🧣｜comfy-ui

[olive yoke]

why?

use the command i sent

[robust bloom]

i am new

[olive yoke]

all of this is in the comfyui readme no?

[robust bloom]

i followd that before.

i will use your command. just wait

now it is 2.2.2,python3.11 -c "import torch; print(torch.version)"
2.2.2

[olive yoke]

okay great

python3.11 main.py --force-fp16 --fp16-vae

also try:

python3.11 main.py

otherwise, a custom node is broken in a way that no one here will be able to diagnose. you should be testing with a simpler workflow

[robust bloom]

thank you for your sharing. it is still not worked .it generate pure color image. what i didn't do is this venv . python3.11 -m venv venv. i don't know if this is issue?

[olive yoke]

okay listen

why did you do that?

you're not at that stage yet

open a new terminal tab

cd then your comfyui git directory

then try

python3.11 main.py --force-fp16 --fp16-vae

then, if that produces a pure blue image, try python3.11 main.py

then, if that produces a pure blue image, your issue is related to the custom node

read what you are doing

don't do anything you don't understand

[sick badger]

anyone try cosxl yet? downloading now but curious what kind of results it yields

[steep marlin]

Probably pretty similar to the results you'd get with playground 2.5, since that seems to be what cosxl is trying to match. Pg2.5 can do pure black and white and has an extremely awesome color range

[sick badger]

similar to what i was building earlier, but yours is way more robust

[robust bloom]

thank you so much.

[olive yoke]

did it work?

[robust bloom]

i just see this message, i will do this.

[olive yoke]

i don't think you are using my fork, but if you are, i merged the cosxl support

[thorny lily]

playground definitely did not invent the proper color range thing

one of the XL release candidates had proper color range via zsnr, and another via offset noise, both of which were community published techniques originally used on sdv1 & 2

cascade also has proper color range, as does sd3

[robust bloom]

@olive yoke it didn't work, this is my checkpoints content: checkpoints % ls
LICENSE safetensors sd_xl_refiner_1.0_0.9vae.safetensors
SVD sd_xl_base_1.0.safetensors xxmix9realisticsdxl_v10.safetensors
put_checkpoints_here sd_xl_refiner_1.0.safetensors ,and this is vae content:vae % ls
put_vae_here sdxl_vae.safetensors, i download fromhttps://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/blob/main/sd_xl_base_1.0.safetensors and,https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/blob/main/sd_xl_base_1.0_0.9vae.safetensors, and,https://huggingface.co/stabilityai/sdxl-vae/blob/main/sdxl_vae.safetensors. and https://huggingface.co/stabilityai/stable-diffusion-xl-refiner-1.0/blob/main/sd_xl_refiner_1.0.safetensors, this should be ok?

[olive yoke]

do you mean it's sitll blue?

can you try a basic workflow?

you don't need to use an sdxlvae

just use a demo workflow from the examples

and see if you get blue

[robust bloom]

ok. what i didn't do is python venv environment. how i can start from a basic workflow? is there an example .

[steep marlin]

Didn't say it did. We are talking about cosine SDXL vs playground 2.5 SDXL. Betterto compare apples to apples

[robust bloom]

yes, it is blue.it seems that no data was generated.

can you please share me a url of a basic workflow? i try that.

[grand pike]

hi does anyone know where i can find the bnk_unsampler custom node?

i get an error when trying to load a workflow designed for re imagining existing pictures

[fierce hornet]

This I think: https://github.com/BlenderNeko/ComfyUI_Noise

btw, github search is great for finding this kind of info https://github.com/search?q=bnk_unsampler&type=code

(assuming ofc the extension is on github 😛 )

@grand pike

[grand pike]

Appreciate it. Unfortunately I have all that installed already, still getting a missing bnk_unsampler error.

@fierce hornet I'll try out that search too. Github is just a little intimidating lol

[fierce hornet]

np, from the manager as well there's a "Nodes Info" that you can search through as well 😉

But yeah, that's the only one I'm seeing with that name

[grand pike]

your advice helped, found a different unsampler called easy unsampler. apparently bnk_unsampler is part of the comfyui-noise package. but it just doesnt work for me. google found one reddit thread mentioning the functionality broke after some kind of update a while ago.

havent made an image yet, but i have a good feeling

[steep marlin]

Update it. If you're using comfy manager you can update all your custom nodes. I want to say there was a fix today for the unsampler node, I'd have to double-check though

Otherwise, you have to go into the custom node folder, open a cmd prompt and run git pull to update it

Comfy manager just automates that process and runs the requirements.txt lists as well

https://github.com/BlenderNeko/ComfyUI_Noise/commits/master/

Yeah he updated it today

[grand pike]

@steep marlin oh dude you're totally right, i see them mention it here in the issues news

[plain basin]

Is there a way to see where a node is looking for a file? For example, ...

[lusty pewter]

That node looks in models/ipadapter

[plain basin]

Unfortunately not.

Even with a model in models/ipadapter, nothing can be selected. This makes me wonder where it's searching...

[lusty pewter]

Did you refresh after adding the model file?

[plain basin]

Indeed

If only there were a way to see where a node is looking.

[lusty pewter]

Because like...this works:

I can tell you that is the right place

[plain basin]

Not sure what my issue is then, it just isn't there.

[lusty pewter]

What isn't there?

You have to go get the models and put them there.

Installing the node doesn't do it.

[plain basin]

Let me make this as obvious as possible for you.
When you suggested that path I moved a model there and restarted cui

at this point I'm thinking its an install bug. But at the end of the day I didn't even ask. ALL i initially asked is if its possible to track down where a node is looking

[lusty pewter]

You don't need to talk to me like I'm some sort of asshole or idiot. I'm trying to help you.

What is the file you put in that directory?

[flat talon]

Is your IPAdapter plus the latest version?

[mortal ferry]

Hey guys, first time here, i'm a newbie comfy user and i want to know if i can generate a random seed automaticaly in efficient nodes

[summer fern]

How to solve this situation

[wild thicket]

aa

[steep marlin]

Click the green recycle button to cycle between random or fixed seed

Also, if you generate an image with a seed you like, you can click it again to remain on that seed

[mortal ferry]

Alright dude, I'll try it tomorrow, thanks

[steep marlin]

Np problem, gl

[lost totem]

Is there a version of Comfy UI optimized for AMD?

[sly sundial]

I've been using a workflow that worked perfectly fine. I then loaded up the workflow from @storm folio to merge the CosXL model and had to install the "ModelMergeSDXL" node to complete it. That worked perfectly fine as well. I then tried to use the previous workflow again, and preview windows no longer seem to display anything. I tried disabling the nodes that I just installed, but still no previews appearing. There are no errors and the workflow does complete.
Anyone else seen anything like this?
Edit: I deleted the nodes, re-added them and now it works 🤷🏻‍♂️

[lusty pewter]

@storm folio - When you blow up the IP2PSampler group node in your CosXL example, the reroutes from the prompts aren't connected to anything on the other side.

[storm folio]

update comfyui

[lusty pewter]

I literally just did a few moments ago.

[storm folio]

did you refresh the page?

[lusty pewter]

I didn't even have the browser loaded at all when I restarted after the update.

[storm folio]

do a shift-f5 refresh

[lusty pewter]

No go. Breaking it into nodes still has them disconnected:

That's immediately after the explode.

[storm folio]

that exact bug is something I fixed days ago before I released the workflow

[lusty pewter]

Wild

[storm folio]

so make sure everything is updated

and properly refresh

[lusty pewter]

I promise it is and has been, but I'll clear the cache, close it all down, and try one more time.

No go.

[storm folio]

can you upload the web/lib/litegraph.core.js file to here?

[lusty pewter]

Yeah, gimmie a sec. 🙂

[storm folio]

that seems fine, make sure your browser hasn't cached an old version or you don't have an extension that's messing it up

[lusty pewter]

I dumped the cache for 127.0.0.1 completely before the last test.

Extension-wise, I'll see what I can do.

[lusty pewter]

34 extensions disabled...basically everything but the manager and it is working as intended:

I'll figure out which one.

[hushed pike]

how can i use controlnets in comfyui to remake the anime girl image with an integration of the stuff in white, so itll like overlay the the black picture and keep everything the same except for the stuff in white which itll remake, sorta like impainting but different becuz if u do this on inpainting itll just fill it in in an normal way, even if you type in "black". also that involves drawing a mask, whereas this method involves using a seperate png as the mask or mask equivelant for controlnet. does anyone have a workflow or comfyui image that does this, or comfyui tutorial or explanation for this sort of thing

[lusty pewter]

Found the culprit. It's 378 from yolain, ComfyUI Easy Use.

[maiden laurel]

anyone know a comfy equivalent to this extension? https://github.com/thygate/stable-diffusion-webui-depthmap-script

[sly sundial]

There are quite a few of them. Have you got "Zoe Depth Anything"?

Just type "depth" in the node search

"Depth Anything"

[maiden laurel]

these are the only ones I can see

[sly sundial]

What about searching in the UI, not manager?

[maiden laurel]

the what

o

i already have it

[sly sundial]

Come out of manager and double click on background, type "depth"

[maiden laurel]

all done

any clue how to switch between workflows?

[sly sundial]

https://tenor.com/view/first-time-meme-movie-james-cameron-western-gif-17538858

You can drag and drop the workflow into the UI, or just click on the "Load" button.

[sly sundial]

It's also possible to have multiple tabs open, with a different workflow in each. They won't all run at the same time, but in the order you run them in.

[maiden laurel]

so what's the "send to workflow" button?

[sly sundial]

Where?

Do you mean "Save..."? That saves in a list that you can choose from when you load.

[hushed pike]

u can also drag and drop other peoples images made in comfyui to copy the workflow that was used to create the image

[late bronze]

#dream

[lusty pewter]

No, Send to Workflow is a right-click option that looks for an input image node on a workflow you select. It's rather handy when you want to pass something from one workflow to another like into an img2img or upscaling workflow.

