🏞｜general-with-images

[lean talon]

lol the tiny human running away

[south temple]

animations aint easy to hold consistancy

[nimble mason]

thats the entire battle

[arctic laurel]

it’s looking like the inside of a half life 2 citadel

[grizzled sage]

past 16 frames yeah. you need controlnets and input frames

loras help too. especially ones that are over fit to a single style

[arctic laurel]

okay yeah that’s half life citadel meets kate bush music video

perfect album cover for a vaporwave resurgence album tho

[clever oar]

help in styles window i don't see save style button

how i make style

[south temple]

fixed the tiling

now its trying to crip up the images

crisp*

[grizzled sage]

miles teller infected me. anytime i hear about kate bush i think of this dumbass scene from snl now. https://www.metatube.com/en/videos/514717/SNL-Rooftop-Bar/ @3:30

[jovial tiger]

it's really too bad all this stuff is for sd 1.5 mainly. there's always loss between it and sdxl because of the difference in training.

[languid pebble]

Coffee?

[jovial tiger]

which workflow is that?

I don't think that's the same one i'm using.

[languid pebble]

SVD from Stable Diffusion Forge

[jovial tiger]

oh ok that makes sense.

it was alwasy better at slow pans

[languid pebble]

I tweaked it a bit ...

6FPS->24FPS + 2xSloMo

[jovial tiger]

yeah i have to see if this thing can make longer animations.

or at least copy a couple of itself onto the mp4

[languid pebble]

In my workflow more than 25 IMG doesn't really work ...

[jovial tiger]

yeah, tried 48, and it's all wonky immediately.

[nimble mason]

yeah contexnt window is short

[south temple]

trying 64 at the moment

not for SVD

anything over 25 in svd gets.... off

[languid pebble]

I wonder why ...

[south temple]

SVD has no current way to batch video

you have to run it, steal the last frame, try and run again

[languid pebble]

The videos don't show a reason why it has to stop ...

[south temple]

just that the model was trained on 25 frame

you could extend the clip with interpolation. .

but thats just slowing it down

[languid pebble]

I was trained for 25 frames a Second ... 😄

At least the red one is a bad one ^^

[jovial tiger]

really good

[languid pebble]

red one becomes ugly ^^

[languid pebble]

More steps didn't help

[south temple]

nope.

25-80 is a fun experement

thats why I went back to animatediff

[south temple]

somoene WILL!!! come up with a better SVD its just time

its almost like they wanted to shaer whats coming

[jovial tiger]

yeah, but probably not stability. article today said a year ago they had $100 million. now they have 4.

[south temple]

right now im seeign what happens if I pass 64 images in the latent vs 32

The assumtion is more latent images allow for a smoother render with more frames to work with

but now we are dealing with 40+ minute renders

nope

[south temple]

clean

[clever oar]

moderator ban my friend for pos image with his lora

[south temple]

getting trash at the moment

[clever oar]

its not normal bAN for it

[languid pebble]

Might be fun after some herbes 🙂

[nimble mason]

depends what it was i guess

[clever oar]

only girls with tatoo

[languid pebble]

Where's the floppy?

[arctic laurel]

wrong universe

[nimble mason]

was it sxually suggestive? nips? young looking? etc

[languid pebble]

Damien should be OK ^^

[jovial tiger]

i' loving how clean that is, i just wish they finally did some motino with the subjects.

[languid pebble]

Time to say good nite!

[nimble mason]

send help

[jovial tiger]

you sir, need to stop throwing spaghetti at your monitor.

and jesus, this auto regional prompter is gold.

[cyan shoal]

auto regional prompter?

[jovial tiger]

it goes out to claude to make regional prompter prompts, then generates 3 variations and a whole bunch of non regional promper ones with that giant prompt to see what happens. it's awesome about 60-70% of the time

[cyan shoal]

ohhhhhhhh

[jovial tiger]

i input "man throwing spaghetti at his computer monitor" in response to clownshark's pic

[nimble mason]

amazing

[jovial tiger]

plus it's using that proteus-rundiffusion/dark arts images/andrea75c cute3d lora merge I did.

this was pretty good too. i like how the monitor is filling up with tomato sauce

[south temple]

ooo thats cool

[nimble mason]

wow

with luminosity flippei

[clever oar]

no more sexy than usual

your blood car must more trigger than girls with piercing

is strange

[nimble mason]

that's paint

[calm canopy]

Kinda begs the question what the usual is

[clever oar]

images what he posted not delete

[jovial tiger]

haha wow

he looks like he's tired of this s***

[clever oar]

he promises he won't post anymore if you unban him

[jovial tiger]

lol can't stop laughing at his face expression

"soon, everyone will benefit from AI at home"

[jovial tiger]

the shark on the right is your watermark. I'm going to save that to my ai folder because it's awesome, but then I'm going to be showing my pics to the wife and this is gonna come up. Thanks.

[south temple]

cant figure out why im still so blury

[jovial tiger]

that's svd?

[south temple]

animate

[jovial tiger]

the resolution on this stuff isn't that high, it's mostly sd 1.5 generated.

[south temple]

right

[jovial tiger]

is this sharper than yours?

[south temple]

which is why I want sdxl in the worst way lol

MUCH

[jovial tiger]

you want my workflow?

