyone says otherwise

for SDXL?

maybe the nodes are not configured correctly

nodes that are responsible for samplers

[native halo]

try asking in sdxl,theres a lot of comfy users there

[brittle rampart]

I'm thinking of making settings that meet the requirements. try using a different sampler. lower the values.

Based on these numbers, I think the guys will tell you more. For me, as a girl, these numbers are a mystery. 🤪

sorry....

[native halo]

sorry no comfy support here

try to ask in #✨｜sdxl almost everyone there uses comfy cause sdxl on a111 sucks

[brittle rampart]

cute!!!!! but... I can’t “pull out” the illustration because of the extra drawn hands... SAAAAD!!!!

[steady grail]

charming

there is this ai to cancel very easily and in an inconspicuous way some details calle lama. usually works as a charm

you shouuld try it

[native halo]

lorenzo lama?

[steady grail]

i don't remember, on the directory i only have lama

my lazy way of naming stuff always backfires me XD

[brittle rampart]

@steady grail @native halo Can tell me more about this lama?

[steady grail]

you cover stuff like with inpaint and it automatically delete it, using what's around as a sample

[native halo]

Lorenzo Lama is an American actor. He is widely known for his role of Lance Cumson, the irresponsible grandson of Angela Channing—played by Jane Wyman—in the soap opera Falcon Crest, for which he received a Golden Globe Award nomination for Best Supporting Actor – Series, Miniseries or Television Film.

[brittle rampart]

👍

[steady grail]

yeah. this stuff send a queue to this lorenzo so that he can repaint the weird details. clear, right?

[brittle rampart]

in ipaint i am only tweak some things but never tried to remove them completely

[steady grail]

is totally another app with his gui and model

it's just a comparison

[brittle rampart]

another ap!?

[steady grail]

application

[brittle rampart]

this is not SD? Now I'm completely confused...

[steady grail]

not idea what the model is based on. but it have nothing to do with auto1111 or comfy

it's a stand alone

[brittle rampart]

link?

[steady grail]

i need to look for it, i downloaded it years ago

[brittle rampart]

oki-doki... thanks

[zinc flax]

Squish

[brittle rampart]

I couldn't resist☺️

[summer night]

novelAI v3 is so cool

it can do prompts of more underrated characters and doesn't even need loras

[silk narwhal]

probably uses a database with thousands of Lora's and apply them from the prompt given.. closed source and no process shown .. so who knows what kind of tricky stuff they apply as it is online only

[bitter latch]

hmm ok what is with my auto1111 today? it's wanting to show me boobs, even when I put nsfw in neg and covered in pos. wtf. I mean I don't mind but I can't post that here

per server rules

hmm ya know I bet it's the "kemono style"

[steady grail]

changed model?

[bitter latch]

finally

nah same model I've been using

renamed it tho that shouldn't change it

[steady grail]

supposed something messed up nega

are you using xl?

[bitter latch]

hell no

[silk narwhal]

not what i expected but i take it 😄

[bitter latch]

BOOM!

[unreal ridge]

Dorothy, nice 😊

[summer night]

is it fine to post girls in bikinis if there is no sexual nature to the image? for example a girl just chilling on the beach

[unreal ridge]

I think so

[summer night]

i guess we'll see

[unreal ridge]

Is she from Blue Archive?

I swear I've seen her before

[zinc flax]

https://tenor.com/view/kayoko-blue-archive-new-year-gif-27711546

Cute

[summer night]

ya

[summer night]

novelAI is awesome but god it's expensive
i hope local models can catch up sooner

[summer night]

can you run SDXL on 8 GB VRAM cards if you set --lowvram?

or would just --medvram be enough?

[bitter latch]

this whole time I meant to do genshin, but it's genshi.. it works

[broken osprey]

Hey ppl what is the appropriate size to create anime images, should I follow the model base 512512 or use a standard like 512896 ?

[summer night]

most commonly i use 512x768 for portrait and 768x512 for landscape

i think one of the 2 dimensions should always be the base resolution

[broken osprey]

Okeee okeee and upscale of 2.0 or of 1.5?

