🤝｜tech-support

[vocal burrow]

no, just the whole txt2img page

[brave frigate]

those are my setting

[vocal burrow]

ponydiffusion recommends using clip_skip 2.
What kind of result do you get with those settings because they look ok to me

[fair oxide]

i use clip skip 2. let me write a prompt and see

ill keep it classy

[vocal burrow]

looks good to me ?
you're wasting resources by letting hires steps at 0. As 0 means "use the same ammount of steps as for the low res pass and when upscaling/hires. fixing you usually need way less (10/15 is enough most of the time)
Also you're not using a specific VAE here ?

[brave frigate]

well those are my vae i tried

[primal glen]

trying to get inpainting to work. What am I doing wrong here?

[brave frigate]

anythingModel just crashes everything and freezes my pc

but its with every model

they just either dont do anything or crash it

[vocal burrow]

sdxlVAE_sdxlVAE.safetensors should work
klf8 , vaeftmse8400000 vaes won't work for sure
the other ones I don't know

basically like I said you can't use vae designed for sd1.5 with sd2, sdxl, ponyxl models

[brave frigate]

so i need to get one for sd2

i'll try

[vocal burrow]

example

[vocal burrow]

if you're using sd2 or sd2.1 models sure.... Pretty unlikely tho, you're probably using sd1.5 or sdxl models.

[vocal burrow]

in this screenshot you're using ghostmix => sd1.5 => every vae in your list should work appart from sdxlVAE

[brave frigate]

where could i find the one that im using?

[vocal burrow]

it should be stated wherever you downloaded it from

[brave frigate]

ah well with this one i used ponydiffusion

and i tried the sdxl_vae but it just crashed again

[vocal burrow]

can't say without seeing the settings you've used but it looks like you're using img2img with a mask ..... and probably have not defined any mask

[brave frigate]

CLIP is at 2

[vocal burrow]

show the console log, copy paste the whole thing in a file and drop said file in here

[primal glen]

I drew a stick figure in paint and drew white on the stuff I want to be the torso.

[vocal burrow]

what's your gpu ?

[primal glen]

here's the settings I used

[brave frigate]

20GB AMD Radeon RX 7900 XT Hellhound

[vocal burrow]

ok, can I still see that console log please ?

[fair oxide]

is this correct

ever since i have been trying to generate images ive been crashing a lot

[vocal burrow]

depends, what's your gpu ?

[fair oxide]

nvidia 3060

[vocal burrow]

you can probably drop --medvram-sdxl then

[fair oxide]

alrighty

[brave frigate]

sry my pc just froze

[vocal burrow]

can you show the whole page and console log please ?

[brave frigate]

here

[fair oxide]

thank you ary

[vocal burrow]

the whole log please

[brave frigate]

there's not just this one in it

the last two crashed and i used a vae

[vocal burrow]

try this one https://civitai.com/models/140686/fix-fp16-errors-sdxl-lower-memory-use-sdxl-vae-fp16-fix-by-madebyollin

[fair oxide]

took me awhile hahah sorry

[vocal burrow]

sdxl model with sd1.5 vae

like I said

[fair oxide]

okay. i think i see what you're talking about

what do i need to change? my VAE or my sd version

[vocal burrow]

most likely if there's "xl" on the left you want something with "xl" on the right

[fair oxide]

probably whatever i want to change

ah i see

[primal glen]

@vocal burrow I exited the window. What information are you looking for?

[vocal burrow]

just keep the same architecture across the model and vae.
If you're using sd 1.5, keep everything in your pipeline on the sd1.5 architecture. Be it controlnet models, vae, etc
if you're using sdxl model => keep everything sdxl
if you're using etc => etc

[brave frigate]

on it rn

hmm now its just very blurry

[fair oxide]

i see

[vocal burrow]

model, vae, the footer containing information abour your SD setup, any error showing up in the console, etc. Because while those testing will probably give you bad results, they should still work.

[fair oxide]

so 1.5 architecture would ssay something along the lines of sd1.5 and for sdxl it would say sdxl? in the file name?

[vocal burrow]

if the dude who made the file is a nice guy, yes.

[fair oxide]

hahahah

[vocal burrow]

If not, welp just read the description / readme / etc that comes with it/its download page

[fair oxide]

gotcha

okay

sweet. you are awesome thank you

imma generate it with a new VAE see what hapopens

[vocal burrow]

I'm about to go to sleep, it's 3am here.

good luck

[fair oxide]

awesome. thank you so much

[vocal burrow]

And yes David I saw you've been struggling with it for some days (didnt have much time to hang in discord for the past weeks/months°. I'll try answering tomorrow if I see you posted screenshot / console logs.

[primal glen]

good night. I'm booting Stable Diffusion back up, I'll post screenshots with that information. 🙂

[vocal burrow]

simply screenshot the whole webpage

and for the console log, copy paste everything into a file and drop the file in here

[fair oxide]

ary it;s 10000x better THANK YOU 😭

[vocal burrow]

Same thing, show webpage screenshot please

(pc is shut down, so it should be my last minutes in here)

Cause it looks like you zoomed in the picture

[brave frigate]

it was the same error code as before

i put the hires steps to 1 which was the reason

basically it now uses the vae but nothing actually changes

its 3:27am for me too so im also going to sleep

[vocal burrow]

Hires step at 1 means it s gonna upscale in only one step

That s not enough. At the very least you want 5. 10 should be ok.

Also not sure what you were trying to achieve initially by using a VAE. But yes a VAE is not gonna change your image drastically, it s mostly used for color correction and tiny details.

Good night

[quaint crag]

how do i make stable diffusion available on lan? im running it on ubuntu-server in a vm so i cant access it also i used easy diffusion to install so i dont have or its somewhere but no idea where webui-user file

[lime stag]

hello everyone, is anyone aware of a fix for the img2img generation tab no longer working? i did a git pull update and still no luck, just get a "error connection errored out" code after loading a reference image and clicking generate

[foggy arch]

why did my Loras suddenly disappear from my list? the files still exist but SD isnt registering them, how do i fix?

[terse lichen]

is there to connect comfyui with another driver if one driver is full ? like a path link for comfyui to understand that others models are kept there

[vocal burrow]

Displayed loras depend of the loaded sd model architecture.
If you loaded a sd1.5 model it s only gonna show up sd 1.5 loras. If you loaded an sdxl model, sdxl loras, etc

[vocal burrow]

Yes, check extra_model_paths.yaml.example file

[vocal burrow]

Need to see the full console log to help

[vocal burrow]

--listen command line argument

[little breach]

what cfg and sampler does the sd3 api use?

[terse lichen]

link pls

[vocal burrow]

should be in your comfyui folder

https://github.com/comfyanonymous/ComfyUI/blob/master/extra_model_paths.yaml.example

[vocal burrow]

same things as in previous sd. cfg <=> how close the AI should follow your prompt vs take some liberty with it. sampler <=> what algorithm is used to make sense from the original random noise (ish)

[little breach]

no I mean what values is the api using for them?

cause it is not configurable

[terse lichen]

i dont see any of the extra model paths in my comfyui folder for some reason

[vocal burrow]

I ve shown you mine, show me yours if it s different

[lime stag]

How do I pull that up?

[vocal burrow]

console log is the text displayed in white in the black background window thingy (the console/cmd)

so you just select the whole text, copy paste it in a file, drop said file in here

[terse lichen]

found it

[whole wyvern]

Hi, Im trying to make a lora for one person for sdxl, i made it using autotrain from huggingface, when i select it in fooocus it doesnt seem to affect the result. Any idea how i can fix that?

[sage scroll]

Hi, i am using stable diffusion with the replacer extension. When i try to do anything with animate diff, its always creating artifacts (see picture). Frame by frame is working and its not because of control net or anything. Anyone got an idea to fix it?

[lime stag]

Already up to date.
venv "D:\AI\stable-diffusion-webui-forge\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: f0.0.17v1.8.0rc-latest-276-g29be1da7
Commit hash: 29be1da7cf2b5dccfc70fbdd33eb35c56a31ffb7
Launching Web UI with arguments: --xformers --no-half-vae --pin-shared-memory --cuda-malloc --cuda-stream
Using cudaMallocAsync backend.
Total VRAM 11264 MB, total RAM 65464 MB
WARNING:xformers:A matching Triton is not available, some optimizations will not be enabled.
Error caught was: No module named 'triton'
xformers version: 0.0.23.post1
Set vram state to: NORMAL_VRAM
Always pin shared GPU memory
Device: cuda:0 NVIDIA GeForce RTX 2080 Ti : cudaMallocAsync
VAE dtype: torch.float32
CUDA Stream Activated: True
Using xformers cross attention
ControlNet preprocessor location: D:\AI\stable-diffusion-webui-forge\models\ControlNetPreprocessor
Loading weights [0e9b3b69c3] from D:\AI\stable-diffusion-webui-forge\models\Stable-diffusion\LunarDiffusionXL-1.4.safetensors
2024-06-03 0329,109 - ControlNet - INFO - ControlNet UI callback registered.
Running on local URL:

To create a public link, set share=True in launch().
model_type EPS
UNet ADM Dimension 2816
Startup time: 14.1s (prepare environment: 3.5s, import torch: 4.7s, import gradio: 1.2s, setup paths: 0.7s, initialize shared: 0.1s, other imports: 1.0s, load scripts: 1.7s, create ui: 0.6s, gradio launch: 0.5s).
Using xformers attention in VAE
Working with z of shape (1, 4, 32, 32) = 4096 dimensions.
Using xformers attention in VAE
extra {'cond_stage_model.clip_l.text_projection', 'cond_stage_model.clip_l.logit_scale', 'cond_stage_model.clip_g.transformer.text_model.embeddings.position_ids'}
To load target model SDXLClipModel
Begin to load 1 model
[Memory Management] Current Free GPU Memory (MB) = 10094.996185302734
[Memory Management] Model Memory (MB) = 2144.3546981811523
[Memory Management] Minimal Inference Memory (MB) = 1024.0
[Memory Management] Estimated Remaining GPU Memory (MB) = 6926.641487121582
Moving model(s) has taken 0.34 seconds
Model loaded in 4.9s (load weights from disk: 0.8s, forge load real models: 3.5s, calculate empty prompt: 0.6s).
1006

Deleted the running on local URL line btw

and there's the error box

[vocal burrow]

just to be sure.... you re not closing the window once it s done, are you.

[lime stag]

nope

[vocal burrow]

also with my haxor power I m gonna guess that your local ip is 127.0.0.1:7860 (that s the local ip so it does not matter)

[lime stag]

the img2img was working 2 weeks ago or so

[lime stag]

very close

[lime stag]

there shouldnt be an issue that im running the amd python would there be? even though i have an nvidia card

it used to work

[vocal burrow]

the amd python ?

you mean python 64 bits ?

Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug 1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)] just means you re running python 64 bits. ''AMD64" is just the 64 bits standard name, it can be run on both intel and amd cpu. And I do mean cpu, python itself is not running on gpu.

