🤝｜tech-support

[vocal burrow]

if you want to do some training definitely pick the 12gb, you'll need the extra vram to be comfortable

Otherwise the 4060 will be faster and is newer so you'll get more fancy tech for gaming and it will be supported for longer by video games

[upbeat oracle]

🙏
is there a good info/cheat sheet to check which gpu stats are important for certain stats?

the 4060 looked more attractive because of the more and newer tensor cores

[vocal burrow]

Not really but pretty much everything has been said already. newer tensor cores go faster. But more vram allows for larger operation (such as training, SDXL generation, etc)

There are 4060 16gb version, and nvidia are about to drop their new super / TI / ultra / HD remix / whatever 4xxx gpus

so prices are probably gonna drop for standard 4xxx gpus

[wind sun]

hello, does anybody know how can i train my AI in SD cuz in YT there are tutorials that are not exactly what im looking for, you see when you are in SD web there is a option Train next to checkpoint merger and that is the thing i need to know how it works

[vocal burrow]

that would be a question for #🔧｜finetune . Also check the ressources at #1080946152318443610 message

[mighty sorrel]

@karmic crown dear its performing the same as earlier

no improvement

[karmic crown]

when you run it, do you see parameters i sent you

################################################################
Launching launch.py...
################################################################
Python 3.10.13 (main, Aug 24 2023, 12:59:26) [Clang 15.0.0 (clang-1500.0.40.1)]
Version: v1.7.0
Commit hash: cf2772fab0af5573da775e7437e6acdca424f26e
Launching Web UI with arguments: --skip-torch-cuda-test --upcast-sampling --no-half-vae --medvram-sdxl --use-cpu interrogate --disable-model-loading-ram-optimization

(my params are slightly diffrent since i have more unified ram)

to check if CPU is used and how much, launch activity monitor

and select GPU Processes

then go to window and select GPU History

[mighty sorrel]

let me paste it

[karmic crown]

GPU should go up to the top during generation

[mighty sorrel]

Launching Web UI with arguments: --skip-torch-cuda-test --upcast-sampling --no-half-vae --use-cpu interrogate

[karmic crown]

those are default parameters

[mighty sorrel]

we have changed it in the file but not implementing here

should i paste your givem param in the batch file also?

[karmic crown]

it should look like this

[mighty sorrel]

i have done the same

[karmic crown]

you should get
Launching Web UI with arguments: --skip-torch-cuda-test --upcast-sampling --no-half-vae --medvram --use-cpu interrogate --disable-model-loading-ram-optimization

please send your changed webui-user.sh

[mighty sorrel]

ok

[vocal burrow]

your commandline_args is still commented

[karmic crown]

#export COMMANDLINE_ARGS
remove #

[vocal burrow]

remove the initial #

[mighty sorrel]

ok

[vocal burrow]

#export COMMANDLINE_ARGS="--skip-torch-cuda-test --upcast-sampling --no-half-vae --medvram --use-cpu interrogate --disable-model-loading-ram-optimization"
->
export COMMANDLINE_ARGS="--skip-torch-cuda-test --upcast-sampling --no-half-vae --medvram --use-cpu interrogate --disable-model-loading-ram-optimization"

[karmic crown]

that

with 1.5 model and those settings Steps: 20, Sampler: DPM++ 2M Karras, Size: 512x512
it should take 35-45 seconds on your mac

also, my MacBook works slower on battery than on power adapter

[mighty sorrel]

ok i am checking this. thanks

[ornate elk]

Hey, what was your GPU again?

I know hot to not get the Gray tiles

[rose skiff]

I am having a problem using a mask image in deforum. I made an image with the text2img tab and set it to init image (that part works) and I used gimp to fill all the parts I want to change white and all the ones I want to stay the same black, then pasted the absolute filepath in the mask file box. When I press generate, the first frame is fine, the part that I want to keep the same is fine, and the rest of the image is all the same color grey. I have tried changing settings and running it multiple times but it keeps having the same issue.

How do I make it draw in the areas that are white in the mask file instead of filling the whole area in the same grey each time?

[kindred tinsel]

Does anyone know why this error happens in CivitAI Browser plus

I downloaded the version that was out in civit ai too so idk

[upbeat oracle]

Have chosen now the nvidia geforce RTX 3060 12GB for the new system, since it's a good limited offer. thanks again 😁

[fair oxide]

For Automatic1111

If you have multiple graphics cards, how do you tell SD which one to use?

It seems to be pointing at the wrong one.

[ornate elk]

Hey, you then need to edit the webui-user.bat
At the line COMMANDLINE_ARGS=
You add --device-id 1 or 0 or 2
Depending on the Number in the taskmanager under performance tab GPU

[fair oxide]

Cool thank you!

[ornate elk]

No problem, also checkout my install guides in the pinned messages of this channel. There you also find the best commands for Max performance depending on the vram

[fair oxide]

Great thank you very much!

[ornate elk]

Your welcome 🙂

[fair oxide]

So this might be a dumb question, but I have two devices at ID 0.

I have an Integrated GPU at ID 1.
I have an Laptop 4050 at ID 0.
I have an eGPU 4070 at ID 0.

Not sure how to pick the 4070.

[ornate elk]

Oh okay thats interesting

I would say set it to 0 and watch in the taskmanager which one gets used

[fair oxide]

Hmmm. Looking at performance, I don't think the 4050 is being used. So there is likely some NVIDIA voodoo going on.

[ornate elk]

Maybe it priorises egpus

[fair oxide]

Hmmm.

[hot spindle]

I'm not a windows person but wouldnt that be a choice at driver/windows? They probably cant work at same time

Dont Win Laptops have that choice in Power settings or similar name?Performace and Energy saving settings? Something along those lines

I do remember if youhave diferent vendors you could set one for video output and another for "computing" , but same vendor was a no-go

[strange thunder]

Does anyone know where the preview images from the ksampler for comfy ui are stored? And how can I edit how often I want to see a preview? because right now comfy is showing me literally every step and I don't need all of that for several reasons.

And if there's a better place to get comfy ui help, please let me know.

[outer rapids]

i am in the right place but i am having trouble installing stable diffuion after i launch webui-user.bat i get an error of Looking in indexes: https://pypi.org/simple, https://download.pytorch.org/whl/cu118
ERROR: Could not find a version that satisfies the requirement torch==2.0.1 (from versions: none)
ERROR: No matching distribution found for torch==2.0.1 how do i correct that?

it lead me to a link but i do not see cu118

which one do i select??

would this be the correct one? torch-2.1.2+cu118-cp311-cp311-win_amd64.whl

and if so how do i install it?

ok i just tried pip install and got this C:\Users\KITTY>pip install torch-2.1.2+cu118-cp38-cp38-win_amd64.whl
ERROR: torch-2.1.2+cu118-cp38-cp38-win_amd64.whl is not a supported wheel on this platform.

[karmic crown]

#🤝｜tech-support message <- nvidia guide

#🤝｜tech-support message <- amd guide

[outer rapids]

thanks!

[lyric vector]

Is there a fix with controlnet im trying to inpaint and if i used inpaint masked only its making a whole image in the mask right now im just using inpaint whole picture

[upper thicket]

i just downloaded the program but it gives me this crappy pixelated pictures why does it not work?

[karmic crown]

your dimensions are the problem
use 512x512, 512x768 or 768x512

1.5 models are trained on 512x512 images, so when you try to generate 200x208, you can only get a mess you did

[ornate elk]

We are on it in #🏞｜general-with-images

2gb GPU mostly without performance settings

[karmic crown]

even if it was properly set he will still get a mess becuse of 200x208

[ornate elk]

Yes I already said that too 🙂

But 512x512 wasn't working

[viral sentinel]

I keep getting errors when creating embeddings. I deleted and reinstalled the venv folder.
Every time I try to train, it stops one step short of where I've set it to save the embedding (every 5 or 10 steps, it stops at 4 or 9).
Does anyone have a solution to this? I updated, and it worked, then happened again.

Logs here:

Applying attention optimization: sdp... done.
Training at rate of 0.004 until step 1800
Preparing dataset...
100%|██████████████████████████████████████████████████████████████████████████████████| 46/46 [00:02<00:00, 16.26it/s]
0%| | 0/1800 [00:00<?, ?it/s]*** Error training embedding
Traceback (most recent call last):
File "C:\Users\user\stable-diffusion-webui\modules\textual_inversion\textual_inversion.py", line 539, in train_embedding
loss = shared.sd_model.weighted_forward(x, cond, w)[0] / gradient_step
File "C:\Users\user\stable-diffusion-webui\modules\sd_hijack.py", line 144, in weighted_forward
return sd_model.forward(x, c, *args, **kwargs)

[ornate elk]

Hey, is your webui updated?

[viral sentinel]

It should be, yes.

[ornate elk]

Check the version at the bottom of the webui in browser

Also on which model do you train it?

[strange thunder]

Comfy ui question: Anyone know how to tell face detailer module which faces to apply the effect to and which faces to skip?

[viral sentinel]

I'm on 1.7 and train with v1-5-pruned-emaonly.ckpt [cc6cb27103]

[ornate elk]

Hmm okay

[vast crest]

Hello everyone, guys, just curious. Can I use 2 checkpoints together on stable diffusion to create an image?

[strange thunder]

You would have to merge them together into a new model and designate how much each is weighted if using a1111. You can do this with a lot more freedom in comfy ui as you can choose two models and weight them respectively in the node. This way you just need to have the models, you don't have to keep large files of each different blend, you're free to blend how you want.

[vast crest]

'd like to try it. Would you be able to find me a tutorial? However, I use the version that opens the browser with table diffusion, and I always forget what it's called XDDD

[strange thunder]

sounds like a1111.

https://openart.ai/workflows/academy This guy has a series of videos explaining comfy ui. He talks about model merging pretty early (though I forget the which video) on but you'll need all the info in order to navigate comfy ui at least semi competently. lol

[ornate elk]

I can give you a quick explanation in the checkpoint merge tab of auto1111 when I'm at home

[vast crest]

Ok,thx 🙂

[vast crest]

Thx 🙂

[strange thunder]

np. my 5th help to someone else vs like 1000 of others helping me XD

[vast crest]

XD

[ornate elk]

Merging:
Here is a quick explanation of the Checkpoint Merge tab of auto1111:
If you have 2 Models (A, B) always use "Weighted Sum":
It will merge them with this formula where M is the Slider Scale: A * (1 - M) + B * M
It will merge the two with the Percentage of the Slider (0.5 = 50%).
If you have 3 Models (A, B, C) always use "Add Difference":
It will merge them with the formula where Slider is M: A + (b - C) * M
In Fact it will erase all of the C Model of your final Merge.

