🤝｜tech-support

[knotty turret]

Ok

[ornate elk]

Oh wait

Here:
#🤝｜tech-support message

[knotty turret]

ok

:D

that one ig

[ornate elk]

I thought I haven't send them in here yet but I did

All 3

Different purpose

[knotty turret]

i just wanna use img2vid for now

not text to vid

[ornate elk]

What's your GPU again

[knotty turret]

RX6600

[ornate elk]

Okay

Let me know if it works with 8gb vram

[knotty turret]

i think it works because dont remeber when but did something like this with my old 6gb vram

looks like need donwload some files

3 ig

[dreamy drum]

Hi! I'm working on a paper that I'm trying to generate synthetic airport runway images with stable diffusion. But I'm new to image generation models and I'm struggling on how to create good images. I'm using the diffusers library, and controlnet for a canny/mlsd mask.

Here's a sample code I'm using:

controlnet = ControlNetModel.from_pretrained(
"diffusers/controlnet-canny-sdxl-1.0",
torch_dtype=torch.float16,
use_safetensors=True
)
vae = AutoencoderKL.from_pretrained("madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch.float16, use_safetensors=True)
pipe = StableDiffusionXLControlNetPipeline.from_pretrained(
"stabilityai/stable-diffusion-xl-base-1.0",
controlnet=controlnet,
vae=vae,
torch_dtype=torch.float16,
use_safetensors=True
).to("cuda")
pipe.enable_model_cpu_offload()

prompt = "aerial view of a runway from airplane in landing phase, realistic high-quality photo, snow-covered mountains in the background"
negative_prompt = 'distorted, unrealistic, missing markings, cropped, worst quality, low quality, multiple, airplane'

controlnet_img = Image.open("datasets/final_canny_edges/EDDF07R1_4LDImage2.png")
width, height = controlnet_img.size
ratio = np.sqrt(1024. * 1024. / (width * height))
new_width, new_height = int(width * ratio), int(height * ratio)
controlnet_img = controlnet_img.resize((new_width, new_height))

pipe(
prompt,
negative_prompt=negative_prompt,
image=controlnet_img,
width=new_width,
height=new_height,

controlnet_conditioning_scale=1.0,
num_inference_steps=50,

).images[0]

---

Image 1 is the controlnet mask
Image 2 is the final generated image
Images 3-4 are the quality I'm trying to get at (a simulator / a real image)

---

Running locally on a RTX 4070 TI

---

If I generate multiple images, sometimes I can get a few good ones, but I'm just new to this and wanted to get more consistent quality. Sorry if this is not the right place to ask. King Regards!

[rigid zephyr]

Comfy UI

ERROR: Could not detect model type of Flux.

It's 100% downloaded and in the correct folder

[lucid yoke]

And whats the correct folder hm

And do you have the proper clip model?

[merry cairn]

anyone here who knows a bit about SimpleTuner ?

i get this error:

with this backend:

and i do have images here:

[maiden basin]

hey guys i havent been using sd for a while is 3.5 the best right now or still sd xl?

[rigid zephyr]

ive got other models in exactly the same folder which are working fine. it makes no sense at all

[lucid yoke]

No, it doesn't it depends on the model

Flux goes in diffusion-models

Same as video generation models, not in stable-diffusion

[rigid zephyr]

oh i see, ok thanks

trying now

[rigid zephyr]

now it's in there, how in the comfyui do I use flux?

[lucid yoke]

Your better off installing comfy manager and stealing a workflow from civitai

Since you need different nodes/clip iirc

[half island]

Is there a command that can install the correct dependency versions? Or do i need to correct these manually?

[ornate elk]

Which webui or tool do you try to install?

[ruby yoke]

this one, probably #🏞｜general-with-images message

[potent fog]

can anyone help me with linking models and everything from one webui to another

i rewrote it like this but it didnt transfer for some reason

[ornate elk]

You can copy the example from my guide.
And there only change the path to the webui

[potent fog]

i can? i thought the file location and name would be different

[half island]

Comfyui. As well as raw mmaudio, but it has it's own issues.

(venv) (base) D:\stable-audiovideo\MMAudio>python gradio_demo.py
Traceback (most recent call last):
  File "D:\stable-audiovideo\MMAudio\gradio_demo.py", line 12, in <module>
    from mmaudio.eval_utils import (ModelConfig, VideoInfo, all_model_cfg, generate, load_image,
  File "D:\stable-audiovideo\MMAudio\mmaudio\eval_utils.py", line 14, in <module>
    from mmaudio.model.networks import MMAudio
  File "D:\stable-audiovideo\MMAudio\mmaudio\model\networks.py", line 9, in <module>
    from mmaudio.ext.rotary_embeddings import compute_rope_rotations
ModuleNotFoundError: No module named 'mmaudio.ext'

[ornate elk]

You have to change one path

The one to your other webui location

[potent fog]

how do i do that

[half island]

@pure tuskAnother compromised account

[ornate elk]

Is your other webui on C:\SD-Zluda ?

[potent fog]

yep

a1111

[ornate elk]

Okay then you can copy the whole example

Without a change

[potent fog]

oh really?

like this?

also should i add --medvram-sdxl and --opt-sub-quad-attention to forge?

[ornate elk]

Oh wait nope

[half island]

Tbh, best to use forge though. Auto1111, but doesn't suck when checkpoint hopping

[ornate elk]

Forge doesnt need the args

[potent fog]

which one?

[ornate elk]

?

[potent fog]

like what do you mean by this

[vocal burrow]

those ones

[potent fog]

ohh the commands

ight yeah my bad

[potent fog]

didnt work

ill just manually switch

[ornate elk]

What does the cmd shows?

[potent fog]

like when it runs or the webui-user?

[vocal burrow]

what's the problem ?

[potent fog]

models and loras not transferring from a1111 to forge

[vocal burrow]

and that's your webui-user.bat ?

[potent fog]

thats when it runs

yeah basically

[vocal burrow]

yup that's normal

[potent fog]

ill just transfer manually

gaeunCOPIUM

[vocal burrow]

nvmd it should work

What's in d:\Programme\AI-Zeug\stable-diffusion-webui-amdgpu ?

[potent fog]

i copied off the example on CS1o's guide

[vocal burrow]

ok but what's in this folder ?

[potent fog]

i did also do this which is the name of the file

[ornate elk]

Oh wait this folder is not like the example

Sry

XD

[potent fog]

[ornate elk]

You have to change it

[vocal burrow]

welp there you go :p

[potent fog]

i did but it doesnt work

[ornate elk]

To C:\SD-Zluda\stable-diffusion-webui-amdgpu

Then it should work

[potent fog]

ight one more time

brah

it aint workin

[ornate elk]

Can you show your webui-user.bat?

[ornate elk]

Oh you have to change the \ to / in the path

[potent fog]

ohhhh

damn

sharp eyes

it worked!!

thanks yall

AstolfoSleep

[ornate elk]

nice np

[potent fog]

one more thing

where do i download controlnet models btw

[lucid yoke]

Huggingface, civitai etc

Depends on the model your using, like illustrious has a different controlnet then like sdxl or flux does

[potent fog]

im using xl

[lucid yoke]

Then for your intents and purposes i think civitai.com or .green (green is sfw) > models > SDXL > controlnet filter

After model its all in filter

[potent fog]

damn so many accounts confiscated

[potent fog]

i see thank you

[lucid yoke]

People keep clicking on links they don't know. Or keep reusing the same password

[knotty turret]

@ornate elk do you know how i can add reference image here?

its possible?

[ornate elk]

i think so but im not a huge comfyui user

check the img2img example and put that together with this workflow:
https://comfyanonymous.github.io/ComfyUI_examples/img2img/

The Comfyui Example page is a good resource to start with basic workflows

[knotty turret]

hmm ok

[knotty turret]

trying to figure out now how i gonna use the empty latent image

because both need to connect latent but only 1 can be used

asking this in comfyui chat too maybe someone know

[potent fog]

should i put this to the max

i keep getting this message is this normal

[formal ridge]

IIRC @ornate elk recommended that be set to 7800, and ignore the warning

[potent fog]

alright thanks!

[slate kayak]

Question i was big into generations then life happened, and I have amd lol. It's there any update for the 7900xt for generations? Last I tried it was horrible and since then my Nvidia laptop crashed

[ornate elk]

hey, yes there are is zluda which provides a good performance with AMD on Windows.
You'll find the Install Guides here:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides

Go with Forge webui with Zluda

Im on a 7900XTX, feel free to ask if any step isnt clear

[slate kayak]

I will look into this tonight. I remember when comfy came out but didn't support amd at all lol.

[ornate elk]

true, there is also a comfyui guide with zluda

but i still recommend FOrge or Auto for AMD Users, as you can better handle the vram usage

[slate kayak]

Ok nice

[ornate elk]

but also with your gpu you can use comfyui if you want

you can also install multiple webuis and link the models between them to safe space

[slate kayak]

Hmm. Ok sounds good. What kinda performance you getting

[ornate elk]

depends on model, resolution etc

[slate kayak]

Let's go just sdxl at 1080

I'm sure they are better models lol

[ornate elk]

1.5 models take (25it/s) and like 2 seconds per image on 512x512
SDXL models take (3it/s) like 11 seconds per image on 1024x1024

[slate kayak]

Ya that's lots faster than the 1it every 2 seconds lol

[ornate elk]

yep, directml is just super slow against zluda

and you also wont have these annoying out of vram errors every time

[slate kayak]

Amazing

[ornate elk]

WQHD Pixelart Wallpaper generated in 50 seconds

Illustrious based model + pixelart lora + hires fix

[potent fog]

forge is crashing my laptop for some reason

[ornate elk]

When?