[lime stag]

ah got it!

thank you

any insight into why the img2img isnt working for me

[vocal burrow]

try to whitelist your webpage if you re using any adblockers

[sage scroll]

There are also no errors in the console. Also, in the frames folder the first frame is what i want. But when its making the actual video, its just giving up and completly corrupting the video

[lime stag]

turned off my ad blocker and anti virus, still no luck

[vocal burrow]

nothing shows up in the log apart from what you ve shown here ?

[lime stag]

weird cause txt2img works fine

[pseudo prairie]

how do u creat better images with ai

its always the face

like everything is great

until its the face

idk how people generate such good images

also is it possible to re-do a specific part of the already generated image

[lime stag]

you can use the negative prompt and same seed

to alter an existing image

also hires. fix with refining and upscaling helps a lot

[gilded nacelle]

Anyone here able to get me going in the right direction for AMD ?

[open fox]

The pinned note i guess

[gilded nacelle]

Doesn't feature a link for process guidance, as im a computer noob 😂 I've always used apps to gen

[open fox]

Pretty sure the message is a process guidance

[gilded nacelle]

Hmm, i didn't see that, I'll check again

[open fox]

Messages in this cannel. Step by Step procedure how to get ZLUDA to work

[gilded nacelle]

Found it thanks

[open fox]

Well if you would not have found it, it would be for the best not to install zluda or pytorch etc. on yourself 🙂

[vast crest]

Hi, guys I can't create a Lora Pony Xl model because it always gives me this error: RuntimeError: Caught RuntimeError in DataLoader worker process 0.
Original Traceback (most recent call last):
File "/usr/local/lib/python3.10/dist-packages/torch/utils/data/_utils/worker.py", line 308, in _worker_loop
data = fetcher.fetch(index) # type: ignore[possibly-undefined]
File "/usr/local/lib/python3.10/dist-packages/torch/utils/data/_utils/fetch.py", line 51, in fetch
data = [self.dataset[idx] for idx in possibly_batched_index]
File "/usr/local/lib/python3.10/dist-packages/torch/utils/data/_utils/fetch.py", line 51, in <listcomp>
data = [self.dataset[idx] for idx in possibly_batched_index]
File "/usr/local/lib/python3.10/dist-packages/torch/utils/data/dataset.py", line 348, in getitem
return self.datasets[dataset_idx][sample_idx]
File "/content/kohya-trainer/library/train_util.py", line 1189, in getitem
example["latents"] = torch.stack(latents_list) if latents_list[0] is not None else None
RuntimeError: stack expects each tensor to be equal size, but got [4, 152, 104] at entry 0 and [4, 144, 112] at entry 2

steps: 0% 0/320 [00:01<?, ?it/s]

[dense jewel]

how to setup

[ornate elk]

Hey, if you have any questions about the Guide, let me know

[ornate elk]

Check the pinned messages of this channel for install guides.

[gilded nacelle]

Cool, I truly appreciate it. I'm currently at work, so i will investigate it when i get home. It will definitely be a few hours from now.

[ornate elk]

You got an connection error.
You may need to use a VPN or disable it if you used one

[zenith sail]

Anyone know why the model class is breaking after updating?

[sinful needle]

how do i go about getting "turbo" sample scheduler? i keep seeing it on Civitai generations but i cant find the actual file

[maiden totem]

Hey, I’m new in this community, I wanted to ask how do you get stapled diffusion on your phone? I’ve been trying to do it multiple times and it didn’t work. I’ve been trying to do a tutorial on YouTube on how to do it and it didn’t help at all so I wanted to ask how do you get stable diffusion on the phone

[cosmic furnace]

Some good news ? About zluda or amd?)

https://tenor.com/view/good-news-gif-20210278

Improvements, achievements or something?)

[bold tangle]

hey i am new to sd i was wonder does anyone having the same problem? when i click copy image to sketch or inpaint or inpaint sketch under img2img the photo become purple (looks like there is a mask on top )but everything is fine with i copy from text2img to img2img

[humble sand]

for some reason when i generate any image no matter what checkpoint i am using it is making abstract art and not what i actually want, ive tried copying everything down to the prompts from the example images and nothing works anyone know how to fix this

[plush maple]

S

[humble sand]

s?

[primal glen]

https://pastebin.com/Yg0bmQah

[primal glen]

looks like it's doing SOMETHING with the image, but only about a quarter of the time is it anything close to what I asked for

[fair oxide]

https://www.reddit.com/r/StableDiffusion/comments/1d6tvh0/vram_help_nvidia_3060/

posted this about my vram issues. i ahve a good GPU idk why im having such a hard time. i should be fine with a nvidia 3060. based on the comments, what should I do? is comfyUI or forge good to switch to?

it seems to be happening whenever i use hires.fx

will using max_split_size_mb do anything?

currently have set COMMANDLINE_ARGS= --xformers --medvram-sdxl --opt-split-attention set. someone with the same graphics card said they run this and it works but im not sure I have the same results. yeah, it just crashed on me. used 30 steps at base 512x1024 and used hires.fx to upscale it to double that and it crashed

[timber oak]

Hi

i turned on cross compatibility in webui a while ago

so sdxl models show non sdxl loras aswell

but now i have too many loras

and i wanna turn it off again

so only loras compatible to my model show up

[fair oxide]

OutOfMemoryError: CUDA out of memory. Tried to allocate 2.00 GiB. GPU 0 has a total capacty of 12.00 GiB of which 0 bytes is free. Of the allocated memory 10.91 GiB is allocated by PyTorch, and 600.00 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

Time taken: 2 min. 42.8 sec.

A: 11.42 GB, R: 12.50 GB, Sys: 12.0/12 GB (100.0%)

ive tried a lot of combos for my set COMMANDLINE_ARGS based on the reddit post but nothing seems to work. it runs out of memory without any edits to that as well. they said they dont have many issues with their 3060 and that their versions of their set COMMANDLINE_ARGS seems to improve their performance but it has not been working for me

[gilded nacelle]

@ornate elk are you available to help with AMD path editing

[humble sand]

if anybody could help me with this I’d big appreciate it

I’ve been trying all day to fix it and nothing is working

[fair oxide]

Make sure your VAE matches your checkpoint

If a checkpoint uses …XL match it with a …XL vae

Or if you have a …3.5SD use a …3.5SD VAE

[humble sand]

could you explain that a bit more my apologies I haven’t used sd in months and wasn’t too familiar last time I used it

[fair oxide]

Your checkpoint is pretty much your baseline for evrything (I believe). Your VAE is basically a tool that enhances your image’s quality

If your VAE doesn’t match the version of the checkpoint, it won’t look good

And if that’s not your issue, check the compatibility of your Loras

[humble sand]

Thank you, I haven’t been using any loras yet because I couldn’t figure it out where do I switch the vae if you don’t mind me asking

Or is it another thing I need to download?

[fair oxide]

Go to settings and look up SD VAE

You can download VAEs on civitai.com

[humble sand]

YOURE literally amazing thank you I’m about to try right now I’ve been lost on this all day 😭

[gilded nacelle]

each model has its own vae to download as well

[humble sand]

How do we find those specific ones I’ve read the descriptions of mine a few times and can’t seem to see anything there I’ve been downloading them on civit I don’t remember doing this all months ago the first time I had SD

Like I’m trying to use HadrianDeliceXL for example and it says base model is Pony so I was thinking maybe that?

Also sorry for all the questions just trying to learn and couldn’t find much when I tried to search 😅 I appreciate all the help

[gilded nacelle]

when you look at the area to download model, directly below it, will be a blue info icon that says "this checkpoint recomends a VAE, download and place it in the vae folder" vae will be highlighted blue, click, download, move to vae folder, found next to where you place your models

[humble sand]

okay amazing I’m doing this all right now

you guys are the best

[gilded nacelle]

glad to help, now im just waiting for help of my own🤣

[humble sand]

I wish I knew how to help lol unfortunately I’m trying to relearn all this because I somehow managed to delete stable diffusion off my pc like 9months ago and now I’m redownloading everything again 😭

[gilded nacelle]

im usimg amd anyways

[fair oxide]

I still have a lot of issues on my own. I hope one of the smart dudes can help me hahaha 😵‍💫

[humble sand]

no i super appreciate the help though like ive been super frustrated over this lol

[vocal burrow]

I had some similar results with some model that required a very specific vae. check the description of whatever model you're using to be sure

[fair oxide]

should i change my VRAM usage polls per second? i am very confused as to why i have so much trouble with m VRAM

i should be pretty good but i haven't found any good solutions yet

and can switching to forge/comfyui help that?

[formal trout]

you are generating in 136 x 160, SD cannot draw much in that resolution, you should generate at least in 512x512 for SD 1.5 or 1024x1024 for SD XL

[fair oxide]

Should I have a base sample generation of 512x512 and then upscale it to 1024x2048? I'm using pony

[formal trout]

I think pony is SD XL (but check your model), so you should generate in 1024x1024, and then upscale

or other recommneded resolutions for SD XL like 832x1152 or 1344x768

anyway if it is a SD 1.5 and you ask for 1024x1024, I think you would get out of VRAM error or something

[fair oxide]

hm okay. i think my problem is just me using upscaling wrong. i think i might be making huge images somehow

so does that mean i cant do 512x1024?

i want a taller image. is that possible or do i stick to the 1344x768 versions instead

or 832x1152, etc

[formal trout]

there's people that say that you can try resolutions like that, I like to stick to recommended resolutions as I'm biased to think they generate better (but it is hard to prove)

[fair oxide]

okay. and are the 832x1152 interchangable or are they locked in their respective horizontal/vertical dimensions?

[formal trout]

no, here are the list, you can install that extension for Automatic1111

[fair oxide]

this is a dumb question.. but is width on the left

[formal trout]

https://github.com/xhoxye/sd-webui-ar_xhox

[fair oxide]

wait

it is

im dumb

[formal trout]

yes

[gilded nacelle]

Are you by chance familiar with the amd setup?

[formal trout]

I have an AMD GPU, I use Linux

[fair oxide]

so when i upscale, should i have the multiplier match these values after it upsacales or should those be my default resolutions before upscaling?

[formal trout]

896x1088 sometimes gives better results for portrait resolution, as 832x1152 is longer and sometimes it makes bodies strech a bit, and 896x1088 is more squarish, closest to 1024x1024, but it depends

[gilded nacelle]

Yoooo, could you help me. On the pinned walk through, im stuck at getting zluda to run, step 3. I dont want to do it wrong and mess up my computer

[fair oxide]

okay so it says base resolutions

i got it

i never saw that

[formal trout]

https://github.com/CS1o/Stable-Diffusion-Info/wiki/Installation-Guides

[gilded nacelle]

Yes. So, step 3 of getting zluda to run, has adding stuff to path. This is my first computer, and first time doing this. Adding to path/environment wrong could ruin it, so i was looking for more of a hands on type of help/ better explanation.