[south temple]

please... would love to look at it and see the differences

Its gotta be something im missing

[jovial tiger]

i'm not rendering btw, i just have a load image that i drop stuff onto, and then provide the same prompt I used to generate the image on the left.

in the efficient loader pos box

I use cute3dcartoon sd 1.5 checkpoint, which is somewhat close to what I use for sdxl.

[jovial tiger]

I went looking for good sd 1.5 checkpoints, and all the top sdxl ones don't have an equivalent in sd 1.5. they're mostly just porn related.

[south temple]

I keep getting this kinda stuff

and I KNOW that pack has more to it

[jovial tiger]

yeah should definitely get better than that.

[south temple]

I see yours

this was mine if you wanna drag it in to look

[south temple]

I see what you were doing I think

[jovial tiger]

that can definitely be part of it. with all this clipvision stuff, if your source image is blurry, it's going to try and replicate that it the result. you have to use a sharp source image to get a sharp result.

[south temple]

thats gotta be a part of my issue

[crisp stream]

I can recommend dreamlike photoreal 2.0 if you haven´t checked on it already

[south temple]

maybe I should load a basic txt to img and see if thats the issu

[jovial tiger]

try putting in one of these ones from clownshark lol, see what comes out (shudders to think)

comfy might ban you.

[south temple]

you are nightmare fuel

[south temple]

what is the monster control net?

[nimble mason]

qrcode one

unfortunately the sdxl one kinda blows

there's two versions of the sd1.5 one, and there's also an optical illusion one that's very much worth using

[vivid frigate]

aw man evven with negative hands

why are my fingers so messed up

[cyan shoal]

bruhh

im having too much fun with Ideogram rn, it's getting harder to get back to SD

I might just wait for Lavi-Bridge or ELLA or just simply SD3

[jovial tiger]

ideogram is awesome. ideogram also has some of the worst graphics quality around. worse than sd 1.5 most of the time.

[cyan shoal]

sometimes it looks like an SD2.1 base image and sometimes like a finetune

and none of them have highresfix so small faces are distorted

but its still a lot of fun to make memes with

[jovial tiger]

I added a prompt of: man with spaghetti for hair is on fire and smoking

[cyan shoal]

nice

I think SD3 2B will look as good as ideogram once it's finetuned tbh

[nimble mason]

some good training data for @junior sky lol

[wispy nest]

when is sd3 coming out?

[south temple]

tje apply advanced control net is broke for me, have to bipass it ...

[nimble mason]

2024

[south temple]

what does the first one do?

bottom right?

[wispy nest]

and since the ceo has stepped down, will sd3 be released with a subscription policy?

[wispy nest]

kinda like other companies?

i read articles that shareholders weren't happy with giving away tools for free

they want money

doesnt matter good or bad, its all about money ^%#^$%@

[nimble mason]

Here is the image you requested.

[wispy nest]

cool pic but i didnt request it lol

[jovial tiger]

the 2 top people of the company were just elevated to co-ceo. if they can save the company from here on, who knows.

[jovial tiger]

thats much better

[south temple]

yup

I need to find a few things to solve the issue 🙂

I need a chekcpoint for the control nets, and a lora

[hazy warren]

I bet that monitor is only 720p

[nimble mason]

Here is the image you requested.

[nimble mason]

Here is the spaghetti and pasta you requested.

[wispy nest]

lmao

[nimble mason]

Here is the shareholder you requested.

[jovial tiger]

this all with ipadapter or back to regional?

[nimble mason]

All ipadapter

[hazy warren]

making some babes I see

[wispy nest]

you can say that, testing ipadapter

[south temple]

I get that... I have days where Im testing steps and stuff

[hazy warren]

my first video

[jovial tiger]

There's some interpolation nodes that'll smooth that out for you.

[nimble mason]

the next great political party

[hazy warren]

oh wow, how did you get the words work so well?

[nimble mason]

rolled a couple hundred snake eyes on something a week ago so the universe had to give me exactly what i wanted on the first try to balance things out

sampler may have had something to do with it

workflow is in the image and this one isn't a shitstorm lol

[nimble mason]

finally tested diff diff with the new ipadapter... works great

just a quick n dirty here but this workflow is comprehensible lol

[wispy nest]

this a factor of mixing shrooms trip with robotics or ai robots the shroom trip is good and bad rarely you will get a surreal creepy image with this if steps are high so it can ""think """ of a bad trip

[nimble mason]

wut

[gusty cloak]

Maybe this guy can be of use here?

[nimble mason]

i was looking at some shit i was making just over a month ago... holy hell have i learned a lot in that time

[wispy nest]

shrumbus

[gusty cloak]

That comment inspired me to take a look at my older images 😄 Found these in my generations from January

[nimble mason]

it's kinda insane how far i've come in 30 days

i'm looking back at stuf and wondering how i didn't see obvious issues with the images, etc

[clever oar]

generate somebody lord of nekron warhammer 40k

[gusty cloak]

Ah shoot I didn't recognize you 😂

[clever oar]

thank you

[wispy nest]

wait a minitue

[clever oar]

it what i need

[wispy nest]

I made avatar yoda

new form unlocked

[grizzled sage]

ai doesn't know what conscious hallucinations would be like. theres never been a painting of that. some artists do psychadellic art but thats just an art style that looks cool while high

ai hallucinations are a whole different thing

[wispy nest]

ok captain kidd

highy unlikely

pun intented]

its trained off art people make art when tripping bad trip art and good trip art

same with skitafranic art

it does know but interpites things ever so slightly

[clever oar]

close

but need without big armor

you make hybrid spacemarine nekron😄

[jovial tiger]

well, if i just put this in with no prompt expansion, it gives this, but that's not it either according to google images. need a warhammer lora i guess.

