🍥｜anime

[worthy delta]

gotcha gotcha

i wanna try n make this

it aint ai

but i wanna make it bro

i should get photoshop again

and edit them

[broken osprey]

Describe it in natural words first and then use keywords

[worthy delta]

pink samurai

[broken osprey]

more detail

[worthy delta]

neon, pink, cyberpunk, samurai

neon, kimono, cyberounk, samurai, mulitple sword, folded arms, oni mask,

[broken osprey]

Try that :)

[worthy delta]

yes sirrr

skeleton one was a flop

[nova remnant]

Pinker samurai

[nova remnant]

[worthy delta]

pinkest samurai

now this is one i gotta make

Spacepunk, armor, medieval armor, futuristic

[broken osprey]

@worthy delta got something, kind of appropiate lmao

[worthy delta]

stanky leg

[broken osprey]

yeah lmao im changing model to see if revanimated works better

[worthy delta]

smash

next

[worthy delta]

thoughts abt this one boys

[broken osprey]

@worthy delta got u

somewhat

[worthy delta]

not bad

[worthy delta]

this one goes hard

[nova remnant]

it is a real good prompt

the spacepunk one is fun too tho

[worthy delta]

the power of my uber ai smarts

[broken osprey]

wait which one are we trying rn, i got stuck on the skeleton lmao

[worthy delta]

steampunk, smaurai, airship, rustic, rugged, 1boy

this one is

[broken osprey]

steampunk? damn Ill give it a try once this one finishes

[worthy delta]

1girl, mature female, japanese armor, steampunk, (masterpiece, best quality:1)

[worthy delta]

SHEESH

try this one

1girl, mature female, japanese armor, steampunk, smaurai, airship, rustic, rugged, (masterpiece, best quality:1)

okay this prompt is cooking

[broken osprey]

damn got this for the prompt :c

[worthy delta]

mine still in the oven

[silk narwhal]

happy new year all

[worthy delta]

happy new year

@broken osprey

[broken osprey]

I think its the model let me try again wiht another

it got this for the last prompt lie ew

[worthy delta]

ewww

man wtf is ur model

[nova remnant]

that hat is sick tho

Is it too cute ?

[worthy delta]

Gothic, steampunk, futuristic, mature, rugged, rustic, vibrant (masterpiece, best quality:1)

[worthy delta]

lads

this

IS PERFECT

@broken osprey

[broken osprey]

i cant generate a good image with your prompts lmao

but it does look good

@worthy delta got this

[broken osprey]

is that an anime figure style hypernetwork?

[worthy delta]

lads

need some help with prompts for a viking assassin

[broken osprey]

so cute i feel threatened by that eye

[silk narwhal]

pvc figurine model SD1.5 model

[broken osprey]

:0 scary

[silk narwhal]

really awesome when you blend it with other anime models in comfy as it pulls the realistic shading a bit more forward

[broken osprey]

hmmmm I guess so..... I have never used model blending tbh, idek what it is

[worthy delta]

1girl, mature female, mole, abs, blond hair, brown eye, beautiful, gorgeous, face scar, blood on clothes, blood on face, barbarian ,viking tattoos, viking clothes, clothed, (masterpiece, best quality:1)

[silk narwhal]

3 more hours and it is already a warzone outside 😄

[worthy delta]

whats a good model for good style but also brings out brutality?

[broken osprey]

no idea, havent even thought about generation....brutality lmao

[worthy delta]

im doing about a viking woman

for a friends dnd character

[broken osprey]

@worthy delta now it worked

my models require a little bit more description lmao

[worthy delta]

looking good

[broken osprey]

imma try the viking one in a second

[worthy delta]

yes sirrr

[broken osprey]

@worthy delta better quality

Tbh this is the first time I try this model and its looking good and the face is well defined

[worthy delta]

i have helped

[silk narwhal]

:/

[broken osprey]

🙏 @worthy delta here you go

[worthy delta]

hawt

needs more clothes

[broken osprey]

more?

[worthy delta]

more viking less anime girl

[broken osprey]

i figured you would say less

[worthy delta]

normally i would

[broken osprey]

let me see if i can add some kind of armor plating to her chest and arms, maybe change the face by using other keywords for age

[worthy delta]

vikings didnt use metal armor

[broken osprey]

now now im not a historian leather then?

now that is a figure I would buy

although...... does she have four ears?!

[worthy delta]

ye ye

[worthy delta]

its the hoodie fr fr

[broken osprey]

@worthy delta got this for armor, on the arms and a little bit more clothing beneath, while retaining the abs 🤤

[worthy delta]

MAMAAAAAA

[broken osprey]

Want me to upscale it?

[broken osprey]

okay now now..... this os not the chat for teasing lmao. but like hot

[worthy delta]

i got some very teasing ones

of the gal i did earlier

this baddie in particular

[silk narwhal]

lol.. that is the most covered one out of the 8 batch 😄

[worthy delta]

send me the in dm

i gotta see dawg

[broken osprey]

LOL, may i ask for the others please? 😉

[worthy delta]

lets just make a group atp

[broken osprey]

bro i am trying to run upscale and ugh its giving errors for some reason :c what does even tuple object has no attribute 'enconde' mean :C

[worthy delta]

idk dawg

i just genetate anime booba

generate*

[broken osprey]

most honest individual that uses stable diffusion

[silk narwhal]

change the upscale model usually helps

[broken osprey]

NOOOOOOOO you got here too late the image is gone

[worthy delta]

i can send you the hella good booba ones iyw

[broken osprey]

here

[silk narwhal]

blergh

SDXL generating 1536x768 on the 3060 is about as fast as generating 512x512 on my 1070 :

Loading 1 new model
37%|██████████████████████████████ | 11/30 [00:59<01:43, 5.43s/it]

[broken osprey]

now the ai made it lewder :c

[broken osprey]

thats why I dont use sdxl !! long live sd1.5

[worthy delta]

momma

but @broken osprey wanna see the spicy ones

[broken osprey]

@worthy delta better

[broken osprey]

yes

[silk narwhal]

cursed 😄

[broken osprey]

when that latent blending doesnt work out

[worthy delta]

dayum a reflection???

