🤝｜tech-support

[gilded canyon]

plain text2img

[wet flame]

and i don't even got any full ram/gpu usage either

it barely using 6 gb vram

but says cuda out of memory wow

m forced to use sd 1.5 and 512x512 images

oh well need to program scriptable objects architecture

[wet flame]

btw are you an artist too 😄

what you selling?

[gilded canyon]

i use AI for visual novel game assets lol

its frustrating because i cant work on anything rn without a1111 breaking

[wet flame]

well m just making things 512x512 rn lol

i can't anything sdxl or such

the xl lora don't show up or no xl models load

as soon i do that error press any button to continue

[wet flame]

btw try some 2D skeletal animation tool to anmate them

[gilded canyon]

do you have any recommendations?

[low basin]

you should go to Settings > System > About (at bottom) > Advanced system settings > Advanced tab > click performance settings > advanced tab > Change (under 'Virtual Memory) and see what it says under there

[wet flame]

anyways i give up on making all these work

i guess m stuck with 1.5 and 512 images 😢

[gilded canyon]

@wet flame I got a workaround solution

Download WebUI forge but DO NOT UPDATE IT

[wet flame]

that don't have seamless tiling sadly

[gilded canyon]

updating it breaks it

oh damn

that sucks

[wet flame]

i need that to make my tilesets xD

[gilded canyon]

my webui is working again now that i reinstalled it but didnt update

hopefully u get a fix soon buddy :(

[wet flame]

the developer said its not their concern on that

its just for hobbies who want to make pretty images for what idk

for me i use AI assisted workflow i can make them manually everything but what is the point here produces similar result and manually takes way too long

[errant tundra]

can always try comfyui

[gilded canyon]

nodes = ew

[wet flame]

no ew

i hate spider webs m good

[errant tundra]

it's so good tho

[wet flame]

its head spinning though

[gilded canyon]

its good but learning curve which isnt worth it for me

i dont use any extra extensions or anything

at most i use hires fix and img2img

[errant tundra]

it's not that bad

I literally started stable diffusion 2 days ago

it's not that hard to learn

[wet flame]

its not about hard its about god damn nodes

screw that

i will start smahing my pc if i do that

[errant tundra]

but the nodes is what makes it so good

[wet flame]

its the node that it makes so much bad for me

i like my words and prompts

i like programming not visual scripting and connecting wires man

[errant tundra]

then you can use a simpler setup you don't have to go as advanced as I have

you could get similar to A1111 ease of use with like 6 nodes instead of my like 30-40

[wet flame]

btw i found this
https://github.com/Stability-AI/StableSwarmUI

[viral knot]

there is even a channel here for stable swarm

[wet flame]

oh i did not notice

[viral knot]

Alex is a pretty cool dude, he can help you out with stable swarm if you have problems

[wet flame]

i just want to make seamless tiles and use sdxl that is all actually

[gilded canyon]

Yes I tried this already, it's pretty good but based on comfyui workflows

[wet flame]

oh no

not good with word promptings?

i basically do this lol
https://youtu.be/Ly6USRwTHe0

i draw my doodles and then make textures with them

[gilded canyon]

its good with word prompting, but you'll still need to work with workflows a little bit

we shouldnt use the tech support channel to talk about this though

[wet flame]

as long it makes me those textures for 2D tilesets m good xD

i use my images as base references

[wet flame]

so you got it working i guess?

[gilded canyon]

yea it works

u should try it

see how you like it

[wet flame]

here we go again lol

2 days lost now more days to come

[wet flame]

well with the help of that it seems i can create 1024 images directly without cuda error

but still i can't use a sdxl model

[gilded canyon]

hmm thats weird

are you running out of memory with SDXL

what model are you using for SDXL

[wet flame]

i can't generate image in the swarm ui itself

juggernaut

[wet flame]

no it says i am only using like 3 gb vram

[gilded canyon]

oh thats weird

i wish i could help but idk much about fixing this type of stuff

[wet flame]

i already had the A111 installed so i put that as ref for swarm

it starts the A111 tab

i can generate there but its not generating on swarm ui

[gilded canyon]

oh weird

try restarting swarm because the first time I tried generating with it, it didnt work

but then I restarted it

and it worked the second time

[brave elbow]

Excuse me , does anybody know how to solve this problem? I have install diffusers using “pip install diffusers” already, but still not work....

1.My system is windows
2. Run the codes from "https://huggingface.co/docs/diffusers/en/using-diffusers/controlnet"
3. Run the codes on Anaconda Vscode jupyter notebook

[light vine]

does anyone here guide me how to train models on any specific person images for free ????

[ornate elk]

Hey, what are you trying to do?

[ornate elk]

Use OneTrainer or Khoya_ss
Both work localy

[light vine]

I used khoya_ss but its not loading in gardio

[light vine]

i just want some images of mine made through **AI **

[light vine]

thank you for helping me out

[ornate elk]

Can you explain that?

[light vine]

I want to train a model specifically on a dataset of only one person to create various images of that same person.

[ornate elk]

Alright, what's your GPU?
For training a lora you need at least an nvidia GPU with 6gb vram

[light vine]

can i train it on google colab ?

[ornate elk]

I think that's possible but can't help with collab

[light vine]

ohh so can you explain it its possible on my computer how can i ?

[ornate elk]

First check what's your GPU.
Open the taskmanager and click on the performance tab.
Then check for GPU0 or GPU1

[light vine]

will try , if not possible on mine i will do it on my friends pc

[ornate elk]

@light vine for google collab I found this, but dont know if it works:
https://colab.research.google.com/github/hollowstrawberry/kohya-colab/blob/main/Lora_Trainer.ipynb

[light vine]

Thanks for your help @ornate elk

[ornate elk]

No problem

[honest tendon]

@ornate elk how do i open comfyui with terminal commands?

[broken tulip]

Anyone around with general Python knowledge who can help me with an error? I'm sure it's easy for someone who knows what he's doing 😛

[delicate citrus]

hi

how can I use the preprocessor depth anything?

when I use it say that I don't have the module

[steady thistle]

@low basin @karmic crown Is there a place to buy partially assembled PC products? I'd like to get a case with a mobo and cpu with a cooling solution already installed, then add my video card and memory, etc.

[ornate elk]

Is it Stable diffusion webui related?

[ornate elk]

Open up a cmd inside the comfyui folder and run the main.py

[broken tulip]

Not directly. It's a video text to image tool. The env is built correctly, it's just a numpy error. So just a general question

[ornate elk]

What's the error?

A full cmd log can help

[broken tulip]

I'll split it into 2 messages sec

C:\st2v.venv\lib\site-packages\scipy_init_.py:155: UserWarning: A NumPy version >=1.18.5 and <1.25.0 is required for this version of SciPy (detected version 2.0.0
warnings.warn(f"A NumPy version >={np_minversion} and <{np_maxversion}"
Traceback (most recent call last):
File "C:\st2v\StreamingT2V\t2v_enhanced\inference.py", line 11, in <module>
from t2v_enhanced.model.video_ldm import VideoLDM
File "C:\st2v\StreamingT2V\t2v_enhanced\model\video_ldm.py", line 4, in <module>
import pytorch_lightning as pl
File "C:\st2v.venv\lib\site-packages\pytorch_lightning_init_.py", line 26, in <module>
from pytorch_lightning.callbacks import Callback # noqa: E402
File "C:\st2v.venv\lib\site-packages\pytorch_lightning\callbacks_init_.py", line 14, in <module>
from pytorch_lightning.callbacks.batch_size_finder import BatchSizeFinder
File "C:\st2v.venv\lib\site-packages\pytorch_lightning\callbacks\batch_size_finder.py", line 24, in <module>
from pytorch_lightning.callbacks.callback import Callback
File "C:\st2v.venv\lib\site-packages\pytorch_lightning\callbacks\callback.py", line 22, in <module>
from pytorch_lightning.utilities.types import STEP_OUTPUT
File "C:\st2v.venv\lib\site-packages\pytorch_lightning\utilities\types.py", line 40, in <module>
from torchmetrics import Metric

File "C:\st2v.venv\lib\site-packages\torchmetrics_init_.py", line 46, in <module>
from torchmetrics.image import ( # noqa: E402
File "C:\st2v.venv\lib\site-packages\torchmetrics\image_init_.py", line 26, in <module>
from torchmetrics.image.fid import FrechetInceptionDistance # noqa: F401
File "C:\st2v.venv\lib\site-packages\torchmetrics\image\fid.py", line 28, in <module>
from torch_fidelity.feature_extractor_inceptionv3 import FeatureExtractorInceptionV3
File "C:\st2v.venv\lib\site-packages\torch_fidelity_init_.py", line 6, in <module>
from torch_fidelity.metric_fid import KEY_METRIC_FID
File "C:\st2v.venv\lib\site-packages\torch_fidelity\metric_fid.py", line 6, in <module>
import scipy.linalg
File "C:\st2v.venv\lib\site-packages\scipy\linalg_init_.py", line 198, in <module>
from ._cythonized_array_utils import *
File "_cythonized_array_utils.pyx", line 1, in init scipy.linalg._cythonized_array_utils
ValueError: numpy.dtype size changed, may indicate binary incompatibility. Expected 96 from C header, got 88 from PyObject

It installed numpy 1.24.4

I uninstalled it to try an other version but can't get it it to work 😛

[ornate elk]

And what's the tool called you try to get running? And what python version do you have?

[broken tulip]

Python 3.10.11

The tool is built via VoC: https://github.com/Picsart-AI-Research/StreamingT2V?tab=readme-ov-file

[ornate elk]

And you installed it inside the venv?

Did you activated the venv before installing it?

[broken tulip]

Usually VoC does everything correct. It's the first time I get an error.

Yeah it built it's own venv instance in a separate folder

It somehow must be connected to just numpy

I found this, but don't understand half of what they say: https://stackoverflow.com/questions/66060487/valueerror-numpy-ndarray-size-changed-may-indicate-binary-incompatibility-exp

[ornate elk]

Have you installed the requirements.txt into the venv?

Check the file and maybe set a version number for numpy or add it there

[broken tulip]

You mean into this one?

Would I just add: numpy -- 1.20.0

For example?