[mortal ferry]

Which could be a good setup for Ultimate Upscaler ?

Because the image output is worse than the normal upscaler

[wild thicket]

like

[pure salmon]

like

[round spire]

Which could be a good setup for Ultimate Upscaler ?

[opaque parcel]

how can you inpaint in comfyui/stabilitymatrix?

[fleet fern]

How do I clear the image inside load image node?

And also. How do I avoid running two node tree at the same time? I have a node tree for txt2image, image2image inpainting and detailer in the same wf. When I want to do txt2img, my image2image tree gets activated as well.

I made this simple "switch" to "activate" which process I want.

[silver nebula]

Control B to switch a node off/on

[olive yoke]

what is your goal?

[mortal ferry]

Nvm i found another way, thanks

[last root]

new to confyui but what is the blue outline that on the grid in how do i make it bigger or remove it

[lusty pewter]

You do nothing and don't remove it. That is the default screen outline that shows where your screen is when you start.

If you didn't move the canvas or zoom out, you'd never see it.

But it doesn't do anything but show you where you would be when you first load up.

[fierce cloak]

How to rotate between terms in a prompt? For example, I want cat and dog to interchangeably be used rotating between the two, I right now am doing it like:
(cat/dog:0.5) Meaning the word cat and dog are interchangeable and it has a weight of 0.5 in the prompt weights
Is that way to use it correct?

https://github.com/asagi4/comfyui-prompt-control

"Prompt interpolation
a red [INT:dog : cat:0.2,0.8:0.05] will attempt to interpolate the tensors for a red dog and a red cat between the specified range in as many steps of 0.05 as will fit."
no idea how to interpret that

Also I see the prompts there are using [ instead of (

[prime tide]

How do i change the transformers/huggingface temp folder on comfyui?

[last root]

can i have 2 confyui instance/folders so that some nodes dont clash when using some specific workflows?

[dapper mountain]

《masterpiece》,《best quality》,8k,high detailed,ultra-detailed,Battle on a desolate battlefield, (clashing of sword and axe), (energy ripples from impacts), (cracks in the earth spreading), (dust and debris kicked up), (intense expressions on warriors' faces), (sound of metal striking metal), (backdrop of a barren landscape).

[sly sundial]

Did you want to randomly choose between cat and dog each time you run the prompt, or alternate during the making of a single image? If you want to randomly switch between them each time you run the prompt, and are using a node that supports wildcard prompts, you can use {cat|dog}

[hoary cradle]

Hello
Does anyone know of a way to fix a design based on two images of the front and back and output images from various angles?

[prime tide]

I figured it out, God bless Symlinks.

[steep marlin]

good ol' mklink /j

[olive yoke]

this is very vague. what concretely are you trying to do? be specific with what is inside the image, and how you want to correct it

[olive yoke]

sure.

[smoky bear]

is there any way to quickly run only certain nodes without having to bypass everything you dont want?
instead of saying what i dont want to run i want to say what i want to run

[hoary cradle]

@olive yoke For example, StabeleVideo3D creates a turntable from a frontal image. This involves directing the design for both the front and the back. If possible, I'd like to include the sides as well, making it a three-sided instruction.

[hoary cradle]

@olive yoke Additionally, when preparing a frontal design and using OpenPose to create a looking back pose, I would like to instruct the back design as well using Image-to-Image (I2I) techniques.

[olive yoke]

only output nodes and all their dependents are run. you only have to mute output nodes to mute all their dependents.

[olive yoke]

what concretely are you trying to make?

what are you typing into the prompt?

[smoky bear]

makes sense ty:) so mostly I just have to mute ksampler to try stuff out

[olive yoke]

you would mute SaveImage

not KSampler

[smoky bear]

I see:)

[olive yoke]

once you mute SaveImage, everything it is dependent on will not be executed

if you bypass KSampler, you are not likely to get what you want

[smoky bear]

okay thank you very much

[olive yoke]

like why are you saying "for example"? just say what it is

[steep marlin]

Because the example they gave might be one of multiple things they want to do

[plain condor]

I can't seem to find an img2img frame interpolation node that actually works on Apple Silicon. Any other Mac users out there having any success with creating frame interpolation animations?

[jaunty hatch]

How to create 360 panorama in ComfyUI? I tried following tutorials from the internet, but it didn't work.

[eager escarp]

is there a way to make the node search window longer? just shows such a narrow range of search hits

[fierce cloak]

I want to randomly alternate between these during the run of the prompt

So far haven't found anything that works

Maybe a different way is to use the prompt timing command, at least there is one in A1111, that allows to switch to a different word in the middle of the generation, for example:
[dog : cat : 0.5]
That in A1111 makes the dog be used for the 50% of the generation and then it switch to cat.
BUT thats not likely how it was typed, I don't remember the exact syntax.
I wonder if there is something like that in ComfyUI

I know there is some extensions that allow to use A1111 prompt reading into ComfyUI, but I am looking to know if there is a way to do that locally in ComfyUI without relying on an extension

[spring jungle]

https://github.com/licyk/advanced_euler_sampler_extension?tab=readme-ov-file#通过绘世启动器安装 how would i add this to comfy

[eager escarp]

as shown in the instructions, git clone in your comfy_nodes folder, or install through the manager

[spring jungle]

i tried doing git clone

it did not show up for me

🤔

[eager escarp]

did you restart comfy

by closing the black console window

or clicking the restart button

and then closing any comfyui tabs you have open in your browser and opening new ones

(you can also hit refresh)

that's fooled me a few times, forgetting that last part

[spring jungle]

nope does not show up for me xD

[eager escarp]

it might be a node only, idk

[spring jungle]

maybe xD

[fierce hornet]

@spring jungle if you're looking for Ella stuff you can try: https://github.com/ExponentialML/ComfyUI_ELLA/

There's also this but I haven't tried https://github.com/kijai/ComfyUI-ELLA-wrapper

(I did succesfully get the ComfyUI_ELLA one working)

[tacit fulcrum]

I am trying to use comfy along with Krita, but this is what I get

[lusty pewter]

It's literally telling you what files it's looking for and points you to where you need to go.

[tacit fulcrum]

yeah but the thing is I dont have a the folder these are to be copied into

I cant find the inpaint model folder

there isn't any

the foocus inpaint models where do I put them

[fierce hornet]

check https://github.com/Acly/krita-ai-diffusion/wiki/ComfyUI-Setup ?

[spring jungle]

thank you i will try installing it now an report back xD

[last root]

is there a node that can help me get image size after i loaded a pic. i saw some other nodes that can display size after i run the promp but dont know if there is any that does it automatically after i load the image.

[quick birch]

Has anyone noticed that Eff. SDXL Ksampler doesn't give the same result as the native Ksampler for the same model, the same prompt and even if the seed is the same. The difference can be the color, the style, than the adequation to the prompt.

[eager escarp]

i've come across some issues with workflows using those... forgetting what now cuz i stopped using them much as a result

[fierce hornet]

Yeah, me as well I stopped using it in favor of the fooocus eff ksampler which is nearly identical but didn’t seem to have the issues

[steep marlin]

Are you also using the efficiency sdxl loader? If so, it likely has differences in the sdxl encoder sizes. Like if you use a regular checkpoint loader node and then pull out a clip text encoder node vs a clip sdxl encoder node. You'll notice the sdxl one has extra stuff to it that you can use like target size

that's what I'm talking about btw. I'm guessing that your non-efficiency workflows are likely using the simple prompt encoding, while the efficiency nodes use the sdxl encoder under the hood, so that's where your differences come from

[nocturne prairie]

why is it that everytime i use animediff i still get controlnet model lines on the generated content

[steep marlin]

probably doing something wrong with the workflow. this is one of those test questions where you'd answer "C) Not enough information"

[nocturne prairie]

turns out ive been using 2 of the same controlnet openpose model on 2 seperate node

too much weight apparently effects the generation itself

[nocturne prairie]

after a couple of tweaks with the controlnet models i managed to get it to generate a more fluid character

[pallid cape]

how to fix it ?
When loading the graph, the following node types were not found:
IPAdapterApply
Nodes that have failed to load will show as red on the graph.
i installed ip adapter
https://cdn.discordapp.com/attachments/1008814691079958548/1229160805367287908/image.png?ex=662eac34&is=661c3734&hm=10b7633b5fb76fff44bb754abd86fbd1b8a17422938384cd178732a89a361300&

[nocturne prairie]

rerunnin comfyui didnt work?

download ipadapter from github

refer to manual install

[fleet fern]

where the fk is this thing?

[sly sundial]

Have you restarted?

Also, is you ComfyUI updated to latest?

There is a basic one built in to ComfyUI now.

[fleet fern]

Yes, everything is updated

Guess I know why.

[sly sundial]

If it was updated, you should see the new, inbuilt PAG node.

It was added 8+ hrs ago

[sharp raptor]

Anyone got a img2img upscale workflow that works with the new ip adapters? https://github.com/cubiq/ComfyUI_IPAdapter_plus

[proud flume]

Hi everyone, Any good tutorials (video or text) on how to work with ComfyUI to recommend ? And specifically on how to code extensions for it ?

[quick birch]

This is what I used. Normally I don't use it, but since I need to create a grid for comparing images, I decided to test it.

[candid flicker]

ELLA now only support sd1.5models？

[cedar mirage]

for now the only model released is for the 1.5 models. The sdxl is not released....

[stoic rock]

Any recommended node for loading a model, or Loras, that helps remind of an ideal prompt structure and sampling settings for working with the particular model your loading?

[steep marlin]

Yeah the efficiency loader uses the encoder sdxl under the hood instead of the simple encoder, that's why you're seeing differences with the same seed.