And with wich Settings

[potent fog]

i dont know

is there any way to check why

[ornate elk]

Did you used hires fix or not?

[potent fog]

nope

normal settings

with adetailer thats it

[ornate elk]

Hmm which resolution? And model

[formal ridge]

The only thing you changed was the 7152 to 7800 right? Were you able to use it without it crashing after you changed that setting?

[potent fog]

yeah i think after i changed it it crashes

[formal ridge]

If you want it to stop yelling at you, the nominal value is 6464, but remember 7152 in case 6464 doesn't work (out of memory errors)

[potent fog]

thank you

it refreshes to 7152 when i run it anw so no worries

[tepid hearth]

Access Denied
Couldn't launch python
exit code: 1

keeps appearing

It worked the first time as Sharkie told me but I opened the webui the 2nd time and it still does that

[slate kayak]

@ornate elk hey quick question. how do i know what amd hip to use?

[lament temple]

Hi guys👋
Please help me to solve a problem in Stable Warp Fusion.
I will detail the details below:

1. Need to make a picture similar to the one at number 1. That is, need the same style, clarity, drawing realistic-anime, cartoon, etc.
2. I get a picture number 2.

I used a lot of models such as revAnimated_v2Rebirth, realisticVisionV60B1_v60B1VAErealisticVisionV60B1_v60B1VAE, realisticFantasy_v20, juggernaut_reborn, faetastic_Version2, dreamshaper_8, Anime_style.
Also tested different settings, strength, strength in promts, tried lors, samplers, etc., but the result did not change or was worse and not at all like what I need.
The picture remains dark, fuzzy, poor quality, smoky, well you can see everything yourself.
Picture number 1(frame from the video), was made on the model Revanimated(most likely).
Also the background is not drawn (I need an abandoned building with mirrors on the walls), but in the first picture the background is drawn well, although the original video is just white empty walls and floor.
Under number 3 will be the original picture from the video (original).

[ornate elk]

With a 7900xt you need the latest one which is 6.2.4

[ornate elk]

Can you show the cmd log?

[slate kayak]

i wont have to redo anything to continue gaming right rocm should be golden?

[ornate elk]

Hip SDK is only used for ai stuff.
Gaming needs the latest AMD Adrenalin driver

[slate kayak]

can i have both?

[ornate elk]

Yep

[slate kayak]

side question. is there any way i can expand my partition drive size without reformating the whole damn computer?

[ornate elk]

In the drive manager you can shrink a partition and then add the free space to the other one.
But I think the data musst be moved or saved before doing this.

[slate kayak]

ok ill check it out. im an idiot not thinking of updates and partitioned my main where windows is too small haha

[ornate elk]

Oh damn xD for real I dont recommended partitioning at all if your not planning on dual booting linux

[slate kayak]

im on windown haha

[ornate elk]

Only maybe if the main drive is to big

[slate kayak]

i have 2 1tb drives

[ornate elk]

Yea then no need for partitions

Better backup the drive before trying stuff like shrink or reformatting it

[ornate elk]

Hey, this is the tech-support channel. Your question goes more onto the actual usage side.
So can you move your question to #📝｜prompting-help ?

[lament temple]

Oh, thank you

[ornate elk]

No problem 🙂

@slate kayak I'm off for now, let me know how it goes or if you have any question about the install steps. Will take a look tomorrow

[copper horizon]

ive been trying to get any version of SD working with my rx7700 XT for like 18 hours.

im starting to go insane

please anyone just spoonfeed me

how the fuck

do i do it

if i see ModuleNotFoundError: No module named 'torch' one more time

im gonna flip

[copper horizon]

ive tried everything

the furthest ive gotten is i generated one image on some repo

and then every other time i get some exception and my entire pc crashes

[formal ridge]

I can't spoonfeed you as I don't have an AMD GPU but from what my Google Fu brings up, they recommend Linux over Windows for AMD GPUs.

[ruby yoke]

did you happen to read through all of the guides pinned in this channel?

[tepid hearth]

I'll send it tomorrow 👍

[slate kayak]

any idea y im stuck a this point?

To create a public link, set share=True in launch().
Startup time: 13.8s (prepare environment: 1.2s, launcher: 0.4s, import torch: 6.5s, initialize shared: 0.8s, load scripts: 1.7s, create ui: 2.0s, gradio launch: 1.1s).
Environment vars changed: {'stream': False, 'inference_memory': 1024.0, 'pin_shared_memory': False}
[GPU Setting] You will use 95.00% GPU memory (19440.00 MB) to load weights, and use 5.00% GPU memory (1024.00 MB) to do matrix computation.
Loading Model: {'checkpoint_info': {'filename': 'E:\AI\stable-diffusion-webui-amdgpu-forge\models\Stable-diffusion\babes_123p.safetensors', 'hash': '5bc15c26'}, 'additional_modules': ['E:\AI\stable-diffusion-webui-amdgpu-forge\models\VAE\vaeFinalPruneVAE_v10.pt'], 'unet_storage_dtype': None}
[Unload] Trying to free all memory for cuda:0 with 0 models keep loaded ... Done.
Global Step: 274189
StateDict Keys: {'unet': 1680, 'vae': 304, 'text_encoder': 197, 'text_encoder_2': 518, 'ignore': 0}
Working with z of shape (1, 4, 32, 32) = 4096 dimensions.
K-Model Created: {'storage_dtype': torch.float16, 'computation_dtype': torch.float16}
Model loaded in 1.1s (unload existing model: 0.2s, forge model load: 0.9s).
[Unload] Trying to free 3051.58 MB for cuda:0 with 0 models keep loaded ... Done.
[Memory Management] Target: JointTextEncoder, Free GPU: 17576.02 MB, Model Require: 1559.68 MB, Previously Loaded: 0.00 MB, Inference Require: 1024.00 MB, Remaining: 14992.35 MB, All loaded to GPU.
Moving model(s) has taken 0.78 seconds
[Unload] Trying to free 1024.00 MB for cuda:0 with 1 models keep loaded ... Current free memory is 15797.60 MB ... Done.
[Unload] Trying to free 7390.16 MB for cuda:0 with 0 models keep loaded ... Current free memory is 15798.92 MB ... Done.
[Memory Management] Target: KModel, Free GPU: 15798.92 MB, Model Require: 4897.05 MB, Previously Loaded: 0.00 MB, Inference Require: 1024.00 MB, Remaining: 9877.87 MB, All loaded to GPU.
Moving model(s) has taken 2.46 seconds
0%| | 0/20 [00:00<?, ?it/s]Compilation is in progress. Please wait...

[copper horizon]

ive tried every AMD Zluda tutorial in the pinned comment

no success

trying the forge one again for fun

(it wont work)

[ruby yoke]

you ahve an AMD cpu or gpu?

[copper horizon]

both

[ruby yoke]

okay, disable your integrated gpu on that AMD cpu chip

[copper horizon]

yeah i have in bios

[ruby yoke]

maybe vast.ai is a better option for you then

[copper horizon]

ill take a look

ok actually

this forge install might be working

im gonna kiss CS1o if this worksa

gpu coil whining

fans spinning

compiling

well ill be damned. it works.

[slate kayak]

lol

im stuck with it on compiling is in grogress

[tardy oyster]

0.0

[tepid hearth]

@ornate elk There it is

[ornate elk]

Compiling is good. That means it will most likely work if its done

[ornate elk]

Can you open up a CMD and type
Python --version
And then
Where python

[tepid hearth]

I'm not on my PC rn

but python is in 'Folders'

[ornate elk]

Ah okay. Then uninstall python and install Python 3.10.11 64bit into its default place. You should not install programms onto desktop

[tepid hearth]

Ohh okay

I'll try, thanks man

[granite spade]

how can i convert a video model to gguf?

specifically wan2.1

i want to convert the 1.3b version

[restive sparrow]

In what way does ram affect generation, would more gb help you?
Only thing I know of benefit is the [store multiple models] in ram feature. (especially with these pony/xl models that are 6gb ea) ty

[ornate elk]

If you have 32gb ram your fine.
If you only have 16gb ram or less, increasing the Windows Pagefile or adding more ram can increase the speed on sdxl and bigger models

[lament temple]

Hi guys👋
Please help me to solve a problem in Stable Warp Fusion.
I will detail the details below:
Need to make a picture similar to the one at number 1. That is, need the same style, clarity, drawing realistic-anime, cartoon, etc.
I get a picture number 2.

