🤝｜tech-support

[urban pollen]

yeah the audio part is local, the gemini nodes just do the lyrics/planning

just match it to your main cfg scale to get the same results as auto1111

[celest delta]

thanks

seriously though, i just can't get the pictures in forge neo as good as my settings in auto1111 and its getting me crazy... style lora isnt as close and there's more weirdness overall

no matter what i try, which seems like im the only one affected, lol

[sudden pike]

ok wtf. its 10x better

thanks guys

man i was like 2 months not using SD and much happened

AI going faaast

[ornate elk]

Yea AMD is working hard to get us the good speed ^^

[urban pollen]

two months is like forever in ai time, everything moves so fast

[celest delta]

see, i want neo to get the good speed like we all agree it has, but my results just doesnt hit right for me i dont get it... also is there a good civitaihelper for neo? the only one i found is glitching and doesnt find the pictures for ALL my lora unlike in auto1111

[urban pollen]

check your clip skip and vae, that usually fixes it

[limber flax]

Thanks @urban pollen & @ornate elk - I've added your workflow and am downloading the ace step 1.5 turbo model now. Should I go ahead and delete the other models for the template that uses Gemini or could I reuse them in another template?

[celest delta]

i use the same vae, and clip skip wasnt a thing on auto1111 so at what value should i put it?

[urban pollen]

it was definitely there in auto, usually 1 for standard or 2 for anime

[celest delta]

maybe i had an older version i dunno

but in neo its clearly there

also its already at 2 so i guess thats not the issue

[urban pollen]

try 1 for normal models and 2 if it's anime stuff

[urban pollen]

which model are you using? that usually matters more than the ui

[celest delta]

yeah i use it for anime, didnt change it

[celest delta]

i tried the exact same settings in both, models snd all that i should get the same result but nope

[urban pollen]

sometimes the optimizations make it slightly different, you using xformers?

[celest delta]

using wai illustrious

[celest delta]

yeah

i tried with cs1o last time but we couldn't figure it out

[urban pollen]

damn, if the seeds match it might just be math optimization differences honestly

[celest delta]

its annoying cause i really get great results in auto1111, i wish i could get the same on neo, but ill try some more stuff this weekend

[urban pollen]

yeah seed parity is basically a myth between different backends, it's super annoying

[celest delta]

even if the seed gave something different i wouldnt mind if the actual quality was the same... less accuracy for style lora and more warped stuff like face or parts of the body

i can still get good pics, but less

you could say since its faster i can generate more to get more chance at good stuff, but eh, lol

[plain crypt]

Hi, quick question: do you know how I can make short, realistic videos on Stable ?

[urban pollen]

check out svd or animatediff, comfyui is usually the best for video stuff

[plain crypt]

I have comfui but it doesn't work

[urban pollen]

what error you getting? video usually needs a lot of vram

[plain crypt]

Oh no, I don't have it anymore, how can I reinstall it?

[urban pollen]

grab the portable release on github, just unzip and run

[plain crypt]

Okay, thanks, but where did I find it?

[urban pollen]

just search comfyanonymous comfyui github, it's under the releases tab

[plain crypt]

ok thanks

[plain crypt]

Which file do I need to launch to open it?

[urban pollen]

look for run_nvidia_gpu.bat, that should launch it right up

[plain crypt]

Okay, thanks. I don't remember how all this works.

[urban pollen]

no stress, just drag a workflow image into the browser window to start

[plain crypt]

a workflow image ?

[urban pollen]

yeah comfyui images save the node layout inside them automatically

[plain crypt]

Uh, I didn't understand anything, and now I don't understand the Comfui page.

[urban pollen]

that looks like a website gallery, try going to http://127.0.0.1:8188 in your browser instead

[plain crypt]

I have exactly the same page, but I have the impression that it wants me to download a template from the page

I have this, then

[urban pollen]

that grok node is an api, try svd for local video instead

[plain crypt]

uh, how do I get an SVD?

[vocal burrow]

not local stuff

[urban pollen]

grab the svd checkpoint from huggingface, then put it in your models/checkpoints folder

[urban pollen]

once you have the model, you'll need an actual svd workflow too

[vocal burrow]

usually nowadays when you open a comfyui workflow it will prompt you about missing models and it should give links to download them.

(you ll still have to manually put them in the correct folders tho)

[plain crypt]

How do I download svd checkpoint from Hunggingface ?

[urban pollen]

search svd_xt on huggingface, but honestly comfyui manager is way easier

[plain crypt]

I don't understand anything and I can't find anything at all.

[urban pollen]

that grok node is paid, you need a local svd workflow instead

[plain crypt]

Okay, and which one should I choose?

[urban pollen]

get svd_xt, it's the standard for decent local video

[plain crypt]

uh, I can't find it in hungging

There's nothing that corresponds to that name.

[urban pollen]

it's under stabilityai on huggingface, look for the img2vid-xt repo

[plain crypt]

Is that it?

[urban pollen]

yeah that's the right page, just click files and versions to find it

[plain crypt]

Do I have to download the whole thing?

[urban pollen]

nah just download the svd_xt.safetensors file from that files tab

[plain crypt]

Is this the one?

[urban pollen]

yep thats it, just drop it in ComfyUI/models/checkpoints/ when its done

[plain crypt]

Okay, thanks.
It's going to take me some time to install it.

[urban pollen]

all good, once it's in the folder just hit refresh in comfy to see it

[plain crypt]

Okay, and what will we need to do next?

[urban pollen]

you'll need an svd workflow next, just drag one into comfy

[plain crypt]

And where can I find them?

[urban pollen]

check the comfyui examples site or civitai, search for svd workflow

[plain crypt]

ok thanks

I downloaded one from civitai

[urban pollen]

nice, just put it in comfyui/models/checkpoints and you're good

[plain crypt]

Okay, and then everything's good?

[urban pollen]

yeah just hit refresh in the menu, remember svd needs a starter image too

[plain crypt]

Ah, I thought it just needed more text. The SVD I downloaded uses text.

[ornate elk]

whats your gpu?

[plain crypt]

nvidia geforce gtx 1650

[urban pollen]

oof 4gb vram is gonna be rough for video honestly

[ornate elk]

with that you can't to video stuff

[plain crypt]

I thought that would be enough for 5 seconds

[urban pollen]

video is like generating 100 images at once, 4gb just isnt enough space

[plain crypt]

Ah, so I can give up.

[urban pollen]

for local video yeah, but stick to images and you're chilling

[plain crypt]

Okay, but I don't dare take a picture; I keep getting this message.

[urban pollen]

it won't break anything, it'll just be slow as hell or crash eventually

you're barely under the limit, it'll still run totally fine without issues

[plain crypt]

Okay, and is it normal that sometimes it doesn't do what I asked or does stupid things?

[urban pollen]

lol yeah that's just rng, try using negative prompts to remove weird stuff

[plain crypt]

how do negative prompts work

[urban pollen]

basically you just list things you don't want, like blurry or extra fingers

[plain crypt]

OK thanks

[slate epoch]

Hello guys please help me out with these images I have a very low Vram PC I can't run stable diffusion locally and I seriously even do not know how to write these image from I tried a online tensor .art website but still there are not I tried to generate at these images with the prompt but I am not getting the same type of quality I also selected the flux dev 2 model but still at that side is unable to generate this type of quality please help me out this

[urban pollen]

oh yeah that's easy, try adding bodycam footage and fisheye to your flux prompts

[celest delta]

another day without CLIP being fixed, surely this weekend :p

[green eagle]

hi all: do you know if this is the right place to ask? I can't get neither forge couple nor regional prompter to work,
did I did something wrong?

[zenith osprey]

Using SD Forge I upgraded from a 3070 ti to a 5080 and get this error... RuntimeError: CUDA error: no kernel image is available for execution on the device 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.

[pallid isle]

anyone want to help me? ill pay

[ornate elk]

you can fix it now by doing this:
after getting the error open up a cmd inside the auto1111 or forge folder by clicking inside the explorer bar and type cmd.
then run these three commands one by one:
venv\scripts\activate
python -m pip install setuptools==81.0.0
python -m pip install --no-build-isolation https://github.com/openai/CLIP/archive/d50d76daa670286dd6cacf3bcd80b5e4823fc8e1.zip
then relaunch the webui-user.bat

[ornate elk]

with a rtx50 gpu you want to use Forge Neo instead, as forge is outdated.
You find a setuo guide in the pinnes messages of this channel.