[steady grail]

for experience, i respect a ration divisible for 256

but that's just how i do stuff

[wraith raptor]

You can do whatever, just make sure you the original size before upscales respect that you are under 768

[unreal ridge]

What's up with the random 💐 attached to her chest?

[supple raptor]

I almost always gen at 960x544 or 544x960. It's low enough res to not cause anatomy or duplication most of the time on most models, but high enough to be pretty good quality. Then once it passes through Hires fix at 2x upscaling it leaves me with a 1920x1088 or 1088x1920 image which is almost exactly 1080p.

[broken osprey]

It’s pretty tho <3

[supple raptor]

flower censor

[unreal ridge]

😔

[broken osprey]

Did you use latent couple for the background or is that just a simple prompt?

[unreal ridge]

don't even know what that is. Just prompts

[broken osprey]

Huh ?!!!!! So a literal prompt did that? Where did you learn to prompt that well

[subtle thistle]

Can someone help me with my prompt vocabulary? I have been prompting for anime girl or similar and usually wind up wiht two-different styles of art. I am insufficiently skilled to be able to determine which of the two is which style.
One is more spritely and one is more formal, I suppose.

I feel like there is a subset of anime styles in here that is beyond my ken.

[unreal ridge]

Idk, mostly myself and seeing how people prompt in different servers (Mostly here and the Japanese SD server). I've been doing this ever since NAIv1 was peak, so I guess there's that too.

[subtle thistle]

In particular my prompt for the top pair was Hero view, soft focus, anime face. A plus sized, short, female, gnome chef, green curly hair with leaves in it. Wearing a dull green smock. Using an iron skillet holding a knife on a galley aboard a pirate ship. Style: Anime

[broken osprey]

And what is that? NAIv1 is that a model ? Also damn I wanna see that jap sd server smm probably have loads of gud stuff to steal prompts from I mean Learn

[unreal ridge]

That server is more dead than this one nowadays. But I'll send you the invite.

[broken osprey]

The k you

Also again….. what is that NAVv1 ?

[steady grail]

Hero view, soft focus, A plus sized. first time i hear about them

[unreal ridge]

The first model Novel AI released

[broken osprey]

And where can I get that?

And are there new ones?

[unreal ridge]

also the model that gave origin to most 1.x anime models, cuz it was leaked 🤷‍♂️

[unreal ridge]

Yep, there's a paid subscription tho

[broken osprey]

I have to pay monthly for a model? I can’t get it lifetime ?

[unreal ridge]

Nope

V3 is probably the best XL anime model by far

[broken osprey]

Damn so you use it for most of your work

[unreal ridge]

nah

[steady grail]

but nowadays the model is very dated

[unreal ridge]

I don't use NAI

[steady grail]

just my opinion, it was a little dry

ecceptional, just for a starting point

[broken osprey]

Oh I see. I remember you said you had personal models so I’m guessing you have great personal-fine tuned models

[nova remnant]

wish i could animate this and add music for new year

[summer night]

hamburger

[nova remnant]

@wispy canopy in response to my question about segmind vega support for a1111 ...

You can convert to a single safetensors to do that: https://github.com/huggingface/diffusers/blob/main/scripts/convert_diffusers_to_original_sdxl.py. You might have to repurpose the script a bit to make it seamless with VEGA.

i have no idea what it means to repurpose a script tho

[brittle rampart]

but it's interesting...
which is more acceptable (or rather correctly said):

1. Generate low resolution. We repeat the option we like using the Hires.Fix procedure. Then we do Upscaling to favorite size.
   2 Generate low resolution. We “drive” the option we like by gradual Upscaling to our favorite size.

[nova remnant]

❤️

Can I get the prompt for that

[zinc flax]

heart hands, camisole, :d, white background, pink hair, blunt bangs, blue eyes, short hair

[nova remnant]

i got this from that prompt...