[formal trout]

I upscale with Ultimate SD Upscaler extension, there you just choose "scale from image size" and set the scale and that's it

[formal trout]

I have no idea really, we should invoque @ornate elk

[gilded nacelle]

Cool, and thank you for the honest answer. They offered to help earlier, but i was still at work 😂

[formal trout]

or if your GPU is powerful enough you can do resizing in img2img, just setting this

works like Hi-Res Fix. with my 12 GB gpu I can only resize a 1024x1024 to 1.2 or something, not much

[fair oxide]

hm. so i ONLY have used hi-res fix

i also have a 12gb gpu

does that mean im super limited? i dont want to stick to a square box 😵‍💫

or can i use a separate upscaler and not use as much VRAM?

is hires.fix known to be VRAM intensive?

[formal trout]

there are some options:

1. generate with hi-res fix, and see how much you can resize there. As I told in my system I can only go up to 1.2

[formal trout]

yes, because it is resizing in latent (whatever that means), it is resizing with real generation power

2. the same you can do with a normal generation, hi-res fix is like a img2img step at the end of the generation. So you send the image to img2img and resize it

Then you have tiled upscale like Ultimate SD upscale. There you can resize to whatever you want, as it divides the image in tiles and generate for each tile. But there you can have problems like aberrations or, tile seams. You have to play with the denoising strengh. Much denoising strengh will generate aberrations as SD cannot see the whole picture and don't know what it is generating. Low denoising won't add much detail and the upscale will be kind of meh.

in latent upscale you don't have those problems, but needs a lot of VRAM

[fair oxide]

okay this is really awesome. thank you for all the help. is less VRAM used when generating an image and then sending it to img2img to upscale? or is it all the same?

[formal trout]

when using tiled upscale, the VRAM is similar to generating one image with the tile resolution

resizing in img2img (no tile upscale) will need more VRAM

[fair oxide]

hm okay

and just for clarification

[formal trout]

using --lowvram you can try to resize beyond the normal limit, but it will take much longer

[fair oxide]

the base resolution graph is what i generate my initial resolution with right? and then the hires.fix rescales it?

so like for SDXL if i wanted a 704x1408 image, i start there and THEN upscale correct?

or do i upscale TO that resolution?

should i rescale to that base chart you sent me?

[ember mortar]

regional prompter is mixing the prompts of 2 differents character even adding 2 columns and using breaks

[formal trout]

you generate and img, let's say 1024x1024

then you resize that resolution, both sides are multiplied for an x value, it doesn't matter now what result value it is

the generation value has recomended resolution, then when resizing it doesn't matter

the generations recomended resolution have to do with the resolution of the images that trained the models. If you do a larger resolution, you will probably start having multiple generations, like it want to put various 1024x1024 generaation in the same generation (hope this make it clear)

so people will have torsos comming out from his head and that kind of stuff

BUT when resizing at lower denoise level, it doesn't matter as it cannot possible change the image that much

[formal trout]

sadly it is like that, no matter what

In my experience it worked more like a suggestion for composition, but for different characters it was always an inconsistent mess

To get several characters I think you have to give an idea that contains that in a simple way "a damsel in distress saved by a knight" (the more cliche, the easier it will go I think), trying to prompt both characters will usually end up in messed up things. And maybe with regional prompter you can say ADCOL princess ADCOL knight and it may work, but always simple

like a basic idea for an image that would contain several characters

[ember mortar]

the thing is I already used it and had some images from the last month when I started the first time with regional prompter and now I having this problem and sometimes it make both characters but the second one is not detailed

[formal trout]

in that case use inpainting, just having a base for inpainting is very good

[ember mortar]

still don't know how to use inpainting properly

I should check a guide

[glad saddle]

Hi, I did a clean Windows install and I'm trying to install forge.

Is this a common error?

stderr: error: subprocess-exited-with-error

Running setup.py install for basicsr did not run successfully.
exit code: 1

[42 lines of output]

[past onyx]

hey folks, just purchased 10$ credits for API, but still getting this error during API calls. Can someone please have a look? {"errors":["You lack sufficient credits to make this request. Please purchase more credits at https://platform.stability.ai/account/credits and try again."],"id":"72dd8fd69d02d9cf8794e8d95f6c7377","name":"payment_required"}%

[median cosmos]

Nvidia GeForce RTX 3070 Ti GPU

is this chip good for stable diffusion

[ornate elk]

Its okay. More vram would be better but with that card you can use mostly everything

[median cosmos]

Oh okay ,I am new to stable diffusion should I use sdxl or comfy UI next

[pseudo prairie]

guys is it possible to view every sample steps

so i generate this image

with 40 sample steps

but would love to see the previous pnes

also whats a seed?

[vocal burrow]

Yes but it s not recommended as it will take more time to generate an image for each step.
A seed is a random number that generate the initial noisy image. Therefore if you use the same settings with the same seed you ll get the same output

[vocal burrow]

sdxl is a model.
ComfyUI is a program (using models, vae, etc)
They re not mutually exclusive

[vocal burrow]

Try asking on their website / mailing them, that s where you most likely to get stability staff to look at your issue. This channel is mostly for community stuff, guidance, etc.

[vocal burrow]

can t help without the full log

[humble crater]

I am a novice (intermediate?) SD user. I have been using Automatic1111 now for six months or longer (installed locally). I was using the DreamShaper model and I had almost mastered prompts for the type of generation I was doing.

The prompts all followed a format something like this: "Photograph of skinny (21 year old Triqui woman from Newvo Leon, Mexico:1.5)(with face of Lun Bawang woman from Brunei:1.5)"
I have a nearly-complete list of ethnicities from every country and region in the world and this method was creating mixed-races incredibly well.

The images I was generating were also extremely high quality.

I decided to add some more extensions and found out about "wildcards". I added about five extensions. After adding these extensions, the image quality of SD suddenly went down to very poor quality. The images were extremely grainy and the faces had those ugly color-blotches (what causes those??? I get those often with various models...Is it something I'm doing? Adetailer won't clean them up). Faces were also coming out distorted.

I re-installed Stable Diffusion, but the images that Dreamshaper was generating were not nearly as good. Many looked much more artificial and perhaps doll-like (not quite the right description, but you get the idea). The image compositions were totally different. Epyptian women would be wearing headdresses, for example, and set against i.e. pyramids. This never happened with the previous installation. All were in normal settings. There was also a lot of writing AND almost every image was a split image featuring two different women.

If I just gave it a single-word prompt of "woman", it generated a "fair" (but not great) image of a woman. But as I added many more details, the images became "weird".

I tried many different models and none of them even close to decently.

I reinstalled wildcards extension only, and immediately the image generation became horrible (or rather more horrible) again.
Stubbornly, I tried it one more time and wildcards screwed up the image generation again. I'm about to re-install SD again, not using wildcards, but the problem is that I won't get the "nearly-perfect" images I was getting in my first installation.
Does anyone have any suggestions? I've scoured the internet for any clues, but found absolutely nothing.
I am not sure what "Dreamshaper" model I was using originally, but I wouldn't think that it would matter THAT much.

Attached is a sample of what the generated images look like now with the same prompt that produced amazing photo-quality images previously

[median cosmos]

So Comfy UI is like Automatic1111 and sdxl is like stable diffusion 1.5 ?

[vocal burrow]

yes

[fair oxide]

I see

Thank you

I also found a way to turn off VRAM shared memory. I’m going to try that tonight and see if I can actually generate images this time around

[vocal burrow]

There s too many issues here to tackle things properly.
In no particular order;
The images were extremely grainy and the faces had those ugly color-blotches most likely an incorrect VAE issue or cfg value
I re-installed Stable Diffusion [and my outputs were different] could be because of maaanyyyyyyyyyy things ; different settings, different command line arguments, different models, different auto1111 version with seed breaking change https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Seed-breaking-changes, etc
I am not sure what "Dreamshaper" model I was using originally, but I wouldn't think that it would matter THAT much. yes it does, different model => different output

So if you still have an original .png from "back then" use it as an input to make sure that you re using the same settings, check the hash of the model / a1111 version used from the metadata,

and dont install a dozen of extensions at once

[humble crater]

Thank you...How would I use that as an input?

[vocal burrow]

drag and drop the image in the input prompt field or copy paste the description from its metadata, drop the text in "input prompt" and then click the blue arrow next to generate to apply those.

[humble crater]

aha thank you

[pseudo prairie]

why doesn’t it generate normally

like the image is generatering i can see the process

but the moment it hits 100%

it doesn’t give me a finish image

it just goes back to

[humble crater]

Could be a number of different problems, such as Out of Memory Error...if you don't have enough VRAM to handle the size of image you are trying to create. I've had this happen on occasion, just a few minutes ago in fact (out of memory). I'm a novice at this and have had a couple other reasons why that has happened, but I don't remember why.

[brittle lantern]

Guys, a question we may not know yet: Will the SD3 model that comes out on the 12th work fine with 8GB GPUs?

[vocal burrow]

Explained right there why. And there s probably more explanation in the console log. What s your gpu ?

[vocal burrow]

We don t know yet. It s probably not gonna run day 1 on 8gb without some tweaking / offloading based upon the sd1.5 sd2 sdxl sd3 trend. But give the community a few days to optimize and tweak the pipeline and we ll see.

[brittle lantern]

👍

Me and my 6600 xt gonna have to wait a bit

[boreal harness]

Hello, I am new. my audios not generating. Is there a problem with the site?

[vocal burrow]

which website ?

[boreal harness]

https://www.stableaudio.com/generate

[vocal burrow]

Try contacting stability.ai staff directly throught email or their website then. This channel is mostly community driven we don t have access to any backend or staff member access. Contacting them directly is your best shot

For what it s worth the dreamstudio (different website but probably same server farm) API does not report any issues

[boreal harness]

Thank you very much 🙂

[ornate elk]

Make sure you use the zluda webui Backend on windows or rocm on Linux for the best performance and less vram usage

[gilded nacelle]

Hey. I had got stuck on the "getting zluda to run, step 3". I looked at the picture showing how to, but I'm not experienced at all, and dont want to mess up the computer. Could you give an actual step by step, telling me where the items need to be added to the environment.

[hard quiver]

does it make a difference if stable diffusion is in ssd or hdd?

[vocal burrow]

Apart from the initial launch and loading speed of the model. No. Once everything is set up it's all on your gpu's vram. (Assuming you're not offloading with --medvram or --lowvram)

[hard quiver]

ok thanks.

[mossy narwhal]

I am trying to achieve the following with the Stability API:
Use the erase feature but retain transparent backgrounds, it seems like it automatically fills the background.
Does anyone know how to do this?

[wet flame]

how to sharpen or remove blurr from images when generatiing?

[brittle lantern]

Ffs

@ornate elk doesn't seem so bad

[past onyx]

hey folks, when playing with the API and https://www.stablevideo.com/generate, I noticed that passing the same image produces different results. What's the 'default' parameters that stablevideo uses (seed, cfg_scale, and motion_bucket)? i kinda want to get the same quality that stablevideo provided me

[ornate elk]

nice, would be awesome

[ornate elk]

sure, i can screenshot the steps

@gilded nacelle

[ornate elk]

it says cuda out of memory at the bottom so the last step (vae step) failed.
you need to install the tiles vae extension

[gilded nacelle]

Awesome, thank you so much. I will proceed when I get home from work.