[ornate elk]

hey, XL based models need a larger resolution
and then the first line should include 1boy or 1girl, just the lines after that

[green eagle]

ok thanks. think I got it to work...although I still have diffculty with overlapping characters

[celest delta]

i should delete venv again before doing this right?

[ornate elk]

nope

[celest delta]

cool

[ornate elk]

venv needs to be there before

[celest delta]

got this error now after launching webui-user

[ornate elk]

ah okay thats the older bug where stability ai removed stable diffusion from github

switch to the fixed dev branch:
open up a cmd inside the stable-diffusion-webui folder by clicking inside the explorer bar and type cmd.
then run these three commands one by one:
git fetch --all
git checkout -b dev origin/dev
git pull

you then probably need to delete the venv folder and rerun the webui-user.bat
and then you need to redo the fix with the clip again too

[celest delta]

is it right? asking since it didn't do lots of text this time :p

got the exact same error

after redoing the previous clip fix

oh wait...

i tried to copy that line, but it didn't fix it

oh i figured it out, i had to do this for each folders

and now new error

@ornate elk

[celest delta]

ok i got it working by updating numpy

FINALLY

one last issue

it's stil launching the old arguments :

despite the .bat being edited to only --xformers

[ornate elk]

Have you saved the bat after editing it?

Or do you have a second bat on the desktop that you launch?

[celest delta]

nah i launch the bat directly in the folder and yeah its saved

[ornate elk]

Can you recheck how it looks?

[ornate elk]

Save the file again and relaunch

[celest delta]

???

why does it work now???

lmao

i always save when quitting and when i reopened it i didn't have to change anything (after all i had to put that new path in)

[ornate elk]

Windows 11 updated the notepad tool, so it doesn't ask to save when you quit it...

[celest delta]

i know that

but i still save, for forge neo it worked

same as before for auto1111 since i had to put the new path for auto1111 or it wouldn't work

so weird

but thanks anway

so it allocates about half the VRAM of my graphic card now, does it sound right?

with just --xformers

[ornate elk]

What's your GPU again?

[celest delta]

5080

one last question for forge neo this time

when launching i get this :

should i add cuda-mallloc? or replace cuda-stream with it?

[ornate elk]

You can add --cuda-malloc
But it doesn't Improve that much

[celest delta]

lastly, (sorry i keep adding), what civitai helper should i use for forge neo?

the one i found doesn't manage to get all the pictures from it, some models do, but other don'ts

the normal one i have in auto1111 works perfectl and find everything

but that one isn't compatible in neo

[prisma moss]

none because they all dont work with neo/classic

another reason i refuse to upgrade from vanilla forge

[celest delta]

guess i'll just use auto1111 to grab all the things and then use neo, xD

@ornate elk i think i found the issue why my pics look different in neo and auto1111 in term of seed... the hires resolution when put at 1.5 for exemple doesn't give the same resolution

like x864 x 896

[ornate elk]

This one could work:
https://github.com/anxety-solo/sd-civitai-browser-plus

Its the only one that is maintained and not outdated

[ornate elk]

Yea I noticed that too that forge neo changed that behaviour so that the numbers can be better divided by 64

I will ask him why

[celest delta]

it works and get all the images, thanks

is there an extension that would let me change the lora preview image with one of my generations?

i think i'm finally ready to give up on auto1111

how do i cleanly uninstall it?

aside deleting the whole folder?

also is it me or forge neo requires less sampling steps? i get good results on illustrious on 10 steps

dunno if it's related at all or if i just got lucky

[ornate elk]

You should be able to hover over the lora preview and then click the edit button and there you can set your own preview image

[ornate elk]

It depends on the models and samplers used

Mostly Euler a, 30 step is the goto for illustrious, but v-pred models need less steps I think

[ornate elk]

Deleting the folder is enough. And uninstalling python 3.10 if its installed

And if you open up a cmd and run
Pip cache purge
That will delete temp files it used for the install.

[bold orbit]

im not sure if this is where im meant to ask but what words do i use to make the pov not to up close

[ornate elk]

try cowboyshot

or full body

[celest delta]

there's #📝｜prompting-help for the next times 👍

[bold orbit]

so if seedance 2.0 is the best video maker can you run it on the laptop or can you only use it on their website

[ornate elk]

its cloud only so works on every device which can open their website

[short night]

where download SDXL

[uneven heron]

models?

[short night]

yes

[uneven heron]

civitai.com

[celest delta]

speaking of civitai, the file names are fucked right now when downloading a .safetensor, am i the only one with this issue?

it adds random characters and numbers between spaces, and at the end of the name too

[short night]

Who knows Hiresfix or upscale how to use it correctly from the stability matrix application

[rose apex]

Which webui?

[short night]

stability matrix or stable diffusion

[ornate elk]

sounds like a site or download issue

[ornate elk]

as example set hires fix upscale to 1.5 or 2, set hires steps to 10, and denois to 0.5 and upscaler to Esrgan 4x and then test

[short night]

I test

but my image bad see

how fix?

[ornate elk]

show the settings you used, like sampler, steps ,etc

[short night]

i send you dms

[gentle sundial]

@ornate elk Hi Legend, I have a question, can I run ZIT on forge? and what do I need to run it, like clip , encoder , vae? I downloaded it and put it with other models but didn't work, it can't read or recognize the model

[ornate elk]

hey, it wont work in Forge, you would need to use Forge Neo

[gentle sundial]

thank you! I have neo, do I need encoder or something else? is like flux or pony xl?

[ornate elk]

ah okay, yea then you need the clip and vae for it to work or an all in one model

so these 3 from the screenshot or the fp8 model or gguf Q instead of bf16
or the allinone from here:
https://civitai.com/models/2173571?modelVersionId=2448013

[gentle sundial]

I downloaded jib mix zit fp8

[gentle sundial]

thank you! can all zit work with these? qwen_3 and flux vae?

[ornate elk]

yep all z-image turbos work with these

[gentle sundial]

Thank youuuuu sir!

[ornate elk]

no problem!

also select z-image turbo in the ui-preset dropdown, and check the models description on civitai for the best settings

[inner fiber]

Hi, I'm trying to add Control net to my stable difussion but it never works... It never shows

I've been trying everything I could for hours

[ornate elk]

hey, on which webui?

and which models

[inner fiber]

The most recent one, I checked that it was the lastest versión

I can send you the exact ver if you need

[ornate elk]

i mean which one? Auto1111, Forge, ReForge, Forge Classic, Forge Neo ?

[inner fiber]

Auto1111

[ornate elk]

ah ok and what exactly doesnt work?

installing the extension?

[inner fiber]

Nope. I've installed the extension and everything worked like usual. I've restarted the UI, even closed the terminal and still nothing.

I asked chat gpt about this, and the result was that there were multiple incompatibilities with numpy, torch and python I think.

Whenever I get the right one, the other breaks.

Some stuff for controlnet needs numpy 1.x while other stuff for stable diffusion needs 2.x

I checked I wasn't downloading the wrong stuff anddddd nope. The extension is the most recent one for now and it's the correct link

In other words, something breaks yes or yes.

that or... I'm missing something big, which... It doesn't seems like it

Like I said, I've been trying everything for hours-

And I really need control net for setting specific poses

[ornate elk]

can you show the complete cmd log after you restart?

[inner fiber]

Yes

Give me a sec

I'll send the full log when i start webui too

Oh hell no

I have the clip thing again 'cause i deleted my previous venv...

Let me fix that...