It probably uses the target width/height of 4096 but idk, I'd have to check the code

The efficiency sampler itself is identical to the regular ksampler though. You can break the tuple input on it and feed in a regular checkpoint loader and prompt to it and you'll see the same seed results that you were expecting

[steep marlin]

See what I mean? They are identical

Oh and your efficiency loader is set to a clipskip of -3, it normally defaults to -2 and I think -2 is the default for the simple checkpoint loader as well.

actually, i double checked the code on the nodes and yeah, it's definitely your clipskip being set to -3

see, they are identical now when the clipskip is set to -2 on the eff. loader

[eager escarp]

sure, there's one in this image

[stoic rock]

oh wow, I just loaded that for fun... wow, that's intense. 88 Image Loaders... I

[sharp raptor]

thanks 🙂

[lone bolt]

Oh, you must have got the lite version of his workflow.

[eager escarp]

the key nodes are found in europe. the americas can be discarded.

[lone bolt]

Don't forget to mention that he should keep the step count between 1.500 and 2.000 for the first pass.

[stoic rock]

Well, I love the image anyways... I guess if that's what it takes to get there.

[lone bolt]

Oh, that's for sure. His images are great.

[quick birch]

Thanks for the information. As I do a lot of test I tried different clip -1 to -3.

The problem is still present even I do the exact settings for both. Here is the images. The first set ist the native and the second is the SDXL effi.

[stoic rock]

When you have a bunch of nodes to updates… do you do it one by one, restarting in between each… or update each one then do a restart?

[eager escarp]

mass update

i click update all

[stoic rock]

Alrighty, I did it.

My ComfyUI Manager now says “500 Internal Server Error Server got itself into trouble”

[eager escarp]

wtf

[stoic rock]

Do you use, DB: Channel (1 day cache) ?

[stoic rock]

I think this in my logs is related to the issue: “The currently installed ComfyUI is not a Git repository”

Seems like some errors have crept into comfyui-manager in the latest release

It would be cool if there were release cycles and versions

[eager escarp]

you can just roll back, it's github

release cycles would mean waiting months for new features like with a1111

and a month in AI years is, well, years

[quick birch]

Efficient Loader and Efficient SDXL Loader are the problem they don't manage correctly the CLIP and probably the VAE this is why the result is not completly the same. It seems for photo like a portrait or dog the problem is not present (less complexity) that is why I think the problem came from managing clip. What is your opinion ?

[steep marlin]

I'd have to take a look at the code more and you're sure you're using the same cfg for both? I'll also try some more complex prompts to see if it remains the same or not on my end

[steep marlin]

I ran a handful of other tests and the images are still coming out identical on my end. I even used the difference node to make sure(black means no change in a pixel)

So there is likely something else going on in your workflow

make sure you're using the eff. sdxl loader and the ksampler sdxl eff. nodes

[sharp raptor]

why so many image loaders?

[eager escarp]

not really any reason, you can just delete em

[sharp raptor]

Ok, this is part of the portrait maker node, I installed both via manager and git clone. Any idea how to fix? @eager escarp

Also tried to pip install -r requirements.txt from the portrait node folder

[eager escarp]

no idea, i don't use portrait maker

maybe the workflow is outta date

or you haven't restarted it but that's about all i got

[sharp raptor]

deleted node and it works. Is it possible to implement a img2img upscale where u can control denoise start value in these samplers or would u have to swap out to different sampler? I would like the workflow to upscale the image and keep the composition completely. Just improve details

[eager escarp]

there's really no such thing as a denoise value imo

but you can play with the multiplier node

higher values will change the image more, lower less

[sharp raptor]

ok wil try this

Does this node make VAE connect to all VAE input sockets?

[eager escarp]

yeah, all that don't have anything connected

reduces clutter

[sharp raptor]

ok

[eager escarp]

click drag from the vae thing on the left and release

it'll give you options to add those

works with clip and model as well

[sharp raptor]

nice

[fleet fern]

How does perturbed AG compare with self AG?

I'm guessing PAG is an improvement?

[steep marlin]

One uses cross attention and the other uses self attention, they are quite literally apples and oranges. That being said, PAG is pretty awesome if you stick to around 1.5x on it with a cfg around 4-5 on your sampler

With those values, you'll see very positive results and it won't burn the images. You can also couple it with automaticCFG

[fleet fern]

@steep marlin Thanks for the explanation. Would it be possible to use them both at the same time then? or is that overkill?

[quick birch]

First of all, thank you for testing, as it's important when doing analyses to check whether others are getting similar or dissimilar results.
To make things perfect I did exactly the same workflow as you. As you can see, the images are different. And when I add cliplayer the image have more similarity with photography but are still different.
Have you updated ComfyUI and node Efficiency? is it a way to switch off clip in SDXL ?

[steep marlin]

Np. Test it with a non-ancestral sampler and use a batch size of one. I'll run your setup and prompt in a minute.

[steep marlin]

I don't see why you wouldn't be able to, that's the fun of comfyui, you can experiment to your heart's content

[steep marlin]

still identical on my end

and don't use clip skip -1, i think pretty much all SDXL models are trained around -2

the default checkpoint loader and prompt encoder do -2 by default as well

so right now, your eff. loader is on -1 and your regular loader is on -2, so yeah, you'll have differences

even with a batch size of 2, still identical on my end btw

[quick birch]

Eureka ! The problem is the clip ! If you put in the both workflow clip -2 you have the same image but not -1 or over -2. it is a pity as with clip -3 or even clip -5 we can sometimes have very interesting results. To my opinion this is a bug. Clip -1 should be clip -1 like on the native comfyUI , If we can not manage clip, it should not be added

[steep marlin]

The native clip in comfyui is -2 and that's why the eff. Loader defaults to it as well. You can also use a clip node with a regular checkpoint loader, can't remember the exact node name but just pull off the clip pin and search for clip

[quick birch]

I think is clip set layer, there is another clip combine but I don't try yet. The possibilities of ComfyUI are endless.
Do you know if there is a pdf or a reference link for creating nodes ?

[steep marlin]

yeah that's the one. i like comfy a lot, but i've been working with UE4/UE5 for a long time now, so i feel right at home with spaghetti graphs. and i'm not sure about pdf/reference stuff. i know comfyui's github has an example node reference though.

https://github.com/comfyanonymous/ComfyUI/blob/master/custom_nodes/example_node.py.example

[olive yoke]

why are you burning time on efficiency nodes

you don't even want to interact with that part of the ecosystem.

[steep marlin]

Because I experimented enough to know they save me a lot of repetitive hassle. I start most comfy projects from scratch or basic templates that I make that I never remember to use lol

[olive yoke]

what is your goal?

i am eager to help 🙂

[steep marlin]

Plus, they clean up graphs a lot with the Lora and control stacks

Oh I just screw around with AI for boredom and lols. Good for ADHD breaks between other things

[obtuse musk]

Hey guys, is there an online hosted version of ComfyUI?

I don't think I will get very far with my macbook air

[nocturne prairie]

try to keep resetting comfyui

i had the same issue before and it just decided to start working after a while but it was with ipadapter

[obtuse musk]

Has anyone set up a comfyUI server to send api requests to?

thinking of setting one up

[tranquil parcel]

Are there any ComfyUI nodes similar to the Color Balance panel in Adobe Photoshop that allow you to adjust the cyan, magenta, and yellow colour cast in the “Midtones”, “Highlights” and “Shadows” of an image? 🎨

[digital quest]

UNABLE TO CLICK ON THE LEFT AND RIGHT ARROWS IT DOES NOT CHANGE THE VALUES AT ALL

[polar dove]

I have been using the IpAdapter quite a bit, and have had some success in saving embeds, which save to the ComfyUi output generations folder, does anyone know where the ipadpt files it makes need to be placed so the IpAdapter Embed Loader can see them?

So far whatever directory I try to place them in I cannot get the node to see the saved embeddings.

[prime tide]

Is there a comfyui addon or node that sorts checkpoints into folders like i could put my favorites into a favorites folder and ones I haven't tested into a folder for such?

[boreal stirrup]

Hey guys I have a question I use Auto1111 (and I also hope this is the correct channel to ask) on my m2 pro macbook pro and was wondering is it worth switching to comfy UI would I see more negatives performance wise or not and would I see for negatives in the fluidity on how fast I can generate good looking images and another question is for mac do I use the standard comfy UI installation or is there a more optimised one for mac some1 in the community has made?

[pliant igloo]

This is rather simple node, but what node is this I cant find the name? 😅 Im not new, but never did it this way Im comparing a workflow of someone else but they have this connect to clip Text encoder string

[flat talon]

Perhaps it's just a regular string node with a custom title

[buoyant knot]

For comyui, and I have AMD gpu. Do I need to click the top option?

A youtuber said "if you run it on your cpu it will be way slower than gpu"

However I don't have nvidia gpu, could I still try to run comfui on the "nvidia gpu" file option even if my hardware is AMD?

[sharp raptor]

What does the clip vision do?

[mystic plaza]

i keep this error and need some help

[fickle remnant]

Do you have that checkpoint?

Otherwise click on it and select one you have

[mystic plaza]

not sure i downloaded a bunch of stuff

[fickle remnant]

Click on the name and you should get a dropdown list

[mystic plaza]

this is what i am getting

[fickle remnant]

When doing what?

[mystic plaza]

trying to prompt

[fickle remnant]

As I already said twice, youre lacking the checkpoint you selected it seems

[mystic plaza]

also where do i get the checkpoint from then ?

[fickle remnant]

Its the standard 1.5 checkpoint, you can get it from SAIs huggingface

But as I also stated twice you can just select a checkpoint you downloaded and put into the models folder

[pliant igloo]

@mystic plazaYes, just like Boto said. Also you can just use the manager and install it from there

[mystic plaza]

this is what i am getting

[pliant igloo]

Are you in the install models selection?