[zinc flax]

I guess it's heavily model dependant

[nova remnant]

most likely

[zinc flax]

Especially with the heart hands

[nova remnant]

yeah and im thinking of going to bed now so dont have too much energy to work on it atm

[magic wave]

V1 of a new style, kinda unstable
vs
V2 of that same style after changing the dataset a bit, more stable + hair color issues fixed

[zinc flax]

booba

[magic wave]

Booba indeed

[zinc flax]

Nice

[brittle rampart]

yep. but without hires.fix. just fast try...

[vital raptor]

it's too bad that model isn't available to download, it's super cute

mine has a bit of trouble with the heart hands, but it works sometimes

[brittle rampart]

and again I couldn’t resist...

[vital raptor]

ai generated ai..

[lyric zealot]

[brittle rampart]

how serious she is.. Tsunderes are scary!

[zinc flax]

Cuute

[lyric zealot]

[broken osprey]

May I ask for the prompt on that one please?

[unreal ridge]

You'll need a Lora for this one

https://tusiart.com/models/605586978534138789
and iirc, you can't download it without an account

and you need a chinese number to create an account

cuz china is cool

Feel free to try without it tho
1girl, solo, white hair, blue eyes, flowers, loli, glowing flowers, bioluminescent flowers, full body, (goth lolita, short dress, frilly dress, detached sleeves), dark forest, scenery, <lora:梦之花:0.75>

[broken osprey]

I’ll use my WeChat like I did for liblib Ai

[stable crane]

very nice white tiger

[zinc flax]

Super cute

[untold glacier]

I was saying to myself, oh that's a great gen, very nice and normal. Then I realized it was Astolfo.

[summer night]

AND?

he cute

[vital raptor]

astolfo best girl from that series

except rem

[summer night]

wrong series

astolfo is from fate, rem is from re:zero

[vital raptor]

right, I was thinking of Felix from re zero

[unreal ridge]

What's wrong with Astolfo? 🔫

[native halo]

he appears in Fate/Apocrypha thats the worst part

[summer night]

apocrypha has so many banger waifus though

jeanne, mordred, astolfo

i want to get the astolfo casual figma but they're so damn expensive...

[unreal ridge]

You can always play FGO and make it way worse 😄

[native halo]

only play for Shuten Douji

[unreal ridge]

Extremely based

her voice is 🤌

[nova remnant]

[untold glacier]

https://tenor.com/view/rukako-ruka-urushibara-but-he-is-a-guy-steins-gate-gif-20673376

Also @summer night ^

[nova remnant]

[lyric zealot]

[nova remnant]

@stable crane many of you

[nova remnant]

@native halo

[maiden rose]

https://cdn.discordapp.com/attachments/1157939799433429035/1190604107610849331/00026-1631696089.png?ex=65a2677b&is=658ff27b&hm=58faf8739fe9bfd1ae2f1e6e0c6a8a8cde95a9a37199af3d998400d903c8462b&

[near elbow]

a dual blade katana... not so practical i guess

[bitter latch]

it tried

htf does LCM get the holding right but screw the rest?

lower settings.. closer

the dark side is strong with this one

that hilt.. hmmm

well, I guess thats one way to do it, katana at one end , light saber at other

[silk narwhal]

maybe needs Sword,Katana etc in negatives so it will not use every single type of Sword/Saber 😄

[keen hollow]

Just an assortment of generations this morning.

And some more.

[steady grail]

his voice was fantastic

otoko Da!

[untold glacier]

https://tenor.com/view/steins-gate-steins-gate-makise-kurisu-gif-4010510076717045510

[gentle musk]

Been playing around with the new IPAdapter FaceID Plus v2. It works pretty well.

[keen hollow]

Karen!

More Karen.

Also, what is faceid plus v2? 😛

[gentle musk]

https://github.com/cubiq/ComfyUI_IPAdapter_plus
https://huggingface.co/h94/IP-Adapter-FaceID/tree/main

That's the ComfyUI Node and the models

It's only for SD1.5 at the moment

[keen hollow]

I'll have to scope it out, though I'm an Auto user.

