🤝｜tech-support

[solid locust]

aaah 😄

... where do I open the CMD? o.o

[pseudo nebula]

windows key , search cmd

[solid locust]

oooooh

thank you ^^

as an admin, right?

[pseudo nebula]

sure

[solid locust]

sweet ^^

holy crepes! I'm getting a boner!

I feel like I'm doing progress! 😄

[solid locust]

thank you so much ❤️

[primal knot]

I have a hardware question. My new PC doesnt working well while all parts much better than the previous one. I have lot of new problems with Comfy (if boot with my previous Win10 too). But now I guess why. At my previous system in my CPU has internal GPU (Intel), so I used my mainboard's display port for monitor, and nothing connected to my RTX3060 GPU card just used for Comfy and similar apps what needed GPU acceleration. Then in GPU activity monitor never listed more than the python when I used Comfy. But the new CPU is F type, no internal GPU, so I connected my monitor to the 3060 instead of the mainboard. And now lot of process in the GPU activity monitor near the Comfy's python. Sometime the GPU utilization in the task manager (and in the top of Comfy menu) 100% because system or another apps using my 3060 card. I guess this is why comfy slower in the new hardware, and for example the SD ultimate upscaler stopped work, what working well in my old mainboard. Now I have an idea. Because no internal GPU in my CPU now, I plan to buy a cheap video controller card for the monitor only, so the 3060 no need to utilize browser, TC, IDE, and other apps what no need GPU. Im not sure is this a good idea? Anyone have similar experiences with this?

[solid locust]

💩

of course I jinxed things...

[ornate elk]

You shouldnt connect your monitor to your Mainboard if you have a dedicated GPU.
Some games won't even start then (Cyberpunk)
If youre worried about GPU usage of the browser or other apps then simply disable "Hardware Acceleration" in the Settings of the browser/discord etc.
Also 4k monitor for a 3060 is not that good for overall performance of everything.
Also if you switched gpu but not reinstalled comfyui. Then it could be still using the Settings from your old GPU, causing issues with vram management.

[solid locust]

hmmm...

[ornate elk]

If you have problems with forge. Try to install the normal auto1111 webui. You can follow my guide in the Pinned Messages of this channel

[solid locust]

will that help fix the issue with ponyturbo about it being extremely slow?

[ornate elk]

What's your gpu?

[solid locust]

6

the number of NOT the beast.

[ornate elk]

6gb vram?

[solid locust]

that's the GPU, yeah.

[primal knot]

I didnt changed my GPU card, it is same 3060/12 on new (and old) system too. I just upgraded the mainboard and cpu + m.2 SSD. The new CPU hasnt internal GPU, the old has. I just want to disable these apps from GPU, I will try to find hardware acceleration settings, thanks.

[solid locust]

.

[ornate elk]

Ahh okay, now I get it. But yea Hardware acceleration setting it is 👍 np

[ornate elk]

Ah a 1660

Not that great for sdxl in general

[solid locust]

that's why someone helped me code ComfyUI to low VRAM mode.

[ornate elk]

Okay, I think forge webui would be the best in this case

[solid locust]

yup

[ornate elk]

Have you installed GIT?

[solid locust]

I believe I have.

unless it disappeared.

lemme check

I do have git, but where? =.=

[ornate elk]

Maybe reinstall it and then launch forge again

[solid locust]

sure

standalone or portable?

[ornate elk]

Standalone 64bit

[solid locust]

alright 🙂

[ornate elk]

Is that after launching forge?

[solid locust]

no, after installing Git and me starting up Git bash.

[ornate elk]

Okay, then close git bash

[solid locust]

oki

[ornate elk]

Not needed

Now the forge bat should work

[solid locust]

the bat? alright, let's see.

wait, we still have to connect it to Krita.

[ornate elk]

Ah you want it with krita

Dont know if forge supports that

[solid locust]

Alright... at least I got ForgeUI, anything I generate from there, I can upload to Krita and change it there.

[solid locust]

thanks for all the help, though. ❤️

[ornate elk]

No problem, you can try to install the Krita extension in forge

[solid locust]

eh?

[ornate elk]

Nvm I thought you want to get this working:
https://github.com/poipoi300/auto-sd-krita-ext-xl

[solid locust]

I mean I do ouo

if that can connect between forge & krita.

in that case I can just try & find it here, right?

[ornate elk]

Or you can try this newer one, seems like its better for integration:
https://github.com/DrCyanide/cyanic-sd-krita

[solid locust]

ah yes, the cyanic! I have that installed ^^

[ornate elk]

Ah okay, did it worked?

[solid locust]

not sure

not really.

but I'm gonna try & put it in ForgeUI's directory, or maybe in Krita's directory.

oh wait, I DID now add in the --api, so it should work better now, I hope.

unless they're just minuses and I gotta sure the one with the underscore.

[latent wave]

"TakeRep" command returned an error: strconv.ParseInt: parsing "--": invalid syntax

[ornate elk]

Alright let me know ^^ I'm AFK now

[solid locust]

so identical.

alright, see ya

[solid locust]

still the plugins for krita didn't work. x-x

[solid locust]

https://tenor.com/view/paz-white-flag-wave-waving-flag-gif-15666633

[upbeat gyro]

Heya, im trying to train a lora model, but it keeps giving this error message, any help?

[fierce hollow]

Hello everyone, I installed the "deforum-for-automatic1111-webui" plugin, Tell me how to make all the frames the same (Realistic video) Each frame is a little different, but how can I fix this?

[hidden raft]

Possible to train lora on kohya_ss with 3080 10gb locally sdxl? Any good tutorials?

[umbral storm]

Hey, does anybody have a clenched fist gesture that they don't mind sharing from https://zhuyu1997.github.io/open-pose-editor/ ?

I have tried it making it several times but I haven't ended up with anything decent unfortunately.

[hazy wigeon]

ok so i managed to get the comfyui krita plugin running. LCM works fine, but generative fill froze and ate up so much RAM that comfyui paused itself. any ideas of what couldve gone wrong?

[crystal nacelle]

hey, im currently working on my manga and collaborating with a few artists. im an artist myself but its getting quite expensive and time consuming. do you guys know how to use StableDiffusion to make consistent charactes and a Comicbook/manga, finalised?

[haughty scarab]

Guys

I need help

I get error cuda operation not supported

In comfy ui

What should i do?

[azure magnet]

Hi, should I use no half vae even if i don't use vae?

[ornate elk]

Hey, yes if your using sdxl models or anime models

[azure magnet]

Even if i use an anime model without vae?

Will it make a better output?

[ornate elk]

The colors will be better if you use a vae

Some models have a custom VAE already baked in and some dont

[eager tulip]

How u did it?