I used a lot of models such as revAnimated_v2Rebirth, realisticVisionV60B1_v60B1VAErealisticVisionV60B1_v60B1VAE, realisticFantasy_v20, juggernaut_reborn, faetastic_Version2, dreamshaper_8, Anime_style.
Also tested different settings, strength, strength in promts, tried lors, samplers, etc., but the result did not change or was worse and not at all like what I need.
The picture remains dark, fuzzy, poor quality, smoky, well you can see everything yourself.
Picture number 1(frame from the video), was made on the model Revanimated(most likely).
Also the background is not drawn (I need an abandoned building with mirrors on the walls), but in the first picture the background is drawn well, although the original video is just white empty walls and floor.
Under number 3 will be the original picture from the video (original).
Also when loading there is an error on the last picture, but despite this the service still loads and works.

[frigid pawn]

How can i use different models using automatic1111 text-to-img enpooint. current when i set any model on webui the text-to-img endpoint use that model only, how can i will pass the model into text-to-img endpoint request

[ornate elk]

Use --ckpt-dir "D:\PathToModels..."
So it has access to all the models in that folder

[slate kayak]

I let it run overnight and it did nothing on 1 image 20 steps for a test.

I unfortunately can't do anything else atm till tonight

[ornate elk]

Oh okay restart it and try generate an image again.
If it doesn't work after compiling is done, let me know + the cmd log

[slate kayak]

I will when I get home tonight. I tried restarting multiple times. I think I messed up on the install step when you copy the library

[ornate elk]

If thats the case we can see it in the cmd log

[slate kayak]

I will grab it tonight .

[slate kayak]

This was the 1st log last night

[upbeat atlas]

Hi, dumb question but I gotta ask. Is there anyway to run stable diffusion on Android locally?

[formal ridge]

You can probably run an app on your phone that will link back to a server you have somewhere, but if you mean entirely locally, no. An Android Phone will never have enough power for this.

[upbeat atlas]

Oh so it's not possible to do it locally, thank you for the help

[formal ridge]

I've seen people run Stable Diffusion entirely on their CPU, no GPU involved. So it might be possible, but the time needed would be astronomical.

And you're talking a lot of custom coding as well to run it on Android

[upbeat atlas]

Yeah no there's no way I can do that xD
Btw, is zluda still the fastest way to run stable diffusion on AMD GPU? It's been quite a while I've been wondering if they got it to run faster yet

[formal ridge]

I have nVidia. I can't answer that, sorry.

That is definitely a @ornate elk question.

[ornate elk]

there is a faster method on windows by running WSL2 with Docker Desktop, you would have to install Ubuntu with WSL and setup stable diffusion inside a docker container.
But WSL2 does not support 6000 or 5000gen cards.
its a bit faster than zluda but has less memory improvements and is not as easy to setup.
So the best for usage/performance on Windows is still with Zluda

[formal ridge]

That's interesting that a virtual machine is faster than Windows native.

[ornate elk]

The WSL2 can directly use the full ROCm stuff the same way as on linux, in comparisson zluda just translates cuda stuff to a half ready windows rocm version

[formal ridge]

So it's removing the slowest "translation layer" (I can't think of a better word) and putting a different one somewhere else. 🤔

[ornate elk]

ROCm is AMDs equivalent to Nvidias Cuda,
on Linux you can run the most ai stuff nativly with AMD cards as ROCm was made for Linux.
AMD now tries to give Windows a full ROCm support too. But thats not there yet so we need Zluda to translate cuda stuff to AMD or runn a Linux on Windows to use the normal ROCm

[warm magnet]

Hey y'all, is there a tool (or tools out there) for CLIP interrogating a local model? Reverse image prompts? etc?

[formal ridge]

I don't know if it still works but you used to be able to do this easily in A1111 or Forge

[warm magnet]

I'll give it a shot

[ornate elk]

these two buttons in img2img can do that

clip and deepbooru

[fathom summit]

I am trying to download Forge UI. This is what happens when I launch Run.bat.

I deleted all the other weird variables I found. I think this is how it's supposed to look like?

I just deleted both of Nodejs and Git because of this.

[formal ridge]

"deleted all the other weird variables I found" Well, I hope you have a backup of your OS somewhere.

[ornate elk]

hey, can you show the full cmd log?

[fathom summit]

But they were all coding or programming programs.

[fathom summit]

Isn't this all of it?

oh

Was I maybe supposed to restart the laptop after deleting them? I haven't done that.

[formal ridge]

If we never see MonMon again it's because his her laptop is dead after deleting them variables.

[fathom summit]

I am a girl.

Bruh. No, I just got this laptop.

Waaa.

[ornate elk]

can you download Python 3.10.11 64bit and install it? then relaunch

[fathom summit]

Huh, I actually just deleted Python because I thought it could be the cause of this.

But it was a more recent version.

[ornate elk]

its needed to run the webui xD

[fathom summit]

I remember there was something I downloaded and I enabled something that was related to PATH in that install menu? I don't remember which program that was.

[ornate elk]

okay good, please install only 3.10.11 64bit

[fathom summit]

Lol, I didn't know this.

[fathom summit]

Mhm, I'm doing this.

[fathom summit]

Please tell me what to choose, since I've become traumatized because of- whatever this is. 🥰

The last I downloaded it, I didn't install it on Disk C, is that okay?

[ornate elk]

check "add python to path" at the bottom
then click install now

keep everything else at default

[ornate elk]

it should be installed on C at its default path

[fathom summit]

Okay!

[fathom summit]

Should I restart?

[fathom summit]

update.bat

[fathom summit]

run.bat

Should I cry yet?

[ornate elk]

you need to install Git too

[fathom summit]

Any version of it?

[ornate elk]

the latest one as 64bit

[fathom summit]

Just default, right?

[fathom summit]

?

[ornate elk]

Yep defautl

[fathom summit]

Um?

Notepad?

[tepid hearth]

@ornate elk, it didn't work

There is Python

[fathom summit]

update.bat

run.bat

[ornate elk]

Dont run the bats as admin

[tepid hearth]

I didn't run it as admin

[ornate elk]

Okay delete forge and redownload it again

[fathom summit]

hmmm, that reminds me.......of this.

winrar

I forgot that ever happened.

[ornate elk]

Dont place the webui on Desktop

It has to be in a new created folder on a drive

Like C:\Ai\ for example

[tepid hearth]

Ohh

Thanks for much, sorry for so little

[ornate elk]

Np, for a complete install guide checkout the link in the pinned messages

[ornate elk]

Have you redownloaded it?

[fathom summit]

I only extracted it again.

Is the zip file only 1.74 gb for you?

Look at this. This is new.

[ornate elk]

Looks good

[fathom summit]

I changed the location of the folder to D:\Forge UI.

installing requirements.

Maybe the & could do this? Because it was in D:\AI Art & Pictures before I changed its location.

Just maybe.

[ornate elk]

Maybe. Folder names should not contain special characters

It can break the Path

[fathom summit]

I think it's working, yeah?

[ornate elk]

Yep its now downloading a starter model to generate images

[fathom summit]

Thank you!

[ornate elk]

Np, have fun!

[idle mica]

somone can help me? im using a 5070 ti and try gen one img and this happen 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.

[slate kayak]

i got it to work. the model i was using took forever to do its shit.. any suggestion on good models for fantasy

is there any way to free GPU memory up? its saying i have not enough haha. and its a 20gb card

[idle mica]

@ornate elk sorry for tag you,do you know anything about this?

[ornate elk]

You have to use Forge or ReForge webui for rtx50xx gpus

[ornate elk]

Which settings did you used? Resolution etc

And model

[main nymph]

question... do you think i overtrained (finetuned) my model? i just learned how to get this whole training thing off the ground after studying and experimenting for about a whole month and i finally got it working. but apparently theirs a thing as overtraining a model (overfitting)
im training a SDXL realistic human and i have a dataset of 71. the final epoch i have set is 100 and it seems its 36 steps per epoch. im currently at 62 epochs. is that overkill? i have some backup save points as early as 30 epochs and going in intervals of 5 up to 60. chat gpt said that only 5 or so epochs

[formal ridge]

To be honest, your dataset is extremely low for model training. Try a Lora instead (unless that's what you meant instead of model)

[main nymph]

i meant finetuning a safetensor.

[noble idol]

sounds like a lora

small dataset and it being for a specific concept and all

just use the model you made to find out if its 'overtrained' or not, thats what i do

[frigid pawn]

@rocky ore how can i will generate male nude images with sdxl juggernaut. suggest lora model for male nudity.

[noble idol]

typing 'competely nude' is usually all you need

[spice perch]

hey yall. i dont know where to ask this bc im a complete noob, basically trying to generate a video w/controlnet via some code and it's just giving me this. the same exact code worked decently before (2nd pic. sry low quality), so i'm stumped. can anyone just point in the direction of the root of the problem here.. could it be low resolution/non-standard aspect ratio or something?

[lament temple]

Hi, can you please advise me which promt to use to get this kind of picture and image in Stable Warp Fusion?

[noble idol]

#📝｜prompting-help help would be the better channel for this 🙏

[solemn girder]

Checkpoint keeps processing forever and idk why

[lucid yoke]

thats a bit vague, what UI are you using?

Can you post the CMD log?

[solemn girder]

v1.10.1

[lucid yoke]

❓ what

Automatic1111 ill just assume then

[solemn girder]

Creating model from config: R:\Git\stable-diffusion-webui\repositories\generative-models\configs\inference\sd_xl_base.yaml
R:\Git\stable-diffusion-webui\venv\lib\site-packages\huggingface_hub\file_download.py:1132: FutureWarning: resume_download is deprecated and will be removed in version 1.0.0. Downloads always resume when possible. If you want to force a new download, use force_download=True.