[gentle musk]

Probably doesn't work in Auto, as that ComfyUI node only got updated today to make it work.

[broken osprey]

May I ask how you got the character to stand so far away form the viewing angle, did you use a Lora? Or was it the prompt( specific keyword for this) or control net?

[lyric zealot]

Don’t think I did for that pic, it was mostly just full body

And no controlnet

[broken osprey]

:0

[lyric zealot]

I was just randomly genning

And I’m not on computer so can’t share metadata

Sorry about that

[broken osprey]

Oh I was just about to ask that lmao

Well I’ll try to get it closer or download the image and get the metadata

[lyric zealot]

Imma be honest, it’s probably something like 1girl, miku, field, full body, cloud, sky

Something along those lines

Good luck

[broken osprey]

Thenk you! I’ll try to replicate it even with more words. Or perhaps I get most images to be close as hell since the models were trained like that

But in any case thanks for the info

[lyric zealot]

If you are looking to generate the subject a little further away

Maybe try wide shot, wide angle

Or even zoomed out

And full body

Then describe the background a little more

[broken osprey]

Oke oke I’ll check them out in a bit !! ^^

[brittle rampart]

why not just skethching?

Is everything new the well-forgotten old?

[gentle musk]

Because when you do that your just copying the initial image. What if you want a completely different pose?

It works even better with photorealistic models, but it works pretty well with anime as well. You can have whatever image you want to generate with whatever face.

[brittle rampart]

and what prevents you from using controlnet?

[gentle musk]

But what if you don't want to. With this you can just generate an image as normal and assign a face you want it to have.

[brittle rampart]

don't want a pose. use the same sketch

or am I missing the point?

[gentle musk]

I'm not sure you are understanding.

If you are using Controlnet then you are restricted to whatever image you place in Controlnet.

What if you don't have an initial image, what if you just want to straight up generate a new image with that face. With this you can do it, you don't need an initial image.

It's mostly designed to be used with photorealistic stuff, for face swapping. I just liked the affect you can get with anime generations. It still looks anime enough, but it also looks like the face.

[brittle rampart]

sounds interesting. I don't have the original image. but I want to create an image with this face!? in your version, as well as in mine, there is a certain original image. But I touched on the controlnet regarding only the pose.

sketching+controlnet pose?1 why not?!

[gentle musk]

Yeah you can use that if you want as well.

What this does it is looks for a face in the image and will then apply the selected face. Works with just an image generation or with an img2img.

[brittle rampart]

"Only works with image creation or img2img." But what is different in your case?

[gentle musk]

I'm not sure I understand what you are asking

[brittle rampart]

I don’t fully understand what you showed?

i am so sorry

[gentle musk]

Have a look at the documentation here - https://github.com/tencent-ailab/IP-Adapter

I don't know how else to demonstrate what it's doing.

[brittle rampart]

OK. can I just transfer the face onto my image?

[gentle musk]

Thats one of the things that it tries to do yes

You can take an existing image and then use Img2img with that IPAdapter and it will try put the selected face onto the image.

[brittle rampart]

but then your ostentatious examples do not quite reflect this. I am of course very sorry for such a critical remark

[gentle musk]

You can't see the facial resemblence between the source image and the generated image?

[brittle rampart]

yes

[gentle musk]

Not sure what to say then, it looks pretty clear to me.

It will never be exact with an anime style, because that would break the style, but it's pretty close. It gets it a lot closer with a photorealistic image.

[brittle rampart]

for me it looks exactly like generating an image through sketching or imgtoimg

I'm not talking about style.

[gentle musk]

The images I posted are different to the source images, so how could it be sketch or imgtoimg?

[brittle rampart]

I'm talking about image generation using methods based on basic images

[gentle musk]

You're talking about Controlnets or img2img though, which copy the exact image you input. This is calculating features of the face and then using that in the image generation.

[brittle rampart]

to me it looks exactly like generating an image based on a basic one using methods img-to-img or sketching

[gentle musk]

I really don't understand what you mean. If you use img2img or sketching then the resulting generated image is going to look pretty much exactly like the source image in size and shape.