[silk narwhal]

like generating more concept art images .. doesn't matter if they are not super clean or sharp.. as long as the idea comes across 😄

[broken osprey]

cuteeeeeee is that a private model?

[worthy delta]

adorable

[bitter latch]

like a croc-a-fox?

[worthy delta]

wait you can make ur own model?

[bitter latch]

hmm ok the fox didn't come out right

no just a new merge of my fuzzy anime with the azoyrpg and a nms (no man sky model) and a style I made

[bitter latch]

so yeah a lot of merges 😄

[broken osprey]

how did you do that?! ipadapter or smth?

[bitter latch]

and there's a Mdjourney style lora in there that helps

[silk narwhal]

just a workflow 🙂

[broken osprey]

so that is fox, lizard, dragon?

[broken osprey]

may i get the workflow pwease?

[bitter latch]

I call it a Croc-a-Fox

[broken osprey]

what :0 json? damn

[silk narwhal]

is for comfyUI

you can also save and drag the image on it.. has the workflow embedded

[bitter latch]

croc-a-panda

[broken osprey]

oh yeah sorry I forgot about that :c

[bitter latch]

I l ike combining species, styles and such

[steady grail]

even if i'm not in furry, they are pretty cute

[bitter latch]

all furry is different type of cartoon, anime style animals

been around before people named it

she looks cute but don't make her mad she's got claws 😄

and somehow I always go back to miku

[broken osprey]

now i need that model @bitter latch please im begging

[bitter latch]

honestly it's prob the lora

[broken osprey]

how can a lora produce a detailed image like that alone tho :c but i understand if the model is personal and you dont want to share it ^^

[bitter latch]

no lora

[silk narwhal]

NAIv3

[bitter latch]

the lora https://civitai.com/models/1141/mjv4-hypernetwork

[silk narwhal]

i like looking at my desktopPC fans when generating stuff .. they don't have any LED controllers (get powered by the rpm of the fan) but when generating the each light up one by one or random 😄

[worthy delta]

anyone know how to fix hands

[bitter latch]

adetailer ext

[silk narwhal]

with comfy .. use Ydetailer

[bitter latch]

or that

[worthy delta]

i use a1111 web ui

[broken osprey]

damn or impact pack

[silk narwhal]

Adetailer it is

[broken osprey]

or for webui use adetailer, but it mostly doesnt work that well

[worthy delta]

link?

[broken osprey]

maybe use control net for that kind of stuff

[worthy delta]

anything that does?

[silk narwhal]

swap to comfy 😉

[broken osprey]

sd1.5 is not that good with hands, I use loras and embeddings for hands, adetailer sometimes and even controlnet if you want complete control

[bitter latch]

same prompt as above with the lora--> Neko, playing games, next to pool, drinking Tequila, beach in distance, two piece bikini , h-cup breasts, hyper feminine hypernet:mjv4Hypernetwork_v1:0.8 octane render

tho sampler and other settings can change it too

[worthy delta]

smash

next

[bitter latch]

and thats also my merge I call 3xFuzzedanime

[broken osprey]

damn, Imma use furricanes furry mix, that jsut came out later and try to replicate

[bitter latch]

havne't uploaded anywhere yet

[broken osprey]

but if that is NAI it didnt understand the part that says drinking tequila, it looks like a sandwich hehe

[bitter latch]

yeah hands go screwy

to much showing

[broken osprey]

nah but it does look hot haha, i like your style

[bitter latch]

changing from eular to restart with hi-res fix can change it too as you can see

[broken osprey]

im not sure what just happened, my iteration counter is stuck at ?it/s, it says comfy loaded the model but my gpu isnt doing anything and its cold.

comfy hasnt crashed im just lost :c

[silk narwhal]

sometimes it takes a good minute to start the sampler

[broken osprey]

the minute has already passed tho :(( i just resetted the thing

[silk narwhal]

especially on a high res latent input

[broken osprey]

hmmmm idk might be cause of the clip vision thing idk, im just trying ip adapter

[silk narwhal]

yea.. can't get IP adapter or Clip Vision to work on my system too takes 5 minutes to load and then errors

[bitter latch]

and I found the order you put things in , matters alot in the prompt

mdjrny lora in front of style

[broken osprey]

@silk narwhal look halp, its here, thinking or doing smth idk

[bitter latch]

and last in style

huh and it went wonky face, wtf

[broken osprey]

yeah ipadater aint moving ahead at all, nothing happens gpu util is at literal 0% this aint good at all

[worthy delta]

I've made a revelation

Males are neglected in ai modles

We need more men

I want men in armor

Looking hot as they pose with armour

Is this just about armour

Yes

But dosent takeaway anything

[broken osprey]

I mean most models should have a little bit of male representation, but like idk, have never prompted for them, even put them in the negative sometimes lmao

[silk narwhal]

wrong aspect ratio set :/

[broken osprey]

ooop

[silk narwhal]

there we go

[steady grail]

cute

black and white are almost better

[broken osprey]

how did you prompt for that? or like how did you do that?

[silk narwhal]

long_hair looking_at_viewer blush multiple_girls 2girls sitting white_hair heart black_thighhighs black_skirt cat_ears black_eyes animal_ear_fluff cat_tail zettai_ryouiki siblings holding_hands cat_girl pink_nails twins furry_female back-to-back pink_sweater fish_hair_ornament

the workflow i am working with has a desaturation pass and saves that image before continuing to the next pass, face and hands fixes and recoloring

recolor is still a miss 😄

[steady grail]

that's a good help for inpainting

[steady grail]

inpanting, decoloring and recoloring to conform everything

[steady grail]

a controlnet or ip adapter can be a big hand in recoloring without altering the final result

[silk narwhal]

controlnet sadly errors on me when doing anything sdxl and ip adapter errors when going into the sampler

[steady grail]

yeah, we still needs good models compatible with xl

[broken osprey]

are the errors with ipadapter common to comfy or is it because of lack of vram?