[mystic plaza]

yes

[pliant igloo]

Thats weird then you need to drop the model into the folder of your comfyui and refresh.

https://huggingface.co/runwayml/stable-diffusion-v1-5

[silver nebula]

https://www.youtube.com/watch?v=3tI6eg4pWiU

Using the API for SD3 via ComfyUI is the same price as using assistant.stability.ai - so no freebie option for using desktop yet!

[silver nebula]

Using ZHO-ZHO's SD3 Node in ComfyUI

[tribal plaza]

hi, help me understand the error. This happens when the process reaches the sampler

[silver nebula]

Go to pytorch.org and make sure you compile Torch with the right version of CUDA

[slim seal]

Can I edit KJ's SD3 API node so I can send the same seed to an SDXL workflow that'd need to be modified to also accept a seed?

[tribal plaza]

I haven’t been able to solve it yet, I haven’t figured out how to compile Torch with the right version of CUDA? can you be a little more detailed?

[silver nebula]

Go to pytorch.org and read up there 🙂

[tribal plaza]

I read it, it didn't help yet 😦

[slim seal]

I need some help.. has anyone used Ollama with ComfyUI?

[zinc silo]

Can anyone help me out with this?

*Error occurred when executing SaveImage:

Numpy is not available

File "O:\stable-diffusion-comfyui\ComfyUI\execution.py", line 151, in recursive_execute
output_data, output_ui = get_output_data(obj, input_data_all)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "O:\stable-diffusion-comfyui\ComfyUI\execution.py", line 81, in get_output_data
return_values = map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "O:\stable-diffusion-comfyui\ComfyUI\execution.py", line 74, in map_node_over_list
results.append(getattr(obj, func)(*slice_dict(input_data_all, i)))
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "O:\stable-diffusion-comfyui\ComfyUI\nodes.py", line 1407, in save_images
i = 255. * image.cpu().numpy()
^^^^^^^^^^^^^^^^^^^

[eager escarp]

is it just me or is the self-attention guidance node broken now?

[smoky bear]

Where can i find sdxl temporalnet controlnet?

[strong pewter]

  File "C:\Users\Z\VST\Super-Stable-Diffusion\ComfyUI\python_embeded\Lib\site-packages\aiohttp\web_protocol.py", line 452, in _handle_request
    resp = await request_handler(request)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Users\Z\VST\Super-Stable-Diffusion\ComfyUI\python_embeded\Lib\site-packages\aiohttp\web_app.py", line 543, in _handle
    resp = await handler(request)
           ^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Users\Z\VST\Super-Stable-Diffusion\ComfyUI\python_embeded\Lib\site-packages\aiohttp\web_middlewares.py", line 114, in impl
    return await handler(request)
           ^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Users\Z\VST\Super-Stable-Diffusion\ComfyUI\ComfyUI\server.py", line 41, in cache_control
    response: web.Response = await handler(request)
                             ^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Users\Z\VST\Super-Stable-Diffusion\ComfyUI\ComfyUI\server.py", line 53, in cors_middleware
    response = await handler(request)
               ^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Users\Z\VST\Super-Stable-Diffusion\ComfyUI\ComfyUI\server.py", line 468, in post_prompt
    valid = execution.validate_prompt(prompt)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Users\Z\VST\Super-Stable-Diffusion\ComfyUI\ComfyUI\execution.py", line 625, in validate_prompt
    class_ = nodes.NODE_CLASS_MAPPINGS[prompt[x]['class_type']]
                                       ~~~~~~~~~^^^^^^^^^^^^^^
KeyError: 'class_type'```
I accidentally loaded another workflow in the middle of generating and i tried to queue a new thing and now i get this. im not sure what to do

[mortal ferry]

i had this error with Ultimate Upscaler

[dapper sluice]

Hi if I have an AMD cpu and gpu, will it generate images at a similar rate as if I spent similar $ on nvidia cpu & gpu and ran the other file?

[zinc oasis]

could you share the workflow? .json

[south widget]

There aren't many updated benchmarks for AMD cards as far as I know, you can check this https://www.tomshardware.com/pc-components/gpus/stable-diffusion-benchmarks

Which card were you interested in?

Done a quick search, I'm interested too and found this https://www.pugetsystems.com/labs/articles/stable-diffusion-performance-nvidia-geforce-vs-amd-radeon/

It shows that the 7900 XTX (MSRP 1000$) performs similar to the rtx 4900 (1600$) with Shark. With Automatic 1111 only Nvidia cards work good.

I have never heard of Shark before 😄

[silver nebula]

I can attach the json - you must then pay for some credits on an Openai API Key.

https://github.com/ZHO-ZHO-ZHO/ComfyUI-StableDiffusion3-API

[lone cape]

Hi, little risky new development out of my nodepack. I included aesthetic scorer to segment detailer, and if the scorer result of detailed segment of image lower than the original part, then ignore the result. Not enought time for testing yet, but the scorer help to optimize detailer's settings, and as I see during development the a-scorer result most of time acceptable. The feature can be off if too much failed. (The red badge mean the scorer ignored the detailed results)

[verbal sky]

New tutorial to help people in using Perturbed, Advanced Masking, and a Fashion Surprise - https://youtu.be/j3xHNmEWWCI Enjoy!

[lone cape]

I just testing Pic2Story model, and look really good:

[zenith quiver]

Hi! Is there a seperate discord for comfy?

[eager escarp]

don't think there's any need for that

[zenith quiver]

anyone have tips for getting an enviroment to render around a character using a mask?

like keep the character part the same and just render the enviroment

[zenith quiver]

hello i cannot find the sampler for this model in comfy: https://civitai.com/models/112902?modelVersionId=351306

DPM++ SDE Karras

[lusty pewter]

The sampler is dpmpp_sde.

Karras is the scheduler.

[zenith quiver]

thanks

found it

but for some reason my results look poor

[lusty pewter]

Because there is likely some other setting somewhere that differs. Don't try to match Civit; there's usually something else different in the settings of the original generations posted there.

[zenith quiver]

could it be vae related?

[lusty pewter]

XL pretty much only has one VAE.

[lusty pewter]

So not likely...unless you're not using the SDXL VAE.

[steep marlin]

Just use the vae from the model

[lusty pewter]

Yeah, DreamShaper should have the VAE built in.

[steep marlin]

And that's a preview vae

[zenith quiver]

oh okay yeah that was the issue

[lusty pewter]

And be sure to grab the full SDXL VAE here if you don't have it. Some models don't bake the VAE in and you need to use it for XL stuff:
https://huggingface.co/stabilityai/sdxl-vae/tree/main

[zenith quiver]

actualy damn that was not the issue

unless I pluggin the vae in wrong

[lusty pewter]

show an example of the output that you're saying is bad

[zenith quiver]

using controlnet

5 steps 2cfg

I think i fixed it

Does anyone know how i could have controlnet only work on my subject? and have it ignore the rest of controlnet input image?

[naive kelp]

Hello guys! Does anyone know if there is a way to iterate through an input field in node? I want to iterate through a lot of styles in the Prompt Styler node, is there any easy way to do that?

[hardy tree]

im confused... is that considered something new? cause i saw the pic2story model on huggingface, and it seems to have been uploaded like 11 days ago, but then it seems to be based on something from a 2022 paper, so like what... lol

or did they release the weights after 2 years

[lone bolt]

You mean like, converting it to an input and connecting a primitive node?

[naive kelp]

I guess that can work, but would you happen to know how would I fetch the name of the styles from the Prompt Styler node? It has more than a thousand different styles, typing them would be kind of tedious

[lone bolt]

I mean like this

You can set that to "increment" then

[naive kelp]

Oh damn dude, still learning comfy, I didn't know it works like that, I thought it only takes like strings and etc. Thanks a lot! @lone bolt

[lone bolt]

We are all learning every day. Glad i could help.

[fallow jackal]

I update my ComfyUI custom_nodes for reusing data in cycles: https://github.com/Pos13/comfyui-cyclist
Now, with Loop Manager node it's much more transparent and intuitive to use.

[lost junco]

/cream

[cursive ridge]

I finally installed sd/comfy 🙂
How do I add hires fix?
Also, I was able to add esergan so far, is it possible to run hires fix then esergan without messing up my images?

BTW comfy is awesome; I had more trouble getting windows to let me paste my models into a different folder, then I did making images in comfyui lol

[sly sundial]

I'm trying to use these, but having problems.

I'm trying to copy your whole detailer group, but the Image Segment nodes have inputs that I can't convert to widgets, yet they have been in your example workflow. This is causing problems because I don't know how to get around that.
There's something wrong with those nodes, even in the example workflow image they have incorrect parameters (NaN) in some fields.

[fallow jackal]

Something like this? Generate → Upscale → Generate = Hires.fix, basically. You can pipe it in whatever you want order. You can add an extra upscale with ESRGAN, if you want.

[cursive ridge]

Thank you 🙂

[brisk charm]

how are you guys adding all the basic functionality like string literals, etc.? got it. i didnt realize the primitive adapts to what you plug it into.

[brisk charm]

each time you generate you use the same model?

[static tree]

Hey all, I got this error while trying to run a video2video workflow:

Error occurred when executing KSamplerAdvanced:

mat1 and mat2 shapes cannot be multiplied (924x2048 and 768x320)

Anyone know what I can do to fix it?

Here's the workflow I made, The left-hand side is just the video loader and checkpoint loader.

[zenith quiver]

hello! I was wondering if someone could help me understand if my simple inpainting workflow looks correct?

[eager escarp]

yep that's correct, it's the most basic one but correct

[spring jungle]

question is there a way turn off a model at %?

i want to run ipadapt model until 50% switch it to model before ipadapt

🤔

[grand pecan]

does comfy ui require python 3.11?

getting blue bottle issue on apple M2 unless I use --cpu arg

[eager escarp]

i'm not aware of any nodes that allow timestepping with models. i really, really wish that was possible, maybe it's not, idk

[spring jungle]

i could not figure out a way i gave up on it xD

i am sure there some way you could write it becuase denoise value control model

idk xD

[eager escarp]

you can do it kinda by using ksampler advanced chained together

but then you lose the momentum with the sampler

[spring jungle]

yeah i cant change ipadapt between 2 ksampler my limit problem xD

i dont think it would work properly that way

🤔

maybe i test around with that tomorrow

[fallow jackal]

Sometimes. It's not a big deal to connect another model.