[brittle rampart]

I'm not talking about size or shape. I'm talking about content

[gentle musk]

Like with img2img or sketching you can't take this image

And then generate this

[brittle rampart]

nope

[gentle musk]

or this

Both made with that same source face image

[brittle rampart]

for this

i generate this

just fast

[gentle musk]

Yes? I really don't understand the point you are trying to put across here.

The thing I'm showing takes the face from the source image and then you can generate whatever you want from it. You aren't restricted to just changing that source image, which is what you are showing.

[brittle rampart]

I just asked about the essence of using what you mentioned in the images shown.

[gentle musk]

Yeah I don't understand what you mean, sorry

[brittle rampart]

hmmmm

oki doki

soryy

[native halo]

the difference is that with ipadapter you can create the same face structure with different poses,if u just use a photo like u did it will be stuck with same pose and same outfit but with ipadapter it can create the girl in different angles,outfits,lightning etc,etc, but for anime style celebrities it sucks,its better suited when u wanna add certain things to an image by using an exisisting photo of the object u wanna add to the generated pic,for celebrities in anime style its better to just use a celebrity face lora + an anime model

[keen hollow]

aww bummer. Hopefully we'll get an Auto version. This is neat.

seems like so many folks jumping over to comfy though

just feels like it takes 9 more steps to do something in comfy 😛

[native halo]

auto still more popular

[broken osprey]

Comfy seems faster and more stable than webui at least with low to mid vram

[native halo]

yea with sdxl the speed difference is night and day,for 1.5 the difference is not that big

[worthy delta]

had to reset my pc, how do i get a111 again?

[broken osprey]

Yeah but for amd and my 2019 card I get considerable less errors in comfy and faster inference for sd1.5 too

[broken osprey]

Depends on your graphics card vendor, if nvidia just look for the Normal automatic 111 GitHub and follow steps

[native halo]

oh yea for amd sucks 😔

[worthy delta]

3060

re got it but lost all my lora and stuff

[broken osprey]

Yeah just look normal GitHub on google

[broken osprey]

Ooof

[worthy delta]

do you know the thing that lets me auto prompt stuff

type out a couple letters and it shows what it could be?

@supple raptor need some help bro lost everything had to wipe my pc

[native halo]

check the guide on #🤝｜tech-support

[worthy delta]

theres alot ive lost god damn

[broken osprey]

Dambooru auto complete is the extension but yeah #🤝｜tech-support and #📝｜prompting-help for further assurance

[broken osprey]

Civit Ai keeps a download history so just go there

[supple raptor]

https://github.com/DominikDoom/a1111-sd-webui-tagcomplete.git

[worthy delta]

thank god

[worthy delta]

ayee thanks nubby

how do i put it in again, god this is gonna be a pain

[supple raptor]

go to extensions and install from url, paste that in the url field and install.

[worthy delta]

will do after i figure out how to get user.bat back cause it aint there

[nova remnant]

https://tenor.com/view/fast-cat-typing-gif-25546751

[native halo]

https://tenor.com/view/mike-o'hearn-lucaskäck-brainfog-anime-gif-11872632083823530698

[keen hollow]

Today's Tifa

[nova remnant]

Booba

[worthy delta]

YOOO CAT

My pc has froze idk why but it did

Imma leave it cause I clicked on webui so hope it dosent blow up

[nova remnant]

Have I talked to you before

[worthy delta]

You... forgot me??

😢

[sonic bolt]

yo

should i use automaticc1111

or sdxl

[nova remnant]

https://tenor.com/view/cat-death-lol-gif-20542436

[nova remnant]

SDXL is a bigger model

A1111 is the UI to run models

[nova remnant]

Yes but with the power of technology I looked back in chat

It's been many days lmao

[sonic bolt]

This is the reason my father tried to kill me when I was born

I'm so stupid

[broken osprey]

not sure what she is supposed to be holding but daym

[gentle musk]

Neither am I lmao

[broken osprey]

lmao!!. Hey you posted about ipadapter plus right?