Ok, let me try

[ornate elk]

With ==

[broken tulip]

Yep thanks

[broken tulip]

Same error

[broken tulip]

Install bat looks like this

Should I maybe add pip nympy etc and reinstall?

Say add to install:
echo *** VoC - patching datasets
pip install --force-reinstall numpy==1.20.3

[ornate elk]

Where did you get that file from. I don't see it on their repo?

Ah its from VoC

[broken tulip]

Yep

[ornate elk]

I will try installing StreamingT2V manually later

To see if I get the same error or not

[broken tulip]

Ok thank you ❤️

I will try reinstalling it with the modified bat

maybe it works

Will fresh ionstall everything with that fix ontop

[broken tulip]

Just noticed after fresh reinstalling. I put my fix at the end, so it will hopefully overwrite it

[ornate elk]

Also strange that it first installs xformers 0.0.19 and then 0.0.25

But could be because it then looks in modelscopes requirements file

So maybe you need to include the right numpy version there too or check that file too

[broken tulip]

You mean related to the xformers?

[ornate elk]

No related how it installs stuff

If it installs everything from requirements.txt and then installs microscope. It will then checks modelscopes requirements.txt and maybe overwrite some packages

Like it did with xformers

[broken tulip]

I see, thanks

[vocal burrow]

tldr but iirc 0.0.19 was the cutting point in A111 for torch1

[ornate elk]

Yea true

[flat crescent]

For some reason in the last month my img2img generations went from using about 19gb of VRAM to requesting 72gb of VRAM and I have not changed any settings. I'm using traditional SD models, and trying to take a 512px by 920px image and upscale by 2.2 in a single pass. This has historically worked fine with my 24gb VRAM card, but now SD wants to allocate a huge amount of VRAM for this task - what changed? What settings should I be looking at to try and resolve this?

[honest tendon]

https://youtu.be/vCCVxGtCyho?si=J0cUP_knqA9f7KNR

how do I do this 😦

with ZLUDA i aint got the python-embed file

[ornate elk]

Hey what webui do you use?
And what's your GPU ?

[ornate elk]

Do you want to use Reactor extension?

Because I got that running with zluda in comfyui

[flat crescent]

I'm using Automatic1111 running an Nvidia A5000. I had been previously using the Ultimate SD Upscale Script, but it seems that it has memory leaks now so that might be the problem.

[obsidian gull]

how do you use multiple tools and stuff with a base model?

like styles and stuff

[honest tendon]

before than which one do i install, im not sure what model is being used rn

is it 1.5 xl or

[ornate elk]

Make sure the extension is updated.
And you can also try to use the SD upscale script.
Set the resolution to a square and enable the script

[flat crescent]

That script works, but the image quality is lacking. The Ultimate SD Upscaler is installed via Git, does the Automatic1111 interface pull updates automatically or do I need to manually request updates?

[ornate elk]

You need to git pull manually to Update automatic1111 if you don't have git pull in your webui-user.bat.
Also what's in your webui-user.bat?
Make sure your using xformers

[obsidian gull]

hey CS1o can you help me out a lil bit? like this thing, how do i use this in creating images? it says its a tool

[flat crescent]

xformers is setup and is working correctly. My set COMMANDLINE_ARGS=--xformers --listen --port 7860 --allow-code --api --autolaunch

[ornate elk]

Its an embedding.
It goes into the embeddings folder.
Then you have to add the embedding name to the negative prompt and it should work

[flat crescent]

Can I add the Gitpull as a commandline argument?

[obsidian gull]

oh alright thanks

[ornate elk]

Nope it has to be added in its own empty row

[flat crescent]

What code would I use?

[ornate elk]

Just
git pull

[obsidian gull]

also do styles work in the same way or are they different?

[ornate elk]

It depends on the file.
Check always the info box on what type it is.
Styles are Normaly different and go into the styles.csv

[obsidian gull]

it says its lora type

[ornate elk]

Then its a lora. Goes into the models/lora folder

[flat crescent]

So the file should read:
@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=--xformers --listen --port 7860 --allow-code --api --autolaunch
git pull

call webui.bat

[obsidian gull]

alright, thanks man

[ornate elk]

NP, to use the lora click on the lora tab and select the lora there. Then it will be added to the prompt

[ornate elk]

If you never updated the webui before it can help to delete the venv folder

Then it will install the latest packages

[flat crescent]

My files in that folder are dated 8/21/2023, so probably not updated, eh?

[ornate elk]

Yes pretty outdated. Maybe a fresh reinstall would be better

A lot changed since then and it should be much faster and less vram consuming now

[honest tendon]

What version of SD is in ZLUDA comfyui?

[flat crescent]

Thank you for your help. This has me back up and running for now, but I will consider a fresh reinstall when I have time. Thanks again.

What happened to the Karras samplers? They seem to have disappeared - I had been using DPM++ 2m Karras for most of my workflow. Is there a better modern sampler that replaces it?

[soft fiber]

i believe their locations were changed, look for schedulers and choose karras

[honest tendon]

I have a couple questions, is comfyui using stable diffusion or is it just a seperate thing?
Is stable diffusion basically a model checkpoint or something?

[flat crescent]

Found it in options, thanks. Is there an advantage/disadvantage to using the karras schedule sigmas or not?

[soft fiber]

changing the sigmas does different things to the output so have fun experimenting. or dont jf youre not into that

[flat crescent]

I'll play around, thanks.

[honest tendon]

N(OOO

i was reading

[ornate elk]

Automatic1111 and Comfyui are both Webuis to use Stable-diffusion.
Stable diffusion is a process to generate images.
Models (checkpoint's) are needed for image generation.
There are community models.
These are based on 1.5 or sdxl (the two base stable diffusion model versions)

[ornate elk]

Its under schedulers.
Leave it on automatic and it will use karras when 2m is selected as sampler

[ornate elk]

these models are for 1.5 and sdxl. check the end of the file name

[honest tendon]

thanks

but how do i know what model ZLUDAs running?

how do i check

also i noticed im missing the ipadapter file

do i just create a new file in the models folder?

https://youtu.be/C94pTaKoLbU?t=811
im following this guys guide, but im missing a couple of nodes which i cant seem to find online or in comfyui manager specifically
Prepare Image For InsightFace
Apply IPadapter FaceID
Load IPadapter Model and other Load IPadapter ___

heres my workflow

[ornate elk]

i tested it and didnt found that node either

also IP-Adapter-Face-ID wont work with Zluda

but normal IP-Adapter-Plus-Face can work

[honest tendon]

that makes so much sense now

[honest tendon]

noticed that too

[shy charm]

idk why regional prompter is so inconsistent
sometimes it works and sometimes it just does this

it just ignores the regions

also i don't normally use the ui this zoomed out, i only did that to fit the relevant settings into one screenshot

[ornate elk]

you have one BREAK to many

you only need 2 BREAK

[shy charm]

i'm using base prompt and common prompt

[ornate elk]

what do you mean by that?

[shy charm]

i think the base prompt affects the whole image and common prompt affects each of the regions individually

[ornate elk]

yea but you only have 2 Regions

so you only can use 1 base prompt and 2 region prompts

[shy charm]

1 base prompt, 1 common prompt, 2 region prompts

[ornate elk]

i dont think there is a "common" prompt its mostly using the base prompt

[shy charm]

i read this on a guide when trying to figure out how to use regional prompter

but tbh i still don't fully understand it

[ornate elk]

try without the common prompt and it will most likely work

i tested it rn

using 2 BREAK instead of 3 worked really good

but yea its very confusing with base and common

[honest tendon]

heyy, so do i only keep the .pt? or do i keep the -neg.pt part and rename it

[ornate elk]

you can leave the name as it is

.pt are embedding files

[honest tendon]

So i can name it Deformed.pt?

[ornate elk]

should work yes

[honest tendon]

sickk

[young bridge]

hi all

can anyone help me with this error

ImportError: numpy.core.multiarray failed to import when I launch comfyui. I already tried pip install -U numpy and pip install -r requirements.txt but it didnt fix it.

[vast hull]

is it possible to change a photo of myself that is standing, using maybe a controlnet pose? that is sitting down, which in return will keep my face but make my body sitting down from that controlnet pose???

[broken tulip]

Does anyone run demofusion?

[round glacier]

Inpaint, maybe?

[nova parcel]

what checkpoint would be the best for this prompt?
im just trying to do SDXL
SD 1.5 i guess could work as well if i use a SDXL refiner

I may try and do Loras as well

[broken tulip]

Hey. I installed https://github.com/deroberon/demofusion-comfyui for comfy. In which folder do I download https://github.com/PRIS-CV/DemoFusion?tab=readme-ov-file?

[nova parcel]

SD 1.5 675x450 and upscale x16
or
SDXL 1350x900 and upscale x8

[obsidian gull]

@ornate elk hey, im sorry but, how did you use negative prompts again?

i just have to add the entire name of the embedding and it will work?

[ornate elk]

Into the negative prompt yes

[proud nacelle]

hi i just signed up for this but i used google is there a way to still link my discord account?

[obsidian gull]

also where am i suppose to put in character files?

in lora folder?

[misty inlet]

wanted to try Pixart Sigma: where am i wrong?

[karmic crown]

i have no idea what that is, but it is not sdxl

you cant use

this is your problem

[karmic crown]

have you tried to use automatic? what happened?

[nova parcel]

Still curious about this, I just need checkpoint suggestions

I heard zavychroma is good

[ornate elk]

yep

[low basin]

Other than like Facebook Marketplace or similar I haven't seen any. Not that it helps you but if you have a MicroCenter within a reasonable drive, they have CPU/mobo/ram combos for ridiculously good prices although you'd have to assemble and get a case

[steady thistle]

203 miles away, a bit of a trek, but I've seen worse. heh

[low basin]

Yeah... that's a little far... I'd just keep an eye on marketplace and craigslist. Quite often people sell whole pc's minus the gpu as they frequently sell them separate

[fair oxide]

Does anyone know if it's possible to set a specific save directory for generated images that are "before face restoration"? I want to keep both but would prefer to send pre to another directory

[atomic mauve]

How much vram will the sd3large consume comparibg to medium?