[ornate elk]

np, let me know if it worked 🙂

[gilded nacelle]

Will do

[solemn atlas]

what does that mean

[ornate elk]

wrong stuff used together

like controlnet 1.5 with sdxl model

[solemn atlas]

I use ponyxl and sdxl vae and xl loras

I really dont know why that happens

[empty ember]

Hi

[ornate elk]

in txt2img or img2img?

[solemn atlas]

on img2img

[ornate elk]

can you show me the whole settings?

using any script?

[brittle lantern]

I have a question does AMD take more vram while generating than nvidia?

Or both relatively same usage

[solemn atlas]

what do you mean by script

[ornate elk]

depends on the webui version you use.
ive made 2 AMD Guides.
One uses Directml which does consume a lot more vram.
One uses Zluda which uses nearly the same as nvidia.

[ornate elk]

your using a 1.5 based vae
that wont work with pony or sdxl

[solemn atlas]

but here it says sdxl

[vocal burrow]

override settings ... well they override whatever is set at the top (or elsewhere)

and here your vae is overridden

[regal ridge]

hi! i just ran into this error & was wondering if anyone knows how to fix it: "RuntimeError: Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check". I thought about deleting & re-installing the app but was worried that I might get the error again later on.

[ornate elk]

Hey, what was your GPU again?

[grave lantern]

Hi im pretty new to this I download and installed correctly but i don't know how to get the right image for what i want. i've download some lora's (don't know how to use them) still having trouble getting my image to become what i prompt. ^the image in question. im trying to add to the image without changing it.

[mossy narwhal]

Using the Stability API, I am doing remove background and then inpaint, but the main object also gets modified. Any ideas how to avoid this from happening?

[brittle lantern]

So zluda how does it work uses cuda or sum like cuda for amd?

[cunning badger]

hey - my hugginface cache folder is 89GB now and my pip folder is 33GB -- is there a way to clean this stuff up? can i delete them and will it rebuild automatically? this is in addition to my AI folder taking up 240GB, its wayyyyy too much

[past onyx]

hey folks I thought of asking here. I am currently playing with the API, and I notice that oftentimes the resulting video somehow tries to transform the current frame completely. All I want to do is create a slow-moving animation, nothing too sophisticated. I was wondering if someone has a recommendation of passing parameters in the API that would help reduce that 'frenzy' transition. Here are a few examples of what I am talking about, I appreciate your help in advance 🙏

[vocal burrow]

pip cache purge

[cunning badger]

how does one perform such a feat

[vocal burrow]

Open cmd

Type "pip cache purge"

Press enter

Voilà

[cunning badger]

wow that was easy, thanks so much

any idea about the huggingface cache?

[vocal burrow]

Not off the top of my head.

And I m heading to bed for tonight. Good luck :p

[cunning badger]

night, and thanks again!

[regal ridge]

I have Radeon RX 5500 XT (32 ram)

[cursive cobalt]

In what folder, for COmfyUI, do i place my wildcards..?

Google yields little to no text based results

[slow holly]

Is there anywhere I can give sd3 a try for free

[gilded nacelle]

@ornate elk this is what always happens. It tries to gen, then stops within a couple seconds

[sharp lichen]

Hi!

Does anyone have ideas on the current best way to change the background behind a subject?

[pseudo prairie]

how do i do that

[sharp lichen]

While keeping the subject the same

[ember mortar]

is it possible to control the height of a character?

[formal trout]

Inpaint anything extension for masking the background, plus inpaint with fill, medium high denoising (in Automatic1111)

[sharp lichen]

thank you

you mean inpaint anything to make the blank and white mask?

You can this in photoshop right

same thing?

what is inpaint with fill?

[formal trout]

An option for inpaint that tries to add stuff that isn't there

[sharp lichen]

it says it "Inpaints the colors of the image"?

This is what I am trying

I am trying to change the background of the plane to the grand canyon

[formal trout]

yes to avoid masking the background manually

[formal trout]

or yes if importing the mask from photo shop

[sharp lichen]

I guess how is this different than just generating a text2img of the grand canyon and then photoshopping the plane on top?

Isn't that all that is happening?

Having inpaint fill on seems to not work well?

using 1 denoise

[formal trout]

with inpaint SD is wathing the plane and it might try to do a composition that makes sense, or make some shadow, etc, it would suposely blend better. But of course you can do it like that, and then blend the details it with inpaint, then run some img2img or upscale with low denoise to make all even

[formal trout]

this method is great, it just need more judgement from you I think

[sharp lichen]

Ideally it is 'grounding' the subject

like you said shadows etc

I don't understand what fill does? I thought I don't want it to use colors from the input image

Also I shouldn't be mentioning the subject in the prompt right?

Or else it tries to add a second plane

[formal trout]

inpatin "original" will try to respect the colors and shapes from the original and change it not that much. "fill" is much stronger and can draw different things.

[sharp lichen]

Also shouldn't denoise be around 1?

[formal trout]

"fill" won't respect the background much... both work in a similar way but with that base difference. "original" with high denoising might be similar to "fill" with low denosing, but I never found documentationa about the specifics

[sharp lichen]

It doesn't seem to be 'aware' that the subject is there...

ideally it is

or else it is just like text2img plus photoshop

any idea how to do that?

maybe controlnet?

[formal trout]

it depends, in this case I think that denosing with 1 will work as you don't care at all about the background.

[formal trout]

well that could be because of the high denoising, if you see the steps of the generation, the plane would show to be non-existant I think. Try with lower denoising of 0.8, 0.7

[sharp lichen]

what deoise while using fill?

really low?

[formal trout]

like beetween 0.6 or 0.8

well this is horrible xD

[formal trout]

Also mind the resolution of the image you are working

[pseudo prairie]

i think its Radeon RX 6800M

[sharp lichen]

i was getting the same thing!

using fill

it kept doing that

[formal trout]

yes, try original with high denoising

fill is kind of crazy

with original it kind of undertand the original and make something that makes more sense, well you always learn something xD

[sharp lichen]

did that prompt include the mention of a plane?

[formal trout]

yes, it was cinematic photo, a plane flying over a STAR wars land battle

with original an 0.74 denosing. I think it will always keep the land and sky how it is.

with original and 0.92 denosing, as you say, it started to draw another plane

[sharp lichen]

are you able to get it without any other planes?

just one plane

perhaps trying above the grand canyon

this is the best i have gotten

how come the wings on your subject is more clear than mine?

[formal trout]

I did a rough mask around and used soft inpainting

and used "inpaint not masked"

[sharp lichen]

yeah i might do that instead

[formal trout]

fill with 0.77 denosing, there you see is not respecting the background, adds a different angle

latent nothing with 0.7 denoise, it is like generating a background and photobashing

fill

so maybe if you have a precise idea, better photobash the plane in a background and then inpaint with original to make it blend

so fill make crazy but creative things

original will try to keep the sky and the land where it is

[pseudo prairie]

thats the issue i think

[ornate elk]

It makes cuda usable with AMD. Its like an emulation layer between cuda and rocm

[tranquil perch]

how to generate photos?

help!! this is the first time been using this

how to create ohotos?

[ornate elk]

Ah okay, I need the full cmd text to check

[ornate elk]

On the discord or localy?

[ornate elk]

What are your txt2img settings?

[ornate elk]

In the extensions tab. Click on available, then on "load from".
Then search for tiled
Then install the "tiled diffusion & tiled vae" extension.
After a restart, enable tiled vae.

[stone coral]

when installing forge

"A new release of pip available: 22.2.1 -> 24.0
[notice] To update, run: H:\webui_forge_cu121_torch21\webui\venv\Scripts\python.exe -m pip install --upgrade pip"
Should i do this?

[vocal burrow]

you can

[stone coral]

thx

is there any point in doing it? or is it just one of those things that you can't go wrong with but probably won't make any difference

[vocal burrow]

can't go wrong and probably won't make a difference

apart from removing a warning message from your log :p

[stone coral]

ok i cant seem to install this

ModuleNotFoundError: import of xformers halted; None in sys.modules

i keep gettin this message from run.bat

[vocal burrow]

what's your gpu ?

[stone coral]

4070 ti

i mean it looks installed actually

but cant run it

[vocal burrow]

I'd say :
1/ don't instal auto1111 using the .zip but instead use git clone
2/ use --xformers command line argument

cf pinned guide

[stone coral]

this is forge and i tried the --xformers already

i will try the git clone now

this seems to have issues

[vocal burrow]

forge does not "need" xformers

but at the same time forge didn't get any kind of update for 3 months or so

[stone coral]

ill let u know if it works, thank u :)

@vocal burrow i got this error

[vocal burrow]

what's the full log

copy paste everything in a file and drop it in here

[stone coral]

fuck it just closed

ok here it is

[vocal burrow]

put pydantic==1.10.15 in the requirements_versions.txt file

[stone coral]

yessir

[stone coral]

i still get the same error

same log

oh wait

i made a mistake

i put it in requirements.txt

[vocal burrow]

also delete the venv folder

[stone coral]

works now, thank you <3

[ivory abyss]

what is token normalization

[vocal burrow]

it's a parameter tweaking how the prompt is interpreted

hard to be more specific without getting technical

[ivory abyss]

Which one should I choose?

for landscape pics

[vocal burrow]

depends of the model, the settings, the prompt, etc

try it out and see what you like the best

[ivory abyss]

okay ty

[vocal burrow]

there's no golden setting

[ivory abyss]

whats the difference

(im learning comfyui)

[rose pivot]

Found a working LORA to make turnarounds, but is there a way to make a turnaround of a existing image?
Like rotating a image of a character.

[gilded nacelle]

Okay, will be later. Thank you

[stone coral]

Someone know why I can't see ultimate sd upscale here?

[vocal burrow]

cause it's not a txt2img script,

it's an img2img script

[stone coral]

ohh thx

[sage sentinel]

Hey, How can i use a automatic111 extension from command line without the UI. Need some guidance. I am using this extension https://github.com/numz/sd-wav2lip-uhq . I can generate from UI. i Can create and add a new API but i'm not sure how to run the script to run extension through cli. I am on linux machine.

[desert grotto]

Is it possible to make it so sd automatically upscales images it generates as opposed to manually doing so?

[peak thicket]

There is also the following error.

[sullen valve]

Hi!
I'm a complete newbie, i.e.: I messed around with midjourney when it first came out and that's about the extent of my experience.

How do I train a Stable Diffusion model locally on my computer?
• I want to experiment with my own work as a style for image generation
• I don't want my source images nor my outputs to end up on the internet
• I'd like to test my glazed and nightshaded images

[solemn moon]

code bro's, PPPLLLEEEAAASSSEEE can we get some decent control nets

[solemn moon]

check youtube, youll learn more

[solemn moon]

for XL

[solemn moon]

i think you want to make yr own lora

[peak thicket]

Why do I feel like the dude above just wants to open an SD and play around?

[solemn moon]

last canny from xinsir is way better than previous tries but still getting these scribble lines everywhere:

[heavy perch]

@vocal burrow

[cursive cobalt]

(ComfyUI) I'm trying to get IPAdapter to work, but this error keeps popping up. Included is a pic of the only active nodes, in this workflow.