[ornate elk]

you can fix it now by doing this:
after getting the error open up a cmd inside the auto1111 folder by clicking inside the explorer bar and type cmd.
then run these three commands one by one:
venv\scripts\activate
python -m pip install setuptools==81.0.0
python -m pip install --no-build-isolation https://github.com/openai/CLIP/archive/d50d76daa670286dd6cacf3bcd80b5e4823fc8e1.zip
then relaunch the webui-user.bat

[inner fiber]

(venv) C:\Users\usuario\Desktop\LA IA 3>python -m pip install --no-build-isolation
ERROR: You must give at least one requirement to install (see "pip help install")

Ah nvm

im so dumb

loll

Didnt pasted it fully

okay done

installing againn

It broke again... @ornate elk

Launching Web UI with arguments: --skip-python-version-check --no-gradio-queue
W0222 01:12:20.160959 4644 venv\Lib\site-packages\torch\distributed\elastic\multiprocessing\redirects.py:29] NOTE: Redirects are currently not supported in Windows or MacOs.
no module 'xformers'. Processing without...
no module 'xformers'. Processing without...
No module 'xformers'. Proceeding without it.
Traceback (most recent call last):
File "C:\Users\usuario\Desktop\LA IA 3\launch.py", line 53, in <module>
main()
File "C:\Users\usuario\Desktop\LA IA 3\launch.py", line 49, in main
start()
File "C:\Users\usuario\Desktop\LA IA 3\modules\launch_utils.py", line 497, in start
import webui
File "C:\Users\usuario\Desktop\LA IA 3\webui.py", line 13, in <module>
initialize.imports()
File "C:\Users\usuario\Desktop\LA IA 3\modules\initialize.py", line 39, in imports
from modules import processing, gradio_extensons, ui # noqa: F401
File "C:\Users\usuario\Desktop\LA IA 3\modules\processing.py", line 15, in <module>
from skimage import exposure
File "C:\Users\usuario\Desktop\LA IA 3\venv\lib\site-packages\skimage_init_.py", line 122, in <module>
from ._shared import geometry
File "geometry.pyx", line 1, in init skimage._shared.geometry
ValueError: numpy.dtype size changed, may indicate binary incompatibility. Expected 96 from C header, got 88 from PyObject
Presione una tecla para continuar . . .

This is a new one...

Chat gpt told me to downgrade but if i do that more stuff would break

[ornate elk]

whats your gpu?

and which python do you have installed?

[inner fiber]

Rtx 4070 super

12gb of Vram

[inner fiber]

Let me see...

Python 3.10.6 @ornate elk

[ornate elk]

ah okay

with your gpu you should better use Forge Neo with python 3.13.12 instead

auto1111 and python 3.10 are outdated

[inner fiber]

Ohh

I never heard of that!

Oh does it work? how do i install it?

[ornate elk]

you can follow the Nvidia Forge Neo setup guide from here:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides

it has the same UI as auto1111

but already includes controlnet

[inner fiber]

OOHHH

And the stuff i use

are compatible?

Like models/loras

[ornate elk]

yep

[inner fiber]

Yippieee!!

[ornate elk]

everything works there, and more

[inner fiber]

I'll try installing that

[ornate elk]

and it gets updated frequently

[inner fiber]

Maybe regional promter could work better there? Altough I think it depends of the model or something... its really specific-

[ornate elk]

yea the dev of forge neo made an extension called Forge Couple which is better than regional prompter

[inner fiber]

YIPPIEEE

All i needed its there then

https://tenor.com/view/scrumdiddlyumptious-hulkenberg-metaphor-refantazio-atlus-yummers-gif-17445515493455474220

[ornate elk]

and it will be faster too xD

[inner fiber]

Greeat

thanks a lot im so lost with ai stuff lul

[ornate elk]

no problem, yea a lot changes in a short time

[inner fiber]

Indeed

[ornate elk]

auto1111 is outdated since over a year now, and forge too, so Forge Neo is the most updated

[inner fiber]

I used auto1111 like

3years ago

and im back again

it seems very outdated and complicated now lol

[inner fiber]

Back when i had my lil 3060

[ornate elk]

yea you can remove it, but save your stuff like models and outputs etc.
and then uninstal python 3.10.6 too before installing the new python

[inner fiber]

Ofc

What version should i have?

Like from Pyhthon

[ornate elk]

forge neo needs 3.13.12

[inner fiber]

Alright!

If i have more questions i'll ping ya, if you dont mindd

https://tenor.com/view/aria-aria-zzz-zzz-aria-idols-zzz-aod-zzz-gif-16291291823664028998

[ornate elk]

sure np

here is the link for forge couple:
https://github.com/Haoming02/sd-forge-couple

scroll down to see how it works etc

[balmy shore]

is this a good place to just share some projects

https://github.com/kenwud/BIVA-batch-i2i-via-api/tree/main

worked on this today its nothing new really but an easy way to automate i2i , or just experiment with editing images

[inner fiber]

@ornate elk Hii! what's this? this is new to me haha-

Also, does Adetailer works with this?

[inner fiber]

I'll leave this for tomorrow buuut idk what im doing, it doesnt follow the pose...

[hidden night]

I had an RTX 5070 ti and now I upgraded my CPU and Ram from i5 11400f 32gb ddr4 to 7800x3d 64gb ddr5, Will this help with image generation or just the GPU?

[ornate elk]

Leave it on automatic, and yes adetailer works normally

[ornate elk]

Screenshot the whole page pls

[ornate elk]

Only the GPU is used for image generation.
The faster ram and CPU will only help when loading the model from SSD into ram and then into the GPU

[inner fiber]

It doesn't seems to...

[inner fiber]

Also my images look more... Weird compared to auto1111, less detailed. Also, when using controlnet, the black bg influences the image sometimes so it looks way too saturated

@ornate elk ping me when you're free to fix this plzz

[willow venture]

hello , does anyone know how to skip a group when a boolean is true. and run the group when its false. in comfy ui

[prisma moss]

oppai

[inner fiber]

[prisma moss]

no touchy

[royal dragon]

hello huys, do you see anything wrong here? hiresfix doesn't work well for me

[prisma moss]

you dont have any hrfix steps

do 10-15

and 1.5x upscale

[royal dragon]

Ohh i thought 0 sets it to the same as initial gen

trying now

still same issue

[prisma moss]

is that flux

also this?

[royal dragon]

Trying now without that, for some reason it changed a custom lora to "lora". It is FluxArtFusionfp8

[prisma moss]

also, i think this is rather a prompt issue, and/or lora stacking. You got so much fluff n stuff in there, that the thing you actually want to generated gets phased out.

[royal dragon]

it works perfect without hiresfix, I guess I ll do ultimate sd upscale instead of hiresfix

[prisma moss]

also ur upscale model

idk what kinda 3jdA extension is, but that aint it

@vocal burrow

[prisma moss]

https://civitai.com/models/147759/remacri

[prisma moss]

first time seeing that thing

[torpid sapphire]

Just pop in and saw this so might as well ask, what is remacri anyways? What does it even do

[prisma moss]

its an upscale model

[royal dragon]

trying 4x ultrasharp, I hope the upscale model is the issue

[torpid sapphire]

Oh, doesn't feels like it ,but probably need to change size to notice a difference huh

[prisma moss]

ye

[prisma moss]

does it work with latent?

[ornate elk]

What exactly? It works for me without issues

[royal dragon]

didn't try, used that because I wanted to replicate this style https://civitai.com/images/121320793

[prisma moss]

he/she got big oppai instead of the pose xD

[ornate elk]

Change the controlnet strength

[inner fiber]

I did, from 0,6, 0.8, 1...

I've also tied without controlnet

And the generation is... different, even with the same stuff.

[inner fiber]

[inner fiber]

Another account hacked

lol

[long blaze]

One question, I understand that it is best to train a Lora de Z image base with ProdigyADV in OneTrainer, but can Z image turbo also be trained with this?

[lucid yoke]

Your better off asking this in the OneTrainer discord ngl

[long blaze]

I dindt find it

[lucid yoke]

The discord? Its on the github page
https://github.com/Nerogar/OneTrainer?tab=readme-ov-file#onetrainer

Click on the 970 online thing

[pallid isle]

Looking for paid help / guidance

I’m looking for someone experienced with LoRA training + AI video workflows.

Need help with:
• Training a LoRA (dataset prep, captioning, best settings)
• Getting consistent high-quality images from it
• Editing existing video footage with AI
• Background replacement
• Adding masks / face coverings
• Adding clothing pieces/ keeping clothing pieces on an image
• Integrating monsters / VFX elements ive made
• Keeping style consistent across frames

I already have the style dialed in — mainly need technical direction + workflow tips.

Happy to pay for your time.
DM me or reply here

[dusky osprey]

Hi! Im trying to run this comfy build: https://github.com/Comfy-Org/ComfyUI/releases/download/v0.14.2/ComfyUI_windows_portable_amd.7z

When I openned the README file I found that the driver that's needed only works on WIN11. I've WIN10. Comfy started tho, not sure it will work yet, but that could be a reason? Should I be worried? Thanks!!

[ornate elk]

that driver isnt needed anymore

but you need an updated driver so 25.9.2, 25.12.1 or 26.1.1 or 26.2.1

[lucid grotto]

@pallid isle Hi there bis is what i can do

[pallid isle]

please dm i can share project details : )

[dusky osprey]

Thanks!