[gentle musk]

The FaceID Plus one yeah

[broken osprey]

aaaah isee Im trying to install it rn and was having some trouble with the face id thingies.

That dependency that it requires is it something like a pip install?

[gentle musk]

Yeah you need to pip install insightface and it's dependencies.

[broken osprey]

ah okay I'll do that rn, the clip vision models are HUGE

[gentle musk]

it's only 2.5GB

That's nothing if you've been using SDXL models lol

[broken osprey]

I cant use XL models lol, 8gb vram struggles with sd1.5 i dont even want to know what will happen with sdxl

[gentle musk]

You should be able to

I have 10GB VRAM and I can use XL Models Plus IPAdapters and Controlnets all together

[native halo]

is 10 bigger than 8

[broken osprey]

ugghhhhhhhhhh I need to reinstall fricking microsoft visual 14 >:C I uninstalled it since i moved to clion god damn, ill do it later

[gentle musk]

Yeah but it's not that much bigger

[native halo]

you also need ram specially if u use a1111

[broken osprey]

yeah but not too much, the main difference will go down to graphics card

[gentle musk]

Yeah I know the RAM, I had 16GB before and it kept trying to crash my PC even with Comfyui

[native halo]

a1111 sucks with xl 😔

[gentle musk]

Ive not used it in ages, Comfyui just works better and it updates a lot quicker

Some of the things like ADetailer though can be a bit of a pain in the ass with ComfyUI though

[broken osprey]

really 16 gb crashed comfy? Im running 16gb and mostly a111 struggles, well with amd it struggles with everything, but the ui was laggy as fuck

[gentle musk]

It was using so much RAM it started hitting the page file, which makes it start to freeze

[broken osprey]

there are some custom nodes for that I use, but its so tedious to set up I get tired of comfy so much

[gentle musk]

This was months ago though when SDXL first came out, it's been improved a fair bit since then.

[native halo]

if u use xl turbo its prob better

[sonic bolt]

Where do I put the sdxl ŕefiner

[native halo]

same folder where u put the sdxl model

[sonic bolt]

Thx

[nova remnant]

Robot future soon

[native halo]

where bodysuit?

https://tenor.com/view/discord-who-asked-me-looking-for-who-asked-travolta-meme-travolta-confused-gif-27133622

[nova remnant]

Hiring friendly robots to find them

[nova remnant]

Play with her

[nova remnant]

[native halo]

(latex:2.5)

[chrome moss]

Does anyone know what model this image uses?

[nova remnant]

Try looking for cute yuki and it's many variations

Let me check another one that might be similar

[zinc flax]

Don't think it's cuteyuki

[nova remnant]

I mean

There's so much you can do with one model too

Unless you get it exactly from the person who generated the image, it's really hard to pin point what it is, if it's not just using a model with basic prompting

[zinc flax]

It's definitely hard to guess a model without any metadata lol

It could be anything v5 mix number 10

• yet another detail lora

[bitter latch]

sigh hate it when i can't sleep

dangit

so merged my merge and...

hmm not bad

guess everyone's asleep, like i should be

sighs

[steady grail]

a lora? or just prompt? very cute

in my opinion, this is a little borderline

[native halo]

https://tenor.com/view/noted-gif-20569892

[bitter latch]

how? huh?

no lora

[steady grail]

wow, the colors are very spectacular

[native halo]

the best part is when u deleted it

[steady grail]

lol

[native halo]