[steady grail]

to me, it works perfectly

[silk narwhal]

the ip adapter errors are something about wrong shape sizes in base model ..etc etc

[steady grail]

and i have very low ram

[broken osprey]

how did you get it to work?

[broken osprey]

but how can I fix that if it doesnt even trhow out an error, it took almost five minutes there in the ksampler without starting and itkept going like an idiot

[silk narwhal]

don't use it as i cannot get it to work either

[broken osprey]

ah okay :c

[steady grail]

i don't use xl

[nova remnant]

"stop in the name of the law!"

"I notice you're slipping in your grades, hey get up off the floor"

[supple raptor]

Are you using the SDXL controlnet/ip adapter models?

[silk narwhal]

yes

[silk narwhal]

`size mismatch for proj_in.weight: copying a param with shape torch.Size([1280, 1280]) from checkpoint, the shape in current model is torch.Size([1280, 1664]).

File "D:\ComfyUI_windows_portable_nvidia_cu121_or_cpu\ComfyUI_windows_portable\ComfyUI\execution.py", line 153, in recursive_execute
output_data, output_ui = get_output_data(obj, input_data_all)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "D:\ComfyUI_windows_portable_nvidia_cu121_or_cpu\ComfyUI_windows_portable\ComfyUI\execution.py", line 83, in get_output_data
return_values = map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "D:\ComfyUI_windows_portable_nvidia_cu121_or_cpu\ComfyUI_windows_portable\ComfyUI\execution.py", line 76, in map_node_over_list
results.append(getattr(obj, func)(**slice_dict(input_data_all, i)))
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "D:\ComfyUI_windows_portable_nvidia_cu121_or_cpu\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI_IPAdapter_plus\IPAdapterPlus.py", line 636, in apply_ipadapter
self.ipadapter = IPAdapter(
^^^^^^^^^^
File "D:\ComfyUI_windows_portable_nvidia_cu121_or_cpu\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI_IPAdapter_plus\IPAdapterPlus.py", line 272, in init
self.image_proj_model.load_state_dict(ipadapter_model["image_proj"])
File "D:\ComfyUI_windows_portable_nvidia_cu121_or_cpu\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 2152, in load_state_dict
raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format(`

this is the error

[silk narwhal]

Ipadapter models

controlnet models

all give the same error in any combination

[supple raptor]

Yeah those are the right models. I'm not familiar with running controlnet/ipadapter in comfy so I'm not sure what else it would be.

[gentle musk]

Are you using the correct ClipVision model?

Would help if you showed your whole workflow and the models you are loading

That error is because you are using the wrong model in some part of the chain

Or alternatively your ComfyUI is out of date, or IPAdapter nodes are out of date.

[silk narwhal]

either of these downloaded from the manager

[gentle musk]

Wrong model for the 2 IPAdapter models with vit-h after them

You need this one https://huggingface.co/h94/IP-Adapter/resolve/main/models/image_encoder/model.safetensors

Id suggest reading the IPAdapter installation documentation
https://github.com/cubiq/ComfyUI_IPAdapter_plus#installation

ip-adapter_sdxl.safetensors should work with the clipvision model you have. But that model isn't really used anymore, the vit-h ones are better

[silk narwhal]

will test in a bit

wished more nodes in workflows would fetch the models from huggingface more

[gentle musk]

That's just adding an extra point of failure into the nodes though. I'd argue against doing that.

What should happen really is the ComfyUI manager should point people to the github pages with instructions. Forcibly point them, or they won't read it.

[silk narwhal]

better then this 😄

[gentle musk]

There's just lots of different Models for IPAdapter, if you look on the github installation guide it has a table with them all on.

You'd also have to decide how to configure the node to pick which IPAdapter it would download. So you'd need extra code to do that.

Because you wouldn't want to just download them all, because that would be like 30 or 40GB or something daft.

[broken osprey]

hey one question Ive got these two models for the clip vision but I am having the same issues as kuromi, the thing gets stuck on the ksampler, even after five to six minutes it just doesnt work at all. Maybe you know how to fix this?

[gentle musk]

gets stuck as in doesn't do anything, or does it throw an error?

[broken osprey]

doesn't do anything, It passes the section that says ipadater loader and it gets stuck directly on the ksampler.

It shows something like

Total Progress: [ ] 0/30 ?it/s

And stays there for minutes on end

[gentle musk]

What GPU do you have?

[broken osprey]

AMD RX 5500 XT 8GB vram

[gentle musk]

When you run it, is it using up all your VRAM?

The SDXL one might

[broken osprey]

it doesn't even use it, my gpu runs cold, literally as if nothing was working. Also I am using only sd1.5

[gentle musk]

Does it work normally? without IPAdapter?

AMD Cards have their own unique problems with this stuff

[broken osprey]

Yep it works normally, I was running controlnet as a test half an hour ago and it worked, even normal generation worked, only ipadapter didnt

[silk narwhal]

works now .. thanks .. been troubleshooting this for over a week

[gentle musk]

hmm with no error message it's very hard to tell the issue.

This probably isn't the problem, but it's easy to check. Try making the names of those clipvision files smaller. There's a very small chance it's hitting the filepath limit and doing something odd.

[gentle musk]

no worries

[silk narwhal]

now just have to find the settings to speed the workflow up with SDXL models: Loading 1 new model
5%|███▋ | 3/63 [01:06<22:23, 22.39s/it]

[worthy delta]

webui a1111 on top!!1

[gentle musk]

What GPU?

[broken osprey]

I'll try that later thanks

[silk narwhal]

rtx 3060

[gentle musk]

hmm, I'd have thought it would be a bit quicker. 63 steps is a little excessive though, 25 - 30 should be enough

my 3080 with multiple IPAdapters and Controlnets usually gets around 1 - 3 it/s

It taking that long it making me think you are running out of VRAM

But the 3060 has 12GB doesn't it?