[lucid grotto]

Sure

[lucid grotto]

@pallid isle Hello i have message you now

[crystal nacelle]

i laucnhed and it installed things and even launched web-ui cmd but then it gave me a DLL error, says another application might be blocking it? my antivirus opens up and scans the neo forge files sometimes while things are running, cmd will even pause whatever its doing at this time, but my antivirus hasnt blocked anything neo forge/ai related so it cant be that.

[ornate elk]

This sounds like Microsofts windows smart app control stuff that blocks the file

Which antivirus Programm do you have additionally?

[drifting prairie]

Which of these samplers is overall the best DPMPP_2M, DPM_2, DPMPP_SDE, DPMPP_SDE_GPU, DPMPP_2M_SDE. DPMP_2M_SDE_GPU, DPMPP_3M_SDE,DPMPP_3M_GPU

This online site I am using these are some of the samplers you can use.

[silver zephyr]

is this fixed now?

[silver zephyr]

🤔

[ornate elk]

It can be fixed now by doing this:
after getting the error open up a cmd inside the auto1111 or forge folder by clicking inside the explorer bar and type cmd.
then run these three commands one by one:
venv\scripts\activate
python -m pip install setuptools==81.0.0
python -m pip install --no-build-isolation https://github.com/openai/CLIP/archive/d50d76daa670286dd6cacf3bcd80b5e4823fc8e1.zip
then relaunch the webui-user.bat

[silver zephyr]

this works on mac too?

[ornate elk]

python 3.14 is not supported for any webui currently

[ornate elk]

Idk about Apple command Syntax :/

[silver zephyr]

so what now

[ornate elk]

Activate the venv and run the commands

Maybe use ai to get the commands needed for apple

[silver zephyr]

this command counts as one whole line?

https://tenor.com/view/i-could've-sworn-there-was-a-gif-of-this-whatever-squid-game-squid-game-edit-squid-game-edit-meme-i'm-so-scared-gif-12360771953743496060

@ornate elk 😢

[limber walrus]

Hello everyone, i hope all o' ya are doing great. I had a few questions for anyone who tried the Lora Training Colab on google,

https://colab.research.google.com/github/Jelosus2/Lora_Easy_Training_Colab/blob/main/Lora_Easy_Training_Colab.ipynb?utm_source=ainews&utm_medium=email&utm_campaign=ainews-nvidia-minitron-llm-pruning-and#scrollTo=vGwaJ0eGHCkw

I never used it so:

• Can i upload my own tags .txt along with the pictures if indicating the path of the data set?
• At a moment it says it connects to the UI so i can tweak the training parameters, do i have to have SD 1.5 installed on my computer?
• What are the performances of the collab? Is it possible to train a lora with a set of 100 picture x 200 steps each on it?

I kindly thank anyone who will take the time to answer.

[crystal nacelle]

i use AVG.

[ornate elk]

No clue but maybe a numpy downgrade is needed

[silver zephyr]

how do i do that

[ornate elk]

Can you relaunch the webui-user and share the complete log?

Maybe its caused by something else

[silver zephyr]

i do ./webui.sh again while in (venv)?

[ornate elk]

AVG got sold to Avast in 2016, Avast got sold to NortonLifeLock in 2022,
NortonLifeLock literally spy's on their users and even installed cryp-to mining software on their customers devices with Norton360.
And Avast also spies on their users whole PC, so all of these are crap.
Uninstall and then only use Windows Defender. Reboot the PC and try again.

[silver zephyr]

@ornate elk it somehow works now

😭

do i need to be in venv to launch the webui next time

[ornate elk]

Normaly not but doesn't hurt

[inner fiber]

@ornate elk hi there, if you're free, ping me. I want to fix the control net thingy and more stuffie

I haven't touched webui since last time

[ornate elk]

In 30 mins I'm at home and we can look into it

[inner fiber]

Sure!

Sorry for the delay, I was cooking irl

[ornate elk]

No problem

[inner fiber]

Hi there

I'm here

Also. Since this might be a dumb thing I'm forgetting and that creates the issue... Do you mind if I dm you instead?

I'm 90% sure I'm asking help because idk how this works instead of something to document lul

[ornate elk]

Sure you can

[inner fiber]

Alright alright

'Goinggg

@ornate elk done, check your dms

[chilly knot]

i install wangp for instruction but when i trying to start i have error

RuntimeError: Failed to import diffusers.models.transformers.pixart_transformer_2d because of the following error (look up to see its traceback):
DLL load failed while importing _fused:

my python version 3.11

[chilly knot]

i solved it

pip uninstall sageattention and install

[ornate elk]

whats your gpu?

[chilly knot]

rtx 5070 ti

[ornate elk]

ah okay, but that should support any sage attention version

[chilly knot]

strange i follow instruction 100%

[ornate elk]

hmm

[chilly knot]

i try wan 2.2 in forge but result very bad

its like mosaic

I still couldn't figure out why the image was so blurry.

i have 32 ram -often its out of memory

[ornate elk]

let me know how good wan2gp is

[chilly knot]

its good no oom

in forge i can use only 4 step

[ornate elk]

ah nice

[chilly knot]

now ideal quality

if i correct undestand i cant use my models only what UI download

[weak umbra]

I have some question on Stability.ai
I need help for api use

please help me

[ornate elk]

messsages them from their official website

this here is not an official support channel for their products

[brittle lantern]

Anyway to run Forge Neo on MacOs cuz I swear it’s only for Nvidia support

@ornate elk sorry to disturb but any idea on this

[ornate elk]

it works for me on AMD too, you just have to git clone the repo and install a compatible torch for your gpu

maybe checkout this script for mac and change it to install forge neo instead of forge:
https://github.com/viking1304/a1111-setup

[brittle lantern]

I’ve used this before but how do I change it to forge Neo?

Ik it’s -f or something for forge

But idk how to change it to forge Neo

I’m assuming I edit text editor on scrip and change the git pull to the forge Neo

[ornate elk]

yea

probably changing the git repo url is enough

[brittle lantern]

Might be my top 3 big brain moments XD

I’ll come back after trying to see if it works or not

@ornate elk so issue the git is forge classic but in the install page on git hub it tells u to clone and git clone https://github.com/Haoming02/sd-webui-forge-classic sd-webui-forge-neo --branch neo

so what do i replace in it

do i copy and past the link and the branch stuff

[ornate elk]

yes the complete command with branch

because without the branch it would be forge classic

[brittle lantern]

this?

or have i done it wrong

@ornate elk malformed input into url function

Ah no spaces

[ornate elk]

set it into " "

oh it already is

[brittle lantern]

yh im lost on how to fix it

u said at the start isntall the torch for ur gpu

issue im having is theres only bat files

and idk how to do the torch thing on mac

[ornate elk]

idk too, cause mac :/

[ornate elk]

Remove the git clone there before the url

[brittle lantern]

I’ve done a shit ton of changes it don’t work icl

[gaunt iron]

hello guys!

[brittle lantern]

Hi

[crystal nacelle]

My Neo Forge is working, that fix your recommended was the one. That + disabling windows whatwhat thingamabob.
windows defender couldnt catch spyware and malware that I had before, AVG did so ima keep rocking with it for abit longer.

[ornate elk]

Alright good that it works now.

But I wouldn't trust software that spies on me. They often also claim to find malware that doesn't even exist, just so that "they found something" and to legitimise themself

Better setup Ublock Origin as browser extension. That saves you from a lot of malware, tracking and phishing sites.

Also Malwarebytes ADWcleaner is pretty good to find malware on the PC.

[ornate elk]

@inner fiber here I tested it

[lethal tiger]

Can I ask for some help too? I'm having trouble following what I need to do.

[ornate elk]

Sure

[lethal tiger]

Thank you. The install might be scrambled and messy with all my attempts now.

[ornate elk]

What's your GPU?

Auto1111 is outdated

[lethal tiger]

My GPU is "Intel(R) Core(TM) Ultra 9 285K (3.70 GHz)"

I hear that Auto1111 is outdated. To be honest, I saw that someone might have fixed their Neo Forge and I was a bit hopeful.

[ornate elk]

Intel core ultra is the CPU. But has an igpu too.
Do you not have a dedicated graphics card?

[lethal tiger]

Oh, GPU, sorry. That one is a Geforce RTX Nvidia 5090.