[eager escarp]

what are you trying to do exactly?

turning IPA on or off isn't an issue

just change the start or end values

[spring jungle]

o yeah

i forgot that was a thing

🤣

trying to solve problem already exist XD

[eager escarp]

i figured you knew, then thought... ya know, i forget some absurd shit sometimes so maaaaaybe...

😄

[spring jungle]

yeah still pretty new to ipadapt xD

remebering what all nodes do xD

i have been getting some pretty cool results with it

[eager escarp]

oh hell yeah it's great

the style weights are a revelation

[spring jungle]

nice xD

[grand pike]

hello lovely people. what does that share option do in the comfyui manager window?

[spring jungle]

i dont think i know of this share option you speak of

[grand pike]

this one on the left

[grand pike]

figured it out, thanks silly

[spring jungle]

oh i think it metadata

[spring jungle]

that look nice xD

[eager escarp]

started with a similar looking face

turned it into paint

[spring jungle]

here what im working with xD

[lone cape]

Hi, yes its possible, because I push the development first, then refresh and push the workflows some days later. Old workflows have use 'Fix node (recreate)' menu after development if the node inputs changed. But today I pushed recreated workflows, just load the .png or .json for example: https://github.com/CosmicLaca/ComfyUI_Primere_Nodes/blob/master/Workflow/civitai-rewear-rehair-wf.png (source images uploaded for testing)

[silver nebula]

HELP!!!!!!

[silver nebula]

LCM LoRA

[cursive ridge]

is it possible to add esrgan to the workflow in comfy SD? If so, how? Where on the dropdown menu 😄

If not, how on earth can I run esrgan on more than one image at once? This one at a time thing is frustrating!

[sly sundial]

What are you trying to so, upscale images? You can add an upscaler to the workflow, and choose to use esrgan as the upscale model, or create a workflow that will run on all of the images in a folder.

You may need to download the model and put it in your models/upscale_models directory.

The UltimateSDUpscale node is one that allows you to do this.

...or just Upscale Image (using model), and there's also a batched version.

Loads of options!!! 😄

[cursive ridge]

Where do I find that in the dropdown menus?

[sly sundial]

It's a custom node; f you don't have it, you'll need to install it from the manager.

[cursive ridge]

I'm trying to make my images look better. Before I tried installing SD/Comfy I'd either use hires fix, or Esrgan... The hiresfix workflow looks even more confusing lol.

[sly sundial]

Should be easy enough

[sly sundial]

There is an SDXL HiRes Fix node, but I've not tried it.

[grand pike]

hello lovely people. got a simple issue i think. trying to upscale a 152 x 152 portrait but it always generates at that small 152 x 152 resolution. What am i doing wrong in this workflow? It's from the scott deweiller workflow video. I know he used a much larger image in his example so maybe thats it?

oh i think it's because the intbinary nodes were set to add.

yup that was it

[brisk charm]

this normal?

[steep marlin]

Yeah it's normal. I think certain types of flash attention aren't enabled for windows yet

[last root]

anonyone know a easy to use node in confyui to automatically calculate and resize to closest sdxl ratio/sizes

[steep marlin]

comfymath has a nearestsdxlresolution node @last root

[olive yoke]

is the idea with this workflow to select the hair with grounding dino, then generate new hair?

[thick ether]

I have 2 promts and I want to generate images interpolation between the prompts. Can I somehow automate it or do I need to change the conditioning average value manually 100 times?

[sly sundial]

It redraws several different parts and then checks the aesthetics to see if it's better or worse, and then only applies the change if it's better.

The whole thing is custom nodes, so it's not easy to extract parts and just include it in your own workflow, so I gave up with it in the end. Didn't seem to detect anything most of the time, but when it worked it was good.

[last root]

how do i pin and unpin a group of nodes all at once. i selected a couple with ctrl but then i right click and pin but it only pins 1. same goes for unpin i select a group of nodes then when i righ click unpin it only unpins 1

[weak urchin]

https://civitai.com/models/388864/geekyghost-aio-all-in-one-workflow

Muah haa haa haaaa lol

Now you can pose your character, decide the outfit and face

[hoary zinc]

Hi there. I've just converted to ComfyUI, how do you copy a node without losing connection?

[brisk charm]

ctrl-c -> ctrl-shift-v

[hoary zinc]

it was a bug, also how do you avoid CUDA error and where to insert the parameters in ComfyUI?

for A111 I used --low-vram in the user bat file but idk what in comfy

since I have only 4GB of VRAM and even when the comfyUUI automatically set it isn't enough

I scaled an image to 2560x2560 px

[brisk charm]

the bat file included in comfyui is this

.\python_embeded\python.exe -s ComfyUI\main.py --windows-standalone-build
pause

i think adding the parameters and saving should do the trick.

--lowvram

[grand pike]

hi so i'm trying to upscale a particularly blurry image but it's not filling in details. I see something about detailers but not sure how to use those.

[hoary zinc]

thank you

[hoary zinc]

I have just started, but you can try to search for videos on youtube. There is many things you can use to increase a quality. I've tried this: https://www.youtube.com/watch?v=CxB47DMEyYQ. I'm missing only the step above, because Ior don't havve a powerful enough GPU.

[hardy tree]

@hoary zinc im happy you are progressing with comfy 🙂

[hoary zinc]

thx, this too me quite a liong time, but I have yet to fix that out of memory error for upscaling.

[hardy tree]

well it's not gonna be easy to avoid all memory problems with low vram, but try the best you can

maybe also look into other comfy launch arguments to help with memory

[hoary zinc]

I think low-vram is already applied at the beginning when the program starts

[hardy tree]

good

i mean eventually there is a limit to what you will be able to do :3

[hoary zinc]

but that's it, I don't know where to start with this, maybe if there is a parameter to work with smaller chunks of data

[hardy tree]

there is tiled upscaling

that could help

it upscales in chunks

also look into tiled vae and tiled ksampler perhaps https://github.com/BlenderNeko/ComfyUI_TiledKSampler all these things can help with low memory

there are probably other things im forgetting

[hoary zinc]

nevermind, I'll check it out, thank you

[hoary zinc]

comfyUI doesn't recognize xformers argument, any ideas?

Is there any guide to the ComfyUI arguments? I have found one reddit page post, but there isn't everything nor on the official github.

[hardy tree]

what version of python do you have for comfy? i think xformers works best with like 3.10

did you install xformers?

also make sure you dont have conflicting arguments that dont work with xformers

i think xformers doesnt have argument, i think it's default as long as you have it installed

you can disable xformers after it is installed

[hoary zinc]

I'll try to install xformers again, but I didn't unistall xformers

so it should still eb there

and I use 3.10.6

[hardy tree]

good

what is your current comfy arguments?

[hoary zinc]

the default argument, --novram (even lower than lowram, but it probably won't help)

and tried --xformers and it said that it doesn't recognize it at all, I've just checked and xformers is installed

version 0.0.25.post1

[hardy tree]

no read what i said above

[hoary zinc]

.\python_embeded\python.exe -s ComfyUI\main.py --windows-standalone-build --novram

🤷 --xformers didn't work at all, it doesn't recognize

[hardy tree]

there is no xformers argument

if you have it installed, it will use it by default

[hoary zinc]

so that's why, also, is there any guide to parameters? I haven't found an entire list of the commands and what they do.

only in cmd when it finds an error, but there is no description

[hardy tree]

i personally never had problems with xformers, i think but im not sure... that xformers doesnt work with your gpu, i think its only for rtx type gpus

but im not sure

[hoary zinc]

it works then, I thought it didn't apply, but comfyUI just works differently

[hardy tree]

so im confused, do you have it working now?

[hoary zinc]

yes, it works

but the parameter doesn't work

[hardy tree]

no you dont need parameter hahaha

it uses it automatically if you have it installed

you can even see that from the comfy console window, it should mention that it's using xformers

[hoary zinc]

Is there a tile upscaler? Or a way to split the Ultimate Upscaler pieces? I used the one Scott Detweiler uses according to his tutorial, but the problem is right there. When it goes int ultimate upscaler.

TiledSKsampler cannot be used there.

[hardy cloud]

Hi! Please help me how to duplicate workflow internally with different parameters for example 10 times to run for a few hours on Auto Queue and it will spin generating 10 different things with different inputs (ref images, prompts, lora triggers, masks, paths to save image...).

Here I made a workflow, took out all that is common in Anything Everywhere. All the loaders. There I have XL Turbo > Lora > InstantID > IPAdapter > KSampler > Face Analysis > Save Image. If I run 3 duplicated such groups everything is ok. If 4, it sometimes falls to lowvram and performance drops. But I want to run 10 such groups, for example, so that the whole day would spin without me. I have 24 gb vram.

What am I doing wrong and what should be the principle here? I just don't understand where lowvram comes from if absolutely all loaders are common and stand separately as if it were a single group.

[distant pendant]

are you actually observing increased vram usage or is it just comfy randomly switching to lowvram though? if the latter try forcing it to keep the models on gpu with --highvram

[hardy cloud]

Yes I definitely see a difference if I only run one group auto queue then it's 17-20/24 vram, but if 4 groups then it's 20-23/24 vram. But I don't understand why there is this difference.

I thought that the biggest things for vram are models + generation process, but I have it sequentially and not in parallel, and the models are the same from the same loaders, even if 1 group is enabled, or all 4

[hardy cloud]

Look please the first two screenshot shows the full process of one pass through 4 groups active. In the third screenshot, only 1 group is active. Maybe some of this makes sense to you? My skills aren't enough to figure it out.