Unless you have the 8GB Version, then you might be running out of VRAM, check in Task Manager

[silk narwhal]

am testing this: https://comfyworkflows.com/workflows/544fcd98-9fe8-468a-bb9b-509570fdca09

[gentle musk]

Yeah you are running out of VRAM

As soon as you start using Shared GPU Memory it completely tanks the speed

Because it has to keep swapping the data between RAM and VRAM

[silk narwhal]

any suggestions?

[gentle musk]

Use a less complicated workflow

Just looking at that you are loading in an SDXL Model, ClipVision, IPAdapter, Anything Everywhere.

It's also merging 2 models that have run through IPAdapters, so that doubles up again on the VRAM use of your models

For reference using an SDXL Model with ClipVision + 2 IPAdapters and a Controlnet pretty much fills my 10GB VRAM.

Also I've never used that comfyworkflows website before and it's pretty cool

[silk narwhal]

search is pretty bad currently 😄

[gentle musk]

But yeah I'd just try a simple workflow, it's easy enough to setup a quick one, but with 12GB a normal workflow without all these weird extras will work fine

[silk narwhal]

Loading 1 new model
78%|███████████████████████████████████████████████████████████▉ | 49/63 [00:39<00:08, 1.62it/s]

replaced the SDXL empty Latent preset for a regular Empty Latent Preset node so i can do 512x512

[gentle musk]

Yeah that looks about right

[gentle musk]

Now you have to try find SDXL Anime models that aren't bad

[silk narwhal]

might template this wf and throw it as a pass in my massive workflow 😄

[worthy delta]

What is yalls most wild generation?

that blew ur mind

[gentle musk]

Is that generating a background and a character then combining them together (sort of)?

[native halo]

the J

[gentle musk]

Is that pose using a controlnet?

[stable crane]

luck

[gentle musk]

nice then lol

[stable crane]

and wildcards lol

[silk narwhal]

yea .. it generates two different prompts and blends them into a new image

[gentle musk]

At first glance it looks overly complicated to do that. I'm sure you can do something similar with less nodes and less VRAM usage.

[broken osprey]

can anyone explain what wildcards are?

[stable crane]

imagine it like a random prompt component

[gentle musk]

A wildcard is just a selection of words that can be used in a prompt

[silk narwhal]

its an experimental workflow .. so i am sure the creator is working on optimizing it

[stable crane]

__boobs__ will be replaced during generation with sideboob or underboob etc whatever I define

[gentle musk]

So you can have "A Woman wearing *"

And * could be an item of clothing from a list

[stable crane]

__color__ hair replaces with a random color name

masterpiece, best quality, mature, tall, extremely detailed, slight smile, jungle forest, cargo pants, __boobs__, camouflage bodysuit, __hairstyle__, __color__ hair, __pose__, __orientation__, __size__ breasts, __color__ theme, (thick thighs:1.2), narrow waist, wide hips so you can see a lot of randomness can result from this prompt

[gentle musk]

I like using IPAdapters to get some "controlled" randomness

[stable crane]

I still need to try that

[broken osprey]

ooooooh I see so like how do I use that in comfy? or in webui for that matter and do I have to define those or do they come in some kind of model or smth

[stable crane]

https://github.com/AUTOMATIC1111/stable-diffusion-webui-wildcards.git I don't have one for comfy yet, but I use this one

in comfy, you can do {sideboob|underboob|cleavage} for instance and it'll randomly pick one

poor man's wildcards

can be good to quickly inject some randomness

[gentle musk]

I tried to do something similar to your image, but it's got too hard on the crouch lol

[stable crane]

oh cool

[broken osprey]

oh i see so that is what {} mean in comfy

I saw a prompt that had {{{flat chested }}} i wonder if this means it can appear or not.... or does it mean that it appear always

[silk narwhal]

{} is negative weight usually

[stable crane]

is it? it seemed like it randomized things lol

[silk narwhal]

token = weight of 1.0, (token) = 1.1 or so, {token}= 0.9 or so

[broken osprey]

lol they put in the negative that is so weird

[silk narwhal]

or was it [ ]?

[stable crane]

https://github.com/comfyanonymous/ComfyUI

You can use {day|night}, for wildcard/dynamic prompts. With this syntax "{wild|card|test}" will be randomly replaced by either "wild", "card" or "test" by the frontend every time you queue the prompt. To use {} characters in your actual prompt escape them like: { or }.

[silk narwhal]

ah ok

[broken osprey]

so like we can use like

{sakurajima mai | kotono nagase | someone else } like that to select one out of the three?

[stable crane]

yeah

[gentle musk]

Which checkpoint is that? It's been a while since I've used SD 1.5 models

[broken osprey]

same same same

[stable crane]

mine? it's lottamix3

[gentle musk]

ok cool

[stable crane]

I didn't upload it yet though, but lottamix2 is on civit

[silk narwhal]

looks like a mix of a rimlight and sticker 😄

[stable crane]

border, outline 🙂

[silk narwhal]

when the last ksampler steps are too low

it doesn't blend the two results proper

[broken osprey]

yeah

weird

maybe if you vary the weights of the images, there's something about it in the github

[gentle musk]

Hmm weird, I get black images in the vae with the lottamixv2 model

ok used a different vae and it's fine

[silk narwhal]

have you tried enabling a different cross attention to the launch params

[gentle musk]

I've never had this issue on Comfyui before, just seems like it doesn't like the built in vae on that model

[silk narwhal]

--use-pytorch-cross-attention

[gentle musk]

It does that by default

Maybe it's the bfloat16 on SD1.5 models I dunno. I mostly use SDXL on it.

[native halo]

yea i too get black imgs on some models,just change the vae and problem fixes itself

its a vae issue

[broken osprey]

tbh I get those too and it happens after a certain number of iterations in a set of 50 or 60 images, but I haven't tried changing the VAE since it goes back to normal after one or two black images

[native halo]

yea try to change vae

[gentle musk]

Yeah for mine it was definately the vae, because the preview was fine it only went black after it was decoded.