[clever oar]

hm yes ai dont know him

[nimble mason]

finally. a realistic tornado

[potent vortex]

@clever oar 帮我生成一张北京天安门的图片，要有红旗

[clever oar]

ok ok

[wispy nest]

leo died for my sins 😦

[potent vortex]

@clever oar generate a picture

[nimble mason]

Here is the image you requested.

[gusty cloak]

That's a poor user you're tagging

[clever oar]

Clownshark Batwing is our discord bot

he generate pic for you

[nimble mason]

Here is the image you requested.

[clever oar]

today he generate war picture

[nimble mason]

Beep boop! User @potent vortex requested a picture of Beijing. My LLM subroutines generated a prompt for a calm, beautiful day with a distant sunset in modern, clean, Beijing.

[nimble mason]

Here is the image you requested.

[sour ice]

I love the dark humor.

[glacial wren]

An eighteen year old beautiful girl

#🏞｜general-with-images An eighteen year old beautiful girl

[clever oar]

Clownshark Batwing bot activate -theme dark humor

[heady anchor]

https://tenor.com/view/comedy-alert-gif-24123786

[nimble mason]

Here is the image you requested.

[sour ice]

where is the shark?

[nimble mason]

Beep boop! The shark is a subtle element of the composition.

[nimble mason]

Here is the image you requested.

[nimble mason]

Here is the image you requested.

[clever oar]

glazgo smile

[arctic laurel]

CLOWNSHARK! You generate IMAGE for MEeeee

[nimble mason]

Beep boop! Prompt me with "Prompt! :: your prompt here"

[arctic laurel]

“Apologies, NSFW material was detected in alex v’s prompt and has been discarded.”

[arctic laurel]

he has finger growing up his suit

[craggy rune]

attention couple is very powerful

[south temple]

lol

Friends forever

[nimble mason]

love all the weird stuff you post

[crisp stream]

Thank you 🤪 😄

[crisp stream]

nice one 😄

[wispy nest]

this is crazy good

[nimble mason]

comfyui workflow is embedded

[wispy nest]

nice but i dont use comfyui

too technical

[crisp stream]

it s actually pretty easy once you get used to node-based working, especially for standard workflows. You can keep it simple or go wild, whatsoever suits you

[nimble mason]

yup, you can just drag and drop ppl's images into comfyui, and you get their workflow

then you can just type in your prompts, click add to queue, and off you go

you don't need to understand it at all to use it tbh

i leave all of my workflows embedded (view at your own peril)

[nocturne oak]

Part of the fun of comfy is the fact that you can easily crack it open and make images like mad...

OR

You can spend hours upon countless hours refining a workflow without generating a single image.
...not that I, uh...would know...anything...about...

[nimble mason]

OR

You can spend hours upon countless hours generating hundreds of images while absolutely trashing your workflow...

[crisp stream]

what are all those images doing? IPAdspter?

[nimble mason]

imo, comfyui is when it becomes truly addictive

yeah

i'm just swapping them in

i wish we had a nice image browser that could keep them in a nice grid or something

[crisp stream]

btw habe you checked on Chainner for node-based upscaling and shizzle?

[nimble mason]

i haven't, anything in praticular it's good at that comfyui isn't? (i'm guessing the answer is "a lot")

[crisp stream]

I honestly couldn´t tell because I´m not that experienced with comfy yet and used Chainner already before Comfy, I don´t know, I simply like using it, feels good 😄
Currently I let it run in parallel while Comfy is generating and upscale the images I like. That for I got Comfy´s output folder opened in an Explorer window and copy paste them into Chainners Inputimage browser into an "Original" folder while defining an "Upscaled" output folder. Images bascially drop from Comfy, I copy them over and hit play. While Comfy continues its work. Pleasant workflow.

Like on the fly upscaling yet only where wanted

[nimble mason]

Ahh cool

Yeah I do almost all my upscaling tiled

I like getting the extra details, but my goal is to make shit look pretty insane usually

[wispy nest]

this is usually my situation with comfyui cause i want to do things my way

OR

You can spend hours upon countless hours refining a workflow without generating a single image.
...not that I, uh...would know...anything...about...```

its the para after OR

[crisp stream]

Can Comfy do batch processing?

[craggy rune]

yep, can even make it real time auto queue, lol

[crisp stream]

like simply processing images in a folder in a row?

[craggy rune]

yep, you can

[nocturne oak]

Comfy can be setup to do just about anything you want.

You could literally script out an entire movie if you so desired.

[crisp stream]

🙂

[craggy rune]

make 10 svd, and combine it 🤣 , and make infinite animate diff movie

[nocturne oak]

Noodles = Freedom

[crisp stream]

😄

[nocturne oak]

The problem is that Noodles also = Intimidation for the fearful.

But it's not as difficult as people think.

You just have to actually learn something instead of riding the rails.