Perhaps this green line "[rgthree] Using rgthree's optimized recursive execution" may play a role here, perhaps this thing is trying to do something in parallel. I mean [rgthree] is trying to optimize something and maybe on the contrary messes things up.

[distant pendant]

you're using a lot of custom nodes so honestly hard to say, could be a memory leak somewhere, there's a lot of ipadapter images based on the screenshot, these might not be getting offloaded properly in the same run

if you're just trying to get 10 random outputs maybe it would be easier to run 1 at a time and queue them up

[hardy cloud]

No it's a machine to create perfect portraits like from the magazine cover for real people, from crappy photos, I called it Likeness Miner because I generate 500-1000 portraits of one person to get 1-5-10 perfect likenesses. This turned out to be the most reliable way. That is, each group has a different photo ref input, part of the prompt describing the appearance. Randomization style, lora style and checkpoint are common.

[olive yoke]

at this stage your are better off creating the workflow in comfyui as straightforwardly and simply as possible, then asking chatgpt to author a python script that varies all the variables for you

[hoary zinc]

A quick question. I'm trying to setup an upscaler, but i need a controlnet model. I clonned this repository "https://huggingface.co/lllyasviel/ControlNet-v1-1/tree/main" into "models/controlnet" and restarted comfyUI, but it cannot find the model, am I oding something wrong?
Image

here is the image, it cannot find it

[hoary zinc]

actually, I might have figured it out, I didn't download the models as the git clone didn't proceed any furthers

the model folder had 10GB+ of models (https://huggingface.co/lllyasviel/ControlNet-v1-1/tree/main) but for some reason it wouldn't proceed

[cursive ridge]

I even hit the load default button to start over, and still can't seem to get Esrgan to work within SD. What am I doing wrong?

[hoary zinc]

you are missing upscale model in "Upscale Image (Using Model)"

you need to load an upscale model, such as this one

I finally moved from step 1 in Ultimate SDUpscale but chose VAE decode titled ... I wonder what image will it spit in my face

[cursive ridge]

so a .pth instead of a .bin?
(sorry just installed recently!)

[hoary zinc]

yes, pth

models/upscale_models folder here they belong

[hardy tree]

@hoary zinc are you using this one? https://github.com/ssitu/ComfyUI_UltimateSDUpscale i think it does the tiled upscaling

[hoary zinc]

yes, but when I activated that titled_decode option the result was same

no matter what, always ran out of GPU VRAM

so I'm trying my modified version with Controlnet and VAE titled

[hardy tree]

are you using a tiled vae decoder?

[hoary zinc]

yes

I'll be trying.

[hardy tree]

hmm maybe lower the dimensions further, like 256x256 idk

but it's always a struggle with low vram 😦

[hoary zinc]

I use KSampler and downloaded that 1.3GB controlnet model

this is the first time I'm not getting kicked out by that message

[hardy tree]

at what stage does it run out of memory for you? ksampler?

remember there is also a tiled ksampler, i think i linked it

[hoary zinc]

titled ksampler?

I see only the normal one and advanced, then some pipes and provider pipes and things I don't know

[hardy tree]

it's by blender neko

[hoary zinc]

I downloaded that one

[cursive ridge]

I used the esrgan installer, it forgot some? Can I just use ones I find in google searches? It seems to have included a heap of bin files.

[hoary zinc]

https://openmodeldb.info/models/4x-UltraSharp you can get these models here

otherwise just search on google for 4xesrgan base model for example

[hardy tree]

yea https://openmodeldb.info/ has a lot of cool ones

[hoary zinc]

either I set up something wrong or idk why 13th step

it takes way longer than A111 to upscale one image

[lone bolt]

Can you make a screenshot of the Ultimate SD Upscale node?

[hoary zinc]

oh no

it's 25 steps

I've noticed just now

🤣

I should've set max 10 for that image

Can it be undone or not?

although I'm already halfway through, so I'll just do something else I have prepared

[hoary zinc]

Will the resulting image be broken?

because it just cancel the entire prompt

I mean image generation*

[lone bolt]

Yes, as long as nothing reaches the save image node, everything will be lost

[hardy tree]

i never understood why people use save node (at least for a regular image generation workflow), i always prefer the preview node, because i dont want tons of potentially bad pics on my pc, and with preview one, i decide manually if i like it and save it myself

[hoary zinc]

it works

finally

althoguh it itsn't a perfect image, it was one of the first tries

[hardy tree]

nice

[cursive ridge]

Thank you so much! FINALLY it works 🙂
Now I'm going to figure out how to batch use it for the few hundred images I created in the last couple of days 😄

[hardy tree]

😮

[hoary zinc]

finally did upscaling correctly

look at the general channel

#🏞｜general-with-images

[hardy tree]

@hoary zinc btw how you liking scott's videos?

[hoary zinc]

great

[hardy tree]

yea

[hoary zinc]

although I only watched 1 or 2 videos so far

[hardy tree]

haha

well one thing at a time eh

[cursive ridge]

Thank you 🙂 I'm probably going to try them all 🙂

[hoary zinc]

I used the method with calculation

[hardy tree]

there is a lot of cool stuff in there for sure

[hoary zinc]

I'll have to dig into it. I just set parameters according to my intuition without digging into documentation.

🤡

I should've done that from the very beginning. There is no tutorial which focuses on low VRAM

[lone bolt]

I checked your workflow and its this what slows it down

[hoary zinc]

yeah, I wanted it upscaled

[lone bolt]

4 times?

[hoary zinc]

and I have 4x upscaler

I knew I could generate it even on lower VRAM.

[hardy tree]

yea documentation wise, it's not great perhaps, but we learned a lot as a community, so you will find an answer either on this discord, or reddit or just google :3

[hoary zinc]

How does one determine how many steps should a sampler and then upscaler take?

I set it acording to the videos I saw or the pictures (usually they show how many sampling steps they took)

[hardy tree]

upscalers dont need a lot of steps unless your initial image is so bad

[lone bolt]

Thats a brave move. I have 16 gigs of Vram and i am scared of the waiting time for a 4x upscale.

[hoary zinc]

but it's quite sharp

[cursive ridge]

SD is dangerous, x4 images each time, and I can set that to 100 and go wash dishes or something!
(nm the fact that I experimented with pony and bad peramiters the 1st time haha)

[hoary zinc]

well, I survived A111, I must survive this one

[hardy tree]

nah 4x upscale is not that crazy, then again depends on the initial dimensions

[hoary zinc]

640x640

[hardy tree]

well thats not too crazy

[hoary zinc]

but you know, otherwise it'd give me CUDA errors

[hardy tree]

yea you have to find the spot for your gpu

[hoary zinc]

I'll just need to make the generation faster now.

[hardy tree]

there are couple inference tricks.... lcm is one of them

[hoary zinc]

Can you show me?

btw this image was SDXL generated

so it's not impossible it just took hours of pain

[hardy tree]

sad cat

[hoary zinc]

What fixers for hands would you recommend?

for SDXL models, I have tons for SD1.5

[hardy tree]

well lcm has a lot of tutorials, not sure if scott has one, cant remember.. i think there is a hand detailer for comfy

[hoary zinc]

problems are the eternal problem no matter what prompt I use

ok, thanks, I'll check them when I'm free. This is what I wanted to do today and it turned out better than expected.

[hardy tree]

comfy has a lot of tutorials on youtube, you can find almost anything

[hoary zinc]

yes, I plan to look at face tutorial, definitely

[hardy tree]

im happy you switched to comfy haha, a lot of people think it's too hard, but its not really

[hoary zinc]

welll, it is a bit harder, but I'll just need to remember relations between nodes

and I can avoid CUDA errors

[hardy tree]

yea but comfy has the relations color coded 🙂

[hoary zinc]

and can alter any process

[unreal reef]

I have a workflow that runs an SDXL gen and then follows it up with an upscale and face detailer. I've been just monitoring it and canceling after the first step if I don't like the look of the image, but I assume there is a better way. Is there a custom node that can pause and wait for approval or something like that?

[hardy tree]

eventually it becomes second nature, like yea load checkpoint, text encode, ksampler, conditioning goes here, etc, i dont even think about it these days, only on complex workflows do i have to slow it down cause i have to make sure its connected correctly

[maiden wigeon]

Hello I would need some help please. I'm new to Comfy UI and I've learned to use Loras in my nodes, but there's something I need your advice on: I would like my prompt to be split into two parts, part 1 and part 2, and for both texts to be concatenated by Comfy UI before going into production. For example, if part 1 is "an orc" and part 2 is "in the forest", the concatenated prompt would be "an orc in the forest". I can't figure out how to do this while managing the Lora at the same time. Help would be appreciated!

[cursive ridge]

I someow managed to create a few hundred images which automatically included a long number at the end of the filename. Suddenly now, they are all overwriting 0001 and I can't figure out how to get the sequential filenames back 😦
The google answers are confusing!

[hardy tree]

so what is the problem necessarily with just one positive prompt that says "an orc in the forest", why you need to separate it? @maiden wigeon

[maiden wigeon]

Because I need to keep a part of the prompt fixed and another part variable, for a specific use

[hardy tree]

well there are ways to concatenate indeed, i never used any of them, but i know i saw some nodes

[maiden wigeon]

thanks, i can't find on google, i'll wait to see if someone knows how to do

[hardy tree]

would this work?

[lone bolt]

No

[hardy tree]

😦

[lone bolt]

What's your plan lodis?

[hardy tree]

no i was trying to help the other dude

maybe he wants this

[lone bolt]

Just adding text together?

[hardy tree]

well idk

[maiden wigeon]

i'll try thanks @hardy tree

[hardy tree]

kk 🙂

[maiden wigeon]

yes

[maiden wigeon]

it works ! you'are the best thanks ! just one question, i found "string to text" but not "string" as in your example, where is this node in the list ?