[woeful dirge]

import torch
from diffusers import StableDiffusion3Pipeline

pipe = StableDiffusion3Pipeline.from_pretrained("stabilityai/stable-diffusion-3-medium-diffusers", torch_dtype=torch.float16)
pipe = pipe.to("cuda")

image = pipe(
"A cat holding a sign that says hello world",
negative_prompt="",
num_inference_steps=28,
guidance_scale=7.0,
).images[0]
image

this is the error

403 Forbidden: Please enable access to public gated repositories in your fine-grained token settings to view this repository..
Cannot access content at: https://huggingface.co/api/models/stabilityai/stable-diffusion-3-medium-diffusers.
If you are trying to create or update content,make sure you have a token with the write role..
Will try to load from local cache.

HTTPError Traceback (most recent call last)
/usr/local/lib/python3.10/dist-packages/huggingface_hub/utils/_errors.py in hf_raise_for_status(response, endpoint_name)
303 try:
--> 304 response.raise_for_status()
305 except HTTPError as e:

9 frames
HTTPError: 403 Client Error: Forbidden for url: https://huggingface.co/api/models/stabilityai/stable-diffusion-3-medium-diffusers

The above exception was the direct cause of the following exception:

HfHubHTTPError Traceback (most recent call last)
HfHubHTTPError: (Request ID: Root=1-66711541-10dbd59c6f3d6f0f184eed18;9d468f6e-1b0d-4596-bef7-3306c3f8361d)

403 Forbidden: Please enable access to public gated repositories in your fine-grained token settings to view this repository..
Cannot access content at: https://huggingface.co/api/models/stabilityai/stable-diffusion-3-medium-diffusers.

OSError: Cannot load model stabilityai/stable-diffusion-3-medium-diffusers: model is not cached locally and an error occurred while trying to fetch metadata from the Hub. Please check out the root cause in the stacktrace above.

please fix it anyone

[woeful dirge]

this is the code

[woeful dirge]

this is the error

[foggy arch]

how do i use img-img to enlarge+improve a previous generation? its not working on my end

[vocal burrow]

denoiser is too high

[foggy arch]

0.5 then?

[vocal burrow]

0.4 even

[foggy arch]

also the resolution isnt getting higher

its stuck to the original size

[vocal burrow]

you clicked the output didn't you

you're not comparing the preview

[foggy arch]

i checked the sizes of the images from the files

1204x1540 is the target size i placed

images are showing 344x 440

[vocal burrow]

show the full webpage and both input and output image

[foggy arch]

this one?

Negative prompt: comic, blurry, deformed hands, deformed face, deformed fingers, multiple limbs, ugly, bad anatomy, bad hands, ugly hands, deformed, blurry, bad anatomy, disfigured, poorly drawn face, mutation, mutated, extra limbs, ugly, poorly drawn hands, (grayscale), weird colors, censored, deformed glasses, lowres, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, badhandv4, easynegative
Steps: 30, Sampler: Euler a, CFG scale: 7, Seed: 1904565443, Size: 344x440, Model hash: 54ef3e3610, Model: meinamix_meinaV11, VAE hash: 235745af8d, VAE: klF8Anime2VAE_klF8Anime2VAE.safetensors, Denoising strength: 0.4, Lora hashes: "shirayuki-nvwls-v1-final: d8964aa638c9, Loraeyes_V1: da8286d2ab99", Version: v1.7.0```

[vocal burrow]

I meant the full thing not just a croped part of it

[foggy arch]

ahh

[vocal burrow]

also what does the console log says

[foggy arch]

here you go

[foggy arch]

which part do i look at?

[vocal burrow]

it's unreadable

[vocal burrow]

the whole thing

copy paste it into a file and drop said file in here

[foggy arch]

this is the console log right?

[vocal burrow]

yes but again it's not the whole log

[foggy arch]

ill post it in .txt format

[vocal burrow]

(still works just fine for me)

[foggy arch]

is it something i may have forgotten to press?

[vocal burrow]

"oneDNN custom operations are on." what's your gpu ?

[foggy arch]

nvidia rtx 3050

[vocal burrow]

v1.7.0 your webui is way outdated

update your webui and its extensions then try again

(got to go for now)

[foggy arch]

i dont use the webui that much 😅 so idk how to update it

[foggy arch]

alr ty!

[vocal burrow]

git pull in its folder

[foggy arch]

git pull and latest SD github link?

[fair oxide]

just git pull

[thorny dragon]

@ornate elk bro, i installed tilled diffusion as u instructed, but it is also really slow while generating a image in img2img..like 45 mins for 1 img....also when i checked my cmd, it showed that the multi diffusion process wasn't 100%, but it gave me a image anyway....does it mean that the extension is not working?

[ornate elk]

No I said don't enabled Tiled Diffusion.
Only enable Tiled VAE

Then it will be fast as before

[thorny dragon]

ahhhh! my bad...

[ornate elk]

No problem 🙂 ik its a bit confusing with the tiled names xD

@thorny dragon enable tiled VAE when your using Upscaling like hires fix or going for high resolutions

[junior trellis]

hey can i use inpaint with refernce image? for instance i want to render a house with surrounding, i have normal boxes around it but i want these boxes to be inpainted and have a specfik style according to an image.

[steep phoenix]

use canny with ip adapter

[junior trellis]

may you send me a tutorial?

https://blog.segmind.com/ip-adapter-xl-models-canny-depth-pose/ this one explain it well, but can i still use inpaint?

for instance i want a block to look like this or a part of an image not the whole of it

[steep phoenix]

ip adapter also works with Inpaint, try using it

[junior trellis]

yes i will thanks alot sir

[burnt totem]

how do i solve this? {arc a750]

[vocal burrow]

try using a python between 3.10.6 and 3.10.13

and rename comfyui_env folder to comfyui_env_back once you re done changing python

[burnt totem]

got it ,but there no exe for 3.10.13

[vocal burrow]

doesnt matter get 3.10.11

[burnt totem]

ok

[burnt totem]

after installing python 3.10.11, and renaming the env it asking for torch see after it finshes

[vocal burrow]

python 3.10.13 ? how did you install it on windows ? Also, logs please.

[burnt totem]

oh 11

my bad

[halcyon cradle]

First time user here. I followed the 6 day old guide on Reddit and was very patient with the downloads. Now I am in StableSwarm and I am trying to generate my first image. I have been waiting for over 10 minutes now and it still has not finished. Python is running in background and it is using A LOT of ressources.
There is no way this should take this long or am I mistaken?

[vocal burrow]

what s your gpu ?

[halcyon cradle]

My PC just froze and I had to restart. My memory was running on 98% because of python.
My GPU is AMD Radeon RX 5700

[vocal burrow]

what guide did you follow ?

[halcyon cradle]

The "How To Run SD3-Medium Locally Right Now -- StableSwarmUI", which is pinned on the reddit

[vocal burrow]

ok and what does the console log says ?

[halcyon cradle]

Last line is: User local requested 1 image with model 'sd3_medium.safetensors'...

[vocal burrow]

not the last line

the whole log

and I mean whole

put it in a .txt and drop said file in here

[halcyon cradle]

Alright. I will restart another generation and put the log in a text file

[old minnow]

sigh... trying SD3 API, again - "a full body portrait of a man laying on the grass" is okay, but "a full body portrait of a woman laying on the grass" is NSFW. what kind of sexism is that, for sake?

not to mention API itself is nowhere near capabilities we have when using local workflows with ComfyUI... is there any chance we'll get 4B weights, instead of that brain-damaged / poisoned 2B crap?

[soft fiber]

ur using sd3 large in that pic lol
sd3 as a whole used pretrained filters for nsfw/sexual content for the images it was trained on so whatever youre looking for, you wont find it in a 4gb model of sd3

[old minnow]

maybe my wording wasn't clear, so let me rephrase it:

1. NSFW filtering in API is idiotic
2. API itself (only way to access 4B / 8B variants of SD3) is very limiting and crap compared to local workflows
3. only SD3 model we can use locally (2B) is crap

[willow crest]

Hey guys. I paid for 1000 credits on the stability website, made a python script based on their example, and got it running. I can choose from sd3-medium, sd3-long, sd3-long-turbo, ultra and core, I can set the style preset for core, aspect ratio and so on all inline in the terminal.

However I noticed that sd3-medium results were coming back blurry. Not only that but they're not deducting credits from my account. sd3-large gens do deduct credits and come out sharp. Any ideas?

If you want to see my script I'll DM it to you, it does not include my key but it's also not really ready to be released into the wild.

^ Pics related (medium vs. large)

Obviously if they were sharp I would have kept my mouth shut about the credits 😆

[old minnow]

we should send SAI management, legal, and "safety" teams to any art gallery, so they have any idea of what is considered as safe / legal in modern society. or maybe even to elementary school, so they could see biology textbook and stop filtering out content like that

[willow crest]

Lol yeah.

Maybe not a matter for tech support but yeah.

By the way here are some gens from earlier, they came out sharp and I swear I didn't change anything. No blurriness in any of the models

[trail pumice]

hello guys, i tried several times to fix this problem but it didnt work, help please

[halcyon cradle]

After 10 minutes the process crashed and complained about "not enough GPU video memory"

[solemn atlas]

Hi, I have a question. Since I download a lot of checkpoints my hardware is full and I bought a new 4TB SSD. Does anyone one how I can clone all the SD stuff into the new SSD? Do I need a program to clone them? I didn't think checkpoints and all of that would take so much storage

[willow crest]

You on windows or linux

[solemn atlas]

on windows 11

I downloaded Macrium Reflect but it doesn't work now I try with Aomei. I looked it up, over 500GB are just my stable diffusion stuff

[willow crest]

https://www.supportyourtech.com/tech/how-to-clone-a-hard-drive-windows-11-a-step-by-step-guide/

Some popular options include Macrium Reflect, EaseUS Todo Backup, and Clonezilla.

[soft fiber]

you can simply copy and paste them onto the new drive, or cut and paste if you dont want two copies

[willow crest]

If you were on linux I would have just used dd lol

sudo dd if=/dev/sdx of=/dev/sdy

[solemn atlas]

I also did this at first, just copy&paste it but it didn't worked then I found out that I need a program to clone

[soft fiber]

what didnt work about it? curious

[solemn atlas]

Thank you I look it up

[soft fiber]

ohhh i see, you want sd to use both drives?