[craggy rune]

I am still learning comfy ui, and still have new stuff to learn everyday

[nimble mason]

yup

[nimble mason]

yeah i've had it process hundreds of images

[crisp stream]

The entry is pretty low I feel when talking about standard image generation, especially with the possibility of recalling workflows that easily

[craggy rune]

even just prompting, damn

[nimble mason]

measure dimensions, crop, then upscale, pad for outpainting, outpaint eveyrithng and patch up seams, resize, partial denoise, output

(prepping stuff for training loras)

[nimble mason]

man i have been crazy into it and i'm still finding stuff constantly

so many possibilities, it's exciting, always something new to try and explore

[nocturne oak]

I have been spending all day making a clean interface with highways and ran into something that's fucking my cascade stage b that I just cannot find. But that's actually part of what I don't mind...gives me something to figure out. (The workflow will have switching between XL & Cascade at the start and all the noodz are clean with highway pipes.)

[crisp stream]

have you got image adjust at hand like here as well?

[nimble mason]

also most of us alawys leave our workflows embedded in our images so def check those out and we're pretty happy to answer q

[nimble mason]

almost all of that and more but it's clumsily spread via diff nodes

don't mind it too much but what we're really missing is a curves node

[crisp stream]

like what exactly?

[nimble mason]

https://helpx.adobe.com/photoshop/using/curves-adjustment.html

that's the bread n butter adjustment tool

[crisp stream]

ah, like in PS, I see, yes, that would be handy indeed 🙂

[nocturne oak]

Well suck me sideways, just talking about this made me go back and look at my Stage B issue and I figured out what was wrong with my pipe.

[crisp stream]

@nimble mason one of the first things discovered is the Latent Blend node for combining a basically arbitrary amount of input images in a cascading way if it´s more than two, then I guess you already know that one 😄

[nimble mason]

infinite trash

[crisp stream]

@nimble mason btw how does inpainting work by now? Heard it was terrible before yet should be ok now?

[wind egret]

whys the preprocessor making a blank image

[earnest solar]

ask in #🤝｜tech-support

and show them the command line that it generates, might help.

[crisp stream]

just couldn´t stand another anime girl image 😄

[earnest solar]

Hey man, I spend all my time tortuing the robot to make fantasy Mandalorians

[crisp stream]

not my fault 😄

[nimble mason]

In comfyui? Just gotta set up the workflow properly then it's better than anywhere else

I can drop a Json tomorrow

[crisp stream]

sure, would love to check it out, thank you

[nimble mason]

Differential diffusion is incredible for it

Still not in a1111

The new ipadapter is insane

Also not in a1111 and likely never will be

[crisp stream]

yes, haven´t checked it out yet, merely loaded the models so far

[nimble mason]

Mask editing is fine now in comfyui, better than in a1111 at least in my exp... Not that either really have more than like three features anyway

[craggy rune]

can I do solid mask with drawing ? @nimble mason

instead of settings this width and height

[nimble mason]

Yeah

Load an image, any image

Right click on it and click the edit mask thing

Then you can use it directly from that

Left click to draw the mask, right click to erase

[crisp stream]

I´m so far used to using Easy Diffusion that for (haven´t seen better yet).

The inpainting overlay:

And the drawing tool (both can be combined as well):

[craggy rune]

your robot need an anime girl

[earnest solar]

😭

[nimble mason]

I feel like there needs to be a clown shark using the trunk as a saddle

[crisp stream]

Go ahead? 😉 😄

[nocturne oak]

Dat neck tho.

[south temple]

MONEY!

"some men want to watch the world burn"

[crisp stream]

https://cdn.discordapp.com/attachments/990205833235279872/998889294569685062/confused-no.gif?ex=66150f26&is=66029a26&hm=8509cf0194a1e9ed17326425ec28706e40d424580174cc08089ce131dd91094f&

[sterile kiln]

Stable.Art is largely better.

[sterile kiln]

@crisp stream
@nimble mason
I did this video 1 year ago, on 1.5, the original image is here ( i did it too on 1.5).
https://www.youtube.com/watch?v=jtsGTZp14D4

The dino is weird, because it is a a picture of a toy. I did inpaint on him but he's stil weird x)

a recent example, (with SDXL) :
https://youtu.be/utjxgMQbshk

The final image :

[languid pebble]

Clever A.I. has decided: This can only be an Asian 😄

[languid pebble]

Buddha is traditionally portraited in Lotus seat ...

[rustic fossil]

给我一个表示高校汇聚外部资源发展的图片

生成表示高校汇聚外部资源发展的图片

[languid pebble]

No Bot to generate here

[clever oar]

how many vram need for generate this

[languid pebble]

Not sure ... I think I started with 8GB

[clever oar]

i have 4)

[languid pebble]

Give it a try ... noch much more than generating picture after picture ...

[clever oar]

its only comfy?

[clever oar]

can i get same result on automantic?

[languid pebble]

I am using Stable Diffusion Forge, but A1111 also works ...

SVD Plugin ...

[clever oar]

what extension i need diff?

[south temple]

man SVD has got some great stuff in it

I might have to create a workflow for txt -> SVD

cuz.... SDXL.

lol

[languid pebble]

txt2vid is to unsecure ... better see first pic before you decide to create vid ....

[south temple]

im testing steps in animations at the moment

yeah

I need a more real trained LCM model

im almost at 50 steps

[south temple]

isnt there an area of pointless? in terms of step count?

beyond 80?

[languid pebble]

In SVD?