[stark turtle]

#📝｜prompting-help Illustrate a photorealistic cinematic shot featuring Sheikh Mujib, a mid-aged man radiating confidence and sophistication, standing beside a vintage expensive car. Clad in a dark grey suit that exudes refinement, Sheikh Mujib epitomizes intelligence and style, further emphasized by his bold-framed eyeglasses and back-brushed hair. With cinematic lighting setting the scene aglow, Sheikh Mujib leans against the car with an air of assurance, projecting an unwavering confidence. His gaze is directed forward, reflecting his steadfast determination and vision. In the background, a picturesque mountain looms majestically, enhancing the timeless elegance of the setting. This composition captures the essence of a distinguished individual in a photorealistic manner, inviting viewers to appreciate his poise and charisma in this cinematic shot.

[sand pelican]

Canny doesnt work for me can someone help??

[ember mortar]

im having a problem in which SD won't start, im using google chrome

[viral knot]

i vaguely remember using a1111 been a long while, but dont you also need something for the model? says None for you @sand pelican

like right next of the preprocessor

[uncut heron]

So i got this error in the cmd window - File "C:\programs mine\stable difusion everything\Automatic 1111\stable-diffusion-webui\modules\devices.py", line 255, in test_for_nans
raise NansException(message)
modules.devices.NansException: A tensor with all NaNs was produced in Unet. This could be either because there's not enough precision to represent the picture, or because your video card does not support half type. Try setting the "Upcast cross attention layer to float32" option in Settings > Stable Diffusion or using the --no-half commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check. - i have 4060TI 16GB + 4x16 RAM

can someone explain what does this mean?

[untold pine]

No model is selected

[sullen valve]

There's loads on youtube but I'd appreciate a link, I did look over there, but I've been having a hard time figuring out where to start.

Which is why I wrote here.

Is there a video or channel in particular you'd recommend?

[sullen valve]

I think?
Where can I learn to do that?

[sullen valve]

Yes, but it's not as 'plug and play' as I imagined it to be...

[untold pine]

There are several training guides with different level of out of date-ness:
https://civitai.com/articles?tags=128645

[sullen valve]

Thank you!

[untold pine]

Any installation guide will do:
https://www.youtube.com/watch?v=mHvnHA04sbc

[open island]

hey anyone here experienced stuttering/freezing when using sketch or inpaint skecth in img2img?

i couldn't figure out what causing it :/

[formal trout]

Hi, I'm trying to install audio-diffusion on Windows https://github.com/diontimmer/audio-diffusion-gradio

Got an error when launching

I wonder if it has to do with having an AMD GPU and having to install ROCM or something

or if I have to manually compile it for a torch (😕 )

[glad saddle]

--> NEW PC, 3 Windows Installs, installed a1111 and forge 20x times, same result: 'Press any key to continue...'

Hi, thank you for reading. This is going to be detailed and long.

I've been at this for a whole week trying to fix it. Should I just return this PC? Or is this a software situation I'm not figuring out?

I bought a new Alienware R16 computer thinking it would be a great option, with 64GBs of RAM, a 4090 and so on. It came with Windows 11. I installed a1111 and for the first 3 days everything is going good. Then I start getting the 'Press any key to continue' prompt and by checking the Event Viewer on Windows I find that python.exe is crashing, related to a python310.dll and so I investigate further.

So I installed a1111 and forge about 20 times on Windows 11 to no good result.
Then I installed Windows 10 (clean install) and do the same about 20 times.
Then I install Windows 10 (clean install) again and so tons of more research and still the same result.

I've tried:

1. Running diagnostics tool: sfc /scannow
2. Running disks checks: chkdsk /f /r
3. Running hardware tests and benchmarks.
4. Cleared temp files: del /Q /F tmp*
5. Installed every dependency individually.
6. Check dependency incompatibilities.
7. Used virtual environments.
8. Tried it with Python 3.10.6 and 3.10.9
9. Updated drivers for GPU
10. Updated Microsoft Visual C++ Redistributable
11. Added logs to the SD code to see if I can find the problem
12. Adjusting the virtual memory settings

I made a crash CORE dump file. This is the !analysis of it

Part 1:
From the provided crash dump, we can see the following key points:

Exception Type: Access Violation (0xc0000005)
Faulting Module: python310.dll
Faulting Function: PyObject_IsTrue in python310.dll
Fault Address: 0000000000000060
Memory Access Attempt: Attempt to read from address 0000000000000060 (which is a null or very low address)
Interpretation:
Access Violation (0xc0000005) occurs when the program attempts to read or write to a memory location that it doesn't have access to. In this case, it attempted to read from address 0x0000000000000060.
Null Pointer Dereference: The crash happened due to a null pointer dereference, which means that the code attempted to access a memory address that was null or not properly initialized.
Potential Causes:
Bug in the Python Script or Library: There might be a bug in the code being executed or in one of the libraries being used (such as PyTorch, Stable Diffusion code, etc.).
Incompatible or Corrupted Installation: There might be an issue with the installation of Python or one of the dependencies.
Hardware Issues: Although less likely, sometimes hardware issues (like faulty RAM) can cause such errors.

Part 2:
The detailed stack trace provided offers more context about where the issue might be occurring:

Stack Trace: The crash occurs within the Python interpreter, specifically in functions related to evaluating conditions and function calls (PyObject_IsTrue, PyEval_EvalFrameDefault, PyFunction_Vectorcall).
Failure ID Hash: {0c0c98c1-d542-8886-8478-aa0a9220ebcf}
Fault Bucket: NULL_CLASS_PTR_READ_c0000005_python310.dll!Unknown
Operating System: Windows 10 (10.0.19041.1)

Three things made me think I had fixed it, but then resulted in working for a batch of 20 images then crashing again even after fixing them.

1. Finding a ESGRAN4x corrupted file.
2. Updating Microsoft Visual C++ Redistributable
3. Installing nVidia CUDA Toolkid

At this point I feel I've exhausted all the possibilities of what I know. And continuing would just be to repeat the same things again.

[gilded nacelle]

@ornate elk This is before stable launches. Also, it launches from launch, not webui

[ornate elk]

ahhh, it tries to use your amd cpus igpu instead of your normal gpu

you need to go into the device manager and disable the "Radeon Graphics (TM)" Display Adapter

then relaunch the webui-user.bat

[gilded nacelle]

Okay, trying now

[willow hamlet]

Anyone know how to use image-to-image with node js ? I always get bad request response

StableDiffusion 3

[gilded nacelle]

Didn't work. It didn't load from webui only from launch, and looks like this

And thats the zluda. Should I try the other amd setup?

[ornate elk]

i need a full cmd log

you should launch the webui-user.bat

not zluda or anything

[gilded nacelle]

I click it, it loads, then says click any key to close. Then the launch folder is only one that will launch stable.

[ornate elk]

whats in your webui-user.bat?

this isnt good

[gilded nacelle]

I enabled the graphics tm, and this is web bat

[ornate elk]

whats your gpu?

[gilded nacelle]

AMD Radeon RX 7700s

[ornate elk]

okay, it tried to used your igpu again

make sure radeon graphics (TM) is disabled in the device manager

maybe restart the PC after disabling it

[gilded nacelle]

Ok ill, try. Also, thank you so much for your help.

[ornate elk]

no problem 🙂

your only one step away from getting it to work

[ornate elk]

you dont have the controlnet models

[ornate elk]

more information can help us to help you 😄

[ornate elk]

hey for training use Khoya_ss tool

[unborn bramble]

image spoilered due to it being AI nightmare fuel
I trained a LoRA with one image with 500 repeats per epoch, the text file being only the activation tag, and with 4 epochs. Batch size: 1.
The reason why I did it is so I wanted the AI outputs to look basically like the original image... but I got this nightmare fuel instead
Parameters: 640x800, DPM++ 2M (Automatic), seed 1, 20 steps
Why does it look like this...

[gilded nacelle]

Doing same thing. Starts to gen, then stops immediately

[ornate elk]

can you copy the whole cmd text?

i cant help when seeing half of an error

[ornate elk]

ahhh okay

the 7700S falls into the Step 4 and 4.5 of my Zluda guide you need to download the RocmLibs.zip and follow the steps
then it will work

[gilded nacelle]

okay. will try now, thx again

[ornate elk]

let me know if you need help with the steps

[empty sphinx]

Hi!! I've a question. I'm trying to train the stable-diffusion-v1.5 with severals artist styles. How is the best option for training? Using external captions where describe the style name or name the image with the style? Thankss

[gilded nacelle]

do i rename original library or the copyed folder

[ornate elk]

copy the folder and rename one of them to old_library

[gilded nacelle]

ok

i do not see a zip file

[fair oxide]

whats the best way to fix hands using impainting

[ornate elk]

@gilded nacelle if you cant find it in the guide, i can also send you a direct link

[gilded nacelle]

I got it, copying files now

[gilded nacelle]

i have a feeling im going to have to factory reset this computer, and do an exchange to something with nvidia

[ornate elk]

@gilded nacelle have you restarted the PC after adding the files to library?

[gilded nacelle]

yes

apparently im just to ignorant to do this, as it keeps not working.

[gilded nacelle]

thank you so very much for all of your help.

[ornate elk]

Yea its tricky to get a 7700s working with zluda and I'm sry, but here was no other 7700s user before you which tested it.

It can work but I understand if you don't want to mess with all that xD

@gilded nacelle can you remove the .zluda folder and try again?

Its in the stable-diffusion-webui-directml

[gilded nacelle]

Ok will try

[gilded nacelle]

By remove, you mean delete?

[ornate elk]

Yes

[gilded nacelle]

Ok trying now

Nope, it just says the exact same things in the command box as before.

[ornate elk]

Hmmm

@gilded nacelle what's in your webui-user.bat?