[solemn atlas]

I tried to do that a few days ago, just copy and paste it but when I tried to open it, it didn't worked so I delated everything from my new ssd. I want to clone them so I have storage on my pc. I can't download anything right know.

[solemn atlas]

no only one one, only on my ssd so i can delate it from my main drive

this is how it looks right know

[willow crest]

No idea what was happening to me earlier - I didn't repeat the exact same prompt but I'm getting non-blurry results now, and credits are deducting, guess it was a glitch

[solemn atlas]

I want to move everything from C to D and then delate everything from C

[willow crest]

Partitioning takes care of "deleting" everything on C

Unless you want it unrecoverable

[solemn atlas]

the last one right?

[willow crest]

Should be, don't do it yet though

[solemn atlas]

yes, I want to delate everything from C

[willow crest]

Oh, wait, you're on the clone page... I can't read German but that might be asking you about partitioning the new drive

[solemn atlas]

yes, it says on the right side

on the left it says "clone one hard drive to another"

[willow crest]

Okay well, you can partition the new drive D if you want, it makes no difference, but later when you're all done copying data over, you'll want to partition C too.

That should be a separate program in windows, partitioning (without cloning)

[solemn atlas]

I didn't even know windows has a program for that

[willow crest]

Ah, the correct term is "format", though it partitions as well

It should be in your start menu / programs list

[solemn atlas]

yes, it's this

[willow crest]

Yep!

[celest pulsar]

Hi all, I would like to ask why my sliders are square shaped while other people I have seen them round, is there any way to change this?
I've looked at themes and other than changing the background and colour I haven't found anything.

[vocal burrow]

typical of amd gpus, I d say drop swarm for now and use one of the AMD guide in the pinned messages

[vocal burrow]

depends of your os, browser, etc

[ornate elk]

Can you share the link? Want to check xD. And with your card I would advise following my AMD Guides from the pinned messages of this channel.
I don't have made a swarm guide but thats the next on my list.
Comfyui will work with SD3.

[vocal burrow]

https://www.reddit.com/r/StableDiffusion/comments/1de65iz/how_to_run_sd3medium_locally_right_now/

[karmic crown]

btw. see this https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/16030

[ornate elk]

Oh yea, but thats not even for AMD

[vocal burrow]

yup saw it :p when I tried it, it was still significantly slower than comfyui. First implementation so that s to be expected

[chrome mica]

@ornate elk I was able to install the SD-ZLUDA, thank you very much for the help!!

[queen marsh]

how do i fix no module named pip after i tried to install tourch with webui-user bat

[ornate elk]

Ah nice! Np

[burnt totem]

why am getting this man , anyone know
File "C:\Users\SYSTECH SERVICING\AppData\Local\Programs\Python\Python310\lib\site-packages\torch\cuda_init_.py", line 769, in current_device
lazy_init()
File "C:\Users\SYSTECH SERVICING\AppData\Local\Programs\Python\Python310\lib\site-packages\torch\cuda_init.py", line 289, in _lazy_init
raise AssertionError("Torch not compiled with CUDA enabled")
AssertionError: Torch not compiled with CUDA enabled
Press any key to continue . . .

i did installed this pip install torch==2.1.0a0 torchvision==0.16.0a0 intel-extension-for-pytorch==2.1.10+xpu --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
. {intel arc a750}

[ornate elk]

Thats caused by an third party antivirus Programm.

Yea, you need to uninstall python.
Them disable the AV Programm.
Then install python 3.10.11 64bit
Then delete the venv folder.
Then enable the AV Programm.
Then relaunch the webui-user.bat

[vocal burrow]

full log please

[burnt totem]

here

[willow crest]

Not just... missing pip? 🤔

[queen marsh]

wait isn't it 3.10.6

[ornate elk]

3.10.6 will work too. But its easier to just upgrade to 3.10.11 as it has more bugs fixed

[queen marsh]

thank you for your help it is now downloading

[burnt totem]

hlo

[vocal burrow]

did you follow the intel specific steps for comfyui install ? https://github.com/comfyanonymous/ComfyUI?tab=readme-ov-file#intel-gpus

[burnt totem]

ya

this one i followed

[vocal burrow]

I dont know then, you can try opening a ticket on their github

[burnt totem]

this is stressing man without nvidias gpus

[vocal burrow]

it can be more difficult with AMD gpus.... But with intel ones ..... it s way worse to setup

ease of use scales with the number of users.

[halcyon cradle]

Thank you. I will look into it

[ornate elk]

try SD.Next (vladmanics webui) as it supports intel + sd3

[ornate elk]

np, if you have any question feel free to ask

[burnt totem]

oh, defiantly gonna try thanks

[ornate elk]

Here is a link with Info and how to set it up:
https://github.com/vladmandic/automatic/wiki/Intel-ARC

[halcyon cradle]

yeah I have some.
In your guide it says that ZLUDA is faster and recommended, compared to DirectML. My issue is that I have a Radeon RX 5700, while your list of supported AMD cards for ZLUDA specifically lists the Radeon RX 5700XT. This brings up multiple questions:
Will it work with my RX 5700?
If not, does it make a significant difference if I use DirectML?

Oh and what do you recommend the most: Automatic111, ComfyUI, or Fooocus?

[ornate elk]

hey, good question.
As the 5700 is the same architecture and gfx version like the 5700xt it should work for it too 🙂
And yes it makes a significant difference between running it with zluda or directml!
I would suggest installing Automatic1111 with Zluda, as its the most stable and easiest to use and configure to your likings.

DirectML is slow and will run out of vram often. Zluda is faster and uses less vram so you can generate at higher resolution and use sdxl models too.

[halcyon cradle]

Great thank you, I will try and follow your guide then. I suppose I will be able to just delete my previous stable diffusion folder from the previous Swarm setup, to start fresh.
Oh and does sdxl only work with Zluda? I heard that the best general art models are 1.5 and XL, and if so, does that mean that I will be out of luck, if I have to use DirectML?

[ornate elk]

no problem, yea as people with vega gpus got it to work im sure it will work for your gpu easily too.
And nope, sdxl can work with directml too. but sdxl models are 6gb in size and do need a lot of vram and RAM to work without problems.
DirectML can work with sdxl but it will be very very slow and could crash when going out of the basic resolution

[burnt totem]

hy ive arc a750 8gb, dont i need to like med vram ? and others commands? and how do start after installation like by clicking webui.bat or creating own launcher?

[ornate elk]

It should use the right stuff automatically

Webui.bat

[burnt totem]

and the checkpoint folder?

[ornate elk]

What's with it?

[burnt totem]

i cannot find it

is in it suppose to be there

[ornate elk]

Its called stable-diffusion here

There your checkpoints go in

[burnt totem]

oh

got it

[burnt totem]

theres launcher.py and webui.bat, shoulnd i create custom environment like comfy D:
cd "D:\sd\orginal comfy UI\Comfy"
call comfyui_env\scripts\activate
python main.py --bf16-unet
pause this was my comfy ui launcher

its downloading again everything when i launched webui.bat

[burnt totem]

hlo?

[ornate elk]

It has already a custom venv

You see it in the screenshot

[burnt totem]

no i mean the launcher like after instalation, i want to create one in desktop which will launch sd.next

[ornate elk]

Right click the webui.bat and click ,"Send to" Desktop

[karmic crown]

btw. have you seen this? no need to rush with sd3 anymore

https://civitai.com/articles/5732/temporary-stable-diffusion-3-ban

[burnt totem]

thats not it, after i click it its downloading the same files like when i installed it, it isnt launching

[ornate elk]

It should be launching after installing the files

[burnt totem]

but i did install all of them once before, do ive to download everything every time i want to launch it?

[karmic crown]

do not mess with venv activation

call comfyui_env\scripts\activate

just run it directly

[ornate elk]

No it should do that only one time

[karmic crown]

comfyui_env\scripts\python something.py

[ornate elk]

Hes not using comfyui rn

[burnt totem]

but its twice man

[karmic crown]

python is python, no mather what 🙂

[burnt totem]

yah

two times

[karmic crown]

venv_name\scripts\python is a way to use python in that venv without activating it

btw. you can also do this

#!venv/Scripts/python

just replace the first line in something.py
and run that py file driectly

@burnt totem better use "Start in" then using cd somefolder

so you can actually remove half of that with those two things

[burnt totem]

oh. so like how comfy ui launcher works right, thannks man

[karmic crown]

when you create shortcut on deskop, go to properties

i am on mac, so i am finding images on google

i used this many times on windows before, to simplify my scripts

so if you put for example d:\folder in start in
that is same as cd d:\folder at the begining of the script

so you are sure that you are aleady in the right folder at start

[burnt totem]

ya i created thanks to you, by the way is there any command nedded like med_vram or for 8gb gpus bf 16 those?

[karmic crown]

so "D:\sd\orginal comfy UI\Comfy" in start in

and just run this

comfyui_env\scripts\python main.py --bf16-unet

instead of

call comfyui_env\scripts\activate
python main.py --bf16-unet

[burnt totem]

got it

[karmic crown]

https://github.com/lllyasviel/stable-diffusion-webui-forge/discussions/801

in case someone haven't seen this yet

[burnt totem]

any recommended command argument for sd.next?

[willow crest]

I'm still having issues with some gens that don't consume credits and come out blurry, if anyone associated with Stability or any experts could weigh in I'd appreciate it.

• This time I got blurry gens for sd3-large too, so it's not just sd3-medium like I thought it was
• Repeating the same seed and everything seems to repeat the blurry result (several minutes apart)
• No mention of soft focus, depth of field, etc. in the prompt
• The gen takes about the same time as always whether it's blurry or not
• 200 status every time, provided the prompt is clean

The images look like they were hit with a heavy uniform gaussian blur, see pic

[ruby yoke]

it's censoring them when it's doing that'

[queen marsh]

well i live in a kinda of power inefficient city so the power went out while it was loading by the time it came back its started showing me this i tried deleting the stable diffusion tab and starting again but when i tried to redownload pip it just lags out for 30 min and then shows me this , is there anyway i can restart from the beginning of downloading pip cause im acctually kinda sure it wont go off again for a while

[ornate elk]

hey, delete the venv folder and relaunch the webui-user.bat

[willow crest]

Thanks, makes sense. Gotta say, the filtering is more strict than midjourney

[ruby yoke]

oof - you probably have corrupt files. i would delete everything, uninstall python itself, and just reinstall fresh. turn the machine off and leave it off for a full minute after you delete stuff to make sure cache is cleared.