[chilly knot]

4060 ti 8 gb vs 3060 12 gb for sd what best

[ornate elk]

12gb is better for SD and ai stuff

[ornate elk]

Check your python version

[chilly knot]

hm only vram important

[ornate elk]

The 4060 ti would be faster

But for sdxl it would be slower

Because sdxl uses more vram

[chilly knot]

can you h help unban a man he was banned for nothing)

he posted pictures of dressed girls from his lore a couple of times

without any bad subtext

I'm trying to understand what he did wrong

[eager tulip]

3.10.6

This is the final part of the error

Erased venv 2 times and then loaded again webui-user

Will try this:

Pip cache purge
Then delete the venv folder
Then relaunch the webui-user.bat ```

[eager tulip]

Seems it working for now at least for RuntimeError: Couldn't install torch.
Steps to follow:

1. Enter to your Stable Diffusion folder
2. Press Windows Explorer direction bar (where you see something like C:\SD\stable-diffusion-webui)
3. Type cmd and then hit Enter key on keyboard
4. Once Terminal open type: Pip cache purge
5. It says an amount of elements deleted (mine were 2)
6. Delete your venv folder inside your SD folder
7. Run webui-user.bat and wait

[ornate elk]

So its working now?

[chilly knot]

standart encoder is good for ip adapter?

sd try to download annotator\downloads\clip_vision\clip_h.pth but fail

pth files its safe for system?

i read that need only safetensors

[dark barn]

Does anyone know a plugin for UE or Blender that use OpenPose to create animation in UE or Blender?

[eager tulip]

Yes but no.