😔

[bitter latch]

well that one yeah I can see it

[steady grail]

looks like an xl model

[bitter latch]

the others are washed out drawn pastel style like in older style pinup cards style

nope isn't

[steady grail]

every details is so clean

[bitter latch]

oh my bad there is a lora in there, mdjrny 4 lora

and thats also using a merged model with rpg 3d Style. that might help it

[steady grail]

expected so, the was some coherence in every image

[bitter latch]

well that and didn't change much between renders

[steady grail]

without a lora, it's hard to maintain some details between gens

[bitter latch]

sometimes

with a lora

saved style plus the lora

hmm ok point made

[worthy delta]

hello fellas

[bitter latch]

morning

oh that arm. wtf

think i'm finally getting tired enough to sleep

[worthy delta]

boob

[bitter latch]

goes to bed

[worthy delta]

where do i put loras again dear god

[silk narwhal]

models/Lora

[worthy delta]

thanks got it

he is handsome aughghapojfpo

[worthy delta]

just figured out why it kept doing same position

it was locked onto the same seed

why does the female version look better than my boy

[nova remnant]

inpainting myself with 0.5 denoise on BluePencilXL v2.9

forgetting to prompt for 'man' and making myself into a waifu where the AI turned my beard into a choker lol

[worthy delta]

damn

dat you?

[worthy delta]

booba

[nova remnant]

yup lol

[nova remnant]

BluePencilXL v3.0 (and LCM) dropped a few hours ago! https://huggingface.co/bluepen5805/blue_pencil-XL/tree/main

BluePencilXL v3.0 test

[silk narwhal]

Positive Latent prompt blending test

[nova remnant]

BluePencilXL v3 does slightly better at faces even when farther from the camera

hmm the unaestheticXL_Sky3.1 gives more realistic color but dang the hk1 variant gives way more bright colors

[nova remnant]

still prefer sky3.1 tho

[stable crane]

Happy New Years!

[nova remnant]

Needs egg on head

[worthy delta]

gimme some nature prompts

[nova remnant]

(nature, nightmare:1.3)

[worthy delta]

nightmare??

[nova remnant]

Yes

[worthy delta]

only reason i got back into ai art was because i wanted to make stuff for my dnd character

[steady grail]

yeah, i suppose with ai, you can give a cool flavor to the story

too bad i never had chance to play it

[worthy delta]

its hella fun

this is my guy

[broken osprey]

What if we all try to comu up with a visual story told through ai images, like a little test to the imagination :)

[worthy delta]

Ooooo

hot lady

1st place

[broken osprey]

lmao

[worthy delta]

anyone know what this art style is?

[broken osprey]

What happens if I send a prompt for hands into the face yolo model :0

[nova remnant]

@worthy delta this just the usual park, nature background but with nightmare

[worthy delta]

goth momma

[broken osprey]

Really? The Ai came up with that just through the keyword nightmare?!

[nova remnant]

Yes

AI does AI things

[broken osprey]

Scary yet howt

[nova remnant]

[broken osprey]

It knew what you wanted deep down

[worthy delta]

a hot tall goth momma

[broken osprey]

Yes

[worthy delta]

tis why ai exists

for hot women

[broken osprey]

Or is it one of those models that are mixes of a bunch and are amazing at creating women and nothing else

[worthy delta]

this failed badly

[broken osprey]

Not sure if that is sarcasm buttttt I have to agree

[worthy delta]

it aint sarcasm

[broken osprey]

[broken osprey]

What’s the prompt lemme try :)

[nova remnant]

It's pikasnewgen model, really good at many things

[worthy delta]

a closeup of a skeleton holding a sword, skeleton, dead, holding weapon, broken armor, nature, vines, vibrant colours, pink flowers, tarot card enviroment, (masterpiece, best quality:1)

[broken osprey]

Alright alright imma take a look

[broken osprey]

Well it didn’t pick up on the broke armor hah

Let me try

[worthy delta]

yeah idk why

[nova remnant]

@worthy delta do you like cute robots

[worthy delta]

yes

[nova remnant]

glowing eyes with curiosity

[worthy delta]

itll crush me

the anatomy is fucked up

[broken osprey]

Okay sent in the prompt let my tile upscale finish and I’ll give you the results

[worthy delta]

yes sir

idk what tile is

[broken osprey]

Bro? How the hell

[broken osprey]

Ah im using a custom node o found for comfy that up scales the image through tiles

[worthy delta]

whats comfy

[broken osprey]

Another ui