"No Interpolation" is only used when you have 1 model (its for converting ckpt to safetensor, baking in a vae or just renaming).
ckpt and safetensor are the output files you can choose for the model.
Choose "dont copy config" if your models are based on 1.5 or sdxl
fp16 is for converting a fp32 model into fp16 (used for amd or new nvidia cards(30,40).

[gritty ridge]

If you have to choose between Razer Blade 17 64GB Ram with NVIDIA RTX 4090 and M3 Max Chip 128 GB RAM and 8 TB SSD also for model training and video etc what are recommendations? Thanks 🙏🏿

[zealous vale]

I reinstalled the webui to update recently. Anything I have loaded besides the 1.5 safetensor gives me something like this. No settings have been changed.

[gritty ridge]

@karmic crown If you have to choose between Razer Blade 18 64GB Ram with NVIDIA RTX 4090 and M3 Max Chip 128 GB RAM and 8 TB SSD also for model training and video etc what are recommendations? Thanks 🙏🏿

[karmic crown]

i am late for a meeting, i will answer later

[proper pike]

Comfyui constantly crashes after trying to render more than 50 frames of an animation in animatediff, does anyone know why? 32gb ram, rtx 4090, 13900k
sd 1.5 not sdxl. upscaled to 1024x1024

[fair oxide]

https://civitai.com/models/16014/anime-lineart-manga-like-style?modelVersionId=28907

I've been using auto1111 for awhile now, but Im struggling to figure out how to get imgtoimg work on something like this model

its like 16 megabytes so I dont understand how it works with sd 1.5 unlike other models that are much larger

and use base sd 1.5

am I supposed to mix this somehow with another model?

I added the safetensors file to my E:\AUTO1111\stable-diffusion-webui\models\Stable-diffusion (personal model folder) and I cant figure this one out

I have coherent 3d model renders, I just wanted to play around with this like a filter for some edges

I think it'd be called a controlnet lora?

sorry its been so long since I many any change to automatic1111, installed awhile ago

don't have time to install updates and learn the landscape / fall down rabbit holes on cutting edge software stuff unfortunately

Im hoping to use either https://civitai.com/models/16014/anime-lineart-manga-like-style?modelVersionId=28907 or https://huggingface.co/lllyasviel/control_v11p_sd15s2_lineart_anime

processing should be done in 16:9 if possible, if not I just need some of the line processing to avoid rotoscope for some of this, that's the main goal, some batch ai processing basically

[vocal burrow]

It's a lora, so that's not a stable diffusion model per say

LoRa/Lycoris quick guide :
1/ Where do I put my LoRa : stable-diffusion-webui\models\Lora, Loras are NOT stable-diffusion model but rather patches for said models.
2/ How do I use my LoRa, you can either :

• Invoke it directly in your prompt by using the following syntax <lora:filename:weight> (or <lora:alias:weight> depending of the setting in setting->extra network). eg : <lora:animeTarotCardArtStyleLora_v20Offset:1> for using the Lora animeTarotCardArtStyleLora_v20Offset.safetensors
• Use the UI "Extra Networks" / card system by doing as described in my screenshot. Clicking an associated lora card should automatically insert the lora prompt liked mentioned above. https://cdn.discordapp.com/attachments/1002602742667280404/1191779742534348870/loraHowTo17.jpg
  3/ Be careful, LoRa are trained specifically for some SD version. Lora trained for SD1.5 (and derivatives) models will not load on SD2 models and vice versa (same goes for SDXL). Also even though Lora trained on SD1.5 models should work for every SD1.5 derivatives, they'll probably work best with the model that has been used for its training.
  4/ LoRa, LoHa, LoCon, what are these ? They're all commonly refered as "Lora" but differ slightly technically :
• LoHa : Lora with hadamard decomposition
• LoCon : Lora extended to redisual blocks.
• More explanations : https://stable-diffusion-art.com/lora/, https://www.sdcompendium.com/doku.php?id=lora_0014, https://www.sdcompendium.com/doku.php?id=locon_training_0071, etc

[unreal terrace]

I'm using A1111 generating at 960x540, and just noticed that the actual images generated are 960x536 -- anyone have any idea what might be doing that?

I checked discord history, the SD reddit, and A1111's git issues, but can't see anyone reporting something like this

I only backtracked it to this because all my upscales to 1920x1080 were shading the top 8 pixels gradually darker and only switching my upscape to "1080x1920" (which still generates landscape) removes the shading, but the resulting upscale is then 1920x1072

I know it's not the ideal aspect ratio for SD, but from research alot of people are genning initially at 960x540

[vocal burrow]

540 is not dividable by 8

536 is and it s the closest value so it uses that.

[unreal terrace]

but that's 1080p / 2 -- is there a better initial resolution to end up with 1080p upscaled?

I figured using clean ratios was best

[karmic crown]

NVIDIA RTX 4090 works extremely fast with stable diffusion, but I honestly have no idea how fast M3 Max with 128GB of unified RAM works compared to that, so I would go for RTX 4090 if your main goal is stable diffusion

I know that M2 Ultra with 128 GB works relatively fast, but it's slower than many Nvidia GPUs
@hot spindle can probably give you more information about times needed for some tasks on that Mac when he find some time

or it was 192GB? i forgot

[vast crest]

CS1o is there a video tutorial on YouTube?

[hot spindle]

Blender render benchmarks puts the lower M2 ultra (the less number of GPU cores - 60 -but mine has 192GB VRAM/RAM) in line with a 3070 (or maybe 4060 cant remember), but Blender has optimized Metal processing done by non other than Apple own developers (Apple contribution to Blender), of course the beefed up M2 with 76 GPU cores will score higher , cant remember exactly but its easy to search in Blender benchmark page

But that's half the story, there are some renders that are off-limits to GPUs because they are so texture intensive, a M2/M3 with large RAM will come very handy there , why Pixar (RenderMan) is mostly done on CPU render farms

[gritty ridge]

Thank you all @hot spindle @karmic crown I see that apple will soon improve PyTorch and find solution with not access CUDA in relation to model training so just checking if there are really dismissive views about Mac and risks from people working on … as I am apparently more keen on buying MAC in the moment and 128 GB RAM

[vocal burrow]

In the end mac machines represent a minority of the stable diffusion population. So you ll have less support regardless of how capable it is.

[karmic crown]

my m1 macbook pro works like a snail compared top MCurots mac

i have only 16GB of unified ram

[gritty ridge]

At least there are two:) anyway info about blender quite precious , and the goal that it is moving to Metal completely… well not the matter of really crazy fast model training for me, more fear of insanely slow and lack of access to some platforms that could mean potentially more skilled programming power… to replace it or whatever… so i double check… but it’s hard to believe that Mac could stay so long behind in the near future …

Thanks a lot

[dusty meteor]

is it possible to simply resize an image using A111 SD UI, without making any changes to it, apart from perhaps cropping? i don't even want to use a model, since i don't wish to change the image at all, only improve details if i'm enlarging. maybe there's a specific SD model solely for resizing I don't know about?

[sonic ferry]

If my KSampler is at 3cfg then my Ultimate SD Upscaler has to be at 3cfg too? I got a big mess at 12 steps, 7cfg on the Upscaler, using the same sampler, same scheduler.

[karmic crown]

Steps: 20, Sampler: Euler, Size: 1024x1024
Time taken: 3.7 sec.

NVIDIA A100 80GB

now let's cry together 🤣

[karmic crown]

if you are talking about images you generate with SD, then use Hires fix with proper upscaler and settings
but if you want to upscale any image, then you need to use img2img with a model and probably an IP adapter and some more controlnet adapters
probably someone else has a more accurate answer than me

[gritty ridge]

CFG scale max / Denoising str min img to img

[wind sun]

Hello so i preinstalled my windows and now when im getting reactor extension its not working for some reason

[karmic crown]

try to remove reactor from extensions to some other place
does it work than?

have you installed this
VS C++ Build Tools (if you don't need the whole Visual Studio) and select "Desktop Development with C++" under "Workloads -> Desktop & Mobile": https://visualstudio.microsoft.com/visual-cpp-build-tools/

[wind sun]

yup

i installed it

oh shit

wait i selected only python as option in vsc

i guess its not a problem if i change the directory

[karmic crown]

i would install VS C++ Build Tools instead

it is smaller and faster to install

i am using xcode command line tools instead of full xcode on mac for the same reason

[wind sun]

right

[karmic crown]

i guess, i am on mac 🤷🏻‍♂️

[wind sun]

now works, thanks

[fair oxide]

rx580 series 8gb

sorry couldnt answer yesterday

[ornate elk]

Okay then you should try 512x768
Upscale by 1.5
Tiled VAE decoder to 1024 and Encoder to 128

[umbral plover]

What am i doing wrong? Trying to apply pose from image so new one is similar but i get nothing even remotly close

[astral comet]

So i've tried creating a model with pictures of myself, following two different tutorials that had me go through a process within a Google Colab document. In both instances, I got blocked somewhere along the way because of some thing being outdated, or incompatible, or whatnot. Both tutorials were over a year old, anyone knows a more up to date tutorial?

[fading ferry]

hi

something went worng and i only got these kind of results:

i use stabilty matrix

[vocal burrow]

sd21 model (therefore trained for 768x768) combined with 512x512 output

[fading ferry]

still bad

and every model gone crazy

not just 2.1

[vocal burrow]

what's your hardware and settings (screenshot your whole webpage)?

what command line args ?

[fading ferry]

i have an rtx 4060ti

[torn lotus]

what do you do to cool-down your GPU while running SD? I get peak temps of 90*

[vocal burrow]

physically clean it ?

[fading ferry]

uh, nothing?

[torn lotus]

it's pretty much clean at 1st inspection

[fading ferry]

do you mean its overhetaing and bacuse of that i got bad results

[torn lotus]

I don't know your case

[umbral plover]

maybe wrong vae?

[fading ferry]

where can i change that?

[umbral plover]

Hmm not sure didnt use matrix, can you give screen shot of whole window? Maybe it's somewhere

[fading ferry]

sure

[umbral plover]

Disable refiner and check again

[fading ferry]

nothing happenned

[umbral plover]

Have you tried any other prompts? Did they produce same problems?