[broken osprey]

damn if it works it works hehe

[stable crane]

ooh nice

there's probably no vae built in

idk wtf I was doing when I made that model 😂

[stable crane]

your pic has me going to try lottamix2 again lol

[gentle musk]

I used an IPAdapter for that 🙂

[broken osprey]

hot DAMN

[stable crane]

damn lottamix2 colors pop more

been a few months lol

[worthy delta]

Large

God bless ai art

[worthy delta]

And ai hot buff men

Amen amen

[gentle musk]

Think I've sorted my settings out now for SD1.5 models

[silk narwhal]

lol .. the prompt blending with SDXL models is about early Craiyon ai levels unless i set the empty to 1024x1024 and all ksamplers 50+ steps

85 s/it

[gentle musk]

lol

[silk narwhal]

trying with SD1.5 now

better

Leftover VAE keys ['model_ema.decay', 'model_ema.num_updates'] 100%|█████████████████████████████████████████████████████████████████████████████| 25/25 [00:06<00:00, 3.88it/s] Requested to load AutoencoderKL Loading 1 new model 100%|█████████████████████████████████████████████████████████████████████████████| 25/25 [00:05<00:00, 4.87it/s]

[gentle musk]

lmao, this is not what I asked for

[nova remnant]

[native halo]

speakin of toes

[gentle musk]

I still don't understand why it made feet lmao

[native halo]

even sd loves 🦶

[silk narwhal]

alright..learning how this works slowly

[gentle musk]

it looks like it's just creating 2 images and then using IPAdapters to merge them together.

Well merge the ideas of them

[silk narwhal]

yea

very little prompting needed

[broken osprey]

@silk narwhal can you send the workflow please? or the image ?

[native halo]

who sneezed on your pic

[broken osprey]

lmao

[bitter latch]

ponders going to bed now and waking up next year

haha

[broken osprey]

2004? Cute

[native halo]

cant believe its 2004 already,only 8 more years for mayans to destroy the world 😔

[bitter latch]

o.O

[nova remnant]

2004

we are young again

[raven oasis]

another bunch of my cutest renders

[worthy delta]

very cute renders

how lewd can we get here fellas?

@gloomy scroll sorry to bother ya but how lewd can we get in here just to be safe

[gloomy scroll]

[bitter latch]

little bit of skin like bathing suits and such but becareful of showing any body parts you aren't suppose to or suggestive tattoo's or what can be seen as sexual poses

[native halo]

burqa only

[bitter latch]

I would say rated R at most

it can be subjective

[native halo]

no ankles

[bitter latch]

the bot sometimes doesn't like armpits.. shrugs

tho I think they fixed that

that's miku I wouldn't make mad

[native halo]

too much skin

heres a proper miku

[bitter latch]

haha you funny... as a tooth ache

[native halo]

https://tenor.com/view/mike-ohearn-gif-5978288995841027384

[bitter latch]

😄

[worthy delta]

yall really love miku dont you

[bitter latch]

shrugs, got fixated, could use some more to do but not sure which one, suggest some

[native halo]

since u like Miku try with another vocaloid, Rin & Len

https://tenor.com/view/vocaloid-len-heart-rin-kagamine-twins-gif-17436000

[keen hollow]

Well, I've generated a few thousand different Tifas across multiple model and Lora combinations.

looking for my favorite releases and combos

getting close 😛

[worthy delta]

do show

[keen hollow]

lol well i'm not gonna post 4000 Tifas... but I'll select some once I narrow in. 😛

[worthy delta]

four thousand????

[native halo]

post the ones when we were doing the slime Tifa

[keen hollow]

yeah I've probably generated about 4000

[worthy delta]

why?

[keen hollow]

21 models and 25 Loras

to find my favorite permutations

narrowed it down to 4 Loras

still working on models

[keen hollow]

that's a Tifa!

[native halo]

old slime tifa

[keen hollow]

I'm still running a few more grids, gotta finish these before slime mode lol

[keen hollow]

trying to do slime but my prompting skills suck 😛

Slime Tifa

her tights just became slime

I need better slime prompts

[broken osprey]

@keen hollow how did you get the legs to be crossed like that? Lora? Also how did you get the head resting on Hand image. Is it just prompting ??

[keen hollow]

"sitting" lol

I'm definitely not a master prompter.

[broken osprey]

LOL

[keen hollow]

I think it was "sitting in a tub of slime"

[broken osprey]

I try sitting and it never gives me crossed legs, not even crossed legs as a keyword works damn: imma try a batch

[keen hollow]

that one is ok... like slime sleeves

[lyric zealot]

“Crossed legs” should works

But depends on model as well

[bitter latch]

well.. then.. ummm slime

[broken osprey]

where is that slime coming from?

[bitter latch]

the ground.. yeah.. the ground

[broken osprey]

Oh of course of course it’s climbing up

[nova remnant]

I am the lewd inspector

You can throw them in my DMs at any time

👍

[stable crane]

me too

[left galleon]

i am open as well to it

[stable crane]

you could also post them in my server

[broken osprey]

you got a server for lewd??? may I ask for an invite?

[stable crane]

check my profile

anyone know how to achieve that soft effect you can get when you hit "Interrupt" like 50% through a gen in A1111?

now I like it stylistically lol

[steady grail]

with comfy is pretty easy

with auto..... maybe using a lower denoise?

[unreal ridge]

👁️👄👁️

[steady grail]

lewd and discord are not very compatible

[unreal ridge]

actually real

[stable crane]

😰

[unreal ridge]

Can't send messages... 10 minutes cooldown 😭

[stable crane]

oh really?

I forgot about that

[unreal ridge]

Ye 😔

[stable crane]

soon~

[unreal ridge]

soon™

[steady grail]

what is it saying?

[unreal ridge]

.