[hardy tree]

hmmm it should be a primitive type node

for example if you search for primitive and connect to the text button, it should automatically convert the node to a string type node

i have so many custom nodes these days its hard to tell where everything is from lol

but i think that one is just comfy

in comfy, you can also drag an input button to the workflow and it gives you suggestions as to what can be connected, for me, one of them was the string node

you should have it too

[maiden wigeon]

that's my list ^^

[hardy tree]

well you dont always need this list btw

you can also like left click to search for nodes by typing

or like i said, drag the buttons so comfy finds a potential match

[maiden wigeon]

Thanks for the search fonction ! i have maaaany string nodes but not this one, thanks anyways for the help

[hardy tree]

i actually found it in the menu, its under utilities -> primitives -> string

but i think thats a custom node, only problem is, i have no idea where it comes from LOL

[hoary zinc]

btw what are those split clips? Or why split them?

the only use could be for Lora's to make it clear which one does which action.

[hardy tree]

split clips?

[cursive ridge]

Apparently, don't select that auto queue checkbox ever 😄

[sick badger]

you can use Show Text to see your combined text

[deft saffron]

I want two person standing together but using photomaker ,how to do that so that two person have different face ?

[quick birch]

There is different way to combine images
The to- from works only with words Example First prompt House and Second Prompt Garden. You will a house in a garden change garden for Forest and you will have a House in a Forest. The first prompt can be more elaborate. This function can touchy
Now there is other Conditioning Combine , this add the two prompt a + b In the first Prompt is your principal Subject and the second it is the environment
You have another Conditioning Set Area with this one you can position an object in a special place in the picture.

[quick birch]

I don't know photomaker but have you tried to make to distinctive workflow for each person with their own face and then combined the two latent to make a single image ?

[deft saffron]

Can we talk on private chat,I have some questions

[deft saffron]

Can we talk on private chat,I have some questions

[quick birch]

ok can we do it with discord ?

[deft saffron]

Sureeee sir

[quick birch]

Lets go 🙂

[deft saffron]

And using photomaker

I generated my profile pic ,it looks just like my face

[hoary zinc]

the lcm models are really great

I can generate low quality preview in the speed of light

[sick badger]

https://replicate.com/philz1337x/clarity-upscaler

would be great to see phil setup some comfy nodes for this

[mystic plaza]

where can i get more models for comfyui ?

[last root]

https://civitai.com/

[mystic plaza]

for some reason i cant seem to get this image generated properly

[last root]

well depends on the prompt/model you used and settings. some models requires specific settings for it to work well

[mellow loom]

/model

[mystic plaza]

its the basic 1.5checkpoint model

[mystic plaza]

@last root and where should i add the downloaded models in the models folder of comfyui or lora ?

[sly sundial]

models/checkpoints/

[delicate snow]

Chinese Style Year Wheel Tai Chi Ripples Golden Sunshine Ancient Streets

[cedar mirage]

Well with the prompt: "6 grey cats on an abstact silver artwork under three moons, star sky, drawing" it would be a nice result. Without knowing the prompt it is kind of difficult to judge the result 🙂

[mystic plaza]

did i not descibe the glass properly since it got wrapped

[night sigil]

hi

[mystic plaza]

am i doing something wrong since most of my images are coming out bad

[lone bolt]

Turn down the CFG between 5 and 7.5 and change the scheduler to karras.

[tranquil parcel]

Problem in ComfyUI Text strings enclosed in curly brackets {} no longer work at all, is anyone else having this problem and knows how to fix this?

[mystic plaza]

so in comfyui do we need to descriptive like i want to create this kind of image but its not giving me that kinda of image can i pls get some help

[hoary zinc]

or you can always try to use AI to do the description of an image

[hoary zinc]

{} = () brackets, but {} are older, at least according to AUTOMATIC111 documentation

[mystic plaza]

meaning ?

[hoary zinc]

AI Copilot to make a description of a picture

and just make a prompt out of it, add things like: natural soap, turmeric roots in the background etc.

[mystic plaza]

its sandalwood actually

[hoary zinc]

lol

yeah, my dormmate told me it looks like a caramel 🤣

[mystic plaza]

bots dont understand human language

[hoary zinc]

well, it certainly is interesting, although even if I make it write a prompt it'll never use the form I give it to.

such as: write a prompt for the appearance of an anime character (find it by yourself on the internet) and then make a prompt in a form: "glasses, long black hair, ..."

it'll just output: upper body: long hair: green hair with ...
lower body:
hair: etc.

[grand pike]

Hello lovely people. Does anyone else have trouble with open pose creating poses from existing images? It's always missing like half of the limbs and it even changes the style of the picture, very blurry.

[hoary zinc]

Although I'm new in this can I look at it?

What exactly are you trying to do?

so far I've tried upscaler and face detailer

unfortunately I hit the CUDA error in face detailer when I tried to generate 2nd face detailer layer with pipe

haven't tried that, but you could add prompt such as "full-body, portrait"

[eager escarp]

it looks like TwoAdvancedSamplersForMask from Impact Pack is broken - the first step looks like a blurry image, second step you can see a lot of noise... it just adds noise. anything much beyond that looks like random noise.

[hardy tree]

these days you can have an open pose editor for comfy, where you can manually make the pose based on the image @grand pike

as for the blurry part... hmm not sure

maybe settings have to be adjusted

also, do you specifically mean OpenPose? cause i think DWPose is better

and maybe others, i didnt check everything myself :3

[grand pike]

@hardy tree Oh? That's sounds dope. Where can I learn more?

@hoary zinc sorry not at home right now. Will try that open pose editor though

[hoary zinc]

I'll try it too when I'm free. I haven't tried to add Lora as a node yet

[hardy tree]

well dwpose is part of the controlnet preprocessors (with openpose, etc), so when you use it the first time, it will download all the models for you. as for the editor, i think if you search for openpose in the manager, you can find couple, and also on github

[hoary zinc]

Does anyone have an idea how to make more than one Face detailer result? I got out of CUDA memory and can't include tilted decoding like I did with the SD upscaler's situation

https://www.youtube.com/watch?v=ekofgf9T-9c It's according to this video, but swapping doesn't work for some reasons.

because the node is updated or something it wrote last time

so if there is a possibility to do the thing with the pipe without running out of VRAM

[steep marlin]

Maybe some precision got changed under the hood. Like from fp16 to fp32

Or the wrong model file is being loaded that is in fp32

[hoary zinc]

because it seems like the VRAM doesn't get flushed and everything stays there (every single step) resulting in memory overflow => CUDA VRAM error

so if there is a way to do 2nd processing of the face through the face detailer 2nd time without doing it all over again.

What exactly is SUPIR and what does it do?

all I see is praise on YT, but I'd like to know if it can run on 4 GB VRAM. So far I had to change img generation to tiled which was easy, upscaler was a bit tricky but I managed to do it too.

[fickle remnant]

Workflow to place product shots into a generated scene, with basic colour adjustment, subject interrogator that automatically places it into your prompt, made simple with group nodes, so not much spaghetti. Workflow should be in the image

Original Image

You need to use CosXL_edit model

[hoary zinc]

What nodes/plugins do you need for it?

[olive yoke]

comfyui has changed to not treating { } specially in all nodes by default. in your workflow, make an ordinary text box by deleting the connection to text multiline, then creating a string primitive, not text multiline, and curly braces will work. it will be multiline anyway.

[olive yoke]

this isn't a supported sdxl resolution. use 1344x768

[mystic plaza]

i see

[hoary zinc]

Where can I find info on supported resoultions of the base models? SDXL and SD 1.5

[olive yoke]

you can punch that exact query into google it's just a list

[hoary zinc]

ok, thx

[olive yoke]

i knew it was an issue because bad resolutions often yield doubles

[hoary zinc]

sdxl has 1024x1024

[olive yoke]

due to the "wrapping" that occurs to implement different target resolutions

[hoary zinc]

good to know, I actually used SD 1.5 with 640x640

[olive yoke]

what is your goal?

you are making a very complex workflow. you don't need any of those extra custom nodes.

[hoary zinc]

just wanted to try it in more steps

yeah, but it's kinda needed for 4 GB VRAM

[olive yoke]

hmm

[hoary zinc]

I had to modify upscaler too or else I'd run out of VRAM

[olive yoke]

but you don't need to use upscaler at all

it doesn't really... do anything

what is your goal?

[hoary zinc]

upscales images

[olive yoke]

what are you trying to generate? what is an example of a prompt?

[hoary zinc]

just some random stuff and trying out if I can make better faces

[olive yoke]

do you have an education in any visual artistic practice*?

[hoary zinc]

nope

just wanted to learn this since I got into AI with prompts and it's been very useful

[olive yoke]

so in what sense would you know if a face is better or worse? it's all placebos to you

you haven't even said a prompt

[hoary zinc]

well, disproportioned eyes or just some little details

[olive yoke]

i mean putting in all these custom nodes and all this complexity feels good, it feels like an improvement, and i think that's why people do it

[hoary zinc]

probably

[olive yoke]

you still haven't written an example prompt

we're almost 20 sentences in, no example prompt

[hoary zinc]

it differs from model to model 🤷

I'll try one in the evening

[olive yoke]

i feel like stability is doomed.

it would be effortless for you to write a prompt

every time someone uses DALL-E3, Bing Image Creator, etc., they get to see the prompt. they can improve things

you can't even write a prompt to me

even though it is effortless! it is completely effortless!

it's copy and paste!

how is stability suppose to improve anything without the prompts

how would it know

you can go get a prompt you want to improve

you have it. i mean maybe the way you've been interacting with this stuff is copying prompts from civitai or whatever

that makes some sense

and instead of telling me a story about how you can't get this prompt

[hoary zinc]

a girl with a long red hair, blue eyes, clear eyes, full-body, holding out a hand, on the road

[olive yoke]

okay so

you know, what's the idea here?