[fair oxide]

any way to get preprocessor controlnet outputs only in batch?

comfyui doesnt support it even according to the developer

and previewing preprocess working amazingly well in auto1111

[vocal burrow]

definitely disable refiner it doesn't make much sense to use a refiner with a non sdxl model, and try another sampler

[fair oxide]

but not in batch

about ready to make a macro in firefox to do this instead already stuck for 2 hours

seems like no one saves the preprocessor output for images only for controlnet or lineart models

im surprised outputs like that are niche

[vocal burrow]

If there is a way, I don't know about it, sorry

if you want, it seems to be exposed in the API https://github.com/Mikubill/sd-webui-controlnet/discussions/1807

but yeah it will require some dev

[gusty solstice]

Could you tell me if a RTX 4070 TI could be enough for model training (image generation)

How much would it take for training a human face? For example mine.

Some hours? A day? More?

[vocal burrow]

Yes.
It depends, hours.

(would be easier with a 16gb vram gpu for sdxl training, but should be possible with 12gb with some trade-offs)

[steady thistle]

Is there some way to find either a very very small checkpoint or use no checkpoint for face swapping with a 0 denoise? I want as much memory as possible dedicated to the face swapping.

[astral comet]

anyone?

[stray tulip]

I'm trying to train a sdxl model using webui and dreambooth but after the first few phases it just stops with no error just "initializing bucket counter" 0/89000

Using 4500 images, tried it twice 1hour each run. Anyone know how to fix this?

[zealous vale]

I get stuff like this when using checkpoints other than 1.5. I have not changed any settings after installing

[ornate elk]

What's the checkpoint?

And how big is the checkpoint file?

[zealous vale]

animagineXLV3_v30 safetensor (6.46gb)

[ornate elk]

Thats an sdxl based checkpoint, these files are trained on a 1024x1024 resolution

So dont use 512x512 as thats for 1.5 ones

[zealous vale]

Oh, thanks

[lucid vessel]

Excuse me , how to control the potential of the Lora?

[upper thicket]

jeez the end resoult was awful

[lucid vessel]

you are doing what?

[lucid vessel]

where i can see the pictures that i make on stable diffusion?

[upper thicket]

i tried to take the lineart of picture one on lineart mode and add the colors from picture b in shuffle mode, but the resoult was awful, dont know what im doing wrong

[upper thicket]

heres the sample pictures

[lucid vessel]

hmm

never tried this

[lucid vessel]

i guess its more easy to ,describe the colors

[upper thicket]

yeah maybe i should add more prompts

[lucid vessel]

i am trying to make a good anime design

[sharp lichen]

Whats the best way to sub out the background of an image but keep the subject? some sort of controlnet?

[shrewd solstice]

I just made the upgrade to a 7900XTX and was wondering if there was an optimal way to run SD on these RDNA3 cards.

[shrewd solstice]

I could be totally wrong here, but my first thought is to use inpainting around the subject.

[sharp lichen]

I think that is the basic way to do it, but I remember seeing a controlnet for it

[shrewd solstice]

I'm certain there's some automated method for it yeah. This stuff has progressed a lot since I last ran anything locally.

[thin quarry]

There's probably something wrong with your gpu cooling, 90*c is too high.
until you'll figure it out I'd suggest to not run SD, that might end your gpu, do quick tests in FurMark , on lower resolutions first.

[shrewd solstice]

Yeah that sounds like it's time for a cleaning at least.

My trusty Titan Xp got a repaste, new pads, and a good wash this year before it was retired to the player 2 system.

[karmic crown]

do not use all, try to use line art for unit 0 (your drawing) and reference for unit 1 (reference image)

did you download controlnet models?

i am sorry, but i can't stay longer and help you more since i am currently experiencing severe neck pain 😦

[ornate elk]

You dont have set the lineart model and preprocessor

[ornate elk]

Hey, I have the same GPU,
the best way currently for AMD cards is to run them on Linux.
But if you want to use it on windows its slower but works too.
For an SD Windows install guide check the pinned message of this Channel

[potent hatch]

Hello, guys when installing SD on pc, i am getting this error, can anyone help?

@ornate elkyes i am in Asia, and also using VPN, but still this error popping each time! any idea how to resolve this?

[ornate elk]

Try change your DNS

[potent hatch]

checked with closign firewalls, turning off antivirus software, same message keep coming!

[potent hatch]

change to what?

[ornate elk]

8.8.8.8 (google)
Or 1.1.1.1 (cloudflare)
Or 9.9.9.9 (Quad9)

[potent hatch]

let me try

[potent hatch]

same! 😦

[ornate elk]

Open up a cmd and type

Ipconfig /flushdns

If that doesn't help then maybe use an other VPN
Like Proton VPN

Or try without a VPN one time to test

[potent hatch]

tried all that already, w/o vpn, with vpn, different dns settings, running cmd with admin, etc.. not sure whats the issue here!

is it github has issues? i see others having same problem?

[ornate elk]

Did you connect the VPN to Europe servers?

[potent hatch]

Japan, then US, both not working!

[ornate elk]

Strange

[potent hatch]

??

seems i am not alone!

but how to resolve this! i cant find any answers

can anyone help

[potent hatch]

is it about heavy traffic on there site?

[ornate elk]

Hmm I dont think so
For me the Github page loads normaly

[potent hatch]

is there any other way to download the files manually? not with cmd?

[ornate elk]

Can you open the automatic 1111 github page in browser?

I still think something blocks your connection

[potent hatch]

yes i can

its working fine

idk people saying i need some token or api to download it? is it?

[ornate elk]

No

But if you download it manually you won't be able to update it via git

[potent hatch]

i just followed a tutorial to dl SD just stuck on this setp

step*

[ornate elk]

Do you have Git installed?

[potent hatch]

yes

[ornate elk]

You can also checkout my install guide in the pinned messages of this Channel

[potent hatch]

https://archvizartist.com/article/how-to-install-stable-diffusion-on-windows-automatic1111/

i have use this guide

[ornate elk]

Is the VPN you use from your country ?

[potent hatch]

thanks, i'll look at it now

[ornate elk]

Some vpns can still filter

[potent hatch]

yes i was using Astrill before, just switch to the local one now

[ornate elk]

Would suggest protonvpn as its Switzerland based

[potent hatch]

but if github site is working why the files not?

[ornate elk]

Some port is blocked (443) that is normaly a firewall port

[potent hatch]

well i am based in Shanghai, i am not sure if proton can work here good?

[ornate elk]

It won't be fast but the files are small for the download

[potent hatch]

hmm may be thats the issue then

[potent hatch]

i'll check it out, do they have a trial period?

i could just use it to download files 🙂

[ornate elk]

There is also HideMe VPN that has a 10gb limit

I would restart the PC, install a different VPN and then connect to a Europe based server

[potent hatch]

i'll look into it, hope this resolves the problem!

what about the graphics card? i dont have very powerful card? will SD work fine in gtx 1080ti

[ornate elk]

Yes that will work good

In my guide you find the best performance commands for it

[potent hatch]

perfect, i'll look into your guide now! thanks for all the help.

[ornate elk]

No problem hope it works 🙂

[potent hatch]

yeah, me too! 🙂

[fair oxide]

Application not responding why is that?

[ornate elk]

Do you have a screenshot of the error?

[ornate elk]

Oh the bots

[fair oxide]

Can't create anything

[ornate elk]

The bots are off rn. The Team is informed and will restart them soon

[fair oxide]

Thank you 🥰

[ornate elk]

No problem 🙂

[stray tulip]

I'm trying to train a sdxl model using webui and dreambooth but after the first few phases it just stops with no error just "initializing bucket counter" 0/89000

Using 4500 images, tried it twice 1hour each run. Anyone know how to fix this?

[umbral plover]

Im getting error when trying to run prompt, propably something to do with models but not sure with which

[steady thistle]

I got an error with ReActor. It was working earlier today, had a freeze, restarted and I'm getting this error:

08:51:15 - ReActor - STATUS - Working: source face index [0], target face index [0]██████| 8/8 [01:46<00:00, 14.63s/it]
08:51:15 - ReActor - STATUS - Analyzing Source Image...
2024-01-20 08:51:15.8425698 [E:onnxruntime:, inference_session.cc:1799 onnxruntime::InferenceSession::Initialize::<lambda_9423999eb39fb4b5a7f23bdd285c385e>::operator ()] Exception during initialization: bad allocation
*** Error running postprocess_image: D:\AI\stable-diffusion-webui-directml\extensions\sd-webui-reactor\scripts\reactor_faceswap.py
Traceback (most recent call last):

[unique wedge]

Hey ı got that problem how can ı fix that. Anyone know about it ? My computer property is gtx 1660ti Mobil

[unique wedge]

Thanks a lot, I didnt see that

[nocturne sleet]

Hi everyone, I want to use the program, but my laptop has an amd processor and amd video card, can this laptop run the program or is there an amd version (Laptop Dell G5 Se AMD Rx5600M 6gb)

[vocal burrow]

cf pinned posts

[nocturne sleet]

thx

[split sorrel]

Hello, I am encountering this error while using REactor Is there a way to fix this error?

my gpu amd 6700

[ornate elk]

Hey, Can you show your txt2img settings while using it?

[umbral plover]

bump

[split sorrel]

ofcorse

reactor on stock

[ornate elk]

Reactor is not for img2img

If you want to generate an image with a face then use txt2img.
If you want to swap faces between two images then you need to go in the Extras tab

[split sorrel]

Yes, I want to change the face in the image. What should I do? Actually, I can do it, but when I use risize to, I get the error.

[ornate elk]

you need to use the Extras tab

then use Reactor in there

you cant resize an image to that high resolution, the out of memory error is normal

you need to upscale images using the sd upscale script or the ultimate upscale extension

[split sorrel]

so thank you

I want to change the face in the video using Mov2Mov, I also get an error, what should I do?

[ornate elk]

whats the error?

[ornate elk]

resolution to high

[split sorrel]

I was able to run it once but it doesn't work now

[ornate elk]

everytime you get that error you need to restart the whole webui

[split sorrel]

yes that's what i do

[split sorrel]

This is the video resolution but can I do it any other way?

[ornate elk]

you need to use a smaller resolution

as output

[split sorrel]

Even 512x512 didn't work

[ornate elk]

oh okay

what was your gpu?

[split sorrel]

6700 😦

[ornate elk]

hm that should work

[ornate elk]

maybe you can convert the video into single frames and then batch faceswap them, then convert them back to video

[split sorrel]

Yes, actually mov2mov does this, but the resolution of the video is also very important. Isn't there an alternative way?

[ornate elk]

lol i already said the alternative above

an other alternative is an nvidia gpu with more than 12gb vram

but i would just lower the end resolution

you can even upscale it after that again

[ornate elk]

you could try remove --no-half from the webui-user.bat if its in there, but i still think the resolutoin is to high

[split sorrel]

Thank you very much

I managed the low rates.