1. I open webui-user
2. Terminal opens showing following message:

Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug  1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
Version: v1.8.0
Commit hash: bef51aed032c0aaa5cfd80445bc4cf0d85b408b5
Launching Web UI with arguments: --xformers --autolaunch
Calculating sha256 for C:\SD\stable-diffusion-webui\models\Stable-diffusion\v1-5-pruned-emaonly.safetensors: Running on local URL:  http://127.0.0.1:7860```
3. A browser tab opens with the http://127.0.0.1:7860 url
4. Browser doesnt load automatic 1111 showing the following message ```ERR_CONNECTION_REFUSED```
5. Terminal show the following code:
```To create a public link, set `share=True` in `launch()`.
Startup time: 32.7s (prepare environment: 9.7s, import torch: 7.1s, import gradio: 3.0s, setup paths: 3.7s, initialize shared: 4.9s, other imports: 2.1s, load scripts: 1.0s, create ui: 0.6s, gradio launch: 0.7s).
6ce0161689b3853acaa03779ec93eafe75a02f4ced659bee03f50797806fa2fa
Loading weights [6ce0161689] from C:\SD\stable-diffusion-webui\models\Stable-diffusion\v1-5-pruned-emaonly.safetensors
Press any key to continue . . .```
(1st picture is what browser shows giving ERROR)
6. I delete --xformers & --autolaunch from the ARG command lines
7. relaunch webui-user

8. Automatic 1111 works and then giving error (2nd picture) this is what terminal shows:

Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug  1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
Version: v1.8.0
Commit hash: bef51aed032c0aaa5cfd80445bc4cf0d85b408b5
Launching Web UI with arguments:
No module 'xformers'. Proceeding without it.
Calculating sha256 for C:\SD\stable-diffusion-webui\models\Stable-diffusion\v1-5-pruned-emaonly.safetensors: Running on local URL:  http://127.0.0.1:7860

To create a public link, set `share=True` in `launch()`.
Startup time: 21.0s (prepare environment: 4.5s, import torch: 5.4s, import gradio: 2.3s, setup paths: 2.7s, initialize shared: 2.2s, other imports: 1.9s, load scripts: 1.0s, create ui: 0.6s, gradio launch: 0.4s).
Calculating sha256 for C:\SD\stable-diffusion-webui\models\Stable-diffusion\v1-5-pruned-emaonly.safetensors: 6ce0161689b3853acaa03779ec93eafe75a02f4ced659bee03f50797806fa2fa
Loading weights [6ce0161689] from C:\SD\stable-diffusion-webui\models\Stable-diffusion\v1-5-pruned-emaonly.safetensors
Press any key to continue . . .```

(I installed SD Automatic1111 on other pc (older one but dif specs) and runned perfectly, this have being a pain in the coconuts...)
This is AMD processor with gtx 1660, older one is Intel processor with gt 1050 ti

[chilly knot]

not see error

[ornate elk]

Have you installed any third party antivirus program?

[ornate elk]

Maybe that site can help:
https://posemy.art/

It has different poses and you can extract pose, normalmap and depth images

[dark barn]

oh nice, i will check how it work

thx

[ornate elk]

Np 🙂

[chilly knot]

help me install encoder manualy

for ip adapter

bad connection

and fail

i download model.safetensors where i need put it

[eager tulip]

I just have windows defender as well as my other pc

[ornate elk]

Okay good

Also use --xformers --medvram --no-half in your webui-user.bat

Then try download an other model from civitai like Dreamshaper v8 and put it into the models/Stable-diffusion folder

[iron hamlet]

any body knows how to set up controlnet reference only with diffusers??
no pc so im on colab with commandline diffusers but i cant seem to find the damn controlnet model
if anybody can reply with code thatd be gr8

[sleek nexus]

I'm starting to get into stable diffusion and I'm using an amd GPU, is it normal for generations to take 5 to 10 minutes?

[eager tulip]

How much Memory does your GPU have, what model are you using and what is the size of your comp?

[sleek nexus]

It is an RX580 with 8gb of vram, I am using the animaginexl v31 model, what is a comp?

Now I'm experiencing an issue where stable diffusion is causing too much load I believe and the monitor is displaying "no signal"

[carmine canyon]

i was tackling this issue today.

i have a gtx 1650 and I was using sd webui. Inferences took around ~8 minutes.

I switched to sd webui forge and image generations now take ~20 seconds

for a 512x512 img, 20 sampling steps

stated from the sd webui forge readme, since you have 8gb vram, you can expect a ~30~45% speed up compared to using the original webui

[sleek nexus]

Ah ok, that may fix my first issue, now I have to see if I can run it again without the PC / GPU crashing

[eager tulip]

XL Models ussually take longer, try 512 instead of 1024. 8 gb is fine but for XL I have seen 12 is recommended for speedbust, seems like 8gb kind of basic for XL model

I am using 1.5, I have just 4GB will try forge later

[eager tulip]

Any tip or suggestion to use forge? I dont want more issues with the actual setup

[carmine canyon]

sorry i dont have any tips. I didn't have much difficulty when i set up sd webui forge.

however, i'll be glad to help if you have a specific issue.
@eager tulip

[iron hamlet]

any body knows how to set up controlnet reference only with diffusers??
no pc so im on colab with commandline diffusers but i cant seem to find the damn controlnet model
if anybody can reply with code thatd be gr8

[hazy wigeon]

using the krita ai diffusion plugin for inpainting, but it hangs and starts maxing out my memory. any suggestions?

the prvious prompts were from live LCM. the dip in memory is because i killed the program

[pliant cipher]

Hey , one question, why on Google Collab with the T4 I get 1 it/s and 30 seconds max per image but on my rig with 1070ti and 3.4 it/s I get like 1 min 15 secs per image?

On fooocus GUI and sdxl checkpoints

[sleek nexus]

I'm installing forge right now and I came across an issue: "RuntimeError: Torch is not able to use GPU" know anything about this?

Now it has an error where it found no Nvidia driver on my system

[ornate elk]

Forge doesnt work with AMD that good.
You should checkout the pinned messages of this channel for my AMD guides

You need some specific arguments for the best performance

[sleek nexus]

Hmm, in that case, is there any advice on how to prevent SD from causing my (what I presume) GPU to crash? When I try generating in a resolution higher than 512×512 with Euler A, halfway through the process the screen turns off and the monitor displays "no signal"

The PC is running but there's no way to fix the issue besides turning the PC off and on again

[vocal burrow]

sounds like it could be a power supply issue and/or over heating.

could be bad drivers too, or overclocking cranked too high

lots of things.

so yeah, clean your pc, turn off overclock if you have any on your gpu, update drivers and try again

[sleek nexus]

I don't have it overclocked so I'll try updating drivers

[cinder ermine]

attempt at installing, first time running bat file

installed git and python already

what am i doing wrong

[chilly knot]

can somebody know how install manualy image_encoder for ip adapter in a111

[sudden hawk]

Hello, when I generate image on SD it takes me a lot of time (20-30m for 1024x1248).
when i install A1111 i put in webui —medvram-sdxl. what i didn’t know is that i dont have enough vram to use SDXL. So now i have deleted medvram-sdxl of my webui, but it still take a lot of time to generate. can someone help me please ?

[vocal burrow]

you re using a fat32 partition => no permissions support on filesystem level => you have to tell git to ignore permissions errors/warning for every folder it manages.... Or reformat in a proper filesystem (ntfs)

[vocal burrow]

what s your gpu ?

[sudden hawk]

i have RTX 3070 with 8 vram 16 ram and intel 7

[ornate elk]

Then you need --medvram-sdxl

Or it would take ages

[sudden hawk]

Ok, but if I put --medvram-sdxl does it means that I use SDXL ? or just SD ? And I have to ajust with checkpoint

[cinder ermine]

how do i tell git to ignore perms

[ornate elk]

That means it uses the --medvram command but only for sdxl models.
So they generate faster.

The 1.5 models are not affected by that

[sudden hawk]

Ok, but before i deleted --medvram-sdxl It was already taking me a long time

[vocal burrow]

it s telling you what command to run in the very screenshot you sent and I croped back at you

at the end

call : xxxxxxxxxxxx

[vocal burrow]

define "a long time"

[ornate elk]

Then try again with --medvram-sdxl
Always launch the webui-user.bat instead of webui.bat