Right click and edit it

[ornate elk]

Okay, looks Alright, but in the last error messages it didn't showed the launch args. Make sure to only run the webui-user.bat and not the webui.bat

[gilded nacelle]

venv "C:\SD-Zluda\stable-diffusion-webui-directml\venv\Scripts\Python.exe"
WARNING: ZLUDA works best with SD.Next. Please consider migrating to SD.Next.
Python 3.10.11 (tags/v3.10.11:7d4cc5a, Apr 5 2023, 00:38:17) [MSC v.1929 64 bit (AMD64)]
Version: v1.9.3-amd-24-g2c29feb5
Commit hash: 2c29feb50e5cd3592b3ea831fe20b17588a2edb4
Using ZLUDA in C:\SD-Zluda\stable-diffusion-webui-directml.zluda
Skipping onnxruntime installation.

| You are not up to date with the most recent release. |

Consider running git pull to update.
no module 'xformers'. Processing without...
no module 'xformers'. Processing without...
No module 'xformers'. Proceeding without it.
C:\SD-Zluda\stable-diffusion-webui-directml\venv\lib\site-packages\pytorch_lightning\utilities\distributed.py:258: LightningDeprecationWarning: pytorch_lightning.utilities.distributed.rank_zero_only has been deprecated in v1.8.1 and will be removed in v2.0.0. You can import it from pytorch_lightning.utilities instead.
rank_zero_deprecation(
Launching Web UI with arguments: --use-zluda --medvram-sdxl --update-check --skip-ort
Traceback (most recent call last):
File "C:\SD-Zluda\stable-diffusion-webui-directml\launch.py", line 48, in <module>

main()

File "C:\SD-Zluda\stable-diffusion-webui-directml\launch.py", line 44, in main
start()
File "C:\SD-Zluda\stable-diffusion-webui-directml\modules\launch_utils.py", line 689, in start
import webui
File "C:\SD-Zluda\stable-diffusion-webui-directml\webui.py", line 13, in <module>
initialize.imports()
File "C:\SD-Zluda\stable-diffusion-webui-directml\modules\initialize.py", line 39, in imports
from modules import processing, gradio_extensons, ui # noqa: F401
File "C:\SD-Zluda\stable-diffusion-webui-directml\modules\processing.py", line 18, in <module>
import modules.sd_hijack
File "C:\SD-Zluda\stable-diffusion-webui-directml\modules\sd_hijack.py", line 5, in <module>
from modules import devices, sd_hijack_optimizations, shared, script_callbacks, errors, sd_unet, patches
File "C:\SD-Zluda\stable-diffusion-webui-directml\modules\sd_hijack_optimizations.py", line 13, in <module>
from modules.hypernetworks import hypernetwork
File "C:\SD-Zluda\stable-diffusion-webui-directml\modules\hypernetworks\hypernetwork.py", line 8, in <module>
import modules.textual_inversion.dataset
File "C:\SD-Zluda\stable-diffusion-webui-directml\modules\textual_inversion\dataset.py", line 12, in <module>
from modules import devices, shared, images
File "C:\SD-Zluda\stable-diffusion-webui-directml\modules\images.py", line 22, in <module>
from modules import sd_samplers, shared, script_callbacks, errors
File "C:\SD-Zluda\stable-diffusion-webui-directml\modules\sd_samplers.py", line 5, in <module>
from modules import sd_samplers_kdiffusion, sd_samplers_timesteps, sd_samplers_diffusers, sd_samplers_lcm, shared, sd_samplers_common, sd_schedulers
File "C:\SD-Zluda\stable-diffusion-webui-directml\modules\sd_samplers_kdiffusion.py", line 3, in <module>
import k_diffusion.sampling
File "C:\SD-Zluda\stable-diffusion-webui-directml\repositories\k-diffusion\k_diffusion_init_.py", line 1, in <module>
from . import augmentation, config, evaluation, external, gns, layers, models, sampling, utils
File "C:\SD-Zluda\stable-diffusion-webui-directml\repositories\k-diffusion\k_diffusion\config.py", line 6, in <module>
from jsonmerge import merge
ModuleNotFoundError: No module named 'jsonmerge'
Press any key to continue . . .

[ornate elk]

That looks better

[gilded nacelle]

nothing still

[ornate elk]

What you can try is to delete the venv folder and then relaunch the webui-user.bat

And then please test with an 1.5 based model before trying an xl model

Also don't change any settings, samplers, steps, resolution etc

Its really late here I need to go off sry

[gilded nacelle]

Ok, will do. And no worries, thanks again.

[regal ridge]

got this error: "RuntimeError: Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check". I had a different error last time so i re-installed my program. My pc is: Radeon RX 5500 XT (32 ram). any help would be appreciated 👍

[hallow gust]

did u try copy and pasting --skip-torch-cuda-test into the the commandline_args in webui-user.bat?

keep getting this error, and whenever i try to select a VAE it just switches back to none

[regal ridge]

It worked!! thanks a lot! 👍

[vocal burrow]

@regal ridge please.... No

By doing that you re now running on CPU only mode

Please check pinned message for proper install guides according to your hardware

[subtle quarry]

can anyone help me figure out why I can't get dw pose to work in comfyui?

it just prints out black.

nvm! Got it!

had to shange a setting

[analog spire]

is this where i come if i have problems? I have been trying to get stable diffusion working for a while but have been encountering problems

[subtle quarry]

yeah, but it may take a while for someone to help you

[analog spire]

oh is it a dead chat

[subtle quarry]

kinda sorta. depends on the time of the day you ask for help.

[analog spire]

ah ok thanks

[fair oxide]

any tips for hleping remove minor imperfections in an image using impainting? i turned denoising strength down to .5 and masked padding pixels to 100. should i upscale the impainting at all or nah>

honestly i just turned up the sampling stemps and it did a lot better 🤷

[gloomy bolt]

i just switched from auto1111 to forgeui

but it doesn't seem to transfer over VAEs

and also some loras i recently put in aren't being shown

[ornate elk]

Make sure when Inpainting to always set the inpaint area to "mask only" and then set a square resolution

[south vigil]

yo Gs i can see on civitai.com/video that there are some very cool ai generated videos. How are they made as i dont think that all of them are completely mad eby a text prompt but more like converted from a normal video. Was that made by stable diffusion and if yes, can you maybe send me a tutorial? I researched but didnt find anything about that sadly as there are many people talking about other text to video ai platforms

[wide pecan]

Hello there， May I know if there is anyone know any contact information of stable audio despite the email [e-mail]? I have been trying to get in contact with Stable Audio for API integration of Enterprise Plan but there is no feedback, wish to get feedback asap...thank you in advance if any advice can be given.

[subtle quarry]

this is what i could find https://www.youtube.com/watch?v=3eAZkmXZlXM

just put in "video2video a1111" (or "video2video comfyui", depending on what you use) in youtube

does anyone know why openpose is printing out just black?

i've tried inverting it too, but no luck...

[ornate elk]

Does the cmd shows anything when pressing the explosion icon?

[maiden totem]

I’ve tried getting the stable diffusion on my phone and somehow it doesn’t work

[ornate elk]

Stable diffusion with google collab won't work without the Pro Subscription of Collab

[maiden totem]

Ohh so do you have to pay to get across with stable diffusion?

[ornate elk]

For google collab yes.
There are also other services for stable diffusion. Most cloud based ones require to pay. Or to host your own.

[maiden totem]

Ohh okay now that makes more sense now thanks for that explanation

[ornate elk]

No problem, if you have the hardware you can install stable diffusion for free localy on your PC at home and connect it to your phone.

There are also free discord bots and websites that use the StableHorde Backend to generate images.

And then there are also paid options like the official Stable diffusion discord bot on this Server or websites liked Leonardo.ai to generate images

Other paid cloud based solutions (which require more setup knowledge) are:
Vast.ai, runpod.io, and paperspace

[fair oxide]

How can I use Stable Audio Open???
I want to use a audio sound effect ai generator for free

[livid lark]

the nightmare of everyone trying to isntall Sd locally.

Python will git you, like the snake got Eva.

[gilded nacelle]

Yeppers lol

[cursive cobalt]

Is there a node for ComfyUI, for wildcards, where it will suggest things with those words in it, similar to this? My Dory goldfish brain can't remember what the wildcard names actually are, so this helps me quite a lot

[cursive cobalt]

Ok... Something is wrong here.
1080 Ti.
Almost 7 HOURS for a batch of 6 gen?

Bypassing the ControlNet Nodes cuts that down to ~5 min

So why is controlnet adding over 6 hrs to 1 gen?

[terse lichen]

how to install this or fix it bro ?

[stone blaze]

@ornate elk I'm trying to use google colab, I have some of the free credits, but getting this error:

_MultiThreadedRendezvous: <_MultiThreadedRendezvous of RPC that terminated with:
    status = StatusCode.UNAVAILABLE
    details = "connection error: desc = "transport: Error while dialing: dial tcp: lookup stable-diffusion-xl-1024-v1-0.tenant-stabilityai-corweave.svc.tenant.chi.local on 10.135.192.10:53: no such host""
    debug_error_string = "UNKNOWN:Error received from peer ipv4:104.18.34.224:443 {grpc_message:"connection error: desc = \"transport: Error while dialing: dial tcp: lookup stable-diffusion-xl-1024-v1-0.tenant-stabilityai-corweave.svc.tenant.chi.local on 10.135.192.10:53: no such host\"", grpc_status:14, created_time:"2024-06-06T14:13:28.815710532+00:00"}"
>

[fair oxide]

Does anyone know a way to dictate the CFG value from the prompt text itself using ComfyUI? There are custom nodes that can extract LoRA's from the prompt text (Load Lora Tag) and I would like something like that that lets you encode the CFG into the prompt and then uses a float node output that can be connected to the CFG input on the sampler node

[hard quiver]

anyone knows why this is happening? they were coming fine a moment ago.

its happening with a particular S.D model. so not all of them

[stray latch]

Hi everyone, is there a way i can use cloud services for SD and Automatic/Comfy via an API relatively simple like i can with LLMs in Sillytavern?

All i saw was using GPU for rent services like Vast.AI but that's.. kind of a lot

[sage scroll]

I use AnimateDiff and its only producing random noise. Can somebody help me?

[ornate elk]

What's the resolution?
Sdxl and pony need around 1024x1024

[ornate elk]

Does image generate normaly?

[sage scroll]

yes

everything else works normal too. Its just AnimateDiff

[zealous salmon]

Hello I contacted https://stabilityplatform.freshdesk.com/support/home 7 days ago and no one is responding back, is this a scam? or can I get some customer support please.

[vagrant notch]

Hi everyone, I'm using A1111 on Colab. can anyone explain to me this error message and how I can fix it? "NansException: A tensor with all NaNs was produced in Unet. This could be either because there's not enough precision to represent the picture, or because your video card does not support half type. Try setting the "Upcast cross attention layer to float32" option in Settings > Stable Diffusion or using the --no-half commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check."

[ornate elk]

Maybe you tried to use a lora as model

[gilded nacelle]

So it finally worked last night, but animagine xl is producing kinda trash images. I do have its vae on.

[ornate elk]

It worked? Nice!
Can you show me an example of an image?

[gilded nacelle]

Ah, im currently at work. Basically, it looks almost like a small child drew it. I even used the character prompt that is worded in it huggingface info.

[ornate elk]

Ahh, that effect appears when generating at a lower resolution as the model got trained on.
Like using 512x512 with SDXL models.
Because they are trained on 1024x1024 base resolution

[ornate elk]

Maybe use a smaller Controlnet model. If you use a 2gb sdxl Controlnet model that overfills your vram

[cursive cobalt]

Those r hat i has

What i have*

[ornate elk]

Check the file sizes

[cursive cobalt]

Ah. I was using the 2.4 gb one

[ornate elk]

Yep, these slow down

[cursive cobalt]

Also. Canny vs depth. What difference?

[ornate elk]

Different use cases

Canny is flat, depth is more for 3d and layering

[cursive cobalt]

Oh so Canny for 2D images, Depth for 3D / Realisitc?

Hard to type with cat on keyboard

[gilded nacelle]

Hmmm. I tried 1024x1024 and it was still mediocre. Any other settings that make a difference?

[gritty bane]

You do not have a valid subscription. Please login with your Discord account while signing up to https://stability.ai/stable-artisan#choose-stable-artisan-plan.
Important: Make sure you click 'Continue with Discord' at the login screen!