[ornate elk]

best way would be: convert video to frames > batch faceswap in img2img > convert them back to video
That works better than mov2mov because it doesnt need to recreate the whole imge just for the faceswap

the only problem is just the high vram usage of the resolution + extensions

[split sorrel]

Yes, actually I tried from there but I'm getting an error.

[ornate elk]

also using smaller controlnet models can help in less vram usage

[split sorrel]

When prices are low the outcome is very bad

[ornate elk]

thse for example:
https://civitai.com/models/38784/controlnet-11-models

[split sorrel]

ı try but I don't understand why I can't do it here while I can do it in the extra tab.

I think I will do it manually, but 😄 450 files will be a bit difficult.

[ornate elk]

It will take time xD

[sonic ferry]

How do mutations start happening on the upscale, too many steps? too high a cfg?

[ornate elk]

Too high denois

[sonic ferry]

0.56 too high? Actually I think that's what came default on the Highresfix so I thought it'd be okay there too. :/

[ornate elk]

Depends on the upscaler

But mostly with 0.5 your fine

Also upscale by 2 and not by 4

Lower denois = less changes to the image

[sonic ferry]

2x UltimateSDUpscaler using 4xNMKD_Siax

[ornate elk]

So your not in txt2img?

[sonic ferry]

Uh, I am...am I using something that's usually img2img?

[ornate elk]

Ultimate SD upscaler is an script for img2img

Hires fix is for txt2img

[sonic ferry]

Really? So I'm always gonna get strange results with it hooked up to the KSampler?

I'll try Hires Fix again I guess, I should be able to do 2x on it?

[ornate elk]

Ah your using comfyui

Yea you first should use hires fix

Then after that you can use the ultimate upscale

[umbral plover]

Is hiresfix just latent upscale?

[sonic ferry]

Now I'm more confused 😄

[ornate elk]

Yes but depends on the upscaler

[torn spear]

Hello, I have a question

[ivory bramble]

Sorry wrong chat

[sonic ferry]

Am I using HighRes fix wrong? Cause I can do 2x with UltimateSDUpscale in like 1/8 the time that Highres says it will do 2x, both at 10 steps.

[umbral plover]

I think hires works on latent and USDU on pixels, maybe difference comes from here, but yeah its strange

[ornate elk]

If its technical you can ask here

[torn spear]

Sorta, I was using Stable Diffusion, and it wouldn't generate the image I want. There was something about --disable-nan-check

[ornate elk]

Okay, then you mostly used a lora as model and that won't work

[torn spear]

Oh.

[ornate elk]

Make sure every file in the models/Stable-diffusion folder is larger than 2gb

Everything below 2gb isnt a model

[torn spear]

Okay let me see

[torn spear]

I'm looking at the file within the model folder, and it does not display the file size.

[lucid vessel]

whats is the better to have a good background on normal models?

[ornate elk]

Inside models/stable-diffusion

[torn spear]

Ahh, I see it. They are lower than 2GB

I see

[ornate elk]

If you downloaded them from civitai check the info box on their site

[torn spear]

Oh okay

[ornate elk]

Every file there has an info box what it is

Lora, checkpoint, embedding etc

[torn spear]

Thanks for the information. Let me check civitai again

I will come back if I need more help

[ornate elk]

Loras go into models/lora

[torn spear]

Thank you, I was just begging for something to help me

[ornate elk]

No problem 🙂

Also its faster to ask the question instead of "I have a question"

[torn spear]

I see

Sorry about that. Many people told me that, and I keep repeating that mistake

[ornate elk]

This Channel is for all technical questions about SD, usage, extensions etc.
And then there is #📝｜prompting-help for image related questions

[ornate elk]

Yea, people won't respond if they get no information

[torn spear]

Yeah lol

[sonic ferry]

What the?! I have seed locked, got an img from Highres, reenabled UltimateSDUpscaler, but it's regenerating the Seed locked KSampler? Or is it because I don't have the Highres locked.

[reef smelt]

Hey, a year ago I couldn't run 1024x1024 on medvram with 6gb vram but now I can? and 1024x1536..?

tho I might be mixing it up but

[sonic ferry]

The updates maybe? Cause I too am seeing some speedups, seemingly running the same stuff I am right now generating images in 1/4 the time.

[reef smelt]

Yea it's unfortunately hard to figure out what is changed

[reef smelt]

shouldn't high res follow normal seed?

[sonic ferry]

That toggle to use the same is there...I should try that next time lol

[swift kayak]

what version of stable diffusion is the amd pin using? it can't be sdxl

not sure what parameters I should be tweaking but the images hardly look anything like what I describe...

my gpu matches the requirements for sdxl in #1072229020520947753 so I was hoping to use that one

[ornate elk]

Hey, did you used my guide?

[swift kayak]

yes

it looks like it's using 1.5

just trying to find how to use sdxl

[ornate elk]

You just need to download the sdxl model

And put it into the models/Stable-diffusion folder

[swift kayak]

just all of the files in the directory?
https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/tree/main

[ornate elk]

No only the .safetensor file

[swift kayak]

alright, will do that 😁

[ornate elk]

Sdxl 1.0 with 0.9 VAE for example

[swift kayak]

i understand, thanks

[ornate elk]

No problem 🙂

For any other SD amd related questions feel free to ask

[swift kayak]

alright, will do. super glad I found this discord, wouldn't have gotten this far without your guide 😛

[ornate elk]

Glad to hear 😄 Thx, there are manny guides out there but manny of them are outdated or missing some essential steps

[swift kayak]

i think that was the problem when I tried to get SD working a few months back. i was probably following an outdated tutorial

[odd rivet]

@ornate elk should i unistall other versions of Python?

[ornate elk]

Yes if you dont need them

[umbral plover]

Im trying to inpaint in fooocus but i dont know if its not working or is it just because of model. There is not even a hint of color difference. Can you help?

[astral comet]

So i've tried creating a model with pictures of myself, following two different tutorials that had me go through a process within a Google Colab document. In both instances, I got blocked somewhere along the way because of some thing being outdated, or incompatible, or whatnot. Both tutorials were over a year old, anyone knows a more up to date tutorial?

[torn spear]

Sorry if this sounds like a dumb quesiton, but to re-access stable diffusion , do I just go back to the folder and click on the user ?

[odd rivet]

i installed everything and the new python 3.10.11 from your link, did everything as you say on the pinned message, but when i start the bat file it says "python not found"

i'lll try to restart the pc

[ornate elk]

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

[ornate elk]

yes, always launch the webui-user.bat to start it

[torn spear]

Thanks. I got it to work now too

[astral comet]

anyone?

[odd rivet]

ok i didnt add the .exe path when installing now i re.installed now its working

[karmic pike]

@ornate elk Hello I read your tutorial what do you recommend I do if I have another version besides 3.10.6 in my path and I cant change it. This is the current path of my 3.10 python.exe "C:\Users\USER\AppData\Roaming\Microsoft\Windows\Start Menu\Programs\Python 3.10\Python 3.10 (64-bit).lnk"

[ornate elk]

.ink is not .exe

thats not the right path

[karmic pike]

"C:\Users\USER\AppData\Local\Programs\Python\Python310\python.exe"

[ornate elk]

thats the right one

[karmic pike]

so how would I do this. I would clone the repo and before launching I would switch the python version and add these params Edit the webui-user.bat (right click, edit), At the line COMMANDLINE_ARGS= You add: --xformers --no-half-vae When would I create my venv? where would it be located at and when would I delete it. I am using anaconda right now

@ornate elk

@oblique hearth Hello zinger sir, can you help me

[ornate elk]

you clone the repo, and then edit the webui with the args + the new python path

then launch

[ornate elk]

you dont need to create a venv, it will be created automaticly

[swift kayak]

when trying to inpaint with seemingly any model, I run out of vram immediately

sd inpainting runs out within a second

if I make the inpainter downscale it to 512x512 it works.

[ornate elk]

you need to select inpaint area "only mask", then set the resolution to 512x512

[swift kayak]

ah ok

[ornate elk]

also for AMD you need to add --no-half to the webui-user.bat
for Inpaint to work

but that will increase the vram usage a lot

[swift kayak]

but mine is 6xxx, the guide says that's for 5xxx?

[ornate elk]

the guide dont covers this special case for AMD inpainting

[swift kayak]

ahhh

now inpainting is working, thanks much 🙂

[molten meteor]

I'm curious if there is a solution where:
you run model locally e.g. have control for all things like Loras, checkpoints etc, for example run ComfyUI ** but use remote computing power**

[ornate elk]

The webui needs to be installed on the remote pc and than can be set to use it from anywhere else

[molten meteor]

something like Collab for example, but on collab u can;t have 24/7 session

[ornate elk]

runpod.io, paperspace, vast.ai for example

[karmic pike]

ok i finished the install @ornate elk

is that it no deleting of venvs?

[ornate elk]

okay, if your done, you dont need to delete the venv

[karmic pike]

can u help me how do I finetune with LoRa

I want to finetune LoRa realisticvision is that possible

@ornate elk

[ornate elk]

do you want use a lora, or do you want to create one?

[polar cliff]

should sdxl be taking significantly longer than the base model to generate images? It's taking 15s/it for me when i usually get 3-4 for others (even on the same resolution). I know it produces larger images but from what ive seen (and from what I've done in the past) it doesnt take THAT much longer, is this true?

using the base model at 1024x1024 with 30 steps

turned on medvram and it decreased the time drastically

[ornate elk]

whats your GPU?

[polar cliff]

2080 super

[ornate elk]

ah okay, then you should checkout my install guide in the Pinned Message of this channel, there youll find the best performance settings you need to do for sdxl and in general

then it will be much faster

spoiler you need: --xformers --no-half-vae --medvram-sdxl

[polar cliff]

Alright sweet
Is --xformers still real? I heard that people were using --opt-sdp-attention instead, i have no idea the difference though

[ornate elk]

xformers uses less vram than --opt-sdp while both are the same speed

[polar cliff]

Alright

[shrewd solstice]

Thanks for pointing me in the right direction!

[karmic pike]

I want to create my own lora on my own person

[shrewd solstice]

This is actually something I'd like to do as well. Not urgently but I would like to feed my own portfolio through and see what I can get.