[willow crest]

I hit the filters for the phrase "wicked grin" earlier... remove wicked, presto

[ruby yoke]

slight smile has the same effect as wicked grin, usually

[willow crest]

It seems to be filtering anything with a woman in it. 4 attempts with this prompt ./imagine "peter mohrbacher, full-length, reptilian female, scaly" -m sd3-medium --ar 9:16

Guess it's still lifes only

I'll go generate a bowl of fruit and cry myself to sleep

[ruby yoke]

try just using the word 'she'

[ruby yoke]

what about "female alien"

[willow crest]

Great

[jade surge]

anyone knows how to add 2 diffrent characters ?

[willow crest]

I mean look, idk how crass I should be in a tech support thread, but these are not sexy pictures or anything with a sexual/objectifying bent to them.

[ruby yoke]

okay now try "female alien, blonde hair, ringlets, tan skin"

[willow crest]

I should not need a klingon forehead crest to prevent the filter.

[jade surge]

sadly bc it would be easier for me

[willow crest]

Good lord, it's almost the female form. I'm gonna get a ban for that

[ruby yoke]

now go from there and tweak the prompt. try your normal prompt, just put the word 'alien' at the end of it for example

[jade surge]

someone can help me with adding 2 diffrent characters ?

[willow crest]

What if I don't want aliens? My example was reptilians. They're both inhuman humanoids with female figures, what's the difference?

[jade surge]

hey you know maybe how to add 2 diffrent characters ?

[ruby yoke]

i know, but the token for alien seems to be disabling the filter

[willow crest]

Interesting. I thought it was a visual filter.

[ruby yoke]

just see how far down the prompt you can put it before it stops working, and it should just enhance the look of a reptilian character

[ruby yoke]

use SD3 or use photoshop

[willow crest]

This isn't working btw

Alien must need to be the subject, the subject is the first noun

[jade surge]

i am trying to make pic which needs 2 characters

[ruby yoke]

drat

[ruby yoke]

the only AI image gen that handles that well is sd 3. but it's a beast, i'ts not a "drop in a prompt and go" AI right now. so you'll probalby be better off to generate 2 images, one for each character, and then composite in photoshop

[jade surge]

the thing what i am doing needs 2 characters if you know what i am saying🤫

like idk how to make prompt to show me 2 diffrent characters with other clothing and ect

[ruby yoke]

i do know. go to the #🗣｜artisan-support-feedback channels, read the support channel i linked first, and use those channels with Ultra as the model. you should be able to prompt for that just fine

then say something like "a woman in a red dress and a man in a white business suit"

[jade surge]

kk i will try

eyes now are broken

[jade surge]

kk its working thx

[ruby yoke]

welcome 🙂

[jade surge]

😉 👏

[halcyon cradle]

Everything seemed to work out great until the very end ;C
No clue what this is trying to tell me, any ideas? This is the log of the webui-user.bat

Oh and what is step 2 for, if it adds a .zluda folder itself?

[ornate elk]

yes thats possible with the regional prompter extension

[ornate elk]

can yous show me the full cmd log?

[jade surge]

but how to use that ? you got any video ?

[halcyon cradle]

Its python -.- But did it not install the correct python version itself?

Just installed it, but for some reason it still says 3.8.10

[ornate elk]

you need to install python 3.10.11 64bit

enable "add to path" when installing

then delete the venv folder

and relaunch the webui-user,bat

[halcyon cradle]

is this option supposed to be here?

[halcyon cradle]

are you refering to the venv folder under Python310\Lib or Python38\Lib?

[ornate elk]

add pythoon to environment variables

[ornate elk]

venv inside the stable-diffusion-webui folder

[ornate elk]

@halcyon cradle did it worked?

[mortal compass]

I'm back bby

Are you familiar with comfy CS1o? My guess is yes 😄

[ornate elk]

a bit

[halcyon cradle]

Still on it. Takes sooooo long to download. But I did get passed my previous issues and the UI popped up in the browser. Thank you

[mortal compass]

Should be easy enough for you! I think I'm close to the answer, but not fully there yet. I have my loras on my D:drive and the rest on my C:drive. There is a yaml file called "extra_model_paths", and it looks like this, and this is where you can change the base path to your A1111, for example "base_path: C:\Ai_Root\Stable-Diffusion_Forge\webui". I'm not quite sure if I can edit anything else in here to change my lora folder?

It does say "#all you have to do is change the base_path to where yours is installed". I figured that is the only thing that can read the path, or am I taking it too literal?

[ornate elk]

okay good 🙂 remeber the first image gen takes 15-40 minutes

[halcyon cradle]

Spoke too soon...

[mortal compass]

This will probably not solve this specific problem, but never have spaces in your folder/file names! If it hasen't already created problems for you, it will 😄

[halcyon cradle]

https://tenor.com/view/kid-look-stare-awkward-guilty-gif-5289290

Is it safe to rename or will everything break?

[mortal compass]

At this point, I'm not sure actually. I'd wait for CS1o to get back to you

[halcyon cradle]

Even tried changing the model, but it still says it failed to load

[mortal compass]

I'm thinking, you can always change the name back later

If needed

[random pagoda]

i cant upscale anymore in auto1111 always get errors

*** Unable to load ESRGAN model E:\stable-diffusion-webui\models\ESRGAN\4x-UltraMix_Balanced.pt[00:11<00:12, 1.51s/it]
Traceback (most recent call last):
File "E:\stable-diffusion-webui\modules\esrgan_model.py", line 31, in do_upscale
model = self.load_model(selected_model)
File "E:\stable-diffusion-webui\modules\esrgan_model.py", line 49, in load_model
return modelloader.load_spandrel_model(
File "E:\stable-diffusion-webui\modules\modelloader.py", line 150, in load_spandrel_model
model_descriptor = spandrel.ModelLoader(device=device).load_from_file(str(path))
File "E:\stable-diffusion-webui\venv\lib\site-packages\spandrel__helpers\loader.py", line 41, in load_from_file
state_dict = self.load_state_dict_from_file(path)
File "E:\stable-diffusion-webui\venv\lib\site-packages\spandrel__helpers\loader.py", line 58, in load_state_dict_from_file
state_dict = self._load_torchscript(path)
File "E:\stable-diffusion-webui\venv\lib\site-packages\spandrel__helpers\loader.py", line 89, in _load_torchscript
return torch.jit.load( # type: ignore
File "E:\stable-diffusion-webui\venv\lib\site-packages\torch\jit_serialization.py", line 162, in load
cpp_module = torch._C.import_ir_module(cu, str(f), map_location, _extra_files, _restore_shapes) # type: ignore[call-arg]
RuntimeError: PytorchStreamReader failed locating file constants.pkl: file not found

[jade surge]

what i can do to make pictures looks better on stable diffusion ai ?

[mortal compass]

That's a very wide question. If you can be more specific I can try to help you out

[jade surge]

less buggy like blured eyes,half arm,no leg and etc

[mortal compass]

there's hundreds, if not thousands of lora's and textual embeddings, that improves or removes those issues. If you're not using them at this point, you really should get into it

[jade surge]

i am using only artstyle lora

[mortal compass]

You can also basically put anything in your negative prompt. Such as "blurry eyes, bad eyes, mutation, bad hands" etc etc

You can use more than one lora at the same time

the more loras you are using at once, the higher the chance is that the image breaks. But I usually use about 2-4 loras with no problems

[jade surge]

1 lora is enough for me

[mortal compass]

alright, enjoy wonky eyes and hands then lmao

[jade surge]

idk which lora i can add

[jade surge]

bc you said if more using at once then the higher the chances are that the image breaks

[mortal compass]

I also said that I usually use 2-4 for my img gens, no need to worry

Even if an image breaks, you can just remove the loras from the prompt easily to fix it

[jade surge]

kk but which should i add ?

[mortal compass]

There's tons and tons of checkpoints, loras, embeddings and more on civitai.com. https://civitai.com/search/models?modelType=LORA&sortBy=models_v9 here are some of the more popular ones, but you can basicaully just search for hands, eyes and stuff like that

[jade surge]

kk i have another guestion

[mortal compass]

shoot

[jade surge]

how to add 2 diffrent characters with diffrent clothing and diffrent poses

[ornate elk]

dont use SD3

it wont work in auto1111 for now

[halcyon cradle]

1.5 also did not work, or I might be using the wrong version

[ornate elk]

try any other model, for example dreamshaper v8

[halcyon cradle]

what is your most recommended model for brought art, mostly objects and fantasy creatures, that would work with this setup?

[mortal compass]

I'm usually generating one character at a time, so I'm not the best person to ask. But that is kind of advanced stuff if you're a beginner, I'd wait with adding more characters if I where you.
Something called "wildcards" can change the clothes and poses for example for you, and it's quite simple. You should look it up, I use them everyday

@jade surge

[jade surge]

you got any video how to use that ?

[mortal compass]

When I first learned about wildcards, I watched https://youtu.be/s-1L6MCVh-E?si=9d-bbWovkziZdS67 and thought it was decent. You only need to pay attention to:
what to install, what a wildcard is, how it works, and the synthax. Then you're good to go

[jade surge]

i don`t have anything better to do so i will try this

[ornate elk]

ouh for fantasy idk, but dreamshaper v8, or Unstable DIffusers are good fot that

[mortal compass]

It's very fun, you won't regret it. You won't understand how to use these now, but maybe for later. Here's a bunch of camera wildcards that I made, enjoy.

[ornate elk]

nope sry

[jade surge]

i am making mostly 18+ stuff on it

[mortal compass]

The ones I posted will work very well with all kinds of pictures, even explicit ones

[jade surge]

ok i see on this video that they don`t say anything about adding another character so i just need to make 2 diffrent wild cards and add it to prompts ?