[ornate elk]

Ah okay then you should install Forge Neo for sure. Its perfect for the rtx50 series

Follow the setup guide from the pinned messages of this channel

First link

[limber walrus]

Hello everyone, i hope all o' ya are doing great. I had a few questions for anyone who tried the Lora Training Colab on google,

https://colab.research.google.com/github/Jelosus2/Lora_Easy_Training_Colab/blob/main/Lora_Easy_Training_Colab.ipynb?utm_source=ainews&utm_medium=email&utm_campaign=ainews-nvidia-minitron-llm-pruning-and#scrollTo=vGwaJ0eGHCkw

I never used it so:

• Can i upload my own tags .txt along with the pictures if indicating the path of the data set?
• At a moment it says it connects to the UI so i can tweak the training parameters, do i have to have SD 1.5 installed on my computer?
• What are the performances of the collab? Is it possible to train a lora with a set of 100 picture x 200 steps each on it?

I kindly thank anyone who will take the time to answer.

[lethal tiger]

Thank you so much, I'll give it a look. I have to go to work now (used up my chance to sleep) and will be sure to update you later once I'd tried it.

[ornate elk]

Alright, let me know then if you have any questions about the setup steps then

[torpid sapphire]

So... I tried installing the deepdanbooru extension on forge neo, but it's not showing up, any where really, not even in settings under extensions, does forge neo not support it? is there a replacement for it?

[ornate elk]

Nope its outdated since over 2 years and won't work on forge neo

[torpid sapphire]

darn

anything that came to mind that works similar to it? a lot of things I found really doesnt work or from 2 years ago

[ornate elk]

Here is a fork but idk if it works:
https://github.com/kamuridesu/DeepDanbooru

[torpid sapphire]

Alright, ill try

[torpid sapphire]

nope, doesnt work, oh well

trial and error i guess

[ornate elk]

Found this site:
http://dev.kanotype.net:8003/deepdanbooru/

[atomic zenith]

I have stable diffusion webui forge neo, I just installed the update and my gpu weight slider is gone, can someone please help me get it back up and working

[torpid sapphire]

It's in the settings now, search GPU weight using the search bar

[torpid sapphire]

This one is weird, perhaps there's a size limit because some images it just cant search

Thanks though, ended up finding a ChatGPT that can analyse image and produce prompt instead

[ornate elk]

yea it says on the site that 10mb is the limit

[torpid sapphire]

Oh I'm talking resolution size, even if the image is like 200kb it still wouldn't search, even if it's sfw, but yeah

[ornate elk]

oh ok

[torpid sapphire]

The ChatGPT one someone made work decent enough

[ornate elk]

yea its okay

[torpid sapphire]

Since I'm here, might as well ask

Is there any way to prevent error 9009 to appear again

Do I have to edit the user.bat file to target python path?

[ornate elk]

yeathat helps

the 9009 only appears when python is not set in system path

so you can force a python version in the webui-user.bat to fix

[torpid sapphire]

That damn window update must have broke something

But yeah, good to know

[quick fiber]

hey @ornate elk do u do comfy support?

i was trying to download it off of amd's ai bundle

[ornate elk]

hey, dont do that xD

i have an extra setup guide for comfyui on amd

[quick fiber]

i was following a tutorial cos when i used the desktop version it kept either not working or crashing my pc 😭

[quick fiber]

ah

i can try it then

how do i delete the amd one..

[ornate elk]

idk, check if there is a uninstall button in the adrenalin ai thing

or check if you can uninstall it trough system control panel

[quick fiber]

there was a file to delete it

is this the one i should use or is there an updated version

[ornate elk]

the Comfyui with ROCm from my guides list:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides

[quick fiber]

tyty

[quick fiber]

should i be using zluda or rocm or directml

[ornate elk]

rocm

[quick fiber]

okay

[ornate elk]

let me know if any step isnt clear

[quick fiber]

ive downloaded it all

and ive tried to do a img to video and ive crashed so im trying to use a resource monitor however it doesnt work 😭

https://github.com/iDAPPA/ComfyUI-AMDGPUMonitor

HIP out of memory. Tried to allocate 17.80 GiB. GPU 0 has a total capacity of 15.92 GiB of which 0 bytes is free. Of the allocated memory 16.28 GiB is allocated by PyTorch, and 1.29 GiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)
This error means you ran out of memory on your GPU.

TIPS: If the workflow worked before you might have accidentally set the batch_size to a large number.

How does one decrese vram alocation

[ornate elk]

that error means your workflow uses to much vram

[quick fiber]

yeah

[ornate elk]

which models do you use?

[quick fiber]

wan 2.2

[ornate elk]

which ones? Q5, FP8, FP16?

14b or 5b?

[ornate elk]

this extensions requires Linux

[quick fiber]

5b

[ornate elk]

use the Q8 model from here:
https://huggingface.co/QuantStack/Wan2.2-TI2V-5B-GGUF/tree/main

[quick fiber]

okay

thanks

[ornate elk]

and here an example workflow for 5b

[quick fiber]

in workflows in there a fsat way to download all the files other than looking them up urself and downloading them

[ornate elk]

sometimes workflows include links to the files

[quick fiber]

I see

[quick fiber]

which file do i put it in

unet right?

ok well

im dumb i fixed it

[quick fiber]

also is this supposed to say that

[quick fiber]

damn that sucks

[vocal burrow]

most likely you re not using it correctly (incorrect output resolution ?)

[ornate elk]

yes, as the amdgpu monitor extension requires linux to work and wont work on windows

[quick fiber]

oh yeah oops u said that before

thought it was something else mb

[quick fiber]

how much noise do u get when u use that?

[ornate elk]

not that much

[quick fiber]

oh im getting so much

[quick fiber]

damn

[quick fiber]

all is working now thanks for ur help

[ornate elk]

no problem

[prisma moss]

I moved my comfy to another drive, but now i get:

I replaced the path in webuiuser

k so i missed a letter sec

great its more fked now

[ornate elk]

you can try recreate the venv by deleting it

[prisma moss]

i did delete it but python3.13 -m venv venv isnt doing anything

if i dont specify the version its just gonna use 3.10.9

[ornate elk]

you need 3.12.10 or 3.13.12

[prisma moss]

3.13.12

[prisma moss]

fk it i went with protable and got all working

[rose apex]

That's really fast. What's your video resolution and duration?

[quick fiber]

i knocked it down a lot tbf

to like 578 by 800 and something

[rose apex]

Portable has it's own separate issues Might wanna switch back

[rose apex]

I see. If you want good quality then go for 720x1280 or something. Also, the 5B model sucks. Use the Q5 or Q6 14B img2vid models

[quick fiber]

yeah it just takes forever

especially if i wanna do more than 80 frames

[rose apex]

Yeah. I have 9060 XT and it takes around 15-16 mins for 4s at 720x1280. I just assume it's because I only have 32 GB ram

[prisma moss]

nah this actually works

i did use portable before, went to manual, fucked it, gone back to portable with new knowledge and got it working first try

[rose apex]

I was using portable as well a month earlier but then it started causing issues with custom nodes. I don't remember exactly what but never had any problems with the venv method so

Oh and I had always been using the manual one. Just tried Portable for like a month

[quick fiber]

yeah i also have 32gb ram

[weary glen]

anyone that can help me out and tell me more about if i'm able to run bf16 models on 9060xt in stable diffusion?

[prisma moss]

@ornate elk u on?

[ornate elk]

Yes

[prisma moss]

which qwen image edit model would you recommend for a 4090?

[ornate elk]

Which model? Do you have a link?

[prisma moss]

im downloading a Q8_0 (21gb)

[ornate elk]

That could work good

[prisma moss]

do i not even bother with the bf16? sincei ts 40gig

[ornate elk]

Nope I wouldn't

[prisma moss]

ait

is qwen uncensored? when local?

or are there specific uncensored versions

[ornate elk]

Should be uncensored

[prisma moss]

ait

[weary glen]

https://huggingface.co/black-forest-labs/FLUX.2-klein-9B

[ornate elk]

Try go for a smaller variant like a GGUF or fp8 version

[prisma moss]

is this one mandatory? or can i use the qwen_3_4b?

or is that one for something else

[chilly knot]

wangp -model wan2.2 i2v 14 b

when I add frames the length of the video increases but the video is created with an additional 5 seconds of length, how to avoid this
maybe incorect frame number?

one 5 second video if i set 81
5 second and 7 second if i set 120

can i get one 7 or 8 second video?

[gentle sundial]

@ornate elk Hi man! a question , I have 8gb v ram , ZIT is really slow , what could I do to make it faster?

I'm using the smallest ZIT fp8 ofcourse

[ornate elk]

hey, which webui do you use and which gpu?