[ornate elk]

for now thats not possible on AMD, but maybe later in the months, or use google collab or other web services

[ornate elk]

for that you should use the khoya_ss tool

its an standalone Lora creation programm

[karmic pike]

ok would u recommend that over something like dreamscape

or whatever the dream version is

[karmic pike]

arent you able to train using google collabs though

[ornate elk]

if you mean Dreambooth, thats only for models not loras

[karmic pike]

ok

what other extensions do you recommend

I see the human skeleton one

[ornate elk]

Controlnet

[karmic pike]

or better img to img tools

yes

which one is the human skeleton one

is that controlnet

[ornate elk]

Controlnet

[karmic pike]

ok

[ornate elk]

Its a very powerfull tool

[karmic pike]

any other recomendations

[ornate elk]

booru autotag extension

very handy

[karmic pike]

how would I install these exactly

just download through chrome and place in respective folder in a111

or inside the UI

I download and install

[ornate elk]

click on the Extensions tab, Click on Available, click on "load from", then you get a list
search for booru

[karmic pike]

thankyou

[ornate elk]

its called Booru tag autocompletion

[ornate elk]

controlnet is named sd-webui-controlnet

you can install them and then you need to restart the webui

but for controlnet you need some additonal files
you find them here:
https://civitai.com/models/38784/controlnet-11-models
these models (you can select them at the top) (each 700mb) + its config file, need to be placed inside the models/controlnet folder

[karmic pike]

where do I get the config file? @ornate elk

and I dont understand which model should I use and their differences I see no documentation

which one would you recommend? openpose is the one that I believe I see YTs use. Am I supposed to use more than one model?

[proper pike]

Does anyone know how to get better prompt traveling results?

beyond blurry and noisey

[ornate elk]

You can start with openpose only
Select it at the top of the link I gave you. Then look below the download button there is the blue word "config"

Download both and put them into models/Controlnet

[karmic pike]

can you help me after installing both control net and webuitagcomplete

• "styles.csv" missing
• Tag Autocomplete issue

venv "C:\Users\USER\Programming\AI\Photo Generation\stable-diffusion-webui\venv\Scripts\Python.exe"
Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug  1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
Version: v1.7.0
Commit hash: cf2772fab0af5573da775e7437e6acdca424f26e
Launching Web UI with arguments: --xformers --no-half-vae
Style database not found: C:\Users\USER\Programming\AI\Photo Generation\stable-diffusion-webui\styles.csv
Tag Autocomplete: Could not locate model-keyword extension, Lora trigger word completion will be limited to those added through the extra networks menu.
ControlNet preprocessor location: C:\Users\USER\Programming\AI\Photo Generation\stable-diffusion-webui\extensions\sd-webui-controlnet\annotator\downloads
2024-01-20 19:20:58,619 - ControlNet - INFO - ControlNet v1.1.432
2024-01-20 19:20:58,683 - ControlNet - INFO - ControlNet v1.1.432
Loading weights [15012c538f] from C:\Users\USER\Programming\AI\Photo Generation\stable-diffusion-webui\models\Stable-diffusion\realisticVisionV60B1_v51VAE.safetensors
2024-01-20 19:20:58,991 - ControlNet - INFO - ControlNet UI callback registered.
Creating model from config: C:\Users\USER\Programming\AI\Photo Generation\stable-diffusion-webui\configs\v1-inference.yaml
Running on local URL:  http://127.0.0.1:7860

@ornate elk

[karmic pike]

@eternal elbow can u halp me plz

[eternal elbow]

please stop pinging everyone and be patient for a response

[fair oxide]

i used to have stable diffusion but i had uninstalled it cause at one point it took to much space but i got it back and the only problem is now when i was isntalling it i followed pretty much everything but im getting exit code 9009 python not found which

[karmic pike]

Stable Diffusion can't find python a needed dependency

1. you need to type python in your cmd to check if you have it
2. can you check your PATH in environment variables both user and system PATH and look for anything python related

3. also right click webui-user.bat in stable-diffusion-webui folder and tell me if on the python line you have any python.exe linked

@fair oxide

[odd rivet]

i've tried the inpaint sketch and it's so so laggy

it goes at like 0.5 frames per second

[lethal nest]

how to fix hand and fingers?

[reef smelt]

I can't merge models? I'm trying to merge two models but it just doesn't do anything when I press merge?

also this happens...

[odd rivet]

where do i find generated images? like in which local folder?

[reef smelt]

it's slower than generating with txttoimg

[odd rivet]

i'm not talking about generating, i'm talking about "drawing" with the brush on the image, it's almost impossible due to the lag

[reef smelt]

Hmm you sure you got the specs for sd?

[odd rivet]

128gb of ram, Threadripper and a 4090, is that enough? 🤓

[reef smelt]

Well how would I know? 😓

Do you have any extensions?

are you using stable diffusion automatic1111?

[worldly jetty]

Wherever you saved your thing

[karmic pike]

yes I would make sure u have everything properly installed

including extensions

[worldly jetty]

Sounds weird, did you update controlnet and update A41

If yes then uh, how are you trying to draw

[worldly jetty]

Just run "git pull" on the folder, it should restore styles.csv, but even then if it's missing it doesn't really matter

Dunno what extension autocomplete is, but try updating the extension

[karmic pike]

ok

[ornate elk]

Make sure the webui is whitelisted in any browser adblocker

[ornate elk]

These two messages can be ignored

Nothing broke

[ornate elk]

You need to reinstall python an check the box "add python to path"

[lost crescent]

If I have this number, how do I find the checkpoint it refers to? 9aa0c3e54d

[tiny garden]

I'm having a rather interesting problem:
I get to run a generation on the newest NoEscape model (which is v10) on setting 512x768, upscale by 2,25 with batch count 9 and it runs fine. Another run on the same setting produces me this error:
Exception in thread MemMon: Traceback (most recent call last): File "threading.py", line 1016, in _bootstrap_inner File "C:\SD WebUI\webui\modules\memmon.py", line 53, in run free, total = self.cuda_mem_get_info() File "C:\SD WebUI\webui\modules\memmon.py", line 34, in cuda_mem_get_info return torch.cuda.mem_get_info(index) File "C:\SD WebUI\system\python\lib\site-packages\torch\cuda\memory.py", line 618, in mem_get_info return torch.cuda.cudart().cudaMemGetInfo(device) RuntimeError: CUDA error: out of memory CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect. For debugging consider passing CUDA_LAUNCH_BLOCKING=1. Compile with TORCH_USE_CUDA_DSA to enable device-side assertions.
I'm using an HP Omen gaming laptop with these specs:
Intel Core i5-13420H (13th generation) 3.4GHz (12 CPU's)
16GB RAM
nVidia GeForce RTX 4050 Laptop with 6GB VRAM
Windows 11

Once this error shows up, my A1111 completely refuses to do anything... Which leads to me doing a full restart of A1111 to have it working again.
Where does this error come from? And how do I fix it?

[ornate elk]

Paste it in the searchbox on Civitai.com

[lost crescent]

Yes thanks, I found it was NightVisionXL 😄

[ornate elk]

You need to edit the webui-user.bat
At the line COMMANDLINE_ARGS=
you add: --xformers --medvram --no-half-vae
Then save and relaunch

[tiny garden]

I had to do something similar on my computer, but had to use the argument --lowvram, as that computer's running only a GTX 1050

[ornate elk]

Yea it depends on the vram.
4-6gb needs --medvram
And below --lowvram

But xformers will help even more

[tiny garden]

xformers is in the arguments, but medvram wasn't

[ornate elk]

Ah okay xD

[tiny garden]

probably missed that little setting while transitioning from my computer to the gaming laptop I bought last black friday

testing it right now

[tiny garden]

@ornate elk it works fine again - thanks

[ornate elk]

Perfect no problem:)

[ornate elk]

Hey, did you get it to work?

[worldly rain]

when I click on generate, process ends in a few simple steps here and nothing changes in my image, why is this happening?

[ornate elk]

Can you show the full settings?

[ornate elk]

Your using inpaint + controlnet?

[worldly rain]

yes

i inpainted the hair

[ornate elk]

For inpainting always select inpaint area "masked only"
And then set the resolution to 512x512 (that won't be the output resolution)

[worldly rain]

alright

also in img2img I set denoising strength to 0 because even if I increase it a little bit I get a result like this:

I used controlnet ip adapter

thats all I used, just wanted to swap the face

should I play around with this to fix it?

[ornate elk]

I didn't tested IP-Adapter in img2img

So I dont know if it will work

But I will test it maybe later

I always use IP adapter in txt2img

[worldly rain]

alright thanks a lot!

I used to use foocus before for stable diffusion, it has a quick precise face swap option, it used to swap the face perfectly but in A1111 its more detailed. is there any way I can find out which face swapper does it use?

[ornate elk]

Mostly they used Reactor

Its perfect for photorealistic images

But nothing more

[worldly rain]

but it used to swap the freckles and small details too

ReActor cannot do that as much as I know

[ornate elk]

Interesting

Maybe the use Facefusion

Or faceswaplab

[worldly rain]

i tried faceswaplab its also different but I will try Facefusion, thank you so much 🤝

[ornate elk]

No problem, I still prefer IP-Adapter for faceswap

As you can also faceswap into any style of a model

Like comic, drawing etc

[worldly rain]

thats true that option is pretty useful 😄

[upper thicket]

is there a way to make a more accurate merge of both pictures?

[hallow latch]

If I notice the page is slow and etc, can I somehow type something in the console to release it from ram and etc?

[ornate elk]

IP-Adapter could help here more than reference

[ornate elk]

Nope

closing is the only option to release vram

[hallow latch]

Okay 🙂

[echo brook]

I can’t use Tile Diffusion, I got this
‘’runtimeerror: expected query.size(0) == key.size(0) to be true, but got false’’
It was fine like days ago but now it just gives me this error

[coral wedge]

Sup everyone. Can someone help me how to fix "indexerror: list index out of range"?
This error occurs on the 5 stage of EbSynth Generation

[lost crescent]

How do I do face improvement in ComfyUI?

[vapid sluice]

Why can’t i load any SDXL models in ComfyUI? It was working few hours ago. Now it keeps saying “press any key to continue” for no apparent reason

[vapid sluice]

No error message, just “press any key to continue”

[upper thicket]

doesnt work. when i choose ip adapter, it completly ignores the lineart and it just make random pictures

[reef smelt]

I'm trying to use openOutpaint
but I get the error
RuntimeError: Sizes of tensors must match except in dimension 1. Expected size 2 but got size 4 for tensor number 1 in the list.

It's strange because using a non-inpaint model works but when I an inpaint model it gives me an error...

[ornate elk]

which ip-adapter model you used?