[lucid yoke]

what model are you trying to use

[solemn girder]

Yeah

[lucid yoke]

hmm

[solemn girder]

None of the models work

[lucid yoke]

well A1111 is probably the issue

[ornate elk]

can you show the full cmd log?

[solemn girder]

Already up to date.
venv "R:\Git\stable-diffusion-webui\venv\Scripts\Python.exe"
Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug 1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
Version: v1.10.1
Commit hash: 82a973c04367123ae98bd9abdf80d9eda9b910e2
Launching Web UI with arguments: --autolaunch --xformers --medvram-sdxl --no-half-vae
Loading weights [3c624bf23a] from R:\Git\stable-diffusion-webui\models\Stable-diffusion\sdxlYamersAnimeUltra_ysAnimeV4.safetensors

To create a public link, set share=True in launch().
Creating model from config: R:\Git\stable-diffusion-webui\repositories\generative-models\configs\inference\sd_xl_base.yaml
R:\Git\stable-diffusion-webui\venv\lib\site-packages\huggingface_hub\file_download.py:1132: FutureWarning: resume_download is deprecated and will be removed in version 1.0.0. Downloads always resume when possible. If you want to force a new download, use force_download=True.
warnings.warn(
Startup time: 22.4s (prepare environment: 4.4s, import torch: 7.6s, import gradio: 1.8s, setup paths: 2.0s, initialize shared: 0.4s, other imports: 1.3s, list SD models: 0.7s, load scripts: 2.3s, create ui: 0.7s, gradio launch: 1.1s).
Loading weights [96dd2f60f3] from R:\Git\stable-diffusion-webui\models\Stable-diffusion\ponymatureSDXL_ponyeclipse10.safetensors
Creating model from config: R:\Git\stable-diffusion-webui\repositories\generative-models\configs\inference\sd_xl_base.yaml

[ornate elk]

thats not the full cmd log

[solemn girder]

Now it is except for the self dox

[lucid yoke]

a local host IP isnt a dox tho

[solemn girder]

ah

[ornate elk]

how much RAM do you have?

[lucid yoke]

^ and what GPU basically

[solemn girder]

64,0 GB

I had an increase a few weeks ago

[ornate elk]

what does the webui show (in browser) when you try to load the model?

[solemn girder]

Just this really

[ornate elk]

are the models stored on an internal drive or external drive,
also hdd or ssd?

[solemn girder]

Hdd

Internal

[ornate elk]

okay, whats your gpu?

and make sure the webui is whitelisted in any browser adblocker and dont use a vpn

[solemn girder]

NVIDIA GeForce GTX 1080

[ornate elk]

okay i have the same in my old pc.
when did you installed auto1111 ? today?

[solemn girder]

Idr when I installed it but a while ago, updated it last year maybe idr?

[solemn girder]

It finished processing

[ornate elk]

okay try to generate

[solemn girder]

It spat out a billion variants of
While copying the parameter named "first_stage_model.decoder.up.3.block.2.norm2.bias", whose dimensions in the model are torch.Size([512]) and whose dimensions in the checkpoint are torch.Size([512]), an exception occurred : ('Cannot copy out of meta tensor; no data!',).

[ornate elk]

okay then you can try to delete the venv folder and let it recreate it at relaunch

[solemn girder]

on it

[solemn girder]

everythings re-downloaded now

It generates now again, just doesnt conform to the loras very well

[ornate elk]

that depends on the model you use

[solemn girder]

Model matches te lora

[ornate elk]

yea but not every pony model works good with every pony lora. also try to adjust the lora strenght

[solemn girder]

hmm

[idle mica]

i was using forge

[solemn girder]

Took a while to gen but it worked, thank you

[ornate elk]

try run the update.bat and then delete the venv folder of the webui folder and then relaunch

[random night]

help, i've downloaded the openpose editer via extension tab, reload the ui but the openpose tab didn't appear

[ashen gate]

Trying to find supported resolutions for Forge's Flux model. Can't find the list.

[ornate elk]

the openpose editor is old and needs you to manually install basicsr into the venv of the webui to work again, or you use the build in openpose editor that comes with controlnet

[random night]

this one?

[ornate elk]

yes this one

[random night]

thanks man

[ornate elk]

this extension has an inbuild openpose editor

[random night]

yeah, i have it, but can't find it anywhere

[ornate elk]

drop an image into controlnet then you should see the edit button

[random night]

still didn't see it

[ornate elk]

oh sry, click on the explosion icon

then on the right of the skeleton click edit

[ornate elk]

preprocessor has to be openpose

or dw openpose

[random night]

oh, it works

thank you

[ornate elk]

no problem

[shut tundra]

Guys I could really use some help. I'm attempting to install sageattention. I have followed all the steps to install triton. CL works, I install python -m pip install triton wheel, and it still says no module named triton found.

[idle mica]

dindt work

[ornate elk]

Can you show the full cmd log?

[shut tundra]

yo @ornate elk it's been a while :)

do you think you could help me out with my sageattention installation? im strugglin :(

[ornate elk]

hey, which node did you try to install?

[shut tundra]

I don't think I tried to install a node, I'm trying to install sageattention to comfyui.

[ornate elk]

cant help much with comfyui, but it seems that the sageattention node is a bit outdated and requires older packages than you have in comfyui

[ornate elk]

that wont work.
Your only way would be to use this:
https://github.com/logtd/ComfyUI-SEGAttention

[shut tundra]

headscratch

[lucid yoke]

while i think the transformers being incompatible doesnt help im not familliar with sage attention.
i think theres a matrix channel for comfyUI iirc

[shut tundra]

what's matrix

[lucid yoke]

oh nvm they also got a discord now

[ornate elk]

we also have #🧣｜comfy-ui

[lucid yoke]

https://github.com/comfyanonymous/ComfyUI
on the github theres a link for it

youd get better results there tbh as this discord isnt filled to the brim with comfy users

[ornate elk]

you have the wrong python version installed.
3.13 is not compatible with ai stuff

uninstall it and install python 3.10.11 64bit
then delete the venv folder from the webui folder again and relaunch

[idle mica]

thank youuuuu

[noble vine]

how do i use inpainting in swarmui?

the outputs i get either leaves the image untouched, greys out where the mask is, or blacks out the full image

[harsh wasp]

Hello people
does anyone know which one should i download wan image2video ?
my specs :
core i5 12th
16 ram (soon it will be 32)
3060ti 8 vram

[ornate elk]

Q3 the 7.93gb one, but i doubt that it will work good

[harsh wasp]

ok what if its 4070 12gb ?

[ornate elk]

Q4 gguf (10gb)

[harsh wasp]

even with CPU usage ?

[ornate elk]

cpu usage would take hours

[harsh wasp]

ok if it will take long its fine
but if it will show BSOD then no

which one ?

[ornate elk]

idk you would have to try

[lucid yoke]

Hi, you did use the edit image page right

What model are you using?

[noble vine]

yes, i use the edit image page and im using pony diffusion

[misty marten]

@ornate elk Is there any way to generate an image with 2 LoRA characters without using comfy-ui?

[ornate elk]

Yep in auto1111 You need the Regional prompter extension.
And in Forge you need the forge couple extension

[sick sparrow]

hi chat, iy said python not found when i clicked the .bat file and i saw fixes saying that i need to ADD PATH

any1 know how i can do this?

oh wait im retarded

sorry

[shut geyser]

installing zluda for automatic 111` with zluda

first time launching

[ornate elk]

Can you relaunch and show the full cmd log?

[shut geyser]

\

i installed the right version of python now

[ornate elk]

Then delete the venv and relaunch

[shut geyser]

worked well

[broken goblet]

Hello! Does anybody here use Forge UI? I have a problem and is that I can't change the VAE when I use a SDXL model, the VAE option only appears when I set it to stable diffusion

[ornate elk]

Hey, is your forge updated?

[broken goblet]

I downloaded it a few days ago from github, so I guess it's the last version? How do I update it anyway?

[solid nova]

Hey, I am running into this weird error when I try to execute this file. I know I have permission to the model as it says that i do on the platform but I am getting an error on the install. Can you someone help please?
I got this error:
TypeError: expected str, bytes or os.PathLike object, not NoneType

[thin reef]

Just wondering, how fast does an RTX 4050 with 6GB VRAM generate images on SDXL? Let's say 512x720 size or sumn

Using a PonyXL model rn and it's taking like 2-3 minutes for that size on 30 steps if that's even normal

[formal ridge]

NGL, that 6GB of VRAM hurts.

[thin reef]

Welp, it's on a laptop

ChatGPT says the speed is unusually slow and it might be because it's not utilizing the CUDA and using the CPU instead

But idk

[formal ridge]

If you're using A1111 instead of Forge or ComfyUI that will also lead to slower gens than most of us.

[thin reef]

Ah yeah Cakesu is using A1111

Will try looking into ComfyUI

[ornate elk]

Run the update.bat

[ornate elk]

Also you have a flux model selected. Not an sdxl model

[broken goblet]

Thank you! I updated it and now the VAE option shows, I have a problem with the Flux model but I'm not sure if I can ask since it's a stable difussion server

[harsh wasp]

@ornate elk hello again my friend
after searching it appears these tiny model can help
which one should i precessed ?