[gentle sundial]

Hi sir, forge neo , Gforce RTX 3060 Ti laptop , total memory 12gb , shared memory 8gb

[ornate elk]

okay, if you have 16gb of ram, then you need to increase the windows pagefile

that improves the performance

and you need these launch args: --cuda-stream --sage --fast-fp16

[wispy elk]

anyone know by the chance how to fix this?

[gentle sundial]

how do I do that? I don't know what that means lol

[gentle sundial]

how about __medvram?

[ornate elk]

not needed for forge neo

[ornate elk]

yes, whats your gpu?

[ornate elk]

Here is the Guide how to increase it:
https://www.tomshardware.com/news/how-to-manage-virtual-memory-pagefile-windows-10,36929.html
Enable it only for C and make sure its not enabled for any other drive.
Set it to 16000 Min and 24000 Max for C
Then apply and reboot the PC.

[wispy elk]

RTX 3090

[ornate elk]

then dont use auto1111. setup Forge Neo instead.
You find a setup guide in the pinned messages of this channel

[wispy elk]

mini thanks

[gentle sundial]

u mean many thanks

[wispy elk]

ye

[ornate elk]

np, auto1111 webui is outdated and broken with the clip error, Forge Neo will work much better

[gentle sundial]

@ornate elk bro, is there any way or lora, that can change the boring default face that pony makes? I looked every where but nothing good so far that makes random faces , I even used wildcards but they work good only with sd and sdxl ,

[ornate elk]

try use the adetailer extension

[gentle sundial]

I did, it just increases the face quality

[ornate elk]

yea true

there are no face loras

[gentle sundial]

it says This resource has been removed by its owner

[ornate elk]

you can play around a bit with the denois in adetailer how strong the facefix is

[gentle sundial]

hmm will do, so nothing really that can generate realistic random faces for pony right?

[ornate elk]

nope

sadly not

[hidden night]

how to train a lora the easiest way?

[lethal tiger]

I no longer see parantheses that Automatic1111 or Forge are no longer working (https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides). Does this mean they can be installed again? I'm still working on installing SwarmUI as an interface for ComfyUI though.

[ornate elk]

yep, I included the workaround fix to get them working again

[cursive hearth]

I am looking for a solution for using Batch Upload in Forge/Automatic1111. I want to upscale a number of images but once I select one image, it won't take any more. I know I can put them into another folder and use the From directory option but it's inconvenient, I want to just keep dragging images into upload (simliar to how you can just continue to drag and drop images into discord).

Anybody have a solution?

[prisma moss]

go to extras tab and upscale there

[white stirrup]

i get this error anyone know a fix

gpu is 3080 and imm on comfy

[prisma moss]

https://tenor.com/view/aqua-konosuba-kazuma-sad-hand-on-shoulder-gif-2472116041386216360

[white stirrup]

so..

[prisma moss]

Do u use --use-sage-attention argument?

[white stirrup]

i have literally no idea what that means

i just click run nvidea gpu and thats it

[prisma moss]

Should rpolly do a clean reinstall, i had that issue myself on a few occassions and no one has ever fixed it

[white stirrup]

alright

[prisma moss]

You would have to make sure to do things in proper order, so check the github page for that

[white stirrup]

yeah i will

thanks anyway

[prisma moss]

Also whats ur gpu

[white stirrup]

^

[prisma moss]

Portable or manual

[white stirrup]

what? you mean if its a laptop or full sized?

[prisma moss]

I mean comfyui

[white stirrup]

honestly i dont remember i installed it a while back

[prisma moss]

Try portable anyway, its easier, some will argue

[white stirrup]

it was portable

[prisma moss]

Good

Also intall cuda dev toolkit

Cuda 13.1

[white stirrup]

thanks i will

[zenith osprey]

Does anyone have a good guide for Wan 2.2 in Stable Diffusion Forge Neo?

[prisma moss]

yea its called better use comfyui

[gleaming stirrup]

what type of stack would be required to read in an image and then evaluate its contents (the use case is for maps)

[fluid bobcat]

Still need guidance

[zenith osprey]

I ended up trying comfy ui its just very slow generation

[prisma moss]

Aint gonna be faster on neo either, if anything it could be slower

[fervent moth]

Okay so I am sorry for coming here just for some advice. I have an AMD PC. I know that the tech is a little more tempremental on non NVIDIA cards but I've been trying to troubleshoot this for about a week to no avail. Ive gone down a dozen rabbit holes and git comment chains etc. For the longest time I just could not get the WebUI to boot at all. I used the DirectML installation method (its possible ive made a mistake?) I was playing Whack A Mole for a while where i'd spend 40 minutes resolving one error code, then click launch and get a new one.

After about 8 days of trying now the WebUI finally boots, but. I think its not using DirectML? as far as I can tell. I am a layperson when it comes to all this stuff so its very possible Ive just done something very silly and not realised but when I click generate it just tells me I do not have a NVIDIA card installed

[limber walrus]

Hello i had a simple question for lora training, does space or return to line influence the learning in anyway when writing a .txt for the images? You know when you have to describe the picture in a txt so the ai recognize what she's learning

[ornate elk]

Hey, what's your exact GPU model?
DirectML should ve avoided if possible as its slow

[fervent moth]

Hey! Thanks so much for replying my GPU is a 9060xt 16gb gram

[ornate elk]

Ah nice, then you totally not need or use directml at all

You should use the ROCm based guides.
Check the pinned messages of this channel for the AMD setup guides.
For example Forge Neo with ROCm
Let me know if any step isn't clear

[fervent moth]

🙏🙏 aaah I thought ROCm was for Linux I must have misunderstood the setup guide

I'll give that a try!

[ornate elk]

ROCm works on windows now too since a few months. The guides are all for windows

[fervent moth]

Hey sorry for asking this question butI don't know where to begin. Im sorry for being such a beginner haha 😅 but im seeing lots of information and I don't know where I should start. Im seeing something called ComfyUI being reccomended? and then something called ZLUDA? 🙏 Do I need to start the entire process from the start again or can I simply "patch" the existing Automatic111WebUI I have installed already

Wait

belay that I forgot to read the full documentation I could expand it for a step by step guide I feel so silly

[ornate elk]

hey, as i said earlier follow the Forge NEO with ROCm Guide. It has the same ui as auto1111 wbeui, but is more updated.
Auto1111 is outdated and broken.
Also Zluda is not needed for an rdna4 gpu like yours. ROCm is the go to

so uninstall python, and delete the stable-diffusion-webui folder
then follow the Forge Neo with ROCm Guide from here:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides
Or if you like ComfyUI, then follow the ComfyUI ROCm Guide.
It has a different UI but also supports audio/video generation and more models, but is not beginner friendly

[fervent moth]

oh oops. I updated my drivers a few days ago to fix a performance problem in a game but now im getting an error message that the drivers won't work 🙏

Is this an easy fix?

[ornate elk]

if your on step 7, its okay, proceed with the guide

[prisma moss]

first u should make sure u have studio drivers instead of gaming (if nvidia)

[ornate elk]

@fervent moth did you got it working?

[prisma moss]

the gremlins took him

[fervent moth]

I did! Thanks so much for your help

[ornate elk]

Ah nice, np 🙂

[abstract peak]

Hey guys, hope y'all are doing great. I installed this upscaler. but for some reason these are my results. It's my first time running anything locally so I don't know much. Any help would be appreciated!

philz1337x (@philz1337x)

How to Upscale Images for Free with Clarity Upscaler:

1. Install A1111 WebUI (three options from hardest to easiest):
a) Follow the official tutorial: https://t.co/BFRKHCrm5z
b) Watch the YouTube tutorial: https://t.co/7PAIqdhQNp
c) Use One-Click start on @runpod:

X · Sep 2, 2024, 7:15 AM

[daring horizon]

Hello i have a problem installing Stable Diffusion for weeks.

and i need help

[prisma moss]

what exactly r u trying to install

not sure what openai has to do with stable diffusion

[prisma moss]

u r not upscaling anything, justu sing img2img

[daring horizon]

I just wanted to install Stable Diffusion and i got this error idk why

[prisma moss]

r u trying to use sdxl models?

[abstract peak]

How can i fix that? Let me know if you need any other screenshots.

[prisma moss]

go to extras tab, thats where the upscale happens

[daring horizon]

yes

[prisma moss]

you can use img2img with same prompt/tag at low denoise

[prisma moss]

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

[daring horizon]

thank you i'll try to install this one

Can i ask another question?

[prisma moss]

i hope u r on nvidia otherwise idk how it is for amd

and ye

[abstract peak]

Okay i'll try both, thanks

[daring horizon]

I have Nvidia dw

[daring horizon]

Is Stable Diffusion portable different with webui one?