[hoary zinc]

for example, but it might be because wrote too complex examples and specified too many things

yeah

[olive yoke]

huh?

i mean what's the real prompt

that you are having trouble with

it is already obvious to you that you don't know how to write the prompts. so why are you adding the complexity of the face detailer nodes or whatever

[hoary zinc]

well, I can't tell you, I'll have to boot up my notebook, I haven't saved the prompts, that's the problem, but some turned out really good

[olive yoke]

the prompt is saved in the images

[hoary zinc]

not face detailer, bruh

[olive yoke]

you can drag and drop in the image into your comfy workspace

[hoary zinc]

you're completely misunderstanding

[olive yoke]

but you don't need face detailer

you have to first define what you want

[hoary zinc]

it's all about the config of workflow 🤷 just wanted to ask for an advice about how not to run out of VRAM that quickly, because when it process once it doesn't flush VRAM and instead keeps everything there

[olive yoke]

which is hard without any art background

the VRAM story is telling you that you are doing something wrong additionally in a broad sense, not just the small sense of, "well i run out of VRAM due to a bug"

[hoary zinc]

yes, it is, that's why I try to observe

I managed to fix the upscaler I told you about before

[olive yoke]

define the final image result you want*

[maiden wigeon]

Hello everyone, how can I make it so that the file created in comfyui has the name of the prompt when saved in outputs? Instead of a generic name? I can't figure out how to do it, thanks!

[olive yoke]

the simplest way is to use a string primitive for your positive clip text encode, then connect it to your saveimage prefix.

[maiden wigeon]

thanks, idk how to do this but i'll check

i haven't option for prefix node connection, how could i ?

[lone bolt]

You can convert the prefix to an input. But be aware that your prompt can contain unsupported characters

I would not recomment to do that

Why would you name the file like that if you can retrieve the prompt from the workflow, embedded into the image?

[maiden wigeon]

thanks for answer, I need to do this to save time, but I don't understand how to make the connection point appear on my save image node to connect it to something else.

[lone bolt]

Rightclick -> Convert widget to input

[maiden wigeon]

hoo thanks ! i have to find this string primitive now

[lone bolt]

Just add a primitive and connect it

It will take the type of the input

For your shenanigans i would recommend the Text Multiline node from the WAS node suite

[maiden wigeon]

it worked, thanks a lot !!!

[lone bolt]

Well, no. I take that back. Maybe not a good idea.

[fickle remnant]

Please use comfyUI manager

[hoary zinc]

I use it.

[hoary zinc]

Is there a way to make meshgraphormer hand refiner to recognize the 2nd hand?

it only shows one, so I wanted to test

[fickle remnant]

[hardy tree]

nice hands

[hoary zinc]

ok, found a solution called krita + ai plugin

gonna try it

[olive yoke]

no

is this SD1.5 or SDXL?

[hoary zinc]

SDXL

[olive yoke]

use the hands fix slider lora

[hoary zinc]

ok, I'll try to find one, actually, I have only SD 1.5 ones

[olive yoke]

meshgraphormer is a good approach but it doesn't detect the hands a lot of the time

as you can see

but generally you should use a negative embedding and pernegguidance to ensure that you don't generate images with problematic hands in the first place

[hoary zinc]

used it and edited the prompt, but it still generates

[olive yoke]

negative prompts only really started working with perpnegguidance introduced what, today?

used what?

[hoary zinc]

incorrect anatomy, low digits, sketch, missing limbs etc... bad hands

and unaestheticXL_Alb2

but I used rgthree power prompt instead of the usual clip

[olive yoke]

you are using too many custom nodes

that are all really radioactive

[last root]

for hires latent fix in confyui what size or % should i increase it by

[rain hearth]

what are you trying to do

[hardy tree]

@storm folio just wondering, will there be native support for TCD/Hyper-SD ? it seems to be some form of improved LCM

[wispy kelp]

Post some screenshots of your most insane convoluted messy comfy workflows. XD

[pallid cape]

sdupscaler takes alot of time for 1 pictre to generate anyway to optimize it ?

rtx 2060 6gb

takes up to 20 30 mins and still not done

[hoary zinc]

depends on how many steps you use

[livid gust]

I'm building my first ComfyUI node, and I'd like it to have a unique color. I notice the Note node does this, comes up brown, but I can't find any information on how to set the node color on creation. Does anyone know?

[clear bear]

If that goes against rules/TOS, please let me know 👼

[livid gust]

@clear bear Shouldn't this be in some other channel? What does this have to do with comfy-ui?

[fickle remnant]

Please use #1092446741984444416

[clear bear]

Thank you very much 👍

[carmine copper]

Anyone here have experience on building a custom workflow

[fickle remnant]

Whats your plan?

[hoary zinc]

How can I install DWPose into ComfyUI?

[lone bolt]

https://github.com/Fannovel16/comfyui_controlnet_aux
Has a DWPose Estimator node

[hoary zinc]

found it, I'll try

[hardy tree]

yea it automatically downloads the models

[hoary zinc]

which one is it?

because I clonned DWopenpose folder and put it into controlnet folder in comfyui

but I saw people having DW Open Pose node which I don't have

so some files must be in custom_nodes folder

[hardy tree]

well in your main menu, do you see something called "controlnet preprocessors" ?

that's how you know it installed correctly

[hoary zinc]

no

which menu

[hardy tree]

most of the time, it's best to use the manager to install things, cause doing manually can sometimes mess it up

well i mean the menu of nodes, when you left click in the workflow

(add node)

[hoary zinc]

preprocessor selector or DWPose Estimator

[hardy tree]

so you do have it?

[hoary zinc]

I have these two

so idk

I saw people having a special node for it

[hardy tree]

well it's called dw pose estimator, not sure what is that other thing you are talking about

[lone bolt]

I posted the node a bit above

[hardy tree]

i personally never seen two for dw pose estimation, only one :3

[hoary zinc]

I'll try to look into it a bit more, but on youtube all I see is A111 where you click a few buttons and have it installed and none for ComfyUI

[hardy tree]

make sure to filter videos by adding comfyui in the search

but you already have it then, if you see the name DW Pose Estimator

[hoary zinc]

ok

[lone bolt]

I guess thats what you mean?

[hoary zinc]

yes, this one

thank you

[hardy tree]

dwpose baby :3 way better than openpose

[opaque sentinel]

Can someone help me fix this problem on ComfyUI please? I really have no idea how to fix it.

[lusty pewter]

Uninstall ComfyUI Impact Pack.
Update ComfyUI to the latest version.
Install ComfyUI IImpact Pack.

[opaque sentinel]

Let me try now

-i didn't install it yet but when im installing, it doesn't wokr
-ComfyUI is the lastest version

[opaque sentinel]

Please help

[hardy tree]

your problem is stated right there: cmd('git') failed due to exit code(128)
do some google search on that and try to solve it

i never seen that problem before, weird

[lusty pewter]

Hence the suggestion about updating. But he's on portable version...might be capped out.

Try what the person in the replies did here:
https://www.reddit.com/r/comfyui/comments/18pc555/error_trying_to_update/
(Just make sure that you use the correct path, not just copy theirs.)

[hardy tree]

oh so this is a problem if his comfy version is too old?

[lusty pewter]

Might be.

[opaque sentinel]

It is possible because I install "git" and "ComfyUI" in different drives, C and D?

[hardy tree]

git should work across drives, as long as it's installed correctly

[lusty pewter]

Doesn't really matter where git itself is installed as long as it's in the path.

But your system sees the git executable, so I doubt that's the issue.

[opaque sentinel]

Thank you so much guys..really appreciated

[opaque sentinel]

Let me check this

[hardy tree]

but im curious, is the problem gonna happen if you install via git outside of the manager?

like directly in the custom nodes folder

[last root]

in confyui when im generating a batch of images do they all have different seed numbers? if im generating like 4 image how do i know which image has what seed?

[hardy tree]

i think it's sequential, first one is the seed you see on ksampler and the others are +1
but not sure

[opaque sentinel]

now only these 2 that i can't install guys... how can i fix it?
i already updated everything

[hardy tree]

its the same error, so however you installed the others successfully, install those two as well

[opaque sentinel]

i did but these two can't install sir

[hardy tree]

outside of the manager?

[hardy tree]

dont use the manager, install them with git but directly in the custom nodes folder

[opaque sentinel]

ohh let me try. im sorry im not really good at computer. i want to use this for my work

[hardy tree]

i mean some of this stuff you just have to learn once and you will know it 🙂

[opaque sentinel]

thank you so much 🙏

[opaque sentinel]

now only this problem sir

ComfyUIStyler

[hardy tree]

you have to find where the node comes from

in manager, install missing custom nodes will tell you where it's from (most of the time)

if not, you need to search on google or github

[tacit fulcrum]

Did they remove (visual area conditioning/latent condtioning) custom node or is it now there by default and named something else

[silver nebula]

I updated ComfyUI and get this error

[opaque sentinel]

thank you

[grand pike]

hi. so how do people make wildcards? Is it as simple as typing out all the descriptions in a text file, and then just using the file name of the text file as a trigger word? I'm making a random ethnicity prompt with 100 ethnicities, but i figure having that many tokens is bad?

[grand pike]

https://civitai.com/articles/959

Found this for anyone else interested

[hardy tree]

the wrong thing was to get an AMD GPU 🙂

[mild ivy]

does anyone know if there is a way to combine stuff like animatediff and layerdiffuse to make animations with the background transparent?

[hardy tree]

considering all the stuff you can do in comfy these days, i would be surprised if there isnt a way, but never did it personally tho

[steep marlin]

Yeah but it would be a multistage process. You'd render the animdiff frame first, then run layered diff to separate it. You would have to do that for every frame of the animation and hope random frames don't bug out on you