[steady grail]

new users are in cooldown i suppose, to prevent spam

[unreal ridge]

ye

[vestal crypt]

https://media.discordapp.net/attachments/1041682785418629125/1191478927294738492/00266-871895399.png?ex=65a59638&is=65932138&hm=d3477282860ea23011994d19845fce404b790682d8fdd1acad1e4b7e78681495&=&format=webp&quality=lossless&width=441&height=662

[untold glacier]

Love this aesthetic

[merry tendon]

thx

[summer night]

why are there hardly any anime-styled SDXL models

blue pencil is carrying local anime SDXL rn

[astral jasper]

there's only 1 guy/team doing any finetuning on more than a million images, the other one is a single guy using a single a100 and has a better looking anime model than bluepencil

which just merged the model that someone else finetuned for a week+ (and every other model and lora too)

[steady grail]

wow

[weak prairie]

We are getting married. How about joining us for an unforgettable wedding night? #aiart #aiwaifu #waifus #brides #weddingnight #hcup #sd #sdxl #stablediffusion #animewedding #kawaii

[magic wave]

And NAI

[finite herald]

can we relax on the hcup500000 gens in here pls

[summer night]

does each lora increase vram usage?

[steady grail]

a little, yes, they need to be loaded

the bigger the lora, the heavier the load

[summer night]

do they take up their filesize in vram?

[steady grail]

i'm not sure it's the same, but it's plausible

[bitter latch]

no more hcup's , got it

grinz innocently

hmm ok no making cups that look like an H.. sigh stupid program

[broken osprey]

Heh just created a little comfy workflow to test samplers and schedulers 4 at a time, finally I understand what the model likes

[bitter latch]

😄

anime Bart simpson

[broken osprey]

damn those eyebrows do be thick

sadly i have no work to show of my process, mostly nsfw :c

[broken osprey]

okay this is as close as I can get to normal

I specially like the way her back is drawn as if it were lit by the lamp as well as her hair

[bitter latch]

uhh not as expected

[broken osprey]

bro looks old :=

[bitter latch]

was suppose to be homer almost running over bart

like in the intro

[broken osprey]

hahaha it looks cool tho, how did you prompt for the mixture of anime and cartoon style?

[broken osprey]

like here

[bitter latch]

got a merge I made with some anime and rpg3d type models in it

haha

tho proble , ---> Finely Detailed, elegant, sharp, focused, Gradient color, soft shaded, UHD, 4k, Hires, HQ, elegantly detailed illustration, black thin outline art style, sharp focus, elegant detail, HDR, detailed shadows, cinamatic lighting, unreal render, realistic

had something to do with it

suppose to have been a fursuit but well .. shrus

shrugs. it went sorta random

[broken osprey]

how many personal merges d you have :000

[bitter latch]

eh not many

Naruto Simpson the Ninja of Fox Village

not what I was expecting, tho still looks good

[broken osprey]

how? gimme prompt please 🙇‍♂️

Also who knew euler exponential and euler sgm_uniform were so good at 25 steps lmao

[bitter latch]

Naruto "bart" Simpson, Ninja of Fox Village, wearing headband and orange ninja jacket and pants, action scene, finely Detailed, elegant, sharp, focused, Gradient color, soft shaded, UHD, 4k, Hires, HQ, elegantly detailed illustration, black thin outline art style, sharp focus, elegant detail, HDR, detailed shadows, cinamatic lighting, unreal render, realistic
Negative prompt: kkw-extreme-neg, blurry, deformed hands, deformed face, deformed fingers, multiple limbs, ugly, bad anatomy, bad hands, ugly hands, deformed, disfigured, poorly drawn face, mutation, mutated, extra limbs, poorly drawn hands, (grayscale), weird colors, censored, deformed glasses, lowres, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark
Steps: 40, Sampler: Euler, CFG scale: 6, Seed: 353700419, Size: 512x768, Model: RpgFuxAnimeNMSV1_Merge, VAE hash: f921fb3f29, VAE: clearvae_v23.safetensors, Clip skip: 2, Pad conds: True, Version: v1.7.0

[broken osprey]

just testing out some prompts hehe, no upscale no face fix, just to see how good they behave

[bitter latch]

hmm suppost to be fur not fux. (facepalms)

ok turned off styles, same seed... DED

eh....

[broken osprey]

caveman bart

[bitter latch]

Naruto "bart" Simpson, Ninja of Fox Village, wearing headband and orange ninja jacket and pants, action scene, finely Detailed, elegant, sharp, focused, Gradient color, soft shaded, UHD, 4k, Hires, HQ, elegantly detailed illustration, black thin outline art style
Steps: 40, Sampler: Euler, CFG scale: 6, Seed: 353700419, Size: 512x768

[broken osprey]

im doing it with counterfeit, lets see what it comes up with

whatever this is, is neither

[bitter latch]

changes from eular to restart

[broken osprey]

well its ++ 2M normal so yeah not a lot to expect, once it gets to sgm_exponential that might do something

[bitter latch]

taps foot

much better

[broken osprey]

whole your merge is amazing

my faces are coming out like shit

[bitter latch]

that one went nuts

hmmms

[broken osprey]

needs upscale but look !!!! Im getting there!

[bitter latch]

midjourney4 lora helps on tht last one

[broken osprey]

wait what is that?

wasn't midjourney like pay model?

[bitter latch]

oh my bad it's a hyper network

[broken osprey]

may i ask where you get that?

civit ai? or personal?

[bitter latch]

https://civitai.com/models/1141/mjv4-hypernetwork

I use Midjourney style, hypernet:mjv4Hypernetwork_v1:0.8octane render

[broken osprey]

okayyy thank you ^^

[bitter latch]

welcome

tho you have to find a mix of other prompting to help it into the style ya want

trial and error

Portrait of Naruto "bart" Simpson, Ninja of Fox Village, wearing headband and orange ninja jacket and pants, Midjourney style, hypernet:mjv4Hypernetwork_v1:0.8octane render
Steps: 40, Sampler: Restart, CFG scale: 6, Seed: 353700419, Size: 768x768

or a good model.. too

[broken osprey]

ye ye rn I dont have much time since Im getting sleepy lmao, but tomorrow maybe Ill work on getting some hypernetworks installed, and use them!!.