And dont set hires Steps higher then 10, if enabled

[sudden hawk]

Ok thank you so much for taking time to understand my pb, and reply me 🙂 i try that, have a wonderful day

[chilly knot]

nobody know where i need put model .safetensors image encoder in auto111?)

o found tutorial only for comfy]\

[ornate elk]

No problem, let me know if it still takes to long

[sudden hawk]

yes it work, i can generate in 3-5m (even less) now 🙏 🥹 Thank you so much

[cinder ermine]

i see

its causing some other complications, probably going to try another drive

[chilly knot]

😭

[ornate elk]

Encoder for what?

[chilly knot]

for ipadapter

[ornate elk]

Models/controlnet

[chilly knot]

it trying to downwload but fail

how i install it manual

[ornate elk]

Ah look at the path

It tells you

[chilly knot]

i download model.safetensfors how i install it

https://huggingface.co/h94/IP-Adapter/tree/main/models/image_encoder

this

[ornate elk]

This goes into models/controlnet

[chilly knot]

and no need download pth?

[ornate elk]

Nope

[chilly knot]

thx ) ivery hope

config json i need to or only model?

i no need rename model.safetensor before use?

[ornate elk]

Only the model

No rename needed

[chilly knot]

i hope it be works

[ornate elk]

What's the model.safetensor?

[chilly knot]

is https://huggingface.co/h94/IP-Adapter/tree/main/models/image_encoder

encoder

i need install image encoder manual

because i cant downloading with bat

no its not work

it try to download again

downloads\clip_vision\clip_h.pth have safetensors version

i downolad it

and try install

i do something wrong?

where it need be placed

[eager tulip]

Noup... I guess maybe is happening between "memory monitor disabled *** Error loading script: lora_script.py" & "RuntimeError: CUDA error: out of memory" (I have 16 GB Ram and 6G on GPU)

Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug  1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
Version: v1.8.0
Commit hash: bef51aed032c0aaa5cfd80445bc4cf0d85b408b5
Launching Web UI with arguments: --xformers --medvram --no-half
Warning: caught exception '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.
', memory monitor disabled```

    Traceback (most recent call last):
      File "C:\SD\stable-diffusion-webui\modules\scripts.py", line 527, in load_scripts
        script_module = script_loading.load_module(scriptfile.path)
      File "C:\SD\stable-diffusion-webui\modules\script_loading.py", line 10, in load_module
        module_spec.loader.exec_module(module)
      File "<frozen importlib._bootstrap_external>", line 883, in exec_module
      File "<frozen importlib._bootstrap>", line 241, in _call_with_frames_removed
      File "C:\SD\stable-diffusion-webui\extensions-builtin\Lora\scripts\lora_script.py", line 101, in <module>
        shared.opts.onchange("lora_in_memory_limit", networks.purge_networks_from_memory)
      File "C:\SD\stable-diffusion-webui\modules\options.py", line 234, in onchange
        func()
      File "C:\SD\stable-diffusion-webui\extensions-builtin\Lora\networks.py", line 247, in purge_networks_from_memory
        devices.torch_gc()
      File "C:\SD\stable-diffusion-webui\modules\devices.py", line 80, in torch_gc
        with torch.cuda.device(get_cuda_device_string()):
      File "C:\SD\stable-diffusion-webui\venv\lib\site-packages\torch\cuda\__init__.py", line 370, in __enter__
        self.prev_idx = torch.cuda._exchange_device(self.idx)
    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.```

[ornate elk]

They are not needed

[chilly knot]

what i need ?

ip adapter cant statw without download encoder

[ornate elk]

Is your gpu driver updated?

[ornate elk]

The annotator clip vith model

[chilly knot]

i cant use only this way?

Downloading: "https://huggingface.co/h94/IP-Adapter/resolve/main/models/image_encoder/pytorch_model.bin

or use model.safetensors

[ornate elk]

I dont think so

[chilly knot]

i have bad connection and cant use autodownloading and install

and need use manual instalation

pth files risk?

[ornate elk]

Pth could be risky if to source isnt trustworthy

[chilly knot]

how you i can trust this link

[ornate elk]

The IP adapter huggingface page seems trustworthy

[chilly knot]

i broken)

[eager tulip]

Gonna check

[grizzled lantern]

I am doing CV, and I am starting with the AIGC algorithm. Can you recommend me how to learn it? Thanks for everyone’s replies.

[vast crest]

Hi, guys, a little help. When I save a sequence on img2img, I press on the icon: open images output directory and then it should take me to the folder: loopback-wave. instead it always takes me to the folder: img2img-grids. You can help me

[eager tulip]

Updated GPU Drivers still same issues (I moved directory to desktop testing if it was somekind of C:/ admin rights restriction)

Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug  1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
Version: v1.8.0
Commit hash: bef51aed032c0aaa5cfd80445bc4cf0d85b408b5
Launching Web UI with arguments: --xformers --medvram --no-half
Calculating sha256 for C:\Users\[username]\Desktop\SD\stable-diffusion-webui\models\Stable-diffusion\anithing_v11Pruned.safetensors: Running on local URL:  http://127.0.0.1:7860

To create a public link, set `share=True` in `launch()`.
Startup time: 24.9s (prepare environment: 8.3s, import torch: 6.2s, import gradio: 2.9s, setup paths: 3.3s, initialize shared: 0.3s, other imports: 2.0s, load scripts: 1.0s, create ui: 0.5s, gradio launch: 0.5s).
Calculating sha256 for C:\Users\[username]\Desktop\SD\stable-diffusion-webui\models\Stable-diffusion\dreamshaper_8.safetensors: ccfb736e73ce6bc115804a43970e651ce62eb7ed26f2de8d82877d571a4a2e1b
Loading weights [ccfb736e73] from C:\Users\[username]\Desktop\SD\stable-diffusion-webui\models\Stable-diffusion\anithing_v11Pruned.safetensors
Press any key to continue . . .```

[chilly knot]

CS i download it file manial and put in right folder then i get

\clip_vision\clip_h.pth
*** -----> !!!! The file is most likely corrupted !!!! <-----
*** You can skip this check with --disable-safe-unpickle commandline argument, but that is not going to help you.

[ornate elk]

Strange

[chilly knot]

its hell how hard start to work

[ornate elk]

Yea you should visit a place with better internet ^^

[chilly knot]

i read that is wifi problem

if i change it to cable with same internet problem fix

but why file corrupted

[ornate elk]

Then try to download the file again

With Ethernet

Corruption appears if the download gets interrupted

[chilly knot]

it can be problem with ip-adapter-plus-sd15.safetensors 90 mb

can this file be normal and problem with ip apter files?

[ornate elk]

Nope I have all IP-Adapter models and they work

I downloaded them from the huggingface site you linked

But you can try an other file

Like IP-Adapter plus face full

[chilly knot]

ip-adapter-plus-face_sd15.safetensors i have

[ornate elk]

Okay

Good

[chilly knot]

remember you give me link to cany

strange why only model from you link work and not from standart place

[earnest jackal]

Hey 1 more question I'm trying to merge the model with other model but its giving me an error, 1 is XL and other is not
does both model needs to be XL in order to merge?

[chilly knot]

cs1 bin file i need only rename ?

to pth

or i need convert

[eager tulip]

I would try searching a variation model to test if that is the problem. I mean, you merging model B XL and model A (not XL), maybe looking a model B (not XL) and trying to merge it would answer that. I really dont have 100% idea, but I think that if there is Dreamshape Model and Dreamshape XL Model as well as Add Detail LoRa & Add Detail XL LoRa, might be for some reason of incompatibility of models, and that may affect merge too.

[chilly knot]

cs1 i redownload and i think is works and i have different error)

[ornate elk]

No need for convert

[chilly knot]

OutOfMemoryError: CUDA out of memory. Tried to allocate 30.00 Mi WTF

for one picture(

[ornate elk]

You can't merge different base model versions

It has to be either two sdxl or two 1.5 ones

[ornate elk]

What settings?

[chilly knot]

with low vram

[ornate elk]

Strange, that should work with your card

Scroll down and check if xformers is installed

It should display the version number

[chilly knot]

Well, our model is on the cuda:0 device but our data tensors (X and y) are still on the cpu device.

[ornate elk]

Looks okay

[chilly knot]