[mortal compass]

As I said before, adding several people to a picture can be hard to make look good, and might need different loras or textual embeddings for it to work. You could try, instead of writing "1girl,solo", you can write stuff like "2girls,duo,clones,several girls," etc

Controlling what 1 character wears and how they look like is easy, but controlling 2 characters in the same image is hard to do

If you're curious about controlling 2 characters at once, go to civitai.com, go to images (or search for what you want) and start looking for images that has several characters that looks good. Check what their prompts are, settings, models and so on. That allows you to experiment with the same things

Good luck, I'm off to bed

[jade surge]

Kk

[queen marsh]

welp now its showing me this

[rose solar]

Hey there
Hope you are doing well
I am a senior full stack developer who has full experience in web and AI development
So if you have some projects, please let me know
thank you
https://github.com/typhon0130
https://figma.com/@typhon0130

[ornate elk]

What's your GPU?

[rancid nymph]

what is going on here

Startup time: 70.2s (prepare environment: 92.5s, initialize shared: 2.2s, other imports: 4.1s, list SD models: 0.6s, load scripts: 3.2s, create ui: 1.5s, grModel loaded in 1030.0s (create model: 2.4s, apply weights to model: 2.5s, apply half(): 0.1s, move model to device: 0.1s, load textual inversion embeddings: 0.4s, calculate empty prompt: 1024.3s).

and now its stuck at
0%| | 0/16 [00:00<?, ?it/s]

[ornate elk]

What's your GPU again?

[rancid nymph]

AMD Radeon 6600 XT

overclocked

[ornate elk]

Ah so your using the zluda version.

Is the generate Button grayed out?

[rancid nymph]

yes

could it have to do with the fact that my drivers updated

[ornate elk]

Yes zluda then needs to load again like the first time you set it up.
just let it load. It can take 15-40 mins. Like the first time

[rancid nymph]

like i had a feeling that its redoing the process when i first installed it and it had to generate the image slower

yeah

[ornate elk]

Thats exactly what's happening

It can happen when updating drivers or the torch version changes

[rancid nymph]

it took about 35 minutes to just load the model and calcualte the prompt

[ornate elk]

Okay, just let it load, I don't think it will take to long now if its already showing 0/16

[rancid nymph]

its been showing 0/16 for about 10 minutes

[ornate elk]

Ok

Still normal. It will take its time

[solemn atlas]

is it normal that my PC is really slow when generating and even Youtube doesn't work properly when it's generating

[ornate elk]

Depends on the webui the gpu and what settings you use

[noble belfry]

you can turn on/off some things in your browser to stop it using gpu. dont forget that AI uses a graphical processor, so its graphical things are busy. 🙂

[jade surge]

someone got lora that makes legs and arms less buggy ?

for pony

[ornate elk]

Anime or realism pony?

[jade surge]

anime

[ornate elk]

Because for anime I likely never get bugged limbs

Pony is to good with hands too

[jade surge]

cuz you are not making porn XD

[ornate elk]

Oh xD yea I dont

[jade surge]

so you got maybe ?

[ornate elk]

Nope, I don't think there are any Lora's made for legs or arms.
Maybe a negative Embedding could help

[jade surge]

what about eyes ?

[ornate elk]

Adetailer extension

Fixes faces and eyes

[jade surge]

its not working for me

it detects eyes on ass .-.

[ornate elk]

Hmm maybe try an other adetailer model

There are several made for eyes

[jade surge]

ik

[noble belfry]

what you can try to do, is use controlnet to assist SD in knowing what goes where. an AI can't really do hands because it thinks in terms of put this-close to that, not logical things like if the thumb is behind then palm then the forefinger cant be here. thats what you get weird deformities of impossible limbs. controlnet can sort of draw everything out first so the model knows where to put the details

failing that, if you like an image, you can use inpaint+masking to select the bad part of the image and get it to do a dozen or so do-over attempts and pick a better one.

but thats just playing dice 🙂

[ornate elk]

True Controlnet openpose can help

[little bough]

hi, any advice here, moved back to a1111 after using forge for months and woop, back to the issues. forge was fine but when i get near the end of my image generation with a1111, i am getting a pause in the percentage bar (last 4%) for about 20 seconds and then my computer is slowing right down with some ram issue i suspect) and then the image comes out. getting a notification i should add the '--no-half-vae' command in to my webui.bat but i dont recall using this in the past, any ideas?

[ornate elk]

Hey, what's your GPU?

[noble belfry]

a great combo are depth images and openpose. that gives it both where things should be and a little logic about where in depth. middling results with drawings rather than photos though.

[little bough]

3070ti .. i'll add i was using this install of a1111 fine before moving to forge after its release.. i've updated a1111 since though etc and gotten latest nvidia drivers

my command line just has these arguments set COMMANDLINE_ARGS= --xformers --medvram-sdxl

[ornate elk]

Okay good.
If your using anime VAE files or sdxl with the official sdxl VAE you need to add --no-half-vae to the webui-user.bat
The freeze comes from the VAE step at the end.
To fix it you can download the sdxl fp16 VAE file and Installing the tiled VAE extension

[little bough]

hmm, ok, im not using other vae's i just have the vae part set to automatic... is tiled vae extension the multidiffusion extension? i already have it installed and activated

i have vae as an option set to automatic

why does forge not have these issues?

getting this too

[fair oxide]

Is it possible to run SD3 on 4GB of VRAM?

[ornate elk]

Yes tiled diffusion is multidiffusion. But only enabled Tiled VAE and leave it on default settings

[ornate elk]

What model do you use?

[little bough]

leosam helloworldxl

[ornate elk]

Okay yea then you need the fp16 sdxl vae

You can download it from here:
https://huggingface.co/madebyollin/sdxl-vae-fp16-fix/tree/main

[ornate elk]

I don't think so

[fair oxide]

Ehh

[ornate elk]

You can still try it but only use the 5gb sd3 model

The 4gb one won't work without the clip text encoders

The 5gb one has them included

[fair oxide]

I dont think it will fit

Maybe later, after upgrade

[lost crescent]

I am using Face Fixer OpenCV (mikey nodes) and get this error - operands could not be broadcast together with shapes (1,4) (0,)

[broken tulip]

Eroor from Comfy. 'transformer_block' object has no attribute 'use_ada_layer_norm'

ANyone knows what that is?

[burnt totem]

File "D:\sd\auto11111\stable-diffusion-webui\venv\lib\site-packages\torch_dynamo\output_graph.py", line 675, in call_user_compiler
raise BackendCompilerFailed(self.compiler_fn, e) from e
torch._dynamo.exc.BackendCompilerFailed: openvino_fx raised RuntimeError: ShapeProp error for: node=%self_norm1 : [#users=1] = call_module[target=self_norm1](args = (%input_tensor,), kwargs = {}) with meta={'nn_module_stack': {'self_norm1': <class 'torch.nn.modules.normalization.GroupNorm'>}, 'stack_trace': ' File "D:\sd\auto11111\stable-diffusion-webui\venv\lib\site-packages\diffusers\models\resnet.py", line 691, in forward\n hidden_states = self.norm1(hidden_states)\n'}

While executing %self_norm1 : [#users=1] = call_module[target=self_norm1](args = (%input_tensor,), kwargs = {})
Original traceback:
  File "D:\sd\auto11111\stable-diffusion-webui\venv\lib\site-packages\diffusers\models\resnet.py", line 691, in forward
    hidden_states = self.norm1(hidden_states)


Set torch._dynamo.config.verbose=True for more information


You can suppress this exception and fall back to eager by setting:
    torch._dynamo.config.suppress_errors = True

how do i fix it

[lost crescent]

Try updating ComfyUI

[broken tulip]

It|s a fresh install from 1h ago

[burnt totem]

hlo what about my problem? any idea?

[lost crescent]

Sorry no 😦

[burnt totem]

oh ok then

[merry hearth]

On SD3 launch, people had trouble making human shapes with proper limbs, has that been resolved?

[regal slate]

Hey guys, I'm running into this problem with automatic1111 webui:

You are running xformers 0.0.23. The program is tested to work with xformers 0.0.23.post1. To reinstall the desired version, run with commandline flag --reinstall-xformers.

I tried running --reinstall-xformers but it gave me 0.0.23, not 0.0.23.post1

What should I do?

[ornate elk]

Hey, whats your webui version?

[vernal notch]

i signed up for stability assistant with gmail, but my discord isnt tied to it

is there a way to tie it to it or do i need to make a new account

[regal slate]

@ornate elk is this what you were referring to? thank you for any help 🙏

version 1.9.4...

this is my webui-user.bat commands currently:
`@echo off

git pull

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS= --xformers --autolaunch --medvram --update-check

call webui.bat`

[white tide]

Does anyone have a good workflow that I could simply use to SD3 medium? Becuase this really isn't it.

I can't get good images out of the box even with the example workflows provided by StabilityAI.

https://huggingface.co/stabilityai/stable-diffusion-3-medium/tree/main/comfy_example_workflows

And I heard that negative prompts doesn't work with SD3 Medium, so why is it in the workflows?

[ornate elk]

Looks perfectly fine. I think you can ignore the xformers 0.0.23 post1 message

[white tide]

As stated by Replicate.com at least:

[ornate elk]

I can share you mine later

It uses the 5gb model without clip text encoders

[regal slate]

oh ok, thanks, i noticed that the generation process has become slugglishly slow since receiving the message, is it normal? I have an RTX 4070 laptop GPU with 8GB Vram but generating SDXL images takes sometimes upwards of 3-5 mins

[white tide]

I don't mind. I'm sure it provides better outputs than what I get now. 😅

[ornate elk]

Oh an 8gb GPU needs --medvram-sdxl instead of --medvram
And then try deleting the venv folder to get a clean one when relaunching.