[noble vine]

hey yall, im planning to use onetrainer cuz i heard its good but i did a test run and there's no file extension on the exported file

its suppose to be a lora, but changing the file to add the .safetensors extension didnt work

[lucid yoke]

Iirc KoyhaSS was better. Were you following a guide or?

[noble vine]

yea

also i cant get the requirements to install on my pc for some odd reason

[lucid yoke]

So you tested it without all the requirements?

[noble vine]

no, i was refering to koyhass

ot's github page said that there was no pre-requirements

[restive sparrow]

on an upcoming build, if 64 is feasible price thinks its worth just for ai generation? cheers

[formal ridge]

AI Gen? Maybe, but you'll definitely need 64 for certain games coming out if you're future proofing.

[ornate elk]

You can ask about it here too

[ornate elk]

Fp16

[harsh wasp]

finally the luck helped me out xd

ok where exactly i should place the model ?

[ornate elk]

Sure if you also have a GPU with large vram amount.
If you only have 8-16gb vram 32gb ram would be enough too.
But with 64gb You'll never get a problem anywhere

[ornate elk]

In models/checkpoint
If it doesn't appear then models/unet

[broken goblet]

Well I updated Forge and now the option finally appears but it still doesn't work, this is all I'm using

I get this error

I put ae.safetensors in the VAE folder and the rest in the text_encoder folder

[ornate elk]

What's your GPU?

[broken goblet]

gtx 1660 super

and 16gb RAM

[lucid yoke]

I recommend sdxl personally at most

[ornate elk]

Then forget flux with that GPU :/

[lucid yoke]

My 3070TI had long load times with flux and oom a few times

[broken goblet]

Aww, is there a way to make sdxl generate better text?

[harsh wasp]

ok as per the websites it can only be run usning comfy
i am already loading the model but not sure which one is already selected (is it text to video or image to video)

can you please guide me on this ?

[ornate elk]

Check this out:
https://comfyanonymous.github.io/ComfyUI_examples/wan/

There you get the example workflow for wan

[harsh wasp]

much appreciated , will read them

thx again my friend

[lucid yoke]

Controlnet + photoshop

[celest hound]

Thank you

[harsh wasp]

just want to confirm , as per the guide , the ffeatuer image to video reqires the heave models ONLY , meaning the other model (2.6GB) model wont work , is that correct ?

Or i can run it normally using the FB model ?

[noble vine]

fixed it, just had to add .safetensors in the output file name

[ornate elk]

The smaller should work too

But you have to test

[harsh wasp]

i am testing the text and as predicted its taking time

[harsh wasp]

should i download the workflow and just replace the model ?

[ornate elk]

You can try that yes

[harsh wasp]

nice andi have the pics to try
let me give it a try after this test

Amazing !!

later i can learn how to increase the length and the quality
let me try the image to video

i cant change the model , do i need to edit the json file and paste the name of the model i want ?

[formal ridge]

You can't change the model if you only have one model.

[ornate elk]

Place it in the folder the example says

Models/diffusion_models

Then you have to reload the ui

[harsh wasp]

i just renamed the name of the model from the json file and its working

hoping to see some good results

it did provide a realistic result however not what i wanted 😆

this is the pic i uploaded

if you guys can provide or send link for quick guide for comfy it will be appreaited , and whats the difference between these 2

[formal ridge]

https://blenderneko.github.io/ComfyUI-docs/

[lucid yoke]

Spam dont join chat

Gettin real tired of these scams tbh

[next nexus]

Hi everyone, I wanted to ask if it was still possible to use Google Colab for free like a few years ago? I can't find anything about this other than users banned for violating Colab's terms.

[random night]

how the fuck can i be able to disable controlnet integrate in forge? i hate it so much

[harsh wasp]

Thanks for your help
just wanted to ask how to increase the model efficiency , it generates video fast so either the settings are light or the model very light

should i download any other model ? my vram is 8

[cobalt wing]

Hi I have a problem with comfyUI. I've been trying to run WAN on m 4060TI 16G, and I heard I could use the quantized models from Sebastian Kemph and Aitrepreneur.
So I downloaded wan2.1-i2v-14b-480p-Q5_K_S.gguf and wan2.1-t2v-14b-Q5_K_S.gguf to \ComfyUI_windows_portable\ComfyUI\models\unet
The problem is, there is no unet folder in the ComfyUI left-side bar
I assume I either have a setting problem or it shouldn't be placed under unet?

[cobalt wing]

I guess it just doesn't support gguf

The diffuser models model is empt even tho both files are inside

M config file

[formal ridge]

No idea, I've never messed with video (yet). *cries in 8GB of VRAM*

[hybrid lintel]

can we have a tips&tricks channel for prompts inside here? i am creating all kinds of prompts for sd1.5 since years. i can recreate everything just by prompting and not using any loras. i once was on civitai, but they deleted my articles without letting me know. they even sent users to other places to get me banned there, just because i know how to prompt. i have my own place where you can check my already huge collection of more than 200 articles. but i want to share my knowledge with a bigger community.

https://board.world-of-hentai.to/threads/stapfschuhs-new-home-for-sharing-prompts.225092/

and check "general with images" and my uploads there for some freebies 😄

[formal ridge]

CivitAI really does have the worst moderation I've seen. You would not believe how many images of "that sort" I've had to report and nothing gets done.

[cold echo]

Hey! I need some help with lora training. Is this the right channel?

[formal ridge]

Hey @lucid yoke, you're up.

[lucid yoke]

Yeah but i use civitai and KoyhaSS

Style loras are easier tbh but some concepts are iffy

[lucid yoke]

I recommend fully asking the question you have so if i or someone else knows they will respond!

[cold echo]

I'm trying to train a LoRA model for bald hairstyle for women, to be used for img2img inpainting. The results are good for the most part, but it struggles to generate part of the face in case the face is covered partially with the hairs and to fill the background at times. I have tried training for majicMIX realistic and realistic vision (both have SD 1.5 as base). Is there any other model that I should try training thr lora for. Sdxl results weren't great compared to these two.

Let me know if some part doesn't make much sense :)

[turbid willow]

Hello, wondering if anyone could help me, total noob here using stability matrix. A lot of the models I download and install using stability matrix give me an error when I try to generate images with them.

"You do not have CLIP state dict!"

What do I do about this?

[ornate elk]

You mostly then downloaded flux models which in addition need 3 other files to work.
Clip t5 to Text Encoder
Clip l Encoder
ae.safetensors vae

[turbid willow]

Yeah, this seems to be a problem I run into with flux models.

Where do I get those files, and where do I put them?

[ornate elk]

Flux also works only with Forge Webui, Swarm/Comfyui
And your GPU should have 12gb or more vram

[turbid willow]

I believe I have 12gb vram.

I just installed comfyui.

havn't tried it yet

Where do I get those 3 files though that you said I needed and where do I put them?

[ornate elk]

Did you used forge or comfyui when you got the error?

[turbid willow]

No.

So maybe I wont get the error now that I have comfyui?

I'll try it

[ornate elk]

You will still get it

[turbid willow]

Oh lol

[ornate elk]

You need the files first but I asked which webui you used before

[turbid willow]

Looks like it was webui forge

[ornate elk]

Ok, then read this forge info.
It also links all the needed files and locations:
https://github.com/lllyasviel/stable-diffusion-webui-forge/discussions/1050

[turbid willow]

Thanks a ton! ❤️

[ornate elk]

You can ignore the base flux model as you already downloaded custom ones

You just need the 3 additional files

[turbid willow]

So the clip-I and the t5-xxl, and... the vae(raw float16)?

Thanks. ^_^

[ornate elk]

No problem 🙂

[turbid willow]

Just double checking, but I found the vae folder to put the vae into but I don't see a folder named "text_encoder", I'm assuming that I should just make a folder and name it than?

[thorny iron]

I have a pair of skis.
And I want to make decals for them, via a print shop I got nearby.
How would I make a collage using a particular set of pictures, using SD to kind of meld them all together in a way that makes them belong?
This sticker would be 110mm by 1860mm

[ornate elk]

Yep then create one in the models folder

[turbid willow]

Thanks, got it so far. Trying to figure out which t5xxl to downloadn now, there are 3 of those

[ornate elk]

Go for the fp8

[turbid willow]

the scaled or not scaled one?

I feel like I need a degree in computer stuff to understand all this stuff lol.

[ornate elk]

Idk, seems like the scaled one is newer

[turbid willow]

Yeah, I went with that one. Thanks again!

I bet that it still wont work cus that's my luck with complicated computer stuff. ^_^

[formal ridge]

if it doesn't, change the folder name to "clip", that usually works

[turbid willow]

Weeeeel, it still says I don't have clip. Rip. I even moved it into the clip folder.

I knew there was no way I could handle this XD

I think I might just need to touch grass >_<

[random night]

here the problem i ran into when using reforge with directml

[turbid willow]

I love that you have "cat' in ur prompt, that's always my go to first prompt when trying to test if a module works XD

[random night]

cheers, my fellow cat person

[formal ridge]

Have you set up your extra_model_paths.yaml file yet?