[south temple]

50

animate

[languid pebble]

In SVD more steps doesn't look to improve a lot ...

[south temple]

this is the image im playing with

[languid pebble]

I'm not sure but you ani looks like it's a short one playing 5 times

[south temple]

its a multi 4

so you are correct in 5x

my bad... you are right

should be 5

interesting

[languid pebble]

Not sure what's the benefit for that.

[clever oar]

is svd?

[south temple]

im playing with settings so I can spot the changes

I am playing in animate

but I am trying to understand what all the changes do

[clever oar]

why no secure?

[south temple]

its not bad... its just too dream like

steps at 35 - running through CFG 1 step a time going up to 12

[languid pebble]

Text you will no know what you get ... and calculate till you get the video ... starting with an image you have a sharper idea

[clever oar]

in automatic 111 what i need for generate this

[south temple]

gen what?

whats your end goal?

[clever oar]

like your video

and what there posted often

[south temple]

we are both using 2 different things to get to the same result

SVD is a img or picture to video.... tons of settings to play with

AnimateDiff is ... SD

[clever oar]

svd not aval for auto 111 yes

[languid pebble]

SVD not really many settings to play with 🙂

[languid pebble]

It is

[south temple]

SVD - I know what I want, I am starting with my idea... and I want the computer to get there.

Animate - Cluster FK

[languid pebble]

There's also deforum ... much more complex ... but you can use Video as input

[south temple]

I havnt gotten into deforum

which people swear by

[languid pebble]

I have used it to create a music video once ... a lot of time calculating and slow learn effect ...

[south temple]

after I run some more tests Ill try SDXL SVD

[languid pebble]

Last version under 30 MB 🙂

I stopped at version 12 ^^

No video input, everything just prompt generated

[south temple]

oog

thats a little... out there

high CFG is trash on this

for LCM

step count is about 30 for a target of good quality... CFG .... users taste

8-10 is target

[languid pebble]

more than 1 year old 🙂

[south temple]

high CFG

[tired yacht]

generate clouds

[dull flame]

this is sopposed to add detail, but i dont see any difference = (

[languid pebble]

I have had a problem like that the old days ... first tries with SDXL ...

[south temple]

I see some clarity

[dull flame]

im trying a 2048x2048 to see if theres a difference

[south temple]

details are seen in the shadows and lighting

its doing what it thinks is right

[dull flame]

apparently, after some reading, its not sopposed to add a lot of detailes, just a little

like, fine details, like u said

[south temple]

looks like its just adding reflection mapping

stones, shins

shadows

seems worthwhile if it doesnt add millions of years to the render

[sterile kiln]

You're talking about svd? What workflow are you using?

[south temple]

animate

[sterile kiln]

Do you have a link? Or a picture to share the workflow please?

[south temple]

whats your graphics card?

[sterile kiln]

I already use svd, i have a 2070s

But i have doubts on my shit personnal workflow

[south temple]

right now im playing with animatediff and comfy

and I cant offer much advice on it since Im not fully understanding all the things right now either ;P

[sterile kiln]

Was thinking you were using a good workflow

I will dig, because when i see some videos/gif from civitai, i think mines look like dogshit

[dull flame]

tried another detiler, seems to somwhat work, but idk

[south temple]

its not great yet

light all over hell

[buoyant moss]

ight all over hell

[languid pebble]

slower is a bit better ...

[south temple]

OK, I need to find a new SVD flow since the one I had was now broken

[languid pebble]

Multipass ...

[south temple]

dang son

if you have a workflow in comfy lemme know.

[languid pebble]

Using simple forge ...

[south temple]

curious if I could link it to the still from promt

Time to build an svd 😛

[languid pebble]

btw heared that Forge and A1111 have a comfyUI support add on

Didn't try but could be fun

ComfyUI I failed to install IP Adaoter Plus 😄

[unborn falcon]

i don't remember if i posted these or not, but i made some wallpapers for my iphone that were abstract expressionist oil paintings of the moon at night using comfyui on a t4 gpu and the sdxl model with a high-detail lora. i wanted to emphasize the texture of the paint, especially the paint-strokes and palette knife.

[languid pebble]

Nice

[leaden fern]

yea

[languid pebble]

Imagine these pictures would paint itself 😄

[languid pebble]

Nini!

[south temple]

im breaking on SVD

Error occurred when executing KSampler:

CUDA error: invalid configuration argument
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1.
Compile with TORCH_USE_CUDA_DSA to enable device-side assertions.

File "D:\Ai\ComfyUI_windows_portable\ComfyUI\execution.py", line 151, in recursive_execute
output_data, output_ui = get_output_data(obj, input_data_all)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

[languid pebble]

I'm using:

[south temple]

ill see if I can find that to see if thats the issue

nope

I have that

neede an update

wierd

life is good

[south temple]

annoying because it crashes so often.

back to animate

lol

[gusty cloak]

In case you had forgotten what base 1.5 looked like 😂

[abstract valve]

product photo flat white background black leather motorcycle racing jacket slim eruopean tailored fit dainese alpinestars style reqalistic photo

[nimble mason]

Here is the image you requested.

[south temple]

welcome back

I cant solve for how to stop images from morphing into each other like there isnt enough frames

[whole garnet]

Una bandera de lila, blanco y negro

[nimble mason]

Here is the image you requested.