[prisma moss]

idk wether or not forge is portable or not

[daring horizon]

Okey ty

[daring horizon]

do i have to delete my python to download it?

[prisma moss]

shouldnt be the case

[ornate elk]

What's your GPU?

[daring horizon]

RTX 3060 12GB

[ornate elk]

Then you can follow the Forge Neo setup guide from the pinned messages.
Its more updated than forge

[daring horizon]

Okey, i hope it doesn't give me errors, bc I'm already so tired of getting errors...

And thanks for the suggestion

[ornate elk]

You shouldn't get any errors if you follow the guide clearly. But let me know if any step isn't clearn.
The setup guides are in the pinned messages of this channel.

[daring horizon]

Ok, I'll let you know

[daring horizon]

@ornate elk Thank you so much, it worked.

I'll try to gen an image later

[prisma moss]

Spicy waifus here they come

https://tenor.com/view/sukuna-sinister-malevolent-evil-gif-15342583514003527998

[daring horizon]

😶

lol

[dusky osprey]

Hi people!
I wanna use Flux-2 klein or any other model that can do a good realistic inpainting on Comfy. I'm going insane trying to make it work on my setup.
I've a 9060XT and 32ram. My PC keeps freezing and I have to hard restart. I tried following multiple tutorials and workflows but idk why the only ones that work are the ultra basic comfy workflows for image-gen that don't really work for the quality of work im looking for. I'm trying to do something more like Gemini or flux. But I don't wanna keep paying the tokens and I try to run it locally.
I'm a total noob with comfy, I used AUTO1111 in the past (years ago) with a worse GPU and always made it work, but comfy is making me go insane. Tried running flux-2klein fp8 but it's still crashes and I have to restart.
Is there a good workflow made for amd users or something similar? Cause im about to go bald if it keeps crashing lmao.
I'm using comfyportable for AMD. Thanks in advance.

[ornate elk]

Hey, make sure to switch VAE decode to Tiled VAE Decode to not get freezes

[dusky osprey]

How do I change that ?🤔

[ornate elk]

you delete the vae decode node, then double click into the empty space in comfyui and search for tiled vae decode

then connect it like before

[tawny saddle]

cs should be getting paid to do dis frfrfr

[limber walrus]

Hello, firstly i agree CS should be paid secondly what could be the cause of this problem in my case?:

I'm training a lora, 4s between each steps and when i restart the training from a state save (in this case epoch 3 of the previous training) the timelapse between steps goes from 4s between each step to 10s, and its stricly the same parameters used in the previous training, what could cause that ?

[tawny saddle]

sounds like a bug to mes, but, def a CS thing

[ornate elk]

Can't help much with Lora training :/

[tulip sedge]

hey,

i help teams fine-tune models for their specific use cases. whether it's adapting llama 3, mistral, or a smaller specialized model, i focus on making it actually improve your workflow—not just look cool in a demo.

i handle the data prep, training loops, eval, and deployment so you don't have to wrestle with the details.

if you've got a dataset and a goal, but aren't sure how to bridge the gap, hit me up. happy to review your approach or jump in end-to-end.

[tawny saddle]

aw, nu worries 💖

[green eagle]

Does anyone know what did I do wrong?

[rose apex]

Are you using the latest 26.2 drivers? I have same gpu and every freezing issues were solved after updating from 25.xx drivers to 26.1

[main rune]

accidentally updated comfyui and now the stupid assets tab no longer show cached or saved images

it's driving me mad looking for either a fix or an alternative

[fresh leaf]

I try to install neo forge, but got error related with python. And dont know how fix it

In app execution aliases, everything enabled

[ornate elk]

You need to relaunch the python installer, then select: Modify, Next, Then select "add python to system environment variable's", then click OK and relaunch

[fresh leaf]

I download python manager, not python installer. But i reinstall and still got error

i think this because username written with сyrillic letters

But after changing username, error still comes out

[ornate elk]

Yes that causes issues

You have to create a new User without Cyrillic letters on the PC and log off and log in to that new user

Then you have to redo the installation

Also reinstall Python completely

[fresh leaf]

Is it impossible to do this on the current user?

[ornate elk]

Yes

Only thing you can try is to install python on a custom path not on C

That could work

The path should not contain your username

[fresh leaf]

I change name with regedit and, it seems, helped

And i have another error in browser

ValueError: Failed to find available model...

[fresh leaf]

As I understand, I need download sd3.5_large.safetensors, and put this in "C:\AI\sd-webui-forge-neo\models\Stable-diffusion"

[chilly knot]

can qwen image edit works with forge neo?
where i can get model encoder and vae for forge?
I try qwen_image_fp8_e4m3fn.safetensors but cmd stop

maybe no enought ram

32 gb

[fresh leaf]

How i can download sd 3.5 то forge neo? I dont understand what need make

[ornate elk]

Sd3.5 is not supported anymore in forge neo

Better use z image turbo, or flux

[ornate elk]

Yes qwen image edit works in img2img

[fresh leaf]

Which of these is better?

[ornate elk]

Depends, try boath

[chilly knot]

i try 5 models ,cmd stop
after this i start qwen image in wan2gp and it give me error

work\1\s\onnxruntime\core\session\provider_bridge_ort.cc:1966 onnxruntime::ProviderLibrary::Get [ONNXRuntimeError] : 1 : FAIL : Error loading \Wangp\wan2gp-env\Lib\site-packages\onnxruntime\capi\onnxruntime_providers_cuda.dll" which depends on "cublasLt64_13.dll

if i undestand right cublasLt64_13.dll must be in cuda folder

i dont have this file

[azure magnet]

@ornate elk Hi, could you help me to fix this?

[ornate elk]

Why does it shoes wan2gp?
What's your GPU?

[chilly knot]

rtx 5070 ti

fisrt i try this model in forge

cmd stop without error

after i try this in wan2gp

also in facefusion dont work cuda generation after install new gpu

maybe i need reisntall Onxx?

[ornate elk]

Yes that could fix it

onnxruntime-gpu

[fresh leaf]

I download z image

TypeError: linear(): argument 'weight' (position 2) must be Tensor, not NoneType

[ornate elk]

Check the wiki:
https://github.com/Haoming02/sd-webui-forge-classic/wiki
Section: Download Models and the Inference References

[fresh leaf]

Works, thx!

[ornate elk]

Perfect np!

[runic jacinth]

Yo mods, survival is down. Yall fix it?

[balmy monolith]

i'm having a lot of issues with the SAM3 node in comfyui, keep getting an error:
Exception during processing !!! 'NoneType' object has no attribute 'data'
the solution apparently is to pull code from the dev branch to a particular commit, but my pull/fetch/merge git command hasn't pulled it correctly. is anyone savvy enough with git to assist please?

here's my git msg's

checking the code in my files, they are different to those in the dac388b branch, which is what i need

my files:

dev branch on git

the two lines are not the same

the code is on different lines, there the file is not the same

[chilly knot]

i have black screen when my image in qwen edit ready and error

wan2gp-env\wgp.py:7035: RuntimeWarning: invalid value encountered in cast images = images.numpy().clip(0, 255).astype(np.uint8)

As I understand it, this is a common mistake of this model, its peculiarity, but how to overcome it?

[chilly knot]

--force-upcast-attention fix it?

[chilly knot]

i fix it -dont know what help me

1. i change resolution
2. i change Attention Type to sdpa

i think 2 help

• change input image

[azure magnet]

@ornate elk Could you help me with this?

[ornate elk]

whats your gpu and which webui do you use?

[azure magnet]

4060 ti 16gb
webui comfyui

[ornate elk]

portable, or desktop or which version?

[azure magnet]

portable

[ornate elk]

try run the update_comfyui_and_python_dependencies.bat

[azure magnet]

Still have it

[ornate elk]

ok then go into the comfyui folder, not where the .bats are.
then open up a cmd there by clicking into the explorer bar and ytep cmd and press enter

[prisma moss]

Did u install the cuda dev toolkit on ur pc?

[azure magnet]

Ok

[ornate elk]

then run these two commands one by one:
.\python_embeded\python.exe -m pip install uninstall torch torchvision torchaudio
.\python_embeded\python.exe -m pip install torch torchvision --index-url https://download.pytorch.org/whl/cu130

[ornate elk]

nope as im on AMD

[prisma moss]

I asked him lol

[azure magnet]

doesn't work

[ornate elk]

?