return t.to(device, dtype if t.is_floating_point() or t.is_complex() else None, non_blocking)
torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 74.00 MiB (GPU 0; 4.00 GiB total capacity;

ok its imposible

[waxen cradle]

Hi guys, I want to buy a new video card since the one I have runs slow on the stable Forge diffusion.

Some say that I should buy the graphics card that has the most VRAM, but I have doubts about the incompatibility it may have with my PC hardware.
I am attaching the characteristics of my PC and I hope you will give me your opinion and suggest graphic cards to buy.

[chilly knot]

simple extension need rtx 4090

[chilly knot]

RuntimeError: mat1 and mat2 must have the same dtype

nice new error

[ornate elk]

Check if you have selected an 1.5 model

[ornate elk]

The only limitation could be your PSU

[waxen cradle]

CPU
?

[ornate elk]

Your Power supply

Better GPUs require more power

Also what's your current gpu?

[waxen cradle]

msi gefotce GTX 1060 3gb

[chilly knot]

after i change arguments

--precision full --no-half --lowvram --always-batch-cond-uncond --opt-split-attention

[waxen cradle]

600w PSU atm

[chilly knot]

cs1 its start generate photo -and in last step out of memory

[ornate elk]

Remove --precision full --always batch cone uncond --opt-split-attention

Only use --xformers --medvram --no-half

[chilly knot]

it works!

low vram

lol

and i close many browser tab

[ornate elk]

Okay you should still use the commands I suggested

Xformers will help you a lot

[ornate elk]

A 3060 with 12gb is a good GPU for SD and price

[chilly knot]

--theme=dark --xformers --lowvram --no-half-vae my settings

[ornate elk]

Thats okay if it works

[chilly knot]

no half or no half vae?

[ornate elk]

A 1660 needs --no-half

[chilly knot]

1050 ti

[ornate elk]

Okay for that use --no-half-vae

[chilly knot]

oh (its like magic that this works

thx for help

[ornate elk]

No problem 🙂

[chilly knot]

ip adapter its nice)

[ornate elk]

Yea its awesome

Checkout IP adapter plus face full for a nice face swap

[chilly knot]

I'm glad but I feel empty after so many unsuccessful attempts 😃

[vast crest]

Hi, guys, a little help. When I save a sequence on img2img, I press on the icon: open images output directory and then it should take me to the folder: loopback-wave. instead it always takes me to the folder: img2img-grids. You can help me 😦

[azure magnet]

Hi, how to fix that?

[ornate elk]

Thats nothing to "fix" its just a mentioning that it won't use a function based on an extension you dont have installed

[azure magnet]

Ok so how i can use the extension with the loras' words?

[ornate elk]

If you install the model-keyword extension

[azure magnet]

I don't understand, need i to install another extension?

[ornate elk]

If you want to autocomplete lora trigger words then yes

If not then no

Right now it should already work normaly for word autocomplete

[azure magnet]

Yes, i'd like, do you know what extension need i install?

[ornate elk]

This one:
https://github.com/mix1009/model-keyword

[azure magnet]

Ok, i'll try, thank a lot

[warped ginkgo]

Where do I put that in the commandline?

[ornate elk]

In the same line as the others

[crude trellis]

hey, I'm trying to install zluda on directml-automatic1111. I've been following the pinned steps and reached the part where I have to add zluda folder to PATH, this could be a dumb question but I have add they as new instead of overwriting any of those, right?

[ornate elk]

Hey, Yes you need to make two new entrys

[fair oxide]

I'm desperately looking for a solution for data export.

I want to download all my images and prompts from DreamStudio using StableStudio. I figure I could probably make a plugin to do this but I'm a bit stumped.

Is there a solution to export all user generated images and prompts? I want these images and prompts, they are mine.

I have spent over $120 working with Stable Diffusion on DreamStudio, so I'd like to get my data somehow.

[upbeat atlas]

Hi, for zluda which is running with automatic111 directml, can I update the driectml normally? Does it mess sth up or not?

[azure magnet]

I have install it and it works, but i don't get the suggestion of the loras in the prompt line, but in the generation tab. It's possible to has the suggestion on the prompt line?

[ornate elk]

Yes you can update it normaly

[ornate elk]

I dont know, didn't tested it yet

[azure magnet]

ok, thanks

[upbeat atlas]

Hi, so I updated my webui and deleted venv as well, but I'm still getting this error

stderr: ERROR: Could not install packages due to an OSError: [WinError 5] Access is denied: 'D:\SD-Zluda\stable-diffusion-webui-directml\venv\Lib\site-packages\~umpy\.libs\libopenblas.FB5AE2TYXYH2IJRDKGDGQ3XBKLKTF43H.gfortran-win_amd64.dll'
Check the permissions.

Is there any other way to fix this?

For the record I can still use the webui and generate images, it's just that now I need to be connected to the internet for it to work and when I do connect I get that error before the webui loads up

[round topaz]

Good evening, can someone advise me how to make a prompt so that 3 people appear in the image? Thank you

[ornate elk]

You can try to edit the folder permission of the SD-Zluda folder ans everything in it

So that your account and applications have read/write permission

[sinful halo]

Sorry to bother you man, could you help me ? you have time now ?

ill let my error here if someone is willing to help, im trying to upscale using ultimate sd upscale and img2img bcs my gpu dont have vram for normal upscaling, but doesnt matter if i use ultimate sd upscaler or the normal one from img2img, it gives that error

i didnt find anything on google

[sinful halo]

i have 2 rx 5700 xt, i cant use both at same time for more vram ?

[vocal burrow]

Nope you can't unify VRAM across multiple gpus

[sinful halo]

/-/ thats unfortunate

thank man

[vocal burrow]

Don't know if you're zluda or not. You might need --no-half for img2img stuff

[sinful halo]

my args> set COMMANDLINE_ARGS=--use-directml --skip-ort --lowvram --opt-sub-quad-attention --opt-split-attention-v1 --no-half --upcast-sampling --disable-nan-check

idk what is zluda lol

the error occurs whenever i try to upscale using any method, in any tab

[buoyant pewter]

I'm trying to setup SD on an Apple M1 MacBook Pro, and the only image that is generated is a all blue result. During the processing the a 'random' image starts, then converges to 'all blue', rather than any thing resembling a bottle. Any suggestions on finding out what is wrong? Thnx.

Here's the node display of what I believe is the install default, using the ComfyUI interface and Stable Swarm install.

[crystal nacelle]

Anyone familiar with using KOYA to train your own models? im using an AMD machine, looks like the installer is asking me questions related to nvida gpu so im not too sure how to proceed. should I tell it to use FP16 or BF16?

[fair oxide]

my thing has been acting weird

I have 2 lora files

I did not create a second one

also loras barely work

[fair oxide]

they appear sometimes in the gui and when I try to use time it varies from doesnt work at all to barely works

and some of my models barely work randomly

usally they randomly decide to work

exept the loras they have yet to work consistantly

[worn frigate]

Hi, please help me if possible.
So on huggingface undercontrolnet category I came across a subject driven controlnet, I tried to install it, but it not like .pth, And on there whole page, there was 0 words about how to setup it on stable diffusion, They do have a well detailed guide on installed but all looks like a language for the developers, and I am not wise enough to understand a line, I still tried but no use, I have the main file of it downloaded, is .bin
Please help me how to i even get it, there is no preprossor.

its supposed to work like this

cool right?

https://huggingface.co/salesforce/blipdiffusion-controlnet

here is the link to it. please help me. i'd really appreciate

inside its controlnet folder, these two files exist, which, the bottom one I downloaded and put in controlnet folder, diffusion pytorch bin. I do not know or have config,json

@green wave do you know anything about it?

[green wave]

I'm pretty sure it doesn't have much to do with Stable Diffusion, and it's somewhat outdated. We'll likely see things like this made for SD3 whenever it comes out

[worn frigate]

thank you so much for replying