[terse sun]

what does that mean? I couldn't find this "clip" option anywhere

[fair oxide]

i uninstalled and reinstalled python and imported it in dif files but still didn't work idk whats going on tbh and i cant finder webui-user.bat i see in stable diffusion that there are python files in that folder so idk why it says its not installed

im honestly willing to hop in c all

if someone wants to help

[ornate elk]

Run a cmd and type
Where python

And
Python -version

[ornate elk]

Go into Settings, user Interface, Quicksettings, and there add
Clip_last_layer
Then hit apply and relaunch the webui

[reef smelt]

Is openOutpaint outdated?

[ornate elk]

Last update 5 months ago

[sonic ferry]

How do you get this? Video made it sound like it's part of ComfyUI's ControlNet Auxiliary Preprocessors But I'm not seeing this.

[reef smelt]

Mine seems broken or my models seem incompatible, do you have any tips?

[ornate elk]

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

[reef smelt]

Yes

[ornate elk]

Also dont use LCM or turbo models

[reef smelt]

hmm what's lcm?

[fair oxide]

i opt out of all those LCM samplers and Turbo models

[ornate elk]

Special models that may not be compatible with the openoutpaint

[reef smelt]

I'll try sd 1.5 inpainting

[ornate elk]

@reef smelt that could be an alternative, its a standalone tool.
https://github.com/Acly/krita-ai-diffusion

[fair oxide]

there is a quite a bit of detail loss with turbo and lcm models

ofc due to their nature of reducing sampling steps

[reef smelt]

CS1o for some reason inpanting models don't work with openoutpaint
"Sizes of tensors must match except in dimension 1. Expected size 2 but got size 4 for tensor number 1 in the list." from "sd-v1.5-inpainting" but my checkpoint might be out of date but I don't think it gets updates

the krita plugin seems nice

[fair oxide]

tried running that last part and uh yeah nothing happened does python need to be installed on my d drive with my stable diffusion cause i tried doing that but im guessing as i see not it really didnt install on my d drive'

[vocal burrow]

you didn't tick the "add to PATH" checkbox during install

[fair oxide]

thank you

gonna try it now

[fair oxide]

anything im missing out? also do i install it into my stable diffuser folder or just my regular d drive

im changing it from my c drive to my d drive

[oblique hearth]

it'd naturally go into your OS drive.

but, doesn't really matter cuz windows will handle the pathing via regedit anyway.

[fair oxide]

yeah but my space is low and im compensating

lol

[oblique hearth]

doesn't matter, worst case scenario you wipe the drive where you have it installed and your apps using Python won't work anymore 😛

[fair oxide]

true

[oblique hearth]

so which UI are you installing?

[fair oxide]

wanna hop in vc? still kinda struggling thinking of just restarting everthing cause its pretty botched lol

[oblique hearth]

no mic but i can watch at least

[fair oxide]

which vc

[oblique hearth]

idk, u pick

[vocal burrow]

wherever you want (I wouldn't install it in the same folder as stable diffusion)

[gloomy wave]

is it possible to update a model/checkpoint in automatic 1111 or is the only way to go from old versions to new ones is replace the file?

[ornate elk]

Every model is a file on its own. There is no way to update a model other than replacing it

[sage escarp]

I keep getting this types of images when using Vision V6.0 B1 model with img2img. Everything works fine with different models. What am I'm missing?
Thanks!

[onyx solar]

What is your VAE setting?

[sage escarp]

This what I'm using -
How can I see the settings?

[onyx solar]

Try setting it to Automatic so that it'll used the baked in VAE.

[sage escarp]

That seemed to fix it. Thanks!

[sage escarp]

Now I tried turning it back on and the images are turning out OK.

Don't really understand the logic 😅

[onyx solar]

Various models bake in a VAE and it's typically better to use them when they do that.

[signal sigil]

Hey guys! I have a question, if you don't mind.
I'm training a SDXL LoRa with 79 images in it. (Concept)
If i was to train this with it's current image count, how many steps do you think i should use to train it?
Also, repeats?
Thx. 🙏

[astral comet]

are a ''model'' and a ''dataset'' the same thing?

[signal sigil]

Nah. A dataset is used to train models.

The model is the thing you end up with, when it's finished.

The dataset is the ingredients, and the model is the cake.
(Basically)

[astral comet]

@signal sigilaight thanks. noob question, how do I manually connect my drive to a google colab?

[copper thunder]

How do I join?

[signal sigil]

That's one thing i don't know, because i don't use colab m8. I'm a "local" user.
I'm sure someone here will know.

[signal sigil]

Join what m8?
You're already here. So you should already be part of the discord channel now.

[zealous cargo]

is there a way to change the vram mode manually on comfyui

[signal sigil]

Not sure.. Been a while since i've last used it. But isn't the command placed in the "Run" file?

By right clicking and then edit?

[vocal burrow]

chances are you don't want to do that.... if it autodetects low vram it's probably for a reason. What's your gpu ?

[zealous cargo]

1650 super

i ran sd with medvram

comfyui works fine but i just wonder if you can change that

[vocal burrow]

it should be --normalvram

iirc

yup that's it

[zealous cargo]

where is the file directed to change it

[bold herald]

What do I do??

[astral comet]

hey guys I have a question for training a lora. i've seen in multiple places that it's good to have shots where we see the whole body. will it still train correctly if the full body images are 512x512, or should I make them in a different resolution?

[vocal burrow]

There's no file to change, it's the command line argument you're supposed to add when calling it main.py

It's easy to make one tho. There you go
(replace the python location with your own or leave empty if your default/system one is compatible)

@zealous cargo

[zealous cargo]

thanks 💯

[vocal burrow]

delete this folder your c:\stable-diffusion\stable-diffusion-webui\repositories\stable-diffusion-ai folder and try again with a proper and stable connection

[astral comet]

anyone?

[bold herald]

Thanks!

[bold herald]

Hey, how do I train stable with my own photos?

[fallow mountain]

I keep getting this message " ModuleNotFoundError: No module named 'jsonmerge'" no matter what I try. I've run install pip jsonmerge, completely reinstalled after deleting repositories and venv. Been at it for hours so I guess it's time to ask for help

[torn spear]

How do I use the Lora ? in Stable diffusion ?

[vocal burrow]

would need a proper full log

[vocal burrow]

LoRa/Lycoris quick guide :
1/ Where do I put my LoRa : stable-diffusion-webui\models\Lora, Loras are NOT stable-diffusion model but rather patches for said models.
2/ How do I use my LoRa, you can either :

• Invoke it directly in your prompt by using the following syntax <lora:filename:weight> (or <lora:alias:weight> depending of the setting in setting->extra network). eg : <lora:animeTarotCardArtStyleLora_v20Offset:1> for using the Lora animeTarotCardArtStyleLora_v20Offset.safetensors
• Use the UI "Extra Networks" / card system by doing as described in my screenshot. Clicking an associated lora card should automatically insert the lora prompt liked mentioned above. https://cdn.discordapp.com/attachments/1002602742667280404/1191779742534348870/loraHowTo17.jpg
  3/ Be careful, LoRa are trained specifically for some SD version. Lora trained for SD1.5 (and derivatives) models will not load on SD2 models and vice versa (same goes for SDXL). Also even though Lora trained on SD1.5 models should work for every SD1.5 derivatives, they'll probably work best with the model that has been used for its training.
  4/ LoRa, LoHa, LoCon, what are these ? They're all commonly refered as "Lora" but differ slightly technically :
• LoHa : Lora with hadamard decomposition
• LoCon : Lora extended to redisual blocks.
• More explanations : https://stable-diffusion-art.com/lora/, https://www.sdcompendium.com/doku.php?id=lora_0014, https://www.sdcompendium.com/doku.php?id=locon_training_0071, etc

[torn spear]

Thanks

[fallow mountain]

what's the best way to send a full log without clogging up the chat?

this is after another fresh install

[terse sun]

how do you have preview images on lora? It would be so much helpful, i usually remember them by names, even then some of the names are so convoluted.

mine looks like this

[upper thicket]

ip-adapter_sd15

[hollow pike]

My suggestion is to get the CivitaiBrowser+ extension and download everything from there - it automatically downloads to the right folders and adds the preview images. Or else you can manually add your own images as the thumbnails in the preview options somwhere there.

[terse sun]

okay, lemme try

[terse sun]

nothing showing up

[hollow pike]

try civitai brower plus

[terse sun]

nvm

found it

had to manuallt scroll and search lol

manually*

[vocal burrow]

Preview syntax rules, put .png next to the [embedding/model/etc] and name as such :

• Embedding : [embedding].pt -> [embedding].preview.png
• LoRa : [lora].safetensors -> [lora].png
• Models : [Model].ckpt/.safetensors -> [Model].png

[abstract gate]

Hi, this is the first time i am posting on here, i have been trying to get the REST API from Stability AI to work for my iOS App. The Text-2-Img and Img-2-Img works fine, it is the masking/inpainting which is giving strange results (i have tried lots of models, and variations including sending a Black mask (original image plus the mask) and the alpha Mask (Just the init image with a transparent Mask). The results being returned appear to be cropped images inside the mask. To me it appears that the Model has generated a full size image (to fit the dimensions of the original image - 1024 x1024 or 512 x 512), and then has only made the pixels over the masked area visiable in the final imnage - it gives a cropped appearance and unusable: here are some examples :

Example 1 - mask is a quarter of the screen and you can see the next image is cropped?
Example 2 - a full screen mask shows the full screen image - uncropped , but obviously a bit pointless
Example 3 - half screen mask - again the image appears to be missing a chunk
Example 4 - the last image is from Dalle-2 API, using a qaurter screen mask, and the dinosaur is added and uncropped the. to correct area.

I am sure this is gonna be something super simple i am doing wrong with the. REST API

Thank you in advance!

[unique anchor]

I get this error:

Type error occurred: unsupported operand type(s) for +: 'NoneType' and 'int'

The WebUI Returns the partial B64 of the generated image then shows this error

[livid wind]

HI does anyone know how to get magic animate working with 8gb of vram

[gloomy wave]

Thank you for answer. Hopefully tools for local ai creations will get easier to use in the future. Is it even worth updating most of the time?

[ornate elk]

Np, i think its already easy to use, depending on the webui. But for Models its always the same as the files are "pretrained", so the file is done.
You dont need to download every new model version, just the ones you like.
for example the new models from end of 2023 are better than the ones from end of 2022.
Whats important imo is to update the Tools for ai stuff. so the latest version of the webui has the best support for files, extensions, models etc

[ornate elk]

someone got it to work with 10gb by lowering the resolution, the team of magic-animate will work on it so you just have to wait a bit. hopefully we see some massive performacnce improvements the next weeks or months

[vapid sluice]

How to shorten model loading time for sdxl in comfyui? I use RTX 3060 12gb vram and it takes tooooo long

[gloomy wave]

This is good to know. The part with replacing is not exactly complicated but you know auto update lazy part ;)- . The webui does get better for sure with each update.