[ornate elk]

oh sry xD

[prisma moss]

I think he should run those in the embed folder

As thats the enviroment for portable

(been there done that)

[ornate elk]

yea or in that folder

but then he dont need python embedd in the command

[ornate elk]

try run it again in the main folder with the bat files

[azure magnet]

Say impossible to find the path

[ornate elk]

can you show the cmd when you opened

[prisma moss]

https://tenor.com/view/popcorn-guy-relaxing-popcorm-popcorn-gif-26404903

[azure magnet]

I try all. This the main

this C:\ComfyUI\python_embeded

[prisma moss]

Lmao bro censoring a screenshot 😭😭

[azure magnet]

this C:\ComfyUI\ComfyUI

[ornate elk]

yea show the cmd please with the commands

[azure magnet]

Here

[prisma moss]

Bro hiding the epstein files or something

Ur command is wrong

[ornate elk]

oh sry my fault, try these commands:
.\python_embeded\python.exe -m pip uninstall torch torchvision torchaudio
.\python_embeded\python.exe -m pip install torch torchvision --index-url https://download.pytorch.org/whl/cu130

im to tired xD

[prisma moss]

Thats basically it

[azure magnet]

ERROR: Could not find a version that satisfies the requirement uninstall (from versions: none)
ERROR: No matching distribution found for uninstall

With the first command

[prisma moss]

Listen

Open python embed folder

[azure magnet]

No worry thank you

[prisma moss]

And in the path, open cmd (type cmd, enter)

When in cmd paste

python.exe -m pip install uninstall torch torchvision torchaudio

Then the latter

[ornate elk]

okay, then you can try redownload comfyui portable, maybe the easiest way

[prisma moss]

@ornate elk also i think ur command is still wrong

U have both install and uninstall in it

python.exe -m pip uninstall torch torchvision torchaudio -y

[ornate elk]

true lol

[ornate elk]

can you try the first command again?

i edited it

[prisma moss]

I pasted the already wrong command by him, try the latter i showed u

[prisma moss]

Use his one

Ok good

[azure magnet]

Done

[ornate elk]

you need the whole command:
.\python_embeded\python.exe -m pip uninstall torch torchvision torchaudio

then this:
.\python_embeded\python.exe -m pip install torch torchvision --index-url https://download.pytorch.org/whl/cu130

[prisma moss]

python.exe -s -m install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu130

[azure magnet]

Y right

[prisma moss]

Go no

Install cu13

Oh thats for uninstalling

Then yes

Press up arrow up on cmd if u pressed no, to go back to the command

[azure magnet]

So always yes, right?

[prisma moss]

Yea

[azure magnet]

Then this: python.exe -s -m install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu130

[prisma moss]

When its cleaned everything run

And install https://developer.nvidia.com/cuda-downloads

[azure magnet]

This give me that

[prisma moss]

Did u run the cmd in python embedd folder?

[azure magnet]

No

[prisma moss]

Well you should

[azure magnet]

I'll try there

[ornate elk]

python installed by windows store causes issues, uninstall it
comfyui portable has its own python

[prisma moss]

Sec

python.exe -s -m pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu130

Try with this

Forgot the pip

[azure magnet]

Done

[prisma moss]

Lmk when it finishes

[azure magnet]

It's download stuff

[azure magnet]

Done, so now?

[prisma moss]

Download the cuda 13.2

[azure magnet]

cuda toolkit right?

[prisma moss]

Y

[azure magnet]

Installed

[prisma moss]

Go to comfyui update folder and run update comfy with dependancies bat file

[azure magnet]

Done

[prisma moss]

Now go to run nvidia bat

And see if it works

Optionally add --fast argument in it

[azure magnet]

Work. What it mean --fast argument, and how do it?

[prisma moss]

Edit the bat file and add --fast next to existing arguments

[azure magnet]

.\python_embeded\python.exe -s ComfyUI\main.py --windows-standalone-build --fast
pause
You mean like this

[prisma moss]

Ye

Ur next step should be to install comfyui manager but i cba atm

[azure magnet]

I already have it

[prisma moss]

Then thats it

[azure magnet]

What it does --fast?

[prisma moss]

Just better it/s

[azure magnet]

K

really really thank you

[prisma moss]

Np

[fathom quest]

hi does someone know how to fix this since python 3.10.6 isnt available ?

[prisma moss]

3.10.9 is available

[tawny saddle]

wonder why they call it torch anyway

🔦 .

[fathom quest]

i've tried with pyython 3.10.6 i managed to found it yet it still not working tho

now it says that

[daring reef]

it's possible to run with a mac cluster, 4x Mac Studio 27 M2 MAX Chip with 12Core CPU + 30Core GPU 32GB 512SSD, that will be 2tb of ram, comfy ui for video?

https://github.com/exo-explore/exo

it's possible for llm models, will it be possible for video?

https://youtu.be/bFgTxr5yst0?si=Cqj_tB1OVhtR1cHF

[tawny saddle]

I dunt even remember what python version I have atm

i reckon wut ever version of python was new back when 1.7.0 auto1111 was new

dint think I ever ever updated it

and all my shi work gud

appears I have 3.10.6

[tawny saddle]

hm, interesting

i dunno, u prolly doin sum wrong xd

could be sum wrong with new stable diff tho, I still be rocking that 2022 version ✌️

or 23

and I only recently as of like a year ago got 1.10.0 or sum

[atomic zenith]

Ok so I only started exoperiencing this issue

[tawny saddle]

just keep trying, prolly sum u overlooking, sum super simple too. happens to the best of us
then when u find it ur like breh, that's all it was fr

[atomic zenith]

Im generating images, precisely 1216x1216 sized using Stability Matrix Webui Forge Neo - alongside this I run SDXL Depth controlnet and adetailer for face details. After that the image upscales using hires fix settings set to use Remacri Original with a 0.4 denoise and an upscale to about 1940x1940, which I believe rests around the upscale of about 1.6?

One generated image at a time

Thing is my Ram usage skyrockets but I should have about 11GB of normal ram left as the image is generating, I have the GPU weight set at around 0.5 usage.

[tawny saddle]

yk that stock animation YouTubers use when something complex is being said and it's like a bunch of equations and shapes and geometry flying past their head

that's wut I see reading all dat xd

[atomic zenith]

After all that my computer freezes by the hires fix, my fans start spinning fast despite the fact that I have an AIO and I keep a good eye on the temperature of my cpu and gpu during each generation

top screen freezes, my drawing tablet displays and usually works mouse and all until I click the mouse or tap a key on the keyboard, then it also goes black, forcing me to manually restart

[tawny saddle]

hmm...maybe ur ram is faulty or has bad cells on the dyes due to over stress or heating 🤔

or maybe u was scammed, maybe the firmware says u have this much but really u got sum fouries in there

[atomic zenith]

My GPU is RTX 5060TI Windforce 16GB VRAM, my CPU is AMD Ryzen 5 5600G, my normal RAM is 32GB

[tawny saddle]

Ayeeee lezz goo, I got a 5600G too, great lil CPU

[atomic zenith]

it only just started happening now, any recommendations on what to do?

[tawny saddle]

yas, id say run windows memory diagnostics tool and wait literal ages for it to finish and see what it says

if u have enough ram and system resources and ur PC acting like a goofball, sum prolly crapped

[ornate elk]

What's your GPU?

Follow the setup guides from the pinned messages. They are updated

[prisma moss]

either an oom, or power related

[prisma moss]

and what is this setting 'gpu weight' ?

[atomic zenith]

its the only few settings you get when you type gpu in the settings tab

was supposed to be a replacement for the gpu weights setting?

[prisma moss]

that should be up to 95% or so not 50%

take the total of ur vram, and remove 1024

[atomic zenith]

Total VRAM 16310 MB, total RAM 32636 MB

[prisma moss]

then deduct 1024 from 16310

[atomic zenith]

15,286

[prisma moss]

and leave it at that

[atomic zenith]

but which is which?

these are decimals not gpu numbers

makes no sense to me

[prisma moss]

set the weights to 0.93

[atomic zenith]

and the blocks?

[prisma moss]

leave them

[atomic zenith]

35 is ok?

[prisma moss]

they are disabled anyway

that is if they are refering to tiles, just leave it

[atomic zenith]

ok what about the setting that says cpu gpu or nv?

[prisma moss]

cpu

[atomic zenith]

Will this prevent hires fix from freezing my system?

[ornate elk]

Make sure you have set hires steps to 10-15 Max and also if the resolution is to large then enable tiled VAE