[south temple]

well that is way better than what I just spend forever rendering

[lean talon]

so many controlnets, but only takes like 5 minutes for a 3 second morphy blend of 2 images

[south temple]

img->img?

[lean talon]

yeah

this workflow

https://civitai.com/models/372584?modelVersionId=416171

its so good

[south temple]

feel free to drop my kitty pic into comfy

[nimble mason]

feel free to drop this into comfy

[crisp stream]

yes, a full photoshop integration most likely is, then it´s tied to Photoshop and while I´m working with PS I am still on CS6. There is something for Kreta as well, yet haven´t tested it yet

I was basically referring to SD browser UI I have seen so far

[thin flume]

Hey guys hallllllp!
Non-native English speaker.
How am I supposed to describe (to prevent) the background spots?
What do you call it in English?
(trypophobia WARNING)!

[south temple]

you could try a prompt "clean white backround"

[steep oyster]

#dragon fish chinesestyle

#🏞｜general-with-images dragon fish chinese style

[south temple]

I need to render a robotic cricket in the woods, chirpin its little butt, spitting all sorta robo colors.

[nimble mason]

Here is the image you requested.

[south temple]

well thats... a place

my files are getting too big to paste

10 minute renders -.-

Got it to choke! Lol wow

Batch size.

120 was much

[sterile kiln]

sorry to tell you this but in 2024, still being on CS6... you're a dinosaur man

I can send you "a link" if you want.. 🙂

[thin flume]

Thanks, but this didn't worked well.

[maiden lantern]

sexy half body

[languid pebble]

Got it running?

[cedar basalt]

a style of Chinese landscape painting associated with the artist Zhang Dachuan

[vestal goblet]

#🏞｜general-with-images coffee shop

[wispy nest]

i’m done making dnd art

[nimble mason]

Here is the image you requested.

[nimble mason]

Here is the image you requested.

[nimble mason]

Here is the image you requested.

[crisp stream]

I´m actually not missing much, except for AI in that sense. Sure some newer stuff is handy, yet for daily work here CS6 is still fine and also licensed 🙂

[languid pebble]

Don't forget all the time to spend to relearn or create new workflows ...

[crisp stream]

wdym exactly?

[languid pebble]

Updating to a new version you have a new UserInterface and need some new workflows ... costs time

[sterile kiln]

when Adobe takes people for cash cows, I personally have no scruples. But if you're one of those people who's self-flagellating about being a huge victim, that's a shame, you're missing out.

As some would say, hacking Adobe software IS ETHICAL

[crisp stream]

I´ve worked a bit while teaching with the latest version and I actually don´t find it that much different, sure a slightly different layout and a bit new functionality, yet quite similar

[clever oar]

hello

[crisp stream]

I´m simply working with the version I bought back in the day, saying avoiding CC. For now it´s working 🙂

[clever oar]

DiffusionMastered make me hybrid two animals

[crisp stream]

Btw almost equivalent to CS6 if you don´t have any access otherwise is http://photopea.com (free + online)

[sterile kiln]

non-legit versions have the formidable advantage of not having CC.
Photopea and others are handy, but after a certain stage they're obsolete.

[crisp stream]

then they are non-legit, and while some might play the if you do it to me I do it to you game, I prefer working legitly

[gleaming dagger]

在月光的洗礼下，一位身着华美唐装的女孩，轻盈起舞，水袖飘扬，如同穿越时空的仙子，将听众带入一个梦幻般的大唐世界

[clever oar]

Clownshark Batwing our friend need picture

you know what to do...

[languid pebble]

wrong model 😄

[languid pebble]

SVD and hands ... not really good

[languid pebble]

Cheesy 😄

[clever oar]

put this woman to a pile of mice

[clever oar]

😃 its sushy?)

[ripe pilot]

It tried to be

[languid pebble]

I like that ... tried to play around with it, too

[languid pebble]

To much party last night? ^^

[crisp stream]

nah, was chill 😄

[languid pebble]

Same here

[clever oar]

nice!

[languid pebble]

Bad luck last seconds only stars move ...

[clever oar]

its svd?

[languid pebble]

Yes

[south temple]

I HAD SVD running, but then it started throwing Cuda errors

Animate works… I just cant get it to stop ghosting between images

I have tried slowing frames, interpolating between. I have tried changing motion, and adding frames to the batch

IM positive I’m missing something with the frame rates to motion in order to get a fluid image without ghostly transitions

[languid pebble]

I don't think I ever tried Animate ...

[south temple]

You have a TON of control…

[crisp stream]

slightly shape-edited

[languid pebble]

Looks complicated ... couldn't install the new IP Adapter today 😄

[south temple]

It is… and it’s not… but god help you if your graphics card is not a baller

10 minute renders - 3090ti

Which leaves me to think I’m making it work too hard

[languid pebble]

I'm chilling ... rendering in silent mode now ^^

[south temple]

SVD ?

[languid pebble]

Yes ...

[south temple]

Part of me might want to pick up on what you’re doing…

I dont know that I want to have ANOTHER program lol

I was hoping the deep dive into comfy would help me really understand what each element is doing. IM LOADS BETTER at understanding… but sometimes I think the machine just does what it does lol

[languid pebble]

And no idea where to put all these models ... ^^

[south temple]

I delete… a LOT

I have learned there are a LOT of crappy checkpoints lol

[languid pebble]

found something interesting but maybe not if you only use Comfy ...