[regal slate]

thank you, I will try that 🙂

[ornate elk]

And make sure your extensions are updated

[white tide]

Worked almost good enough with a long prompt surprisingly

[trim crystal]

@ornate elk

[ornate elk]

Thats me

[regal slate]

seems it automatically went to the post1 version after deleting venv and reinstalling! thanks again

[ornate elk]

Perfect no problem 🙂

[burnt totem]

hy got this error File "D:\sd\auto11111\stable-diffusion-webui\venv\lib\site-packages\torch_dynamo\output_graph.py", line 675, in call_user_compiler
raise BackendCompilerFailed(self.compiler_fn, e) from e
torch._dynamo.exc.BackendCompilerFailed: openvino_fx raised RuntimeError: ShapeProp error for: node=%self_norm1 : [#users=1] = call_module[target=self_norm1](args = (%input_tensor,), kwargs = {}) with meta={'nn_module_stack': {'self_norm1': <class 'torch.nn.modules.normalization.GroupNorm'>}, 'stack_trace': ' File "D:\sd\auto11111\stable-diffusion-webui\venv\lib\site-packages\diffusers\models\resnet.py", line 691, in forward\n hidden_states = self.norm1(hidden_states)\n'}

While executing %self_norm1 : [#users=1] = call_module[target=self_norm1](args = (%input_tensor,), kwargs = {})
Original traceback:
  File "D:\sd\auto11111\stable-diffusion-webui\venv\lib\site-packages\diffusers\models\resnet.py", line 691, in forward
    hidden_states = self.norm1(hidden_states)


Set torch._dynamo.config.verbose=True for more information


You can suppress this exception and fall back to eager by setting:
    torch._dynamo.config.suppress_errors = True, any idea?

[ornate elk]

What's in your webui-user.bat?

And did you used the OpenVino Custom Script?

Also did you even installed the right webui version?

[burnt totem]

i followed this https://github.com/openvinotoolkit/stable-diffusion-webui/wiki/Installation-on-Intel-Silicon - https://www.youtube.com/watch?v=5X0RmlH6JI4 for the webui. COMMANDLINE_ARGS=--skip-torch-cuda-test --precision full --no-half . after it launched there openvino accelerator script needed, so while it was downloading my internet wasnt that good so it took like 5 turns to finsh it, was it corrupted? if so how i uninstall it?

[ornate elk]

You can delete the venv and relaunch the webui-user.bat
Then in the txt2img settings you need to scroll down at script's there select Accelerate with Openvino

[burnt totem]

got it

[burnt totem]

it downloading nvidias torch? Creating venv in directory D:\sd\auto11111\stable-diffusion-webui\venv using python "C:\Users\SYSTECH SERVICING\AppData\Local\Programs\Python\Python310\python.exe"
venv "D:\sd\auto11111\stable-diffusion-webui\venv\Scripts\Python.exe"
fatal: No names found, cannot describe anything.
Python 3.10.11 (tags/v3.10.11:7d4cc5a, Apr 5 2023, 00:38:17) [MSC v.1929 64 bit (AMD64)]
Version: 1.6.0
Commit hash: e5a634da06c62d72dbdc764b16c65ef3408aa588
Installing torch and torchvision
Looking in indexes: https://pypi.org/simple, https://download.pytorch.org/whl/cu118
Collecting torch==2.0.1
Using cached https://download.pytorch.org/whl/cu118/torch-2.0.1%2Bcu118-cp310-cp310-win_amd64.whl (2619.1 MB)
. shoulnd it download like torch==2.1.0a0 torchvision==0.16.0a0 intel-extension-for-pytorch==2.1.10+xpu --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/ intels xpu?

[ornate elk]

You have to run the torch-install.bat after that

That should install the openvino one

[burnt totem]

oh. ok

[ornate elk]

Ah no, it seems like that file doesn't exists anymore

Then open up a cmd and run
Pip cache purge

Then delete the venv folder again and relaunch the webui-user.bat

[burnt totem]

thers first time runner.bat?

[burnt totem]

ok

by the way

i wanted to install forge web.ui rather then auto11. is it possible with open vino like auto 111?

[ornate elk]

It could work yes

[jade surge]

how to make lora for sd ?

[ornate elk]

With Kohya_ss or OneTrainer

[jade surge]

do i need good gpu or cpu ?

i have 1060 gtx

@ornate elk

[ornate elk]

Yea you need at least 6gb vram.
On a 1060 it will take long

[jade surge]

you got maybe one ? XD

[ornate elk]

Im on AMD and need to test training first.
And on my old 1080 it was slow

[jade surge]

i can make you something for artstyle lora other then money bc i can`t pay with anything

[white tide]

Any chance you could share your workflow soon for SD3 Medium?

[ornate elk]

When im at home

[ornate elk]

You could also just remove the Clip text encoder. And connect the clip to the positive and negative prompt

Thats what I did and then select the 5gb sd3 model

[jade surge]

how to make prompt have 100% chance to show up ?

[ornate elk]

@white tide here is the SD3-Fast Workflow:

[noble belfry]

so SD 1.5 with SDXL images? 😄

does woo like "missing fingers" still work given we've been prompting that for ages? i wanna know who the jerk is whos been training hard on missing finger related things to make this still a 2024 prompt to manager 😉

[ornate elk]

i guess not for sd3

[small crescent]

anyone hear know how to use ipadapter

[ornate elk]

Yes

[lost silo]

Hi, I'm having a bit of an issue! Stable Diffusion 2 has been great for me so far, but I attempted to install the optimised stable diffusion package and suddenly I'm being hit with "modulenotfounderror: no module named 'taming'"

However, I've gone through and used "pip install taming-transformers" and it did it's thing but I'm still getting the error... any ideas?

[night locust]

Hello everyone, I've been using Stable Diffusion WebUI Forge installed through Stability Matrix, and it's been working great for a long time. Now when I try to start it though, it seemingly gets stuck on the startup and nothing happens. I haven't updated it or anything so I don't understand why this has started to happen all of a sudden. Here's the console output before nothing else happens:

Python 3.10.11 (tags/v3.10.11:7d4cc5a, Apr 5 2023, 00:38:17) [MSC v.1929 64 bit (AMD64)]
Version: f0.0.17v1.8.0rc-latest-277-g0af28699
Commit hash: 0af28699c45c1c5bf9cb6818caac6ce881123131

Not sure where I'd start troubleshooting this kind of issue. Any help would be appreciated.

[queen marsh]

intel iris pro graphics 6200

[quiet gust]

@ornate elk I was able to launch stable diffusion webui but I now have this problem

[crystal mantle]

Reinstalling Stable Diffusion after having it deleted for a few months, I am stuck with error code 2 no matter what I try I cannot get rid of it somebody help me. I've tried video tutorials and none of them have fixed it.

[ornate elk]

Can you show the full error log?

[crystal mantle]

I'm trying again if it fails I'll send it

[ornate elk]

Can you show the full CMD log?

[ornate elk]

Then you need to install the webui for Intel CPU and GPU.
https://github.com/openvinotoolkit/stable-diffusion-webui/wiki/Installation-on-Intel-Silicon

[ornate elk]

Hey, you can try delete the venv and relaunch it.
But Forge isnt recommended anymore for normal Users because with the next updated it will change a lot and won't work normal anymore

[chilly trail]

So. Vlad automatic just gen broken images. Tried reinstalling 3 times. I got one successful gen, but after that I just get images like these:

Checked all the usual settings. The command window said something about invalid sampler, but when I try to gen again, it doesn't give me back the same error(after changing sampler as well).

[ornate elk]

What's your GPU?

[chilly trail]

rtx 4080

I picked the Cuda/nvidia option during install.

[light glacier]

Does anyone know if it's possible to take a partially denoised latent from SDXL and then refine the last few steps with SD3? I've tried decoding the SDXL latent and re-encoding with SD3, but that just turns black.

[rich garden]

Anyone know what checkpoint/model Ideogram.ai uses?

[light glacier]

It'll be bespoke and have a custom workflow behind it all. Not something that can be run locally.

[lost crescent]

ComfyUI and TensoRRT - I built an SD3 TRT - sd3_clips_t5xxlfp8 - batch 1 - 1024 x 1024 - and using it gives this error = Given groups=1, weight of size [64, 4, 3, 3], expected input[1, 16, 128, 128] to have 4 channels, but got 16 channels instead

[light glacier]

4 Channels will be for SDXL and 16 for SD3, so you've got something mixed there.

[lost crescent]

How do I flush RAM and VRAM at all?

[ornate elk]

With that gpu I Would suggest using Automatic1111

SDNext is always a bit buggy and handles stuff different
You can find my install guide in the pinned messages of this Channel.

[hollow prism]

using automatic 1111 how to set a argument for a different path for the wildcards in the webui?

[little bough]

ok so i already have this but it has always been in my models folder so i guess it's never ever been used? my generations have come out fine (sd vae setting was on 'automatic' ) with all models i've used in forge/a1111 since last year that i've had the a1111 install running. I put the vae in the VAE folder and no difference in speed tbh. I've tried tried with adding "--disable-nan-check --no-half-vae" to my webui bat command line and that removed some of the aforementioned errors but no change in speed. I also changed relevant settings in A1111 to the same ones that I had setup in forge and still no change, tried with the tiled VAE extension on and no change. My A1111 is up to date and graphic drivers are the latest.. I'm now thinking maybe it's just an issue with the latest version of A1111 or something? I am not having these issues with Forge though I think the version of A1111 behind forge is a bit outdated now so that may in part explain it? Thanks

[fervent halo]

my thing gets stuck here. I already got python and set up everything (I think) but when I run the .bat file it just sits here. I left it for like an hour and nothing happened

[ornate elk]

Yea VAE files need to be in the models/VAE folder and have to be selectet in the VAE dropdown then.
The fp16 VAE + enabling tiled vae reduces the generation speed for sdxl+hiresfix by 20 seconds tested on amd and nvidia
I don't think its an issue with Auto1111.

[ornate elk]

Hey, what's your GPU and what's inside your webui-user.bat

[fervent halo]

lagging I cant get task manager up one sec

[ornate elk]

There is one / wrong before python.exe

[fervent halo]

💀

im blind asf

ty

[ornate elk]

You also need launch args based on your GPU vram

[ornate elk]

Also how much RAM do you have?

[fervent halo]

16g

I took out 16 a few months ago like a dumbass

and im too lazy to put it back

[ornate elk]

Put it back in, it will help a lot

[fervent halo]

I prob should its summer no excuses anymore

anyway

I realized the fowrad vs backslash didnt matter

[ornate elk]

Your using an AMD GPU. But you installed the webui for nvidia

[fervent halo]

cause in the origial screenshot it read the "/" as ""

[ornate elk]

So you can delete the whole thing

[fervent halo]

delete the whole like

one sec

[ornate elk]

The stable-diffusion-webui folder

[fervent halo]

gone

[ornate elk]

And then you can checkout the first link in the pinned messages of this channel.
There are all my AMD stable-diffusion install guides.
For your card the best one would be
Automatic1111 with ZLUDA.