[vapid sluice]

I get the feeling that the loading time increases exponentially as i start a new session

[vapid sluice]

Now comfyui is as slow, if not slower than Automatic 1111, it was working fine few day earlier

It always gets stuck at “got prompt”, and I can’t find a single fix on google

[vapid sluice]

Should I reinstall 😭😭😭

[red loom]

How to change the setting using api so that I can apply the color correction functionallity to img2img

[violet plinth]

how to prevent kohya ss saving lora during training multiple times inside one epoch? i have set "save every N epochs" to 1 but it also saves it every 29 steps which slows training a lot

[lost crescent]

Help - ComfyUI Custom Node - "Cannot import B:\ComfyUI\custom_nodes\bilbox-comfyui module for custom nodes: No module named 'colour'"

[lost crescent]

pip install colour - simple answer really 🙂

[fallow mountain]

If I could have some help with this, I’d be forever grateful

[sonic ferry]

1.5x Highres Fix, same seed toggled, is 0.56 denoise too high? I keep getting scars, growths, and tumors seemingly half way through 10 steps. Eh, looks like some of it went away toward the end, now onto the Ultimate SD Upscaler, 1.5x at 0.35 denoise, fixed seed set to 0...not sure if that's the right way.

[steady thistle]

Is there a way to use Openpose with Turbo SDXL?

I'm getting poor results

[ornate elk]

are you on AMD ?

ok, then set the inpaint area to "only masked" and the resolution to 512x512 (thats not the output resolution)

[astral comet]

anyone wanna help a brother out in vc for a couple minutes? honestly my question is too vague to write down, would be super appreciated

[sonic fable]

Can someone tell me, how to use starlight filter in stable diffusion?

[unique anchor]

Anyone knows how to get rid of this error?

"Type error occurred: unsupported operand type(s) for +: 'NoneType' and 'int'" "

[torn lotus]

hi. what does it mean: the control net model is not installed?
I'm using comfyui for krita

[unique anchor]

Man I have been stuck on this error since Friday 😭 Can someone help

[ornate elk]

you need to download the controlnet models needed to use controlnet

[ornate elk]

More information can help, what webui are you using, what are you trying to do? Any extension used? Is your webui updated?

[unique anchor]

I finally fixed it, it was just an error in my code that used SD Webui, my bad.

I forgot that tracebacks exists for logging

[ornate elk]

Ah okay

[formal trout]

hey I'm uploading a 1344x1152 img into inpaint tab (which I outpainted 2 times to the right with the Outpainting Script), and it just load it cropped as the first outpaint only. Img load good in PNG info and img2img.

is this the solution? https://www.reddit.com/r/StableDiffusion/comments/10huun7/inpaint_preview_frame_size_in_automatic_1111/

what should I put change there in that case? And which style file?

[ornate elk]

You need to set the Inpaint Area to "only mask"
And then set the resolution to 512x512 (that won't be the output resolution)

[formal trout]

I did that and still load it cropped in the rigth side, weird

I loads it from left to right?

exactly cropped at 1088 × 1152 pixels, so 1344x1152 looks like 1088 × 1152 pixels

[ornate elk]

What are your img2img settings?

[ornate elk]

And the output image is cropped?

Any scripts active?

Oh you mean its cropped if you put it into img2img

Make sure the webui is whitelisted in any browser adblocker

[formal trout]

Ohhh nevermind I discovered a very tiny thin scroll line bellow the inpaint img frame that let me see the rest of the image

I alt+scroll wheel and only zoomed in that area, only way to access the rest of the image is by that scroll thing

[ornate elk]

There should be other key combinations you can use. Check the little Info dot at the top left corner of the image

[formal trout]

cool, I'm trying S and F that I never tried, ty

[sleek gyro]

Has anyone had this error?

ModuleNotFoundError: No module named 'numpy._utils'

[ornate elk]

The whole cmd log could help

[sleek gyro]

Could you explain a little more, I'm not that technical

[ornate elk]

Sure. You should copy the whole commandline windows output In here so we can find the cause of the error

Also it helps if you describe what you tried that resulted in this error

[sleek gyro]

I'm just launching the webui-user.bat to open the UI and getting this error:

....\Stable D\stable-diffusion-webui\venv\lib\site-packages\numpy_globals.py", line 20, in <module>
from ._utils import set_module as _set_module
ModuleNotFoundError: No module named 'numpy._utils'

[ornate elk]

Is this after you first installed it?

[sleek gyro]

No. I’ve been using SD about a month and all of a sudden this happend

[ornate elk]

Ah okay. What's your GPU and what's inside your webui-user.bat?

You can right click and edit the webui-user.bat

[sleek gyro]

I have NVIDIA GeForce RTX 3060 Laptop GPU

[sleek gyro]

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=

git pull
call webui.bat

[ornate elk]

Okay at the line COMMANDLINE_ARGS=
you add: --xformers --medvram --no-half-vae
Then save and close.
Then delete the venv folder.
Then launch the webui-user.bat
That will rebuild the venv and can take a bit.

[sleek gyro]

Got it. Let me try

[sleek gyro]

It is working now! Thank you so much for your help 🙏 I spent 3 hours googling and trying to solve this before came here

[ornate elk]

No problem 🙂 it should also be faster now too

[sleek gyro]

Yeah, much faster 🔥

[ember venture]

I'll try that for the problem I'm having, my performance fell from 1.1 it/s to 6s/it

[ornate elk]

What's your GPU?

[ember venture]

Tesla M40

I'm not at that computer right now though

[ornate elk]

Ah okay, that card should use:
--xformers --no-half-vae
If sdxl is slow you can add --medvram-sdxl

[ember venture]

It was running fine until two weeks ago ish, then Auto1111 stopped running at all and I'd get dependency errors every way i tried to fix it. SDNext runs fine, then it went from 1.1 it/s to 6s/it a week ago

[ornate elk]

Ah okay, deleting the venv folder in auto1111 fixes a lot

It gets clean rebuild after launching the bat again

[ember venture]

I tried that repeatedly

[vast crest]

Hi guys, sorry to bother you, but I always have problems. I wanted to create a gif with Animatediff but it gives me this error: OutOfMemoryError: CUDA out of memory. Tried to allocate 496.00 MiB (GPU 0; 11.99 GiB total capacity; 9.75 GiB already allocated; 0 bytes free; 10.98 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF
how can I solve it?

[ember venture]

Is that the first thing you try to do after starting?

[ember venture]

Which is a bit annoying because SDNext doesn't have the online integration for remote control over internet

[small ore]

I have a question. How do you batch preprocess an image sequence through controlnet in A1111? I'm testing the new Depth Anything models (That don't work in Comfy), so I'm having to use Auto instead. I basically just want to create a sequence of depth maps from Auto. Nothing else.

This is trivial to do in ComfyUI, but as I said, the depth anything models aren't working in Comfy yet

I just want to load a batch of images, run them through controlnet, then save the output in a folder.

[ornate elk]

Then you can show the error you got here. Maybe a quick reinstall is needed in the worst case

[ornate elk]

Ah thats unfortunate. What I use sometimes is RustDesk for Remote Control the PC

[ornate elk]

What are you trying to do?

[vast crest]

hi, yesterday you told me to ask for help in the tech.support section. so I copied and pasted the error message that appears when I try to create an animation with animatediff

[ornate elk]

Ah okay. That error means your run out of vram.
Are you're using xformers?

[fair oxide]

this works on any AMD gpu right?

[ornate elk]

Yes it will work

Follow the guide and keep an eye on Step 5-7

[fair oxide]

alr ty imma try it

[vast crest]

Yes, I always use the xformers command

[ornate elk]

What was your GPU again xD sry always forget

[vast crest]

My PC: CPU Ryzen 7 5800x - GPU RTX 4070 OC 12 GB - Ram DDR4 32 GB 3600 Mhz

[ornate elk]

And what are your AnimatedDiff settings?

Resolution etc

[vast crest]

resolution 528x768

scale 7 - steps 30

[ornate elk]

Hmm looks normal

[tepid vortex]

I want to use stable diffusion to generate embroidery for a piece of clothing I have with an inpaint sketch. What models would be good for this?

[fair oxide]

what does this mean "no module 'xformers'. Processing without..."

[worldly bough]

it means you don't have xformers module

don't remember how it works under the hood but it reduces memory usage and increases speed

pip3 install xformers btw. Doesn't exist for AMD branches though. Don't do that if you're using rocm it will try to replace rocm version of pytorch I believe.

[ornate elk]

That means you dont have xformers installed. Thats okay as it only works with nvidia GPUs anyway

[fair oxide]

oh alr

[fair oxide]

what are StableStudio | StableSwarmUI, I saw they use WebUI are they local as well? like I can use them unlimited amount of times and fast?

[ornate elk]

Stableswarm ui is local too. But idk if it supports AMD GPUs

[fair oxide]

hmmm I might test it and see

[vocal burrow]

Everything local is pretty much unlimited, that s the point of local instances.

[ornate elk]

It uses Comfyui as Backend so it should work with directml

[unborn mauve]

i'm having this error

while trying to use control net with animatediff

[severe prairie]

Help me please, because I don't see where I can find preprocess images While using train

[sharp lichen]

Is it normal for auto to run relativly fast when you start it and then after a few jobs it gets slower?

in sdxl

[ember venture]

That could mean your gpu is thermal throttling

[west star]

how much vram do i need for animatediff xl? i have 16gb and it crashes (with an out of memory error) before i even see 1% progress

[sharp lichen]

oh really

what is that

how do i fix

rtx 3080

[torn lotus]

hello. I downloaded "ControlNet openpose model for SDXL" but still getting the same error

[ornate elk]

can you show the full cmd log?

[ornate elk]

what yre your animatediff settings and resolution?

[west star]

context batch 16, 8 fps, 64 frames
896 x 1152 50 steps
i tried lowering width and height but it would still oom
going any lower would just make the output bad/not worth it for me

[ornate elk]

are you using xformers?

[west star]

yup

[ornate elk]

hmm

[west star]

is it supposed to work with 16 gb?