[south temple]

Tell me more? 🙂

You have some amazing renders…. And they frustrate me enough to want to try and do what your doing 🙂

[languid pebble]

1 Place for modes for all Ui's https://www.youtube.com/watch?v=qcpfrpMbCA8

For SD Forge there yhould be a 1-click-installer now ...

[south temple]

You have a link where I can learn about it ?

(Also … thank you for being awesome )

[languid pebble]

The one click installer? There are a lot of Youtube videos

[south temple]

I’ll dig through.

[languid pebble]

I've installed it the old way ... so I only know it's out there now ^^

[south temple]

Found some stuff.

[languid pebble]

I think it's worth a try cause it's so easy ...

[south temple]

Smooth renders that are up to 4-8 seconds is all that matters 🙂

[languid pebble]

You could try:

[south temple]

It’s an XL

Animate gets really mad at anything not 1.5

[languid pebble]

But only 8 steps ...

Ahh... ok ...

[south temple]

You CAN use SDXL files… but the animations are trash

I have been pushing the limits of like… 40 steps, and a cfg of 8-12

Which may or may not be a reason for the ghosting

[languid pebble]

Some models want lower steps and cfg ... suer

dreamshaper I'm using 8 steps, CFG: 4

[clever oar]

what i need for use ip adapter model and encoder?

ip adapter plus good to change face?

[languid pebble]

Yes, it is. Install can be hard at the moment. still a lot of problems ...

[clever oar]

why

[languid pebble]

I tried with comfyui without success

latest technology ... it takes time till all the parts work together perfect

The A.I. is a mix of many different modules ... you update one and that causes problems with others ...

[clever oar]

clown shark said that is very goo

[languid pebble]

It looks pretty good, yes! I have used a youtube video from Oliver Sarikas ... maybe you find one from the coder of IP Adapter?

[languid pebble]

Need to sleep! Good luck to you!

[clever oar]

bye good night

[south temple]

I kinda wanna know the prompt for that octopus

Motion bucket is 127

!!!

I never thought about playing with that in a lower bucket number

A lower number leads to a more stable motion I just read

[jovial tiger]

That's just Dalle with happy octopus and people, put through img2img to turn it evil. 🙂

Here's stable diffusion though. grotesque evil octopus squirting blood from its tentacles at shocked people

[south temple]

He’s angry

[clever oar]

pth its bad ? i need safetensors?

[south temple]

For what?

[clever oar]

for ip adapter

i need encoder

sd try to download but fail

its safe for system use pth

[south temple]

I used comfyUi / and it updates automagically

[clever oar]

i readed that cpkt can start virus

and need only safetensors

[dusty kestrel]

Yo guys, I'm trying to use sd-inpainting to fill the gaps in an image using a mask, but I tried playing around with all of the settings and none of it seems to do absolutely anything.
Here's the image and the mask, any ideas?

example output:

what am I doing wrong here

[surreal veldt]

I'm trying to substitute the background of a ballet student using inpainting, but every time behind him come some ghostly dancer like in the attached image. I tried using in negative prompt "no people" or similar, but didn't worked. Wanted to ask if someone else has experience on this and can give an hint.

[south temple]

Some of this stuff is not 100% what you are doing wrong. Some of it is just broken outputs.

I have spent days…. Running a test on perimeters just to see the difference in small / big moves in both ways

Only to find just one more parameter to play with 😛

[clever oar]

how i manualy install encoder for ip adapter

[south temple]

Might be a file you need to snag off the GitHub and drop it into the right folder.

[clever oar]

what folder i need?

https://huggingface.co/h94/IP-Adapter/tree/main/models/image_encoder

this where need put

[south temple]

Im not the best at this, but you’ll have to dig for a folder that would host your ip-adapter. Im not sure if you put that in your checkpoint folder or the ip adapter folder or both

YouTube / google is gonna be more helpful than I

[crisp stream]

From what I know this one is preferrable: https://github.com/cubiq/ComfyUI_IPAdapter_plus

[clever oar]

you put clip_vision in model folder?

[clever oar]

i am broken why is so hard

[crisp stream]

guess so from what I remember, haven´t installed it myself yet

[clever oar]

fail download

i want install it manualy

i read that ph files dangerous

and i want use safetensors

[eternal dome]

create new folder , name it "ipadapter" put it in "models " folder and download ipadapter files , put it in that folder

[clever oar]

i need encoder

clip vision intall

i need install ip adapter plus and encoder in one folder?

[clever oar]

thx i will try

[nimble mason]

Yeah new ipadapter blows the old one sky high out of the water

[cyan shoal]

got deepfloyd working lmao

[nimble mason]

Wayyyy better than clipvision

[nimble mason]

Now that's an achievement

I gave up on that

[cyan shoal]

it was difficult

the worst nightmare was figuring out xnosh and huggingface login

[floral flicker]

For when you need to corp your images

[south temple]

I just found myself in the dev discord for AD

And boy… thats useful for later

[clever oar]

not see encoder

[eternal dome]

What encoder ?

[nimble mason]

AD?