has it stopped working now? I thought its a controlnet model. Is it like a different software?

Ok i will just delete it, I do not think i can ever get it working.

[hollow wraith]

Hello, Everyone.
I want your help.
I am going to convert any dog image into sympsonized one using stable diffusion project.

[worn frigate]

use Controlnet extension for stable diffusion.

[hollow wraith]

I am sorry. I never know AI.

[brisk ermine]

Hello.
I have a question in hopes of getting some help.
I'm using ComfyUI and I want to create an object removal feature using stable diffusion.
So I got the inpaint model, masked the image and ran the inpaint function.
But it keeps creating objects even though I left the affirmative prompt blank.
In this situation, is there anything I can try in general (model suggestions, etc..)?

Thank you.

P.S. It should be a method that doesn't use controlnet.
The inpaint model I used is https://civitai.com/models/139562?modelVersionId=297320.

[pastel glacier]

Hello not sure where these questions should go, but I guess this fits here

I need some assistance and recommendation and some questions:

I want to build my llm on top of some pdfs for certain topics, I have limitation on the computing power, hence I'm considering using inference by open source models (llama2, mistral, phi, etc.)
which way is the lightweight to build the llm inference model?
I am currently trying to run phi-2 on top of my PC with 16GB RAM, it can't run. Unless I have more free space than 10GB
Is the llm model always require minimum of 8 GB for its inference?
as I know inference - CPU or native RAM, fine-tuning GPU VRAM. Please correct me if I'm wrong 🙏

[ornate elk]

Kohya won't work with AMD, but OneTrainer could work

[hollow wraith]

Hello Alain catcher 2!'
So, thank you. you can let me know about the project that you are using now?

[ornate elk]

Hey, you can try follow my zluda guide. It seems your card might be supported trough the community.
The zluda guide is in the Pinned Messages of this channel

With the zluda version you won't get the errors when upscaling

[ornate elk]

Hey you can combine LM Studio + Anything LLM to get a chatbot that can use your local PDF files

Then choose Mistral as model

[ornate elk]

What do you use for that?

[hollow wraith]

Thank you for the reply

I do not know about AI well.

I want to conver any dog image into sympsonized one.

such as this.

could you understand my proposal?

[ornate elk]

Yes I understand.
For that you would need a tool that uses stable diffusion and then you need to use img2img or ControlNet.
Is your PC capable to run Stable-diffusion locally?

@hollow wraith
Start with this guide if you have an Nvidia GPU:
#🤝｜tech-support message

[wide parrot]

👍

[upbeat gyro]

AttributeError: 'NoneType' object has no attribute 'unscale_grads'

Please tell me how to fix this, this happend when i tried training a lora model

Traceback (most recent call last):██████████████████████████████████████████▌ | 4/5 [00:18<00:04, 4.56s/it]
File "F:\AiReps\Stable Diffusion\extensions\sd_dreambooth_extension\dreambooth\ui_functions.py", line 735, in start_training
result = main(class_gen_method=class_gen_method)
File "F:\AiReps\Stable Diffusion\extensions\sd_dreambooth_extension\dreambooth\train_dreambooth.py", line 2003, in main
return inner_loop()
File "F:\AiReps\Stable Diffusion\extensions\sd_dreambooth_extension\dreambooth\memory.py", line 126, in decorator
return function(batch_size, grad_size, prof, *args, **kwargs)
File "F:\AiReps\Stable Diffusion\extensions\sd_dreambooth_extension\dreambooth\train_dreambooth.py", line 496, in inner_loop
patch_accelerator_for_fp16_training(accelerator)
File "F:\AiReps\Stable Diffusion\extensions\sd_dreambooth_extension\dreambooth\utils\utils.py", line 342, in patch_accelerator_for_fp16_training
org_unscale_grads = accelerator.scaler.unscale_grads
AttributeError: 'NoneType' object has no attribute 'unscale_grads'

This is the full message

[pastel steeple]

Hi folks! Can anyone tell me what is the max input token/char/word count for a Stable Image Core prompt? I can't find an answer anywhere on the website

[lament temple]

@ornate elk Hi I have issues installing stable diffusion .
[4:18]
I keep getting the error C:\stable-diffusion\stable-diffusion-webui-master>git pull
fatal: not a git repository (or any of the parent directories): .git

exit code: 1

stderr:
C:\python.exe: No module named pip

Launch unsuccessful. Exiting.
Press any key to continue . . .
[4:18]
I have already installed pip but still get this

[vocal burrow]

fatal: not a git repository (or any of the parent directories): .git```
Most likely you didn't initially install using git. Did you ?

stderr:
C:\python.exe: No module named pip```
python installed directly in c: ?!?!?

[oblique parrot]

Why is the SD3 preview that OpenArt.ai is allowed to use for creative upscale so nerfed and bowdlerised that it makes the Hayes Code look like Larry Flynt?

This thing sees NSFW so often it appears to be an artificial pervert, because no human has a dirty enough mind to think that much is naughty.

(Also, who do you guys think you are telling people what is and isn't naughty?)

[vocal burrow]

1/ SD3 is made by Stability.ai, not by OpenAI
2/ it's not even released yet
3/ There is no NSFW filter for local installs, so do as you wish with it. But this discord is not the place to share smut.

[ornate elk]

Hey follow my guide in the Pinned Messages of this channel

[cinder ermine]

got this on the kohya ss gui

trying to caption images but it appears for training too

what could cause this

[upbeat gyro]

Why do i keep getting ignored :/

[ornate elk]

You have a dreambooth error

I would recommend deleting it from the Extensions folder and then delete the venv folder

After that your webui will work again

[upbeat gyro]

Alright, thanks

[misty marten]

I'm trying to restore faces in my generated images, using Reactor, when I put them in GFPGAN, the images come out with artifacts and saturated points, with some light and dark spots, can anyone help me solve this?

[pastel steeple]

anyone? 😭

[misty marten]

I think there are no limits

[pastel steeple]

are those characters?

[misty marten]

I copied and pasted several words to see if I could reach the maximum, and apparently there is no

[pastel steeple]

I think they provide only a portion of that prompt to the model, 800k characters doesn't seem realistic

[crystal nacelle]

made my own model, however the adetailer results in this opaque square over the faces? anyone know whats up...running zluda amd A1111

[deft geode]

Are there any Docker API Containers Available for the various models?

[iron hamlet]

help

im running diffusers on colab

i try to lora by pipe.load_lora_weights() but it throws "PEFT backend required for this method:

[hazy wigeon]

is this normal for comfyui? did i install ipadapter wrongly?

[buoyant pewter]

Trying to get SDXL-1.0 running on a MacBook M1 laptop. This is the result of 'generate', with an initial install of StableSwarm + ComfyUI. The inital image displayed in the Ksampler window appears to be a random field, which converges to smooth 'blue' for the output.

[stuck cradle]

Try using SDXL text encoder I also get weird results when I use the normal text encoder on a SDXL checkpoint

[ornate elk]

What's your VAE ?

[ornate elk]

What if you use codeformer?

[carmine agate]

Hiya, I'm attempting to get forge up and running but am hitting a snag. After unzipping the ez 7z file, running update.bat & then run.bat, I get the following error complaining I don't have cpuinfo:

  File "I:\ai\webui_forge\system\python\lib\site-packages\transformers\utils\import_utils.py", line 1076, in __getattr__
    module = self._get_module(self._class_to_module[name])
  File "I:\ai\webui_forge\system\python\lib\site-packages\transformers\utils\import_utils.py", line 1088, in _get_module
    raise RuntimeError(
RuntimeError: Failed to import transformers.models.clip.modeling_clip because of the following error (look up to see its traceback):
No module named 'cpuinfo'

Any of you kind folks know how I can get this module? From what I can tell I need to install it using the embedded python stuff but I'm not really sure how to do that.

[misty marten]

the codeform works, there are rarely any errors like this, but it is not as faithful to the original photo, GFPGAN restores the face almost exactly

[severe sand]

I'm trying to install SD locally on my PC with this guide:
https://www.youtube.com/watch?v=n8RhNoAenvM

Getting some errors after running webui.bat. Would anyone be willing to help? Pretty new to all this

[severe sand]

still struggling T_T

[fair oxide]

I've installed IPAdapter Plus, but the Apply IPAdapter node is not showing up for me

All I have access to is the Advanced and the IPAdapter

Has the node flow changed such that I have to apply an attn mask now?

[sinful halo]

Man, thanks for the help, i think i did everything in the guide, bbut it is not working /-=/ sorry

i need to use the skip torch cuda test ? i thought zluda emulates cuda

[severe sand]

I have a Radeon RX Vega 56

[woven sinew]

I try to merge 2 models but I get this error:
Error merging checkpoints: [enforce fail at ..\c10\core\impl\alloc_cpu.cpp:72] data. DefaultCPUAllocator: not enough memory: you tried to allocate 671088640000 bytes.
Time taken: 3 min. 29.2 sec.

[ornate elk]

Hey you need to follow my install guide for zluda. The tutorial is bad.
My guide is in the Pinned Message of this channel

It should work with your card

[ornate elk]

Wrong path settings,
you need to edit the path entry ans there add two new entrys

[severe sand]

Thanks, are you sure though? The error I'm getting seems to be related to rocm

did you watch the tutorial i sent?

[ornate elk]

I watched this tutorial weeks ago and its wrong. It also doesn't mention that it only works for amd RX 6000 series if you dont use the custom rocm library from github.
My guide covers this step.

[severe sand]

Oooooh alright, thanks. Do I uninstall rocm and delete the github repo that I cloned for now?

[ornate elk]

You dont need to uninstall rocm(hip)

But remove conda

And start with a new folder

[severe sand]

meaning what?

these?

I'm basically completely new to all of this^^

[ornate elk]

Thats zluda, no need to delete if you downloaded the latest stable build of it

[severe sand]

alright

[ornate elk]

And update your python to 3.10.11 64bit

[cinder ermine]

help i cant install correct torch ver

on windows with an nvidia card

[severe sand]

Does it has to be that specific version?

I have 3.12 installed aswell

[ornate elk]

3.12 is not supported

And won't work

[severe sand]

Kk

[ornate elk]

So yes 3.10.11 is the latest supported

[ornate elk]

Make sure your gpu driver is updated

[cinder ermine]

just did

[ornate elk]

Do you have git installed?

[cinder ermine]

git --version

[ornate elk]

Did you git cloned Khoya_ss?

From baltmais github?

[cinder ermine]

i downloaded via chrome, gimme a sec gonna try that now

[ornate elk]

Checkout this guide for Khoya_ss install:
https://github.com/bmaltais/kohya_ss?tab=readme-ov-file#installation

[cinder ermine]

following it now

[severe sand]

Wait I can't have it somewhere in documents?

[ornate elk]

Yes because documents folder was created by windows and not by you.
That will result in permission conflicts later

[severe sand]

Alright thanks

but shortcut within documents works?

[ornate elk]

You can make a shortcut into documents yes

[severe sand]

I mean a shortcut to the folder in documents

alr

[jaunty palm]

Hi i hope someone could help me figure my problem out. i keep getting an error when i generate an image at 520 x 768 but when i genrate at 512 x 768 it works as it should.

i am using TensorRT to speed up my work

[severe sand]

@ornate elk Do I need to be concerned about step 6 at all if it's my first time installing?

[jaunty palm]

but i keep running in the issue whn where i change the ratio just a bit and it just gives me a ValueError.

[ornate elk]

You can skip it if you dont have a model ready. It will download the default one

[severe sand]

• as for ROCM, "don't check the pro driver".. what is meant by this?

The installer I have says that it's the pro version but I didn't see multiple options

[ornate elk]

At the point where you installed the HIP sdk (rocm) there is a checkbox at the bottom for the amd Pro driver. That shouldnt be checked

[severe sand]

O
Pretty sure it was checked when I installed, is that a problem?

This is what I have installed rn

[ornate elk]

Check if your adrenaline software still says adrenaline and not and pro

[carmine agate]

Been away from A1111/forge for a long time. Are lycoris now interchangable with lora? Do they need a lycoris:.... & be in a seperate dir?

[ornate elk]

Lycoris work the same like loras now

[ornate elk]

Looks good

No pro driver

[severe sand]

Alright

Although I definitely didn't unchecl that when installing

the only thing i unchecked was the visual studios extension

[ornate elk]

Okay

Maybe its unchecked by default

[severe sand]

I suppose I'll see how it goes from now

is it not supposed to work with the pro version?

[ornate elk]

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

[jaunty palm]

Gpu: RTX 3060 12gb

[ornate elk]

It could work with pro too. But that driver is not needed as Adrenalin is better

[severe sand]

kk

[severe sand]

ROCM or radeon software?

I'm a bit confused

[ornate elk]

?

Adrenalin is the GPU driver

[severe sand]

Yes

And for SD I installed ROCM

[ornate elk]

Rocm is also a driver and library

[severe sand]

An extra driver?

[ornate elk]

Not really

The HIP SDK is like a module

For the adrenalin driver

[severe sand]

I'm confused about why installing ROCM would change my adrenaline cersion to pro

[ornate elk]

Because in the hip SDK installer is a checkbox some people clicked

You didn't so your fine

[severe sand]

Oh alright fair enough

[ornate elk]

Make sure you whitelisted the webui in any browser ad or script blocker

Also which model do you use?

[jaunty palm]

default model v1.5 pruned emaonly

[ornate elk]

Oh I see your using temsorrt

Thats the problem

[jaunty palm]

yeah i was looking around for a faster way to generate images and came across this solution but it does not work for all res

[ornate elk]

It doesn't support every extension in auto1111 and is also more like Experimental feature

[severe sand]

Do I replace these?

[ornate elk]

No you have to rename the folder

[severe sand]

which one?

[ornate elk]

Then drop the other folder in the same location

The old library folder

[severe sand]

isnt that the equivalent of deleting it?

[ornate elk]

You have a 3060 with 12gb, SD should be fast enough for your gpu

Wouldn't recommend using tensorrt

[ornate elk]

Renaming is like saving it if something fails

[severe sand]

Alright

[jaunty palm]

fair enough thought why wouldnt i try it but it only causes me a headace

thanks for helping me out

[ornate elk]

No problem, for faster generations, use DPM++ 2m karras sampler or Euler a.
Also use 30 steps.
And if you try out hires fix, set the hires steps to 10.
Also download a better model from civitai.
Most 1.5 models are 2gb and generate faster than the ugly 1.5 default one

[fair oxide]

im installing ipadapter for the first time, and it seems there was an update that breaks the code

So i updated comfyui in the manager, restarted, and am still running into issues

error message, for reference

[jaunty palm]

Alright thanks man!

[severe sand]

Is it alright that I add this line to make sure the correct version is used?

[ornate elk]

No problem.
Also to revert from TensorRT back to default. You need to delete its extension from the Extensions folder. And after that you need to delete the venv folder.

[ornate elk]

Not in this file

Pls edit the webui-user.bat

There is PYTHON=

where you can add the path

[severe sand]

ah thanks!

Do I need quotiationmarks?

never really used batch before

[ornate elk]

If the path contains spaces then yes

[severe sand]

Kk

Got it, thanks

time for the actual test

been on this all day

[ornate elk]

Mark types as fast as a chat GPT bot xD

[severe sand]

I was so confused

WHAT THE HELL

Might be a bot in disguise

[ornate elk]

Yea its a bot xD

[severe sand]

but the status is online on phone

@proper notch test

AI gaslighting

Is it supposed to be installing more stuff at this point?

[ornate elk]

It does in the background

[severe sand]

I'm on the final step

alr

[severe sand]

Is there someplace I can look for some basic knowledge to understand related terminology, how to get different models etc?

huh

[ornate elk]

Maybe on https://stable-diffusion-art.com/

For everything model or lora related checkout Civitai.com

[severe sand]

thanks
never even heard "lora" before lmao

[ornate elk]

Thats the base 1.5 model

[severe sand]

1.5?

[ornate elk]

Yes 1.5