Excuse me, what is this question

[ornate elk]

A good negative prompt, then trying out different resolutions.
Like 768x1024 for portrait for example.
Then using adetailer extension for face fix.

[gilded nacelle]

Awesome! Thank you so very much! Its really generous and kind of you to give out so much help, its truly appreciated!

[wet flame]

i keep getting this error

which file is the resume dowonload button

and this started appearing

maybe reinstall automatic again

[analog spire]

I am trying to download stable diffusion. When trying to figure out what download to do, it seems I need to use a specific gpu for different versions. When checking task manager to see what gpu i have, it seems i actually have 2. GPU 1 and GPU 0. One is nvidia the other is amd. How do i know which one i should be using, and how do i know that it will be using the rightr one when i download it?

[ornate elk]

hey, whats your nvidia GPU?

the amd one is probably the igpu from the cpu chip

[analog spire]

rtx 4060

[analog spire]

oh shoot your probably right

my knowledge on computer hardware is limited to say the least

[ornate elk]

no problem

[ornate elk]

if your follow my nvidia guide from the pinned messages of this channel, you will install the right version

[analog spire]

great I'll give it a shot and report back. Thank you so much!

[ornate elk]

no problem 🙂

[analog spire]

ah sorry, i have one more question before i get to it. I have other versions of python, and i use pyenv to manage them. Will this be an issue? do i need to manually install the 3.10 version from the python website, or can i just continue to use pyenv to manage it

[ornate elk]

no problem at all 🙂
here is a quick image i made, with settings you can try:

[ornate elk]

hey, you can use pyenvs but you need to make sure its a python version between 3.10.6 and 3.10.13

[vocal burrow]

low resolution output with sdxl model

[ornate elk]

@sage scroll make sure this is enabled when using animatediff in auto1111.

[analog spire]

Alright here is the update on where I'm at. I followed all of your instructions on installing for nvidia gpus. The only change is that I manually added python 3.10.11 .exe file path to my environment variables because i installed it using pyenv. I have about 8gb of vram, so i added the "--medvram-sdxl" to the .bat file. after setting it all up and restarting once, I double clicked on the .bat file. a cmd window popped open for about a second and then immediately closed again. Nothing happened after that, so i tried again with the same result. I might need some insight on what to do next

[ornate elk]

can you show me the file you editet with your path?

(the webui-user.bat)

[sage scroll]

anyone got a possible solution? ive tried reinstalling and every model i have dowloaded

[sage scroll]

its enabled but still not working

[ornate elk]

make sure you also use negative tags

[ornate elk]

and where did you specified the pyenv because your python path needs to be changed inside the webui-user.bat when using a specific version

oh you only added it to the env variables

[analog spire]

ah ok. I added it to the path variable in the system environment variables

[ornate elk]

then yea add it to the PYTHON= line

[analog spire]

ok. 👍

[analog spire]

:o something is happening

that might have done it

[sage scroll]

@ornate elk Can you share a model, generation settings and a promt thats working for you?

[south vigil]

thanks bro, this helps immense!

[ornate elk]

sure:
model from here:
https://civitai.com/models/4384/dreamshaper

[sage scroll]

ty!

But i meant something thats working in animatediff 😅

[ornate elk]

ohhh sry i got confused haha

give me a sec

[sage scroll]

(Epilepsy warning!) Thats like the best that i can do atm xD

[hollow mason]

Hey any idea why when I generate with IPAdapter(comfyui) it just looks unfinished/horrible? No matter the preset I choose with the unified one.

[formal trout]

how can I do this ?

[ornate elk]

sooo im ready

[sage scroll]

wait, do i have to add the lora?

[ornate elk]

only for the 1.5 v3 version

not for 1.5 v2 or sdxl

i see i used v2 by accident lol xD so the lora didnt got used

[analog spire]

@ornate elk It works! Thank you so much for all your help!

[ornate elk]

perfect! np 🙂

[sage scroll]

Ty @ornate elk Its working! Thank you very much. I cant belive it was just a simple lora

[ornate elk]

ah nice np 😄 sometimes its just one little thing

[keen cargo]

So I purchased a subscription to Stability Audio and it says Im still on the free plan. Is this a glitch?

[hard quiver]

i figuered it out. for somereason it went away when I cut some stuff from negative prompts. as for resolution. I generally do 1080p

[ornate elk]

alright, if you used AnimateDiff extension before, you need to revert a setting that can also cause that issue with sdxl
this one:
#🤝｜tech-support message

[hard quiver]

alr. thanks

[solemn atlas]

I'm sorry that I'm asking so much but what can i do here?

[sage scroll]

Try to clean up your Harddrive (Would be my first quess)

[ornate elk]

Delete or move stuff from your harddrive,
clear the dust bin.
open up a cmd and run pip cache purge
get WizTree and check whats taking up so much space

[fair oxide]

what does this mean. i've never had it before

[ornate elk]

mostly your RAM could be filled up to the max and some programm doesnt get its stuff back

[fair oxide]

that is odd

ive never had it happen before. and it's always so early within my generations

like yesterday night i ran upwards of 50 generations and no problems

[solemn atlas]

thank you

[ornate elk]

whats your gpu? and whats inside your webui-user.bat?

and how much RAM do you have?

[fair oxide]

12 gigs

set COMMANDLINE_ARGS= --disable-nan-check --pin-shared-memory --cuda-stream --cuda-malloc

im running forge

im thinking maybe i get rid of disable nan check

[ornate elk]

yea disablee-nan-check hides only errors and fixes nothing

[ornate elk]

and how much RAM?

not gpu vram

[fair oxide]

lol 6 gigs. looking to upgrade soon

it's weird tho because ive never had it happen before

yesterday was perfect

yeah seems to have worked

interesting

[ornate elk]

oh wow, yea

upgrade is needed ^^

[fair oxide]

next paycheck haha

thanks CS

what kind of art do you make?

[ornate elk]

np, i make anything to test and help users. but mostly doing Anime stuff 😄

[fair oxide]

awesome! got a place where you upload your work? i'd love to see it if you share : )

[ornate elk]

currently not 😮 maybe i should start uploading stuff

[fair oxide]

you should!

[ornate elk]

just to make sure:
you have 6gb RAM and
12 gb GPU VRAM ?

[fair oxide]

yessir

[ornate elk]

okay

then yea definitly upgrade RAM

at least to 16gb (with 2x8gb) or to 32 with (2x16gb)

[fair oxide]

will do

[ornate elk]

that are some i made

[fair oxide]

dude those are sweet

holy mola

wow

[ornate elk]

thanks 😄

[fair oxide]

ofc!

hands are so hard hahah

[ornate elk]

true ^^

[fair oxide]

i used to have the textual inversions for bad hands that used to appear but now they won't appear in my list

which is odd

i run SDXL but they ran in the past

even tho they were 1,5 they still showed

[ornate elk]

they are base model related.
so some embeddings only work with 1.5 and some with sdxl

[fair oxide]

hm. ill lok at civitai and theyll use the same checkpoint as them but they can still use it for some reason

[ornate elk]

sometimes it can be buggy when switching models, (also some loras and mebeddings always show, no matter what model)

[fair oxide]

true

[bleak mirage]

Hello everyone. I am having an issue with A1111-stable diffusion. I am running into a problem where I have i stalled it correctly and only have 2 models and one extension. What happens is I boot up the .bat file necessary to start it but once all is loaded on the CMD the website opens but it is a never ending loading screen with the two cubes looping. What do I do. It worked wonderfully before this happened. 😬

[ornate elk]

what do you generate with SD?

[ornate elk]

hey, what does the cmd shows?
and also make sure the webui is whitelisted in any browser adblocker

[fair oxide]

i generate nsfw stuff hahah

[bleak mirage]

Hold on I got you 👌

[ornate elk]

ahh classic xD

[fair oxide]

smoke em if you gottem you know hahah

WOW

first time impaint worked for one of my hands

damn

[ornate elk]

thats rare 😮

achievement unlocked "Inpaint a Hand"

[fair oxide]

fr bruh

first time too

maybe i should go shiny hunt

what's your got to method for hands?

what do you put into your prompt when you impaint?

[ornate elk]

"hands behind back" "hands behind head"
jk xD
mostly when using sdxl or pony the hands are better.
when using 1.5 i try to not get hands in the foreground

[fair oxide]

very valid

[ornate elk]

i dont inpaint hands that much

[bleak mirage]

👌

[ornate elk]

that takes to much time and doesnt work that good

[fair oxide]

true

[ornate elk]

looks normal
make sure the webui is whitelisted in any browser adblocker

also you want to add --xformers to your webui-user.bat (for best performance)

[bleak mirage]

Ok I will but the browser I use is Microsoft edge I do not have an adblocker on it. It also worked perfectly before on the same browser 🙂

[ornate elk]

okay, can you screenshot your extensions tab and also try in an other browser

[bleak mirage]

Ok I can later I got a doctor’s appointment soon unfortunately 😭

[ornate elk]

Sure, np then let me know later

[bleak mirage]

👌

[sick crater]

Hi, I would like to analyze the embedings of this model: https://huggingface.co/stabilityai/stable-diffusion-2-inpainting Do you believe is it possible or is an copceptual error? For example analyze how have change from the embeding of the original picture and with the inpaiting

[sick crater]

Do this makes sense? XD

[azure magnet]

Hi, In terms of quality output is better SD standar or SDXL?

[ornate elk]

Hey, SDXL has better quality because its trained on 1024x1024, while 1.5 is trained on 512x512 and needs upscaling.

[azure magnet]

So, all are moving to SDXL?

[viral knot]

sd3 around the corner :3

[ornate elk]

Nope. Users with lower end GPUs will still use 1.5 and also people who like the specific styled models.
For example there are more anime models and styles, characters out there for 1.5 than for sdxl

[azure magnet]

I don't understand

[viral knot]

yea i still use 1.5 to this day

but i also use a lot of sdxl

[viral knot]

it's the new model that will surpass sdxl

coming very soon

[azure magnet]

Please, tell me more

[viral knot]

well you will know everything June 12

but if you want info right now, there are some links i could give ya, but its more technical i guess

[azure magnet]

I have a 3060 with 12 gb, probabily i'll move to an 5090 if come out in this year. I drawn anime style, should i move to an SDXL ( i think that slowly i'll have all the model in XL)?

[azure magnet]

Yes please, i set the reminder to 12. So if i understand well will come out sd3?

[viral knot]

https://stability.ai/news/stable-diffusion-3
https://stability.ai/news/stable-diffusion-3-research-paper
https://stability.ai/news/stable-diffusion-3-api
https://new.reddit.com/r/StableDiffusion/comments/1d6t0gc/sd3_release_on_june_12/

that should cover everything :3 @azure magnet

[azure magnet]

Can i steal you just a few minutes for some questions?

[viral knot]

steal me? oh my.. haha, i mean you can ask sure

[azure magnet]

steal... cultural bias to say do you have some minutes, for some questions.
So what will be the difference between SD, SDXL and SD3?
I mean the difference between SD and SDXL are how big are images trained. So SD3?