[languid rock]

Hello, i have been tring for hours to get SD to launch. Windows machine, i install Git, Python 3.10.6, launch the cmd and install SD, add models, but when i click webui-user.bat the cmd prompt flickers away and nothing happens. i just click the launch.py file and it's dl'ing 2.6gigs of stuff for some reason, hoping thats an update i need. any thoughts? thank you

i will say, i couldnt get SD to dl from the cmd prompt without starting it in admin, i can't seem to get anything else to work by starting them in admin however

[torn lotus]

this might be an issue, as the error is generated from krita

[ornate elk]

checkout my install guide in the Pinned Message of this Channel, and reinstall

[languid rock]

will do, i'm reading through it and i'm not terribly savvy in certain areas of computing. every time i try to edit my .bat file it requires i save it elsewhere and won't replace the current file due to permissions. any way around that as i am on the admin account? thank yo u

[ornate elk]

check the folder / path your in

for me it sounds like you have git cloned the repo inside the system32 folder (which would be very bad)

[languid rock]

Aw crap! you're right! its in my program files!!!!!!! i see you mention not putting it there!!!! this could be it 🙂 🙂

[ornate elk]

yes :), thats it

[languid rock]

ITS DOWNLOADING!!! thanks a ton!!

[ornate elk]

dont move the folder, always git clone it into a fresh created folder like in my guide

[languid rock]

thanks! found that out the hard way 🙂

[ornate elk]

no problem 🙂 also dont forget to edit the webui-user.bat

[languid rock]

now i probably can haha!

[ornate elk]

yea should work 😄

[ember venture]

It means your system is heating up to the point that it reduces power to not overheat and damage itself. That might not be what you have, but it fits the symptoms. Do you notice the same thing with games or rendering? Runs well a few minutes then slows down?

[bronze leaf]

I have a prompt that is not processing. It has been that way for about half an hour now. Whenever I try to make a new prompt, I get a Pending Request Error. What can I do?

[ornate elk]

check the cmd console

[bronze leaf]

Sorry, I don't know how to do this.

[ornate elk]

oh wait, do you mean the discord bot here?

[bronze leaf]

Yes

[ornate elk]

which bot channel are you using?

did you tried to use an other one?

[bronze leaf]

Bot 8 when it happened, but I've tried using all of them with the same result

[ornate elk]

strange, i tested it rn and it worked

maybe you need to restart discord completely

[topaz maple]

does anybody know why scale by 0.5 is doubling my image size? I thought it should half the image size=

[bronze leaf]

I have refreshed, updated, and restarted discord and the problem persists

I have to go now, I will check back tomorrow. I may need to uninstall Stable Diffusion completely

[arctic fox]

Does anyone have this issue? (https://www.reddit.com/r/StableDiffusion/comments/194838f/issues_with_lora_on_stable_diffusion_with/)
I'm dealing with the same issue.

[ornate elk]

okay, let us know if the error persists

we had some server issues today with the bots

[ornate elk]

the easiest solution is to not use olive

it doesnt support everything

and it also wont get fixed, as it wont get updated

You should use the normal Directml fork of auto1111 without Olive and onnx, for AMD

[arctic fox]

I see, thank you!

[ornate elk]

For an install guide checkout the Pinned Messages of this Channel. There are also the best performance commands included.

If you want to change your version to the non Olive version, let me know

I'm using AMD too

[torn lotus]

i got the log from Cui

Import times for custom nodes: 0.0 seconds: D:\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI_JPS-Nodes 0.0 seconds: D:\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI_IPAdapter_plus 0.0 seconds: D:\ComfyUI_windows_portable\ComfyUI\custom_nodes\comfyui-tooling-nodes 0.0 seconds (IMPORT FAILED): D:\ComfyUI_windows_portable\ComfyUI\custom_nodes\ControlNet-v1-1 0.0 seconds: D:\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI_UltimateSDUpscale 0.3 seconds: D:\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-Manager 0.9 seconds: D:\ComfyUI_windows_portable\ComfyUI\custom_nodes\comfyui_controlnet_aux

clearly the install was erratic. got a guide to install controlnet?

[mystic basin]

can anyone help me with this? i try to generate an image, 1920 by 1080 but this happen in the middle. i don't understand why

like it stop working

the usage of the vram of my gpu increase and the use of it too but then i stop for i don't know what reasosn

[onyx solar]

Well, first off, none of the models handle 1920x1080 natively. Could you make one at (close to) that resolution with an XL model? Potentially...but your success rate would be much lower than staying within the 1 megapixel size for XL models or the ¼ megapixel size for 1.5 models.

You should be generating at lower resolutions and upscaling after.

[mystic basin]

oh that's why

thanks for the info

[ornate elk]

not for comfyui sry

[ember venture]

Back home, wiped out venv and reloading it, let's see if it runs well again

[mystic basin]

i've got another problem. it's not that it's not working but stable diffusion doesn't seem to find my model. i don't know if it's me that have put them in the wrong folder or not

[ember venture]

4.5s/it at 512x512, 6s/it at 512x768. Built-in benchmark still gets 1.1 it/s though

[onyx solar]

After putting them in the folder, did you refresh/restart A1111?

[mystic basin]

reload the webui? user thing

[onyx solar]

You can either reload the checkpoints or restart the entire thing.

[mystic basin]

nope, it doesn't find them

[ornate elk]

What's your GPU?

[ember venture]

Tesla M40

[onyx solar]

From what it looks like, those should be in the right place, but since your file list doesn't show the base stable-diffusion-v1-5 file, I'm guessing you might have dropped those into a different directory. You should do a find for that base file, then put the other checkpoints where that is.

[ember venture]

I thought I had xformers loaded but I actually didn't, that might be part of it, it seems to be installing it now

[mystic basin]

there?

[onyx solar]

Your install is looking somewhere other than the models\Stable-diffusion folder it seems. But I would suggest that you try moving them to see if they show up after refreshing the checkpoints.

[mystic basin]

alright i'll try

[ember venture]

All right, now we're getting 1.4s/it, much better, at x2 batches, let's see if batch size is what does it

[ornate elk]

Did you also add --medvram?

[ember venture]

Pretty sure I have

[ornate elk]

Use batch count instead of batch size

[mystic basin]

i'll reload the web ui user thing sicne when refreshing, it doesn't seem to work

[ember venture]

Before two weeks ago I was getting 1.1it/s on x4 batch count parallel

[ornate elk]

Count or size?

[mystic basin]

nope, it still doesn't show

[ornate elk]

@mystic basin @onyx solar hes using the AMD directml webui BUT with olive. So its normal that is doesn't found the models because they are not converted to the onnx format

[ember venture]

Size. nvidia-smi reports that x1 batch size is only using 1/3 of total memory, each additional one in parallel increases utilization and before auto1111 started refusing to run it wouldn't lose performance on larger batches until it ran out of VRAM

[mystic basin]

what can i do then?

[onyx solar]

Ah...well, that makes sense.

[ornate elk]

Don't use olive and reinstall the AMD webui using my AMD guide from the Pinned Messages in this Channel

[mystic basin]

alright thanks

[ornate elk]

After that the models will work

[ember venture]

Now nvidia-smi says driver/library version mismatch, library 535.154

[ornate elk]

Is your driver updated?

[onyx solar]

Just make sure you move the models you downloaded out to another folder before you wipe out the current install so you don't have to re-download them.

[ember venture]

This morning it was version 535.146, which was the only version that auto1111 tolerated, then it stopped running entirely and I switched to SDNext

[mystic basin]

thanks, i didn't think about that

[onyx solar]

It's a common thing to forget when you're trying to solve problems.

[ornate elk]

Maybe you should check what's the latest driver for that gpu

Also make sure your webui is updated

[ember venture]

535.154 is newest, previous one is 535.146 from December

[ornate elk]

Is your webui version 1.7.0, torch 2.0.1 and xformers 0.0.20 ?

Check at the bottom of the webui in browser

[mystic basin]

nice tutorial

it seem to work for now

[ember venture]

I'm on SDNext, not auto1111, the latter doesn't work at all on this system anymore. SD.Next version from Jan 13. Xformers 0.0.24. Torch 2.1.2+cu121

[ornate elk]

Ah okay

I throught you deletet the venv and it worked again

[ember venture]

No, I was talking about SDNext the whole time. Auto1111 still wouldn't work after reloading venv

[ornate elk]

Ohh ok

Any error?

[ember venture]

Struggling to update drivers, still seeing errors for driver/library mismatch but it is running

And maybe I wasn't paying enough attention before and I only thought I was getting 1.1 it/s, I tend to set a batch then go do something else and check later

It seems to be working well enough now. On to trying to train models

[mystic basin]

thanks for the tutorial CS1o

it work perfectly now

👍

[ornate elk]

Perfect, no problem 🙂

[ornate elk]

If you have 6gb vram or less you can't train models.
But with 6gb vram you could try train a lora

[ember venture]

I have the rare 12GB Tesla M40, I could have gotten 24 but I set aside a certain amount of money and couldn't afford the bigger one

I'm trying to train loras but it keeps halting after a few minutes

[ornate elk]

Ah okay 12gb is enough for both

Can't help with Khoya_ss as I used it a year ago

I'm off now gn

[wide thistle]

Hello, I can't seem to run pytorch on Ubuntu 22.04 with RX 7600 (installed Rocm6.0.0) (torch.cuda.is_available() always returns false)

Had to build pytorch manually

[glossy prairie]

I'm getting an error when trying to use Animatediff with ControlNet; it worked a few days ago but I updated Controlnet and guessing something changed with how Animatediff is calling a ControlModel reset().

Does anyone know if there is a way I can resolve this error, or do I need to wait for Animatediff to make changes? Thanks!

*** Error running process: C:\Users\shaun\ai\stable-diffusion-webui\extensions\sd-webui-controlnet\scripts\controlnet.py
Traceback (most recent call last):
File "C:\Users\shaun\ai\stable-diffusion-webui\modules\scripts.py", line 718, in process
script.process(p, *script_args)
File "C:\Users\shaun\ai\stable-diffusion-webui\extensions\sd-webui-controlnet\scripts\controlnet.py", line 1117, in process
self.controlnet_hack(p)
File "C:\Users\shaun\ai\stable-diffusion-webui\extensions\sd-webui-controlnet\scripts\controlnet.py", line 1102, in controlnet_hack
self.controlnet_main_entry(p)
File "C:\Users\shaun\ai\stable-diffusion-webui\extensions\sd-webui-animatediff\scripts\animatediff_cn.py", line 248, in hacked_main_entry
model_net.reset()
AttributeError: 'ControlModel' object has no attribute 'reset'

[ornate elk]

Hey, best would be to make an issue on the sd-webui-controlnet github page

[ornate elk]

whats your gpu, and whats in your webui-user.bat?

only intel UHD?

no other GPU?

check in the taskmanager under the performance tab

if there is GPU 0 and GPU 1