[turbid willow]

I only speak english, sorry >_>

(That means no and I have no idea how XD)

[formal ridge]

Wherever you installed the ComfyUI folder, there's a file with that name (although the actual .extension will be different). You edit that file to tell ComfyUI where your models, Loras, Embeddings, etc are.

[turbid willow]

I see a file called comfyui_version but it's in python so I can't open it.

I did a file search for anything with the word "comfy" in it, and that was the only thing that showed up.

[ornate elk]

Can you show the full cmd log?

[random night]

here

[random night]

i run it with stability matrix

[formal ridge]

if you're file searching, why didn't you search for "extra_model_paths"? :P

[turbid willow]

Cus I don't know anything XD

I fond a file with that name, but I have no idea what to do with it.

[formal ridge]

rename it to extra_model_paths.yaml (Is your Windows set to hide filename extensions?)

[turbid willow]

Rename it? That's already what it's name is though.

[formal ridge]

Okay, that tells me your version of Windows has extensions hidden. You'll have to change that. Give me a second.

[turbid willow]

See?

[formal ridge]

If you never renamed it, it's not working because you can't see the real filename extension.

So you'll need an open Windows Explorer (any folder will do) before you can mess with the settings to view extensions.

[turbid willow]

Got it, done.

[formal ridge]

Done as in you renamed it, or done as in you opened a folder?

[turbid willow]

I set the viewer to show extentions.

Trying to find the file again.

It's REALLY hard to find.

My search funciton took a solid minute to locate it lol

[formal ridge]

It's probably searching through the embedded python venv (if you're using the portable version of ComfyUI)

[formal ridge]

Okay, good. Now you edit that file in your favorite text editor telling it where you downloaded everything.

[turbid willow]

What do I change?

[formal ridge]

If you have other UIs installed (such as A1111) just edit the first section. The hashtags (#) tell it to ignore that line, delete that only on lines you modify.

[turbid willow]

There are no # in the whole file, so that's easy

But yeah, I have no idea what I need to edit inside this file.

[formal ridge]

As an example, this is from my file. I have all my stuff in one central location.

#     base_path: path/to/comfyui/
     checkpoints: C:/SD/Models/StableDiffusion
     clip: C:/SD/Models/CLIP
#     clip_vision: C:/SD/Models/StableDiffusion
     configs: C:/SD/Models/configs
#     controlnet: models/controlnet/
     embeddings: C:/SD/Models/Embeddings
     loras: C:/SD/Models/Lora
#     upscale_models: models/upscale_models/
     vae: C:/SD/Models/VAE

[turbid willow]

This is mine

Couldn't tell you what any of it means or does.

[formal ridge]

Do you see that folder that says CLIP? That's where you need to put the text encoder files you downloaded.

[turbid willow]

yup, that's where I put them.

models/clip

[formal ridge]

Did you make sure to relaunch the server after?

[turbid willow]

yup

Closed the whole thing down

Computers hate me, story of my life. This whole thing feels a like I;m reliving a thousand other times when people try to help me with a computer thing and it just doens't work XD

I'm not dumb but I think I'm just not smart enough for stability matrix maybe

[formal ridge]

I don't want to make you feel bad but Stability Matrix is the easy way of doing this. 😬

[turbid willow]

yeah lol, I know, that's why I went with stability lol. I can get some of the other models to work just fine, like dreamshaper, but if I try to do anything with flux, it just keeps tellingme I don't have the clip file.

[formal ridge]

Okay then, want to send me your workflow?

[turbid willow]

Wait a second... wtf... This doens't look right XD> I think I did somethign wrong lol

_>

Pretty sure the file should not be named that lol

[formal ridge]

That's just a placeholder. You need to move your files there.

[turbid willow]

That's so weird, I thought I did XD

wait.. no no.. I totally did, these were not here an hour ago.

What the heck happened lol

Yeah, here it is.

THere's TWO clip folders... hmm

[formal ridge]

StabilityMatrix should be symbolically linking the two, but it might not.

[turbid willow]

One was inside comfyui and the other was inside models

I put them in the clip folder inside the models folder.

I didn't put them in the clip folder that was inside the comfyui folder.

So maybe that's why

I have to go do some other stuff, but I'll fiddle with this more later. Thanks again for the help ❤️

[ornate elk]

Don't use SD 2.1

Its the only model version that needs .yaml files and also is very restricted

Go for sdxl if your GPU has 6-8gb or more vram

[plain kelp]

whic one is more accurate

[ornate elk]

juggernautxl or realvisxl

[autumn pendant]

i keep getting this error every time i try to load a model on automatic1111, "PermissionError: [WinError 5] Access is denied: 'C:\Users\cats2/.cache\huggingface\hub'"

[plain kelp]

guys how is possible to keep the same image or style ? from the promp no??

[ornate elk]

where is your automatic1111 stored?

[plain kelp]

a good upscale?

[ornate elk]

Depends on the image subject.
For realism try 4x UltraSharp

[plain kelp]

ok bro

[ornate elk]

In img2img you can generate images based on an input image

[plain kelp]

im on sb

[ornate elk]

Sb?

[plain kelp]

stability matrix

[autumn pendant]

my E drive

[turbid willow]

Thanks for the help before guys, I fixed my problem and got flux working. It takes ages to make a picture though. I have a 4070 GPU.

[turbid willow]

I have flux working now, but it takes forever to do some weird memory thing before it even begins making an image. Anyone know why this step takes so long? 104 seconds oof... Once it's done though and starts making the actual image, it only takes a few seconds.

Gpt is explaining it to me, disregard.

[buoyant mural]

Does anyone know if a old GPU with a lot of Vram would work well with Comfy ? like the "Nvidia Quadro M6000 "saw it on the used market

[formal ridge]

CUDA 5.2, most likely not unless you rolled a lot of your own code.

[autumn pendant]

would somebody mind helping me with my problem from before?: i keep getting this error every time i try to load a model on automatic1111, "PermissionError: [WinError 5] Access is denied: 'C:\Users\cats2/.cache\huggingface\hub'

[craggy orchid]

Hello. Has any developer successfully integrated the image generation function from an image with ControlNet, applying architecture to the web? Which curl command and model produce the most beautiful, sharp, and realistic images? Could you please share the curl command with me?

the quality should be sharp as this ( gen with midjourney)

[hybrid summit]

Hello there, I'm trying to send HTTP requests for text2image to a local model,
I'm using POST request to this address http://127.0.0.1:7860/sdapi/v1/txt2img
also i've set Content-Type application/json in Header
the body is {
"prompt": "A futuristic city with flying cars at sunset",
"steps": 50,
"cfg_scale": 7.5,
"width": 512,
"height": 512,
"sampler_index": "Euler a"
}

but the answer is detail not found

what am I missing?

[lucid yoke]

It depends on the UI and API honestly

[zinc knoll]

Hey all! New to the channel, and looking forward to learning from you all!

Is DaVinci resolve the tool most of you are running right now?

[lucid yoke]

Absolutely not lol, we use stable-diffusion / flux

Most of us run it locally, do you have questions about installing it or? I recommend that general discussion happens in general so people with questions don't get missed

[zinc knoll]

I will. I am between runs for my day job. Will be back in an hour or two.

[hybrid summit]

This is ehat i have so far but no luck

I runned WebUI-user

Here my header

[ornate elk]

Do you have --API in the webui-user.bat?

[hybrid summit]

AS ARGUMENT?

As argument?

[ornate elk]

Yep

At the line commandline_args=

Its needed for the API to work

[hybrid summit]

Tried but still no luck

[formal zephyr]

Is there anything similar to stablediffusion-webui for 3.5? I don't really like comfy UI, but it's the only one I can find people recommending

[ornate elk]

Forge webui should have SD 3.5 support now

[formal zephyr]

Ty!

[zinc knoll]

So, I'm home now. Is there a guide to getting the tools set up on linux mint?

[zinc knoll]

I found the instructions for using Pinokio and am following those.

[zinc knoll]

I see what I thought was davinci, is comfyui. I don't have a dedicated video card, so far all install videos have failed.

[formal knoll]

I want to use this checkpoint https://civitai.com/models/978314/ultrareal-fine-tune , Should I use SD or ComfyUI for best results? I will be running on colab..just started to use checkpoints idk how these work.Had used Fooocus

[random night]

i has generate image successfully but my resuilt is like that, what happen?

[ornate elk]

Make sure the webui is whitelisted in any browser adblocker

They often cause issues in the img2img tab

[random night]

i tried, not working

i still got the image in the infinite browsing extension, so not like it didn't generate any image

[random night]

delete venv setup again doesn't help either

[ornate elk]

Is your webui updated and the extensions?

[random night]

i tried git pull, anything up to date

[ornate elk]

The extensions have to be updated in the extension tab

[random night]

hang on a moment

[random night]

still the same

[ornate elk]

Seems like your browser or an add-on is blocking stuff

[random night]

i block all of that for the webui

[ornate elk]

Which browser and extensions do you have?

[random night]

besides, the problem here just exist the day before. For the last 2 weeks, everything run normally

i run on edge and these two

[ornate elk]

Yea these can cause this behaviour then

[random night]

and i blocked it

[ornate elk]

Hmm okay

Then can you show the full cmd log after you created an image?