[south temple]

Animate diff

[nimble mason]

oh, nice

[nimble mason]

composition, style, output

it's not the same thing

i shouldn't have said "model" cuz it isn't one with ipadapter+ i think

i don't even have that file lol

uh, cuz you said something about it being available on webui, so i used google to see what you might be referring to? jeez

[grizzled sage]

yup. now i got you pegged. you one of those. noted

"no you can't do that"

[nimble mason]

never said a word about composition before

you still can't do what i'm doing in webui

[grizzled sage]

yeah i'm not going to get baited into that . cult of comfy is pretty locked in to their "supiority"

[nimble mason]

@grizzled sage can you take those two images and get an output like the third? i'm genuinely curious what the state of things is over on 1.8 with this

prompt was "high quality, detailed"

that model follows the original composition more precisely, the ipadapter plus version more loosely. so it's looking like there's advantages to both.

[grizzled sage]

i could but this is obvious argument bait. i already showed you examples and you declared it wasn't the same thing and told me it was impossible. hard to believe this is genuine and worth my time

[nimble mason]

composition, style, output

well, it's not the same thing, but believe whatever you want

that's with the model

this is with the ipadapter+ "composition" weights drop down option

both combined.

[south temple]

Please train all future ai on this please and thank you.

I missed the argument.

Did I say something?

[nimble mason]

there wasn't one lol

just playing around with some cool new shit with ipadapter

i didn't realize an actual model had dropped for composition shit

[south temple]

Adapter plus?

[nimble mason]

so now i'm comparing with the new ipadapter plus composition weights

[south temple]

You’ll have to help tell me more about that lol

[nimble mason]

with both, with the model, with the ipadapter plus composition weights

[south temple]

I have a few workflows I’ll drop in here that I’ve seen from the devs. I think that’s permissible

[nimble mason]

input

probably

[south temple]

It’s the weights I think I have to find

[nimble mason]

the workflow should be in these images and yes i finally gave in and cleaned it up lol

prompt was "high quality, detailed" so nothing from conditioning helping guide it really

[south temple]

(Not at the house yet… cuz work life) but ya know lol

Are weights a thing we have to download?

[nimble mason]

yeah, which are you looking for? stuff for ipadapter?

[south temple]

Your screen is like magic the gathering

[nimble mason]

now we're in business

[south temple]

Which honestly… would be a cool ai game to play. Someone creates a thing, and you have to defeat it with another thing

[nimble mason]

i just switched the model for the style weights node

ha yes

...it really does look like i've laid everything out for a card game, yeah

so looks like it's worth trying the weaker ipadapter models when using the style transfer weights with ipadapter

with the ipadapter plus model... the strongest one... this was the result

lost too much of the composition

[south temple]

Real death squad cyborg, inflated with feathers and talons, this robot with fierce god ray eyes, vomits tentacles of slop and glory, baby kitten, evil to all hell

[arctic laurel]

Yeah I’m gonna need an image of the band TOTO performing atop a giant sandworm

[nimble mason]

Here is the image you requested.

[south temple]

Pink Floyd plays with megadeath in the clown cave

[nimble mason]

(i just added your prompt, everything else was the same)

prompt for both above was just "high quality, detailed"

well, both above before the one i just posted right here^^

[south temple]

That dude rages

[nimble mason]

running now lol

[south temple]

Oh god

[nimble mason]

new ipadapter stuff is so addictive

[south temple]

Let’s . Gooooo!!!!

[arctic laurel]

What’s this new adapter busisnes

[south temple]

It’s sick nips

[nimble mason]

it's the new ipadapter nodes in comfyui

great stuff

[arctic laurel]

cannynips?

[grizzled sage]

i blessed the sands down in arrakis

[south temple]

Canny bus

[nimble mason]

Here is the image you requested.

[south temple]

Deadmau5 performs atop the worlds largest dead mouse

I guess that works lol

The head of a cat, the body of Thor, the grunt of Godzilla, he makes the worlds best cheese dip, you should see his glutes, oh lord the power of fart, landslides of gooey peanut butter, oozing out the nose, he has a cold

You may call him bill

[nimble mason]

composition, style

[south temple]

Wicked 🙂

I’ll leave this here so it helps someone

[frosty edge]

1 girl

[nimble mason]

Here is the image you requested.

[fiery whale]

any body knows how to set up controlnet reference only with diffusers??
no pc so im on colab with commandline diffusers but i cant seem to find the damn controlnet model
if anybody can reply with code thatd be gr8

[arctic laurel]

The original Cyber Woman With Corn and two img2img variants

[outer galleon]

brain

[jovial tiger]

yes?

[nimble mason]

that's aicreatorartistic... this is with block 1 in/out from proteus rundiff

got something messed up that led to the extra leg... one sec

[jovial tiger]

booo proteus = extra leg

[nimble mason]

lol

there

[jovial tiger]

i may come back to it, but suddenly i'm soured on proteus after working with it a bunch in a/b testing tonight compared to that ai creator one.

[nimble mason]

i just like borrowing block one from others as it doesn't seem to really derail the model, but can result in some pretty different tones/styles

[jovial tiger]

this thing is wild.

definitely

[nimble mason]

reverse in-out, before was linear

[small cloak]

生成jay chou 的照片

[nimble mason]

Here is the image you requested.

[novel nest]

How to use?

[nimble mason]

Prompt me with "Prompt! :: your prompt here"

[small cloak]

Help me generate a corgi dog

[jovial tiger]

what's the corgi doing?

[nimble mason]

Processing.

[nimble mason]

Here is the image you requested.