I have to organize them as I did with the lorasl

[bitter latch]

huh.. I should name this and post it

[broken osprey]

most of the models I have are dedicated to anime or realism, Im not sure characters like bart enter in them :c

[native halo]

they are there because the base 1.5 has them

[bitter latch]

mine are a mix of furry, anime, a RPG semi-realistic, and a No man sky for variety, in that model

tries it with out the saved style

huh...

[broken osprey]

then putting his name would've worked on counterfeit since its based on 1.5 but it didn't show up at all, only naruto

[native halo]

try to include yellow skin,spike hair

they still come out deformed tho cause it wasnt finetuned with that character but it will try to draw it with few data it has from base

[broken osprey]

aaaaah I see, well I will download a character lora to add more detail, while also increasing the description on him to see if that works out :)

[bitter latch]

Portrait of Naruto "bart" Simpson, Ninja of Fox Village, wearing headband and orange ninja jacket and pants
Steps: 40, Sampler: Restart, CFG scale: 6, Seed: 3768983874, Size: 768x768

scratches head

[native halo]

bart simpson with 1.5 base

[broken osprey]

rook rook cute girl in a park

[native halo]

not as cute as this qt pi

[broken osprey]

no one can be as cute as that, mb

[bitter latch]

took out "bart" simpson

[broken osprey]

thats much better tbh

[bitter latch]

hmm where's the upload to civitai button gone?

[broken osprey]

il barto

[bitter latch]

haha default settings.. suck

[broken osprey]

is this upscaled? it looks so neat for a 671*671 image

[native halo]

anatomically correct il barto

[bitter latch]

euler is some better

lcm ---> img2img restart , from 20 to 40 steps not bad

[bitter latch]

haha

cackles

well that didn't work.. right

looks like he's sick with jonduss

and trust me.. that aint good

[native halo]

speaking of drip

[bitter latch]

haz an idea.. type type

😄

cackles evilly and hits generate

nt the one but .. ok

well I guess Naturo trump does change much...

I had to

ummm

o.O

yeah... no.

falls over DED

ok now thats more anime

moves on. ponders new prompts

wtf does this do so well on naruto?!

Naruto
Negative prompt: kkw-extreme-neg, blurry, deformed hands, deformed face, deformed fingers, multiple limbs, ugly, bad anatomy, bad hands, ugly hands, deformed, disfigured, poorly drawn face, mutation, mutated, extra limbs, poorly drawn hands, (grayscale), weird colors, censored, deformed glasses, lowres, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark
Steps: 20, Sampler: Euler, CFG scale: 7, Seed: 1345220038, Size: 512x768

https://tenor.com/kKRaJi73teZ.gif

CyberTronian Ninja

haha meant post card, but put post cart.. DED

ok now she's in a desk.. wtf

wow

[nova remnant]

@native halo new maid arrived

[native halo]

(skirt lift:2.5)

[nova remnant]

[bitter latch]

😄

goes to pass out... thud

But first. Iron Knight power Ranger Suit

[nova remnant]

👍 family friendly

[bitter latch]

ok thats better and.. DONE

[zinc flax]

Shorts jumpscare

[bitter latch]

Black Knight Power Ranger Batman suit

stupid brain I can't get this out of my head

power rangers

[native halo]

https://tenor.com/view/homer-simpson-the-simpsons-nooo-screaming-shouting-gif-16986595

[native halo]

ankle alert ⚠️

[nova remnant]

that's a really cute face

[nova remnant]

and this

[merry shore]

Hi all! Does anybody have tips for creating consistent characters in the anime artstyle? I've tried keeping the same prompt and only changings words related to expression or outfits / backgrounds, but the character generated always changes a little bit. Also facial features sometimes come out with no detail or poorly drawn. Also, I'm using dreamstudio not A111 web ui

[vital raptor]

you would need a lora trained on the specific character

[merry shore]

Hmmm i see. Is there no way to do so solely relying on prompting? I'm developing an app right now and have opted to use stability api so i don't need to worry about custom infra and have my own standalone A111 webui that can be used as an api

[vital raptor]

if it's a character the model knows then you might be able to get close, for example Hatsune Miku is popular enough that most models know the general look

but, depending on training and merging in the model, it's probably not exact, the only way to get the exact same person reliably (and even then it's not 100%) is to have a custom trained model or lora

[merry shore]

Gotcha - thank you! Do you know if there is a list or doc of popular anime characters that the model has been able to draw well? All good if not, I'll just be experimenting with popular anime characters to find out which ones the model knows

[vital raptor]

well, there's hundreds if not thousands of models and merges of those models, so no, it's not likely you'll find any list

[merry shore]

Makes sense, thanks!

[brittle rampart]

The facial features turned out to be quite cute and delicate

[steady grail]

on comfy you can use ip adapter

[brittle rampart]

btw. in A1111 i can use it?

[steady grail]

no idea

when i used auto, ip adapter still wasan't a thing

[brittle rampart]

hmmmm...

[worthy delta]

anything i could improve on this?

[nova remnant]

Omg such pretty clothing! What's the prompt for it?

[finite herald]

.

[broken osprey]

damn

at least a censor xD

how about this? shouldn't break rules right?

[finite herald]

[broken osprey]

gotta keep the lewd away

[finite herald]

or be b0nked

[broken osprey]

Does anyone know how to save custom colors in comfyUI? Like adding to the predefined list?

[silk narwhal]

editing the py maybe?

[broken osprey]

but which one :c

[supple raptor]

I'm pretty sure it would be done with .\web\user.css but I don't know css.