[random night]

Applying attention optimization: Doggettx... done.
Model loaded in 91.1s (load weights from disk: 0.4s, create model: 81.4s, apply weights to model: 7.1s, apply half(): 0.3s, load textual inversion embeddings: 0.6s, calculate empty prompt: 1.3s).
100%|██████████████████████████████████████████████████████████████████████████████████| 20/20 [00:02<00:00,  9.88it/s]
Total progress: 100%|██████████████████████████████████████████████████████████████████| 20/20 [00:16<00:00,  1.20it/s]
Total progress: 100%|██████████████████████████████████████████████████████████████████| 20/20 [00:16<00:00, 11.74it/s]```

here

[ornate elk]

Full cmd log pls

[ornate elk]

Remove --no-half --lowvram -- opt-split-attention from the stability matrix launch args

And the civitai helper extension outputs an error.
So maybe reinstall that or delete it

Then delete the venv folder again

Also why using stability matrix. Its not needed or recommended for AMD stuff

[random night]

cuz i also using comfyui

[ornate elk]

Whats your GPU BTW?

[random night]

i don't have gpu, only igpu, which is 780m

[ornate elk]

Ah okay

Then you probably need --lowvram

[random night]

yeah, i use civitai helper fork version, that's why it shows the error

[random night]

still show no image

[ornate elk]

Okay, then last thing is to try an other browser

If its broken there too then its the setup itself

[random night]

alright, i'm gonna download chrome real quick

[ornate elk]

Try Firefox 😉

[random night]

okay

[random night]

still the same

maybe something happen with the setup

[ornate elk]

Yep then its the setup

[random night]

so i need to redownload all the folder, not just venv?

[random night]

thanks god that i leave all the stuff in stability matrix

[ornate elk]

You could try to delete the config_ui.json and the config.json

That resets the UI

If that doesn't help you can try delete the extensions and the venv folder.
Or reinstall the webui

[random night]

i will try this one

if not work, i'll reinstall the webui from scratch

[random night]

RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cpu and cuda:0! (when checking argument for argument index in method wrapper_CUDA__index_select)
bruh, get this when delete those two file

never mind, remove the --lowvram and it run normal

fuck

[vast crest]

Hello, lately webui 1111 gives me strange problems.
i have to restart it several times before it opens. here are some photos, what do you think is the problem?

[random night]

try (pip install "tokenizers<0.14")

[ornate elk]

Still the broken image?

[random night]

yes

i'm reinstall webui

[ornate elk]

Hey, can you show the full cmd log ?

After a relaunch

[random night]

oh c'mon

why it still has broken image

[random night]

the reinstall fail

[ornate elk]

Any error?

[random night]

still no error

[random night]

here the cmd log when i fix some of the bugs

still the same as if it didn't reinstall

[ornate elk]

Looks normal

What you can try is to upgrade Python to 3.10.11 64bit

[obtuse drift]

Hello, i tried this guide ( https://stable-diffusion-art.com/stable-video-diffusion-img2vid/#Install_Stable_Video_Diffusion_on_Windows ) with installing SVD under windows and get this error: "ImportError: numpy.core.multiarray failed to import"
D:\svd032025\generative-models\venv\lib\site-packages\torchmetrics\functional\text\chrf.py:33: UserWarning: Failed to initialize NumPy: ARRAY_API not found (Triggered internally at ..\torch\csrc\utils\tensor_numpy.cpp:84.)
EPS_SMOOTHING = tensor(1e-16)
RuntimeError: module compiled against ABI version 0x1000009 but this version of numpy is 0x2000000
2025-03-05 14:11:43.081 Uncaught app execution
Traceback (most recent call last):
File "D:\svd032025\generative-models\venv\lib\site-packages\streamlit\runtime\scriptrunner\exec_code.py", line 121, in exec_func_with_error_handling
result = func()
File "D:\svd032025\generative-models\venv\lib\site-packages\streamlit\runtime\scriptrunner\script_runner.py", line 593, in code_to_exec
exec(code, module.dict)
File "D:\svd032025\generative-models\scripts\demo\video_sampling.py", line 6, in <module>
from scripts.demo.streamlit_helpers import *
File "D:\svd032025\generative-models\scripts\demo\streamlit_helpers.py", line 7, in <module>
import cv2
File "D:\svd032025\generative-models\venv\lib\site-packages\cv2_init.py", line 181, in <module>
bootstrap()
File "D:\svd032025\generative-models\venv\lib\site-packages\cv2_init.py", line 153, in bootstrap
native_module = importlib.import_module("cv2")
File "C:\Program Files\Python310\lib\importlib_init_.py", line 126, in import_module
return _bootstrap._gcd_import(name[level:], package, level)
ImportError: numpy.core.multiarray failed to import

What i am missing?

something with NumPy isnt working ...

thank you in advance

[random night]

bruh

so i need to delete venv again?

[ornate elk]

Yep

[ornate elk]

You need to run:
pip install numpy==1.26.4

[vast crest]

git pull

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS= --xformers --no-half-vae --autolaunch

call webui.bat

[ornate elk]

I mean the cmd log, the black window

[vast crest]

when I start the program do you mean?

[ornate elk]

Yes

[vast crest]

sometimes my part like this, so I have to close and try again

1

2

3

4

[ornate elk]

Oh you have dreambooth installed.
That causes the issues

Delete the dreambooth extension from the extensions folder.
Then delete the venv folder and relaunch the webui-user.bat
Then wait untill it opens in browser.
At the next relaunch it should start normal

[vast crest]

realy,ok one moment 🙂

now it works perfectly. Thanks a lot CS1o

🙂

[ornate elk]

Perfect, no problem 🙂

[random night]

fuck

still not working

[vast crest]

❤️

[ornate elk]

Have you tried disabling the extensions and then deleting the venv again?

[random night]

i dit it from the start

[ornate elk]

Hmm but your last logs included civitai extension's

[random night]

i mean, the civitai i use right now is the main one, not fork

[obtuse drift]

thx, helped 🙂

[random night]

and i don't see any problem with the extension

[obtuse drift]

Next problem is
TypeError: randn_like(): argument 'input' (position 1) must be Tensor, not NoneType
Traceback:

File "D:\svd032025\generative-models\scripts\demo\video_sampling.py", line 190, in <module>
value_dict["cond_frames"] = img + cond_aug * torch.randn_like(img)

do svd also support amd 6700xt and how to activate it?

[random night]

wait, you use amd gpu?

[obtuse drift]

yes

[random night]

why not use zluda?

[obtuse drift]

didnt heared from it %) i will test it, thx

[random night]

check the pin right at this channel

@ornate elkhmm, when i use forge linking to automatic1111 through venv, it generate image

weird

so the problem not come with the venv file

it's something else

i should try deleting all the extension

[ornate elk]

you cant use that with amd probably

but if you want to get into image generation checkout my AMD install Guides:
https://github.com/CS1o/Stable-Diffusion-Info
Go for Auto1111 Zluda or Forge Webui with Zluda

[obtuse drift]

pined is only SD, but no SVD?

[broken goblet]

Anyone knows what "cyberfix" is?

[ornate elk]

if you install ComfyUI with Zluda you can use SVD and other text or img2video models like LTX video

[ornate elk]

a model, or a bunch of models based on noobai

[random night]

i have remove all those extension, still no image...

and delete venv again too

i don't know what to do anymore

[ornate elk]

You could try install the webui normaly , not with stability matrix

[random night]

i install webui normally, git clone, edit webui-user, only when finish downloaded venv then i integrated it to stability matrix

[ornate elk]

Okay and what happens if you dont integrate it?

[random night]

then i have to download all my model again, and double the amount of storage

[ornate elk]

You can link model folders in the webui-user.bat

With --ckpt-dir and --lora-dir

[random night]

how about output sharing?

[ornate elk]

You can define the output in the webuis settings

[random night]

wait, you can do that?

[ornate elk]

Sure

Its in the settings in the first two options

Output paths etc

[random night]

hmm, i will try to disintegrate from stability matrix then

[random night]

still no use

[ornate elk]

Can you show the cmd log again and a screenshot from the webui in browser?

[random night]

yeah, wait a moment, i need to redownload venv again

[ornate elk]

its still launching trough stability matrix

not as a seperate install

[random night]

yeah, when i realize that disintegrate stability matrix won't do anything, i put it back

[ornate elk]

hmm ok

is your gpu driver updated?

[random night]

absolutely

[ornate elk]

do you use this model to generate images ?

because its only made for inpainting

[random night]

Calculating sha256 for C:\Users\nhung\AppData\Roaming\StabilityMatrix\Packages\stable-diffusion-webui-amdgpu\models\Stable-diffusion\sd\perfectdeliberate_v5.safetensors: 1bb210864399f4eeb25db61d82ff63c90cc57d46a311a63919e09e5da2c9f9d7
Loading weights [1bb2108643] from C:\Users\nhung\AppData\Roaming\StabilityMatrix\Packages\stable-diffusion-webui-amdgpu\models\Stable-diffusion\sd\perfectdeliberate_v5.safetensors
Creating model from config: C:\Users\nhung\AppData\Roaming\StabilityMatrix\Packages\stable-diffusion-webui-amdgpu\configs\v1-inference.yaml
Applying attention optimization: Doggettx... done.
Model loaded in 5.4s (create model: 1.4s, apply weights to model: 3.3s, apply dtype to VAE: 0.1s, calculate empty prompt: 0.3s).
100%|██████████████████████████████████████████████████████████████████████████████████| 20/20 [00:01<00:00, 13.41it/s]
Total progress: 100%|██████████████████████████████████████████████████████████████████| 20/20 [00:17<00:00,  1.13it/s]
Total progress: 100%|██████████████████████████████████████████████████████████████████| 20/20 [00:17<00:00, 14.99it/s]```

here

still the same

[ornate elk]

is controlnet disabled?