[fervent halo]

holy crap my gpu is old I wanna upgrade now

[rich garden]

Ah damn.. at least I know now so thank you

[fervent halo]

right spot hopefully?

[fervent halo]

where would you recommend I make the SD-Zluda folder then?

[ornate elk]

Directly on C: drive for example

[fervent halo]

k

[ornate elk]

Or any other drive

[fervent halo]

I fucked up

I forgot to direct it to the folder

do I need to worry?

found it nvm

dw I just need to close the cmd

whoever made the instructions is a fucking genius

I would have said open cmd and then had them use cd this cd that to direct themselves to it losing half the people following it

[ornate elk]

I made these instructions xD and yea I wanted to make it as foolproof as possible

[fervent halo]

good shit

@ornate elk

what are the next steps lol

the error said it couldnt find the python directory do you want me to add that back?

[ornate elk]

You have to install python 3.10.11 64bit
And check "add python to path" when installing

[fervent halo]

oh I have it installed already

I just followed it and copied the path

[ornate elk]

Okay

Yea then it will work with the path you added

[fervent halo]

k

gg computer is in heart attack mode

ty for the help

[fervent halo]

oya that reminds me

ur pfp reminds me*

I need to continue watching one piece

[ornate elk]

Should add that to my guide xD

[fervent halo]

it took me like 3 months to get like 20 eps

I aint hooked yet

[ornate elk]

You started fresh?

[fervent halo]

naruto it took me until chunin exams and then I binged the hell out of it

[fervent halo]

all I know is bro is a rubber guy

and he cant swim

[ornate elk]

Oh wow

[fervent halo]

yup I aint read it

does this happen often?

where it just sits here

I lied im on ep 25

sorry im all over the place I'll slow down

[ornate elk]

Then you have a epic journey in front of you!
It takes a bit time to get hooked because of the old episode styles.
But its totally worth it to watch further!

[fervent halo]

some dumbass spoiled so much for me already

the gear parts or whatever

[ornate elk]

That happened to someone here before. But normaly it should continue

[fervent halo]

like gear 3 gear 5

idk what it does all I know is bro gets op

[fervent halo]

slow and steady

[ornate elk]

Nothing more to say ^^ its gonna be epic

[fervent halo]

alr

[ornate elk]

I think you will have forgotten about it until your actually reaching it

[fervent halo]

nothing I've watched has topped naruto yet

soo

I've rewatched it like 3 times atp 💀

its stuck again on the cmd imma js go get some food tysm for the help

if I get stuck again you'll prob get pinged cause im an annoying turd

[ornate elk]

No problem, sure let me know if any step isnt clear enough. Some parts can be a bit tricky

[fervent halo]

I feel stupid for not being able to install i

[ornate elk]

Nah, its fine you would have been on the right track if you had an nvidia GPU.
AMD needs some extra steps to get the best performance out of it

[fervent halo]

its still on the same part 💀

[little bough]

hmm, well I looked up something and apparently it is just A1111 being not as good as Forge, it had been so long since i used a1111 that i forgot how inferiro it was anyway I saw a suggestion to set the VAE to TAESD and that seems to have cured the lag at the end, speed seems similar to forge now! I didn't know about the tiled vae trick that you mentioned before so i should just use that all the time now by default? Thanks

[fervent halo]

YO ITS OPEN

IT TOOK SO LONG TO INSTALL I FORGOT WHAT I NEEDED IT FOR

[ornate elk]

Enable tiled VAE anytime you upscale with hires fix and sdxl.
And use the fp16 sdxl vae. Also remove --disable-nan-check it won't fix anything.

[fervent halo]

oh question

@ornate elk

do I need to leave the cmd open while I have the localhost running

I'd assume so

[vocal burrow]

yes

[little bough]

ok thanks and does one lose anything from using the TAESD VAE?

[vocal burrow]

also yes

[vocal burrow]

setting TAESD for preview will only impact the previews, so no. You'll just get imprecise previews.

[little bough]

sorry i meant here

[vocal burrow]

you'll loose some little details then. It's your call to decide if it's worthy.

do some comparaison with fixed seed, similar prompt+settings

eg : https://github.com/madebyollin/taesd

[little bough]

hah wow, just did, quite a different composition, only expected a quality difference but the picture is quite different

thanks

[vocal burrow]

did you set a seed ?

[little bough]

yup, exact same settings just the full/taesd option toggled

[vocal burrow]

can you show both results ?

[little bough]

sorry i was trying to figure this out as something weird is going on, i was testing with a gibberish text prompt 'yryt' and swapping between the 2 VAE modes and i got a different image, then i tried a more proper prompt like 'a bridge in california' and swapped between the 2 VAE's and as you said there are just minute differences between the 2 but the same picture and now I am trying to swap between the 1st example i tried but it is giving a similar picture again and i can't recreate the difference so maybe I swapped a setting somewhere despite the 2 prompt/settings files being exactly the same except for the VAE/SD switching so yeah... i have probably touched another setting that i can't recall but is not showing in the text file that has changed the composition

what do y'all think of the "VAE Checkpoints to cache in RAM" any reason to enable this, any caveats etc?

[vocal burrow]

it will store xxxx vae in ram instead of having to load them from the ssd/harddrive

so if you keep switching VAE it will help a little during the initial loading time, but will use more of your ram (not vram). Note that the VAE will still be sent to GPU's vram. It's not "keep it in ram instead of vram", it's "keep it in ram before sending it to vram"

[karmic crown]

"but will use more of your ram (not vram)" - unless you they are on Mac, than it will use their unified RAM in both cases 🙂

[vocal burrow]

yes ofc cause mac does not get proper vram :p

[karmic crown]

which is great when you have 192GB of unified ram, so its like having 128GB of VRAM 🙂

[vocal burrow]

of way slower vram* :p

(just pulling your leg)

[karmic crown]

the real problem are missing cuda cores, not speed of vram 🙂

but ok

[ivory slate]

Hello does anyone know how to fix this issue when i click generate some blue scrambles comeout then the result becomes grey. I've tried other models and its exactly the same

[karmic crown]

hm... this is strange. i saw that problem only on Mac. can you try to use this

i know what is the reason for that on mac, but it should not happen on windows

what is your gpu and what your parameters in webui-user.bat are?

[vocal burrow]

what's that model ?

[ivory slate]

its 3060 but what's parameters in webui-user.bat

[ivory slate]

its a random one i found from huggingface

[vocal burrow]

link please

[ivory slate]

one min

[ivory slate]

https://huggingface.co/alvdansen/Painted-illustration/tree/main

[vocal burrow]

yup that's not a sd model

[ivory slate]

whats the difference

im new to this

which ones should i downlaod

[vocal burrow]

sd models are at least 2gb

this is most likely a lora (aka "a patch for a model")

[ivory slate]

oh ok

so i should download the ones 2gb and more

because yea ive notices that small one dont work

[vocal burrow]

yes you should download a proper sd model.

[ivory slate]

but large size work

[ivory slate]

ok thanks for the help

[vocal burrow]

like I said anything under 2gb is not an sd model

[ivory slate]

yea

[karmic crown]

and try to download "pruned" fp16 versions

smaller filesize

[vocal burrow]

eg of a sd1.5 model https://civitai.com/models/4384/dreamshaper

[ivory slate]

got it thanks👍

[ivory slate]

this one is nice

[ornate elk]

@fervent halo let me know if it worked 🙂

[fervent halo]

It did thanks

[slender yew]

anyone there

can anyone please help me

i am in a big problem

[ornate elk]

What's the issue?

[slender yew]

i had a problem with stable diffusion 1.5 now a bro told me a solution i am trying it bro

[jade surge]

how i can add shading to pictures ?

[honest tendon]

is it possible to download Fooocus after already having Comfyui still in use?

in other words I wanna use comfyui and fooocus simultaneously

[rich garden]

what checkpoint is this?

[jade surge]

isn`t it character lora /

[rich garden]

yeah it is but i wanna know what checkpoint was used to make it

[jade surge]

from where you got this pircture ?

[karmic crown]

do you have civitai link for this image?

it should have all the information if it is from civitai

[ember bough]

How can I use ComfyUI on my RX 580 8GB?

Or any other image generation UI

[jade surge]

you got 580 wtf

[rich garden]

it says "pony" but the base model says "SDXL 1.0" https://civitai.com/models/370998/pony-xl-10-yorha-2b

[karmic crown]

found it

https://tensor.art/images/710006923795253407

[ember bough]

yeah...

heard about some zluda stuff but don't know if my gpu is compatible

[jade surge]

SDXL 1.0

[rich garden]

i thought pony was like more 3d realistic tbh

cuz this looks like 2d anime

[jade surge]

AutismMix SDXL

its for pony

[karmic crown]

check the link i sent

[rich garden]

yeah i couldn't seem to find the model

[karmic crown]

you have all parameters there

[rich garden]

yeah i did thanks

[jade surge]

zluda wtf ?

[rich garden]

i don't think my laptop is powerful enough for pony cuz it'll always freeze when generating

[karmic crown]

CS1o iz Zluda expert here

what is your GPU?

[ember bough]

RX 580 8GB

[karmic crown]

https://github.com/CS1o/Stable-Diffusion-Info/wiki/Installation-Guides#amd-automatic1111-with-zluda

[ember bough]

man im new to this stuff. please

[jade surge]

what do you want to know ?

[ember bough]

yeah but is it compatible with my gpu

[karmic crown]

i am not sure, thats why it is better to wait for CS1o

[ember bough]

i'm trying to use my gpu instead of my cpu

[jade surge]

i don`t think you can

[ember bough]

i tried to create one using directml but it took 30minutes

[rich garden]

i just don't see how this can be a pony image when it looks like anime 2d cuz i thought that pony was more realistic 3d right?

[ember bough]

i guess it used my cpu because an error

[jade surge]

then just say in negative prompt 3d

[rich garden]

oh

[ember bough]

ton of tutorials that says different things and i'm really confuseed

[jade surge]

and source_anime