[broken osprey]

I tried that but it didn't detect it, Im not sure also which class would I have to change too yeah im stuck wiht that one :c

[silk narwhal]

aside from making a template (right click save as template) i don't think you can save customs

[broken osprey]

yeah that might be it, rn im sticking with templating and saving the color in there

[steady grail]

charming

[nova remnant]

they are eating mini foxes

[vital raptor]

nah, those are just napping

[glossy eagle]

I just got my m.2 for Linux and my half dead Linux drive is dying as I’m typing this

The mb transfer of stable diffusion is getting slower and slower

[broken osprey]

somewhat decent shot

[near elbow]

morning all...

what are you looking for in an avatar? and i assume "dom = dominatrix" not "dom = the guy from fast and furious" based on your name? might be too NSFW for this server

i think this is the maximum amount of skin that's appropriate here

[bitter latch]

I think because I put from 1800's saved it from going over the line

anime profile sheets?

[wheat falcon]

ip-adapter is nice 😄

[brittle rampart]

can i see your other settings?

[wheat falcon]

sure

[brittle rampart]

many thanks!

[wheat falcon]

np, also used hires fix with 4x-UltraSharp and 84000-mse vae

[wheat falcon]

😄

[brittle rampart]

How fast is your regeneration with controlnet?

[wheat falcon]

1:45 with hires fix

[bitter latch]

Miku Chibi Robot

[gloomy scroll]

I wish I didn't know css

[glossy eagle]

Ultra-HD

[nova remnant]

used keywords like intricate pattern and details on dress

[nova remnant]

Ending control step: 0,5 or 0.5?

[wheat falcon]

0,5

[nova remnant]

when lowering the value with up and down arrow next to that box i get dot

[wheat falcon]

I don't

Make sure Controlnet extension is updated

[nova remnant]

just updated few mins ago

default value on ending control step is 1

comma works like a separator i think not value

[wheat falcon]

It should work as a value

It stops at half the steps

[nova remnant]

yeah that's what im saying ... half of 1 is 0.5 .. seeing you have comma instead of point seems strange

[wheat falcon]

You can use 0.5 too, it shouldnt matter in that extension

[nova remnant]

actually i can't, it wont let me put a value with , the field only accepts .

[wheat falcon]

Oh thats interesting.
How do you set Denois values in hires fix?
0.45 or 0,45

[nova remnant]

0.4 usually

[wheat falcon]

When using the arrow up and down

[nova remnant]

comma is a seperator

right i get dot

[wheat falcon]

I get comma

[nova remnant]

amusing

[wheat falcon]

I think its a german/European thing xD

[nova remnant]

oh

[wheat falcon]

We say a file has 1,3 Gigabyte in size for example and not 1.3 Gigabyte. And we also type that aswell

[nova remnant]

ahh ok that explains it 🙂

but that aside im curious to know why are you setting ending control step to half the default value 1?

two images .. one on the left 1 and the right 0.5

[wheat falcon]

Did you used Ip-adapter?

[nova remnant]

yes

[wheat falcon]

What was the input image?

[wheat falcon]

Yea that person isn't a difficult character so it won't change much

[brittle rampart]

tell me ppl why controlnet slows down?

[wheat falcon]

If the character wears, glasses, has a beard or recognisable design you need to play around with the end step

Or the weight

[brittle rampart]

same version. 426

very very slowly

[wheat falcon]

What size are your controlnet models?

[nova remnant]

not sure, but could depend on vram

[brittle rampart]

--medvram parametr in bat

[nova remnant]

could try t2i models that are lot smaller in size and supposedly more efficient

[wheat falcon]

yea your using the large models 1.5gb

im using 700mb ones

[nova remnant]

https://huggingface.co/TencentARC/T2I-Adapter/tree/main/models

[wheat falcon]

https://civitai.com/models/38784/controlnet-11-models

@nova remnant ending step: 0.5 and 1

based on

[nova remnant]

ahhh cool, i have to play with more complex images as inputs

[brittle rampart]

@nova remnant @wheat falcon oki-doki... try this is

[nova remnant]

take note some of the CN models are not available with t2i .. to replace the missings ones you could try the 700mb sizes

[wheat falcon]

make sure you also download the .yaml files for the models

[bitter latch]

Miku Mecha Design

[bitter latch]

hmm different?

grinz

not sure what happened to hands

umm hmmm? ok.. she didn't have a good night, ended badly I would say

[unreal ridge]

she rocking the lego hand 👍

[wheat falcon]

https://tenor.com/view/wash-your-hands-chameleon-pet-clean-water-gif-16711643

[broken osprey]

Hellooo does anyone know or have a good workflow in comfy for doing stuff like latent couple in stable diffusion, basically region prompting?

[primal wharf]

chopsticks

[nova remnant]

pat

[silk narwhal]

gotta love the new upscalers:

[nova remnant]

Kuromi

Hewwo

[silk narwhal]

owweh

[nova remnant]

You have such a funny profile picture

[silk narwhal]

thanks

[nova remnant]

What is it called?

[broken osprey]

what's that upscale called?

[silk narwhal]

https://www.upscayl.org/

[nova remnant]

Ohh

[silk narwhal]

She is going to blackmail him afterwards with that selfie

[primal wharf]

Lmao.

Mr black hair red eyes army guy going through multiple dates

I finally got fox ears working, I had to restart SD and the webui lol.

[silk narwhal]

damn Pika's Upscale works good

[bitter latch]

pika pika?

https://tenor.com/4LSq.gif

[nova remnant]

Chu

[bitter latch]

thats what I get with pika mika

😄

Pika Miku chu

Pika Miku Chan

[stable crane]

lol

I'll take that pokemon if that's the evolution

[bitter latch]

mixes up the prompt

[stable crane]

is CN built into comfyui? or requires custom nodes?

[nova remnant]

[primal wharf]

Is that Ivy-chan?

[nova remnant]

I think it's just that smol gen

I took his prompt, removed every embedding and stuff

turned out like that

[primal wharf]

Ivy-chan's doppelganger, without the nasty attitude maybe.

[silk narwhal]

Before the SD Ultimate Upscaler