[random night]

yes, it does

[ornate elk]

hm ok

[jaunty vale]

Is the "NVIDIA GeForce RTX 3090 NON-LHR FHR GFX Card 24GB GDDR6X GPU Dell Alienware OEM" good for stable diffusion?

[random night]

i mean, yes

[jaunty vale]

All that NON-LHR FHR stuff doesn't affect anything?

[random night]

AI stuff in general is only care about vram, bandwidth, so you good to go

and cuda generation

[jaunty vale]

So the nvidia cudas are still there?

[random night]

yes

[jaunty vale]

Thankyou friend

[quick fiber]

SD\webui\stable-diffusion-webui-directml\modules\dml\pdh\__init__.py", line 25, in expand_wildcard_path
        raise PDHError("Something went wrong.") 
modules.dml.pdh.errors.PDHError:``` im recieving this error and im not sure why. it doesnt let me generate anything. btw using the amd version

its all this

[broken goblet]

Is there a way to make SDXL faster on ForgeUI but without losing much quality? I heard about the "--medvram" command but I think it's only for automatic1111

[formal ridge]

You're supposed to use --medvram-sdxl, but if you really want to max out your speed, try this on your commandline.
--xformers --always-batch-cond-uncond --disable-console-progressbars --disable-nan-check --lowram --medvram-sdxl --precision autocast --upcast-sampling --use-ipex

[broken goblet]

Can I do that on ForgeUI? I haven't found any tutorial

[formal ridge]

Yeah, just edit the batch file like you would on A1111

[quick fiber]

can anyone help with the pdh error?

[quick fiber]

directML isnt detected eventhough ive downloaded multiple times through cmd promt

[ornate elk]

Your following an old guide it seems.
What's your GPU?

[quick fiber]

Rx 580

[quick fiber]

Is there a newer one I should follow

[ornate elk]

Yep you have two options.
Installing a webui with directml.
Which is easy to install but is slow and very vram limited.
Or installing a webui with Zluda. Which is a bit more work but significantly faster and not as limited.
You can find all Guides here:
Either go for Automatic1111 or Forge Webui for your GPU:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides

[quick fiber]

I used automatic1111 the first time and it didn't work

I can try again tomorrow with both and hope for the best

[ornate elk]

Sure. They will work. I helped a few people with rx580 and 590 to get it working

Let me know if you have any question's

[quick fiber]

Aight thanks

[ornate elk]

You can try getting it with directml working first and then try the zluda version

[quick fiber]

In the error I sent what caused the PDHError?

[ornate elk]

Its not the full cmd log so its hard to say what caused it

But first issue I see is that you tried to launch it in the downloads folder

And the folder name is old so your most likely followed an old guide

[quick fiber]

Oof

Do I do it in a the user file

[ornate elk]

My guide covers everything

[quick fiber]

Okay cool thanks. I really hope it works cos I was punching myself looking up so many guides

Probably why I used an old version

[ornate elk]

NP, yea there are a lot of outdated guides out there

[broken goblet]

is this good if I have a 1660 super?

[oblique pivot]

Hello, I tried following the guide provided for macOs by Fooocus, but apparently I am not even able to properly install pytorch?

I only have access to a Intel Macbook Pro Retina 15" from 2015 with an i7, integrated graphics card (Intel Iris Pro 1536 MB) and 16GB RAM.

I followed the guide to install pytorch as outlined in https://developer.apple.com/metal/pytorch/ using anaconda:

curl -O https://repo.anaconda.com/miniconda/Miniconda3-latest-MacOSX-x86_64.sh
sh Miniconda3-latest-MacOSX-x86_64.sh
source ~/.zshrc
conda install pytorch torchvision torchaudio -c pytorch-nightly

When this resulted in unsolvable dependency constraints, I downgraded to python3.10

conda install python=3.10
conda install pytorch torchvision torchaudio -c pytorch-nightly

But when running the code for verification, I receive "MPS device not found."

import torch
if torch.backends.mps.is_available():
    mps_device = torch.device("mps")
    x = torch.ones(1, device=mps_device)
    print (x)
else:
    print ("MPS device not found.")

Is there anything I should do differently?
If not, is there maybe another guide recommended for macOs?

I want to try out Stable Diffusion, but online services seem kinda expensive considering the limited number of images that can be generated using these services..
What is a good way to try this without upfront investment into new hardware in my situation?

[hybrid shore]

Are there any tricks to get a pose to take when trying to use openpose? I keep getting a black square with the specific pose I'm trying to copy

[formal ridge]

No idea, I've never used anything lower than an 8GB card for this stuff.

[quick fiber]

i did everything ur guide said and still got the same error

[toxic girder]

stable diffusion was working alright yesterday but today when i went to run the webui bat file, I got hit with this error message and can't access the ui page now. Any help or advice would be highly appreciated.

[random night]

like the cmd said, add "--skip-torch-cuda-test" in webui-user.bat in "set COMMANDLINE_ARGS"

[toxic girder]

do i leave it like this?

[random night]

yes

save that and run the script again

[toxic girder]

dude i fucking love you

thank you so much

[random night]

no problem man

[toxic girder]

if u dont mind me asking do you know how to fix this also?

[ornate elk]

Can you show the full cmd log ans your settings like resolution, steps, model etc

[ornate elk]

What's your GPU?

[toxic girder]

nvida rtx 3080

[ornate elk]

And can you show the full cmd log?

[ornate elk]

Remove --skip-torch-cuda-test
And only add --xformers --medvram-sdxl

Also make sure that your GPU driver ist updated

[toxic girder]

to where do i add it?

[ornate elk]

In the webui-user.bat

At the line commandline_args=

[toxic girder]

is this good?

im gonna save it after i update my drivers

so be right back

ty for helping me

[ornate elk]

No problem

I brb, driving now

[toxic girder]

i got this again after i replaced the line 😭

[drowsy mauve]

Louis Vuitton and IQOS showroom

[ornate elk]

Can you show the full CMD log?

[toxic girder]

I think i fixed it!, it generated something. Thank you for helping me!

[ornate elk]

No problem 🙂

[slow badge]

Anyone knows how to get work kohya? I just got a 5080 but I get several errors

I think I need a specific torch but it automatically downloads an old version

[lucid yoke]

5080 still isnt fully compatible with torch iirc

I had to do some trickery in my comfy env to make it work

[slow badge]

I also got to work with forge but I tried the same trick but didn't worked with kohya : (

[shy charm]

i keep having this issue with reforge where if i add a lora, it overloads the vram on the first image and takes forever to generate, but then all subsequent images generate normally until i load a different lora

[dapper rampart]

hello i get this error how can i fix it i dont understand

[rigid zephyr]

trying flux with ComfyUI. It starts generating and then says 'reconnecting'. I try to start again and it says 'failed to fetch'

Is it an issue with my RAM?

[rigid zephyr]

ok i found out this was a memory issue. any suggestons on how to run flux with lower memory usage?

[ornate elk]

The issue is that you placed the webui in a OneDrive synced folder

[ornate elk]

Hey, what your GPU?

[rigid zephyr]

16gb

[ornate elk]

Nvidia or AMD?

[rigid zephyr]

Apple M2

silicon

[ornate elk]

Ah ok

[rigid zephyr]

its not enough right? should i be getting another PC/chip

[ornate elk]

You should go for smaller flux models but its not worth running them on apple.
Best you should use is sdxl based models

[rigid zephyr]

what GPU am i gonna need to run FLUX models? I need a PC that's not gonna have a problem running high models

[ornate elk]

A GPU with 16gb vram or more works the best with flux

[ornate elk]

You can try to run this flux model:
https://civitai.com/models/141592?modelVersionId=992642
It has a smaller version.

You also need the t5 clip fp8 and clip l text encoders

[rigid zephyr]

yeah i have these from flux

ok thanks, trying

[rigid zephyr]

is schenll even lower in GPU usage? And do i need the same clips for that

[ornate elk]

Schnell just needs less steps

And yep same clips needed

[charred rover]

can't find info. can I run fp8 models on rtx 20xx?

[rigid zephyr]

yeah its not working, i think i need something better sadly

[ornate elk]

Yes but they will run at fp16

[charred rover]

so.. can I run this https://github.com/lodestone-rock/ComfyUI_FluxMod?tab=readme-ov-file on 11gb vram? do I need some GGUF Flux? I don`t know much about Flux

[ornate elk]

Yes but for flux stay with q4 or n4 models

But its not worth on any GPU with less than 12gb vram

Sdxl is much better in terms of speed, loras, controlnet etc

And vram usage of corse

[charred rover]

yes and the quality is perfect

[ornate elk]

Yep and you can upscale it easily

[charred rover]

true

thank you

[ornate elk]

No problem