[viral knot]

i know what steal me meant, was just playing a bit with ya :3

well it's always a progress forward anyway (or at least should be), so
sd3 will train better finetunes, loras, etc cause it captures details better
and for inference it will generate better pics because it follows prompt for
example, etc. so you can very precisely do stuff that otherwise would require
either tons of tweaking, or loras or prompt engineering, while sd3 fixes that.

keep in mind, June 12 is the 2B or medium size sd3 model release, there is also
4b and 8b, so that will technically be even better, but we have to wait for those

[azure magnet]

https://tenor.com/view/excited-adorable-i-cant-wait-agt-americas-got-talent-gif-16683443

[viral knot]

yea and dont forget the community will make tons of cool stuff with it

controlnets, loras, etc

[azure magnet]

But should i change automatic1111?

or is just an update

And so, should we train a new checkpoint?

[viral knot]

well most people these days seem to use comfy i think, (dont have stats but just a guess)
comfy usually has the new stuff very early, you can find on github, comfy itself updates
very regularly, and is also very well optimized for systems so you can run under low vram, etc
very efficiently. most people start with the easy stuff like a1111, but then realize comfy provides
way better workflows or pipelines too, you can do so much more with comfy that you cant
or is very limited with the others

[keen cargo]

Bump.

[serene notch]

hello, so i'm trying to use some Xl models, it works but i'm noticing it is very slow compared to standard models, even though both my cpu and gpu barely reaches 50% utilization, why is that?

[azure magnet]

but what quality are images trained of sd3?

[viral knot]

if you are asking for the base or recommended resolution, it's 1024x1024, but it can be other dimensions

if you are interested in training, check out onetrainer or koyha ss for example

[azure magnet]

So it's like XL 🤯

https://tenor.com/view/hhgf-gif-25031041

Thanks for all your time and news.

[viral knot]

np, now back to work for me :3

[south vigil]

hey so i installed the openpose editor right now and restarted the ui but cant see the tab of it, shouldnt I be bale to find it here?
I actually also checked twice and it is installed for sure
Its just not popping up here, maybe its now on a new place than the tutorial has told from 1 year ago?
Or maybe it snot compatible with lobehub?
can anyone help?

[bleak mirage]

Here are my Microsoft edge extensions and I tried Google chrome with adblocker off. It did not work 😭

[restive sparrow]

https://github.com/facebookresearch/xformers#installing-xformers
Could anyone tell me where I input the command to update xformers new version? Im sure its not in web ui bat right? thats only for re install tyvm
EDIT: oh I guess it does update with just --reinstall-xformers, it didn't a minute ago so was sscating my head

These are my versions are they all in line with yours?tyty
version: v1.9.4  •  python: 3.10.6  •  torch: 2.1.2+cu121  •  xformers: 0.0.23.post1  •  gradio: 3.41.2

[neat cradle]

Anyone know what would cause Ultimate SD upscale to make images just pure noise?

[restive sparrow]

habe you tried a different upsclaer from the list

and sampling mehtod

[neat cradle]

Do some models have issues with certtain upscalers?

I tried Eurlar A didnt try a different upscaler

[restive sparrow]

yea ive seen it happen

[storm basalt]

hi all, is there a plan to add support for avif / jpeg / webp ?

and if so, what Exif tag shall be used to store the Parameters?

ComfyUI already pretty much agreed on using 0x010f: Make and 0x010e: ImageDescription for prompt and workflow, how about you used 0x010d DocumentName ?

[maiden totem]

So I’ve been trying to get stable diffusion on my phone and I pay for the pro subscription Collab and I couldn’t find the stable diffusion link website

[grim reef]

I think you can run it on the civitai website you dont need colab

[ruby yoke]

did you look on github?

[warm leaf]

hello, sorry for my english, i need help, i dont know why prompts/outputs LoRa has effects, i need help ty 😦 i speak spanish

[maiden totem]

Yeah I looked up GitHub and then I clicked the first link at the top

[terse lichen]

this is the cmd log but cant see anything which is causing it to not run @bitter yarrow @timid plaza @lost crescent @light glacier

[light glacier]

See #🧣｜comfy-ui

[ornate elk]

I meant the extensions of the stable-diffusion webui

[lost crescent]

In fact, I cover all these folders by having them remote from my main ComfyUI installation - accessed by SymLink

[ornate elk]

Your on an old webui version

1.7.0 and latest version is 1.9.4

[terse lichen]

meaning ?

[lost crescent]

I cured this by reinstalling - just save these folders before you delete ComfyUI - your workflows, pyssss workflows, models, input, output and custom nodes.
In fact, I cover all these folders by having them remote from my main ComfyUI installation - accessed by SymLink

[lost crescent]

Meaning when you delete ComfyUI - all your important data/large files - are preserved

[terse lichen]

but where do i get a sym link tho ?

[lost crescent]

Make an empty folder i.e. ComfyUI\models - then link that to the remote models folder e.g. ComfyUI_Remote\models - the format is this - mklink /D "ComfyUI\models" "ComfyUI_Remote\models"

So your ComfyUI folder always contains a link to your big files (long times to download) which means they will not be deleted when you want to delete and reinstall ComfyUI

This is for Windows

[ornate elk]

Comfyui also has an extra file for linking models. Its the extra_model_paths.yaml where you can specify paths to other webuis like auto1111 or folders.

[vocal burrow]

by default it s named extra_model_paths.yaml.example and therefore has no effect, you need to complete it and rename it to extra_model_paths.yaml

[lost crescent]

I can never get that Extra Paths file to work 🙂

[vocal burrow]

it explains everything there is to do tho.

What s yours ?

[dull rapids]

Hi guys! wondering if you encountered this error before? How did you fix it?

[solemn moon]

is there a way to see what the model number is without having to put it into your sd folders and then looking through the gui

[timid plaza]

Do you have a controlnet model? Take a look at which it is trying to load and make sure it's installed.

[dull rapids]

I managed to fix it! just had to reinstall the controlnet model

[timid plaza]

I'm really not sure, but it's very sensitive to versions. I ended up creating a Conda environment just for Comfy/SD. Then with all my python environments it was picking up the wrong one from base and I had to use an Anaconda shell to start it, so I made a batch file rather than having to manually start, activate environment, etc.

[terse lichen]

manged to fix it by reinstalling comfyui @timid plaza @vocal burrow @lost crescent @light glacier

[pine moth]

Is there a trick to get sd working with python 3.10.9 or 3.10.11. I'm trying to get kohya working but it doesn't like 3.10.6

[vocal burrow]

sd works with python 3.10.6 to 3.10.13 without any issues

[bleak mirage]

Oops my bad I got you 👌

[vocal burrow]

what have changed since then ?

[bleak mirage]

Well lets see it worked perfectly with the extensions and models for a while. One day I start it up and the CMD process was interrupted by a “CONFIG.json” file always being sent to the “tmp” out of suspected corruption of said file. I have done research and found out I need to delete that .json file and the web-ui config.json file. It will be replaced by new ones.

I did that and now everything works again. However now the website itself has an endless loading with the looping orange squares. That is whats wrong now. My CMD is even normal

My theory is that my SD is corrupted or broken. I am potentially thinking of a fresh reinstall 🤔

[shy river]

Hello, i recently install stable diffusion and SDXL version to try some stuff but i have a problem with it, when i start generating, i have "TypeError: expected Tensor as element 0 in argument 0, but got DictWithShape", and i have this with all SDXL checkpoint. I don't know what it is and i have no result on internet, can somebody tell me what's this and what i must do

[vocal burrow]

try deleting the venv folder too

[vocal burrow]

show the whole page

[bleak mirage]

Ooo ok and start it back up again? 🤔

[vocal burrow]

yes

[vocal burrow]

sdxl model used with sd1.5 lora

dont mix and match

[shy river]

It won't allow me to generate even if i don't use other Lora in prompt?

[terse lichen]

you need to have a compatible lora or a sdxl lora for it to work and vice versa

[vocal burrow]

what does the console log says ?

( I m leaving soon got to get a package )

[shy river]

File "C:\Users\username\OneDrive\Desktop\A1111\stable-diffusion-webui\modules\sd_samplers_kdiffusion.py", line 168, in forward
cond_in = torch.cat([tensor, uncond])
TypeError: expected Tensor as element 0 in argument 0, but got DictWithShape

[bleak mirage]

It reinstalled the “venv” file but it still is endlessly loading 😭

[ornate elk]

try deleting the ui_config.json

then relaunch the webui-user.bat

[bleak mirage]

I will try that after work 👌

[vocal burrow]

the full log

copy paste everything from the console in a .txt file and drop said file in here

[prime chasm]

Anyone know how to make sdxl sketch work? I see a bunch of models and loras on the huggingface page (here https://huggingface.co/lllyasviel/sd_control_collection/tree/main ) but no indication whatsoever of the differences between them or how to use them (is it a "use controlnet + lora at the same time" thing? etc)

[solemn moon]

Hello code bros
Is there a way to get a a notification for automatic1111 after boot up that’s says either “extensions or drivers” are outdated

Spent the day testing control nets, went back to my original settings, still didn’t work, gave up on life, remembered to check out drivers and with an update it was fixed.

A notification would be a lot easier for my laziness than having to check driver app and extensions on every start. Maybe a “git pull” vibe can work 🤷‍♂️😅

[solemn moon]

lora will be for style, control net will be for controlling, i think

[prime chasm]

Still there are a bunch of models for the same thing with no indication as to the difference

[fair oxide]

However now the website itself has an endless loading with the looping orange squares.
that sounds like what i faced once and deleting the style.csv file fixed it for some odd reason.

[bleak mirage]

Noted I will try that once off work thanks. Where would that be located 😁

[fair oxide]

.\stable-diffusion-webui\

[bleak mirage]

Fair got it 👌

[shell locust]

I'm brand new to AI and Stable Diffusion.

• Watched several videos
• installed Stable Diffusion Forge and got it working
• still learning all the settings and stuff

I want to accomplish creating this model, but can't figure out where to start especially since it seems to be based for an older version and it's at high-level.

Can I accomplish this within Stable Diffusion Forge?

Or does this need to be done outside of it?

[wet cedar]

Hello! Sorry to bother everyone.

Do YAML files go in with the safetensor files (models)?

[ornate elk]

whats your GPU? and whats inside your webui-user.bat?

[ornate elk]

hey, yes

and name them the same

[agile burrow]

Can someone help me, the .sh says this, im on Fedora 40 with Gnome 46

(I already update PIP but i dont know why it says that)

[vagrant notch]

Has anyone succeeded working with SDXL Turbo on Automatic1111 Colab ?

[ornate elk]

python 3.12 is not supported with SD, best is to use 3.10.6 up to 3.10.13

[agile burrow]

I installed it but i cant uninstall python 3.12 because it says it will remove gnome-shell

[slow garden]

I'm having this error, could I have some help please?

[ornate elk]

Your C:\ Drive is full

make sure to have 15gb free space

[slow garden]

Even if I have stable on a different drive?

[ornate elk]

yes, because for the installation it will download stuff in temp folders on C