The commonly used models are based on either 1.5 or sdxl

[severe sand]

I don't know what either means

[ornate elk]

Checkpoints = models
Models are used to generate images.
They can be trained on different styles
You can download multiple models and switch between them

[remote ivy]

Anyone here run forge on linux?

[severe sand]

Oh, I know alteast what a model is. Haven't heard the term "checkpoint" before tho.. why 2 terms for the same thing?

[ornate elk]

Checkpoint is the official term, ans model the more common used one

[severe sand]

And what're the differences between 1.5 and sdxl? Are they like the 2 main models provided by the developers or something?

[severe sand]

O

[remote ivy]

Im trying to figure out how to run forge on linux, but it keeps just pulling in the original automatic1111 from its git repo and then launching that instead of forge

[ornate elk]

Yes excactly. The base 1.5 and sdxl model were released by Stability.ai as Opensource. And the community started to train their own models based on them

[severe sand]

Aah alright, thanks a lot for all the help:)

[ornate elk]

No problem, but I need to go off now its really getting late here

[severe sand]

No worries) good night

[ornate elk]

If you have any issues, feel free to @ me, than I can take a look at it tomorrow.
But make sure you set the system variable paths correct. Thats the most common issue when following the guide.

[severe sand]

The variables are both set and I checked the box when installing python, should work. And alright, will do^^

[crystal nacelle]

anyone know how to fix this Ad detailer square showing up in final render?

[fair oxide]

plz help

[shell nymph]

trying to figure out why when ever I enable hires. fix upscaler it takes a super long time sometimes?

not sure what I did to mess it up, it was working fine for awhile now

[fair oxide]

What's ipadapter best used for? Combining images?

[oblique parrot]

Yes, I know. That’s why I am in the Stability server asking. OpenArt.ai got early access to SD3 use in their creative upscale tool but it came with Church Lady level filters. And I have no interest in making smut. I’m just sick of Stability’s NSFW detector going off any time it sees a bare ankle.

[fair oxide]

why do the images I generate look so bad

my result

whats meant to generate
black bar is cuz boobs

[shell nymph]

try to use a similar width/height, start with their exact promp and negative prompt then tweak it from there

[fair oxide]

Tried it some of them work only at 1024 by 1024 for some reason

The one I showed doesn't seem to care

[fair oxide]

bc its an sdxl probably

[real glade]

My first guess would be to try adding "worst quality, low quality" to the negative prompt (or get the EasyNegative embedding). Without those two, your basically telling it you dont care how bad it looks.
Could also try bumping up the sampling steps a bit?
My small brain is also telling me it has something to do with XL models. Ive downloaded a few, and every single gen has looked like trash for me.

[fair oxide]

sdxl works best at 1024x1024

switch to comfyui 🙂

[buoyant pewter]

Well, I tried installing a different version of ComfyUI and had only marginally better results. It was a hassle that led me through a number of Home Brew/Python issues, but at least I got an image that was not just a blank field of color (typically blue) with the SD 1.5 model, rather than the SD XL, which never produced any image but one filled color.

My setup is based on a Macbook Pro M1, running Sonoma.

[fair oxide]

idk what the situation on macs are

you might have to jump thru more hoops to run comfyui

[real glade]

my brother in Christ why has it been CONSTANTLY doing this?
I havent had any problems for the longest time...

[fair oxide]

apple = bad

[real glade]

wdym "OuT oF MeMoRy"

[fair oxide]

sorry cant help ya

[fair oxide]

youre on windows running a 3090?

huh

thats odd

[real glade]

wym

[fair oxide]

well you have a better graphics card than me

so i thought it odd that you're having issues

old card?

[real glade]

i used to constantly make 768x1024s all day, but now its just like nope u cant do that

[real glade]

ok... I literately just closed everything other than SD and now it just magically works again?...

[limber pier]

Are you running a1111

[real glade]

I am

[limber pier]

That's prolly the issue

I had the same problems

Try forge

A1111 is so bad with vram and is slow as hell too

[ornate elk]

Not enough RAM, you need to increase your Pagefile size.

[ornate elk]

What VAE did you use?

[ornate elk]

Make sure to set hires steps to 10 and not to 0

[ornate elk]

You need to use negative prompts or some negative embeddings

[severe kindle]

Seems everyone just looks at the word Mac and assumed that was the issue and didn't look at your workflow.
Don't use the SDXL refiner model so start an image generation use the base model.
SDXL doesn't do 512x512 use 1024x1024 to test.

[keen edge]

Hey, I have an issue where I can't tick NetControll in extenssions, therefore I can't select models I downloaded to OpenPose. Any help?

[grizzled void]

Hello,
I tried to install stable diffusion with automatic1111 on my laptop with a gtx1650 4gb

[ornate elk]

Hey, can you show a screenshot ? Also do you get any error in cmd?

[grizzled void]

And it take a LOT of time for an image to generate, i tried to put some command to optimize the system but it's not working

Can someone help me please ;-; i installed git and python already, if I can generate an image in 5/10 secondes it's fine but if I need to wait too long even with an optimize stable diffusion i give up

And, in the seconde case, Is there a way to run it with an hosting service for free (if it is enough powerful to generate an image in 5/10secondes)

Thanks !

[ornate elk]

Hey, a 1650 needs only --xformers --medvram --no-half

And then make sure you didn't have hires fix or refiner enabled by accident

[grizzled void]

I put those commands but it's not enough

;-;

[grizzled void]

What is this ?

[ornate elk]

Screenshot your webui

With steps, resolution etc

[grizzled void]

For the resolution i have 512/512

Steps : 10

[ornate elk]

Make a screenshot please of the full site

[grizzled void]

I tried lower resolution but it's not working too

I'll do it in a few hours, i'm outside

[ornate elk]

Oh okay

[grizzled void]

Can i contact you in mp ? 🙂

[ornate elk]

Sure

[grizzled void]

Thanks ^^

[sinful halo]

added it to the right path now, sorry for that, but now theres another error

[ornate elk]

Your missing the --

[sinful halo]

lol, im stupid /=/ thanks man

[ornate elk]

No problem ^^

[keen edge]

Screenshot of what precisely? I'm not home currently

And I didn't noticed any errors in cmd, might have to take second look tho

[ornate elk]

Sure, let me know when you can test it again

[keen edge]

For now, this is what I can't tick, it just stays always like this even arter hitting apply

[ornate elk]

Do you use auto1111 or forge webui?

[keen edge]

auto1111

[ornate elk]

Make sure you whitelisted the webui in any browser adblocker

Sometimes adblocker or script blockers can cause issues with the webui

[keen edge]

Okay, I will check it out later and return if the issue prevails, thanks for now

[ornate elk]

What was your gpu again?

You may need the rocm library test zip instead of the normal library

[vast crest]

Hi, do you know another way to create Lora? Because currently I can't create them anymore, since the process always fails. Help me 😦

[carmine agate]

Within Forge(/Maybe A1111 too), Hypernetworks are Embeddings it seems but are .pt no longer supported, only safetensors?

Excuse me, not Hypernetworks, TexturalInversion

[carmine agate]

Also, I seem to get a several second lag when coming back to forge from elsewhere on the computer & clicking on a field. Are there any known tricks to get around this? (My computer & gpu are quite old)

[blazing mantle]

So I'm having a problem with ControlNet. Looked everywhere but couldn't find a fix thats worked for me, so hopefully someone here can help. Whenever I try to use openpose, or several other models for that matter, I can't get any previews in the Processor window and my results don't come out correctly.

I keep getting a error message in the cmd window saying "ConnectionResetError: [WinError 10054] An existing connection was forcibly closed by the remote host"

Any ideas? I just did a clean reinstall of SD 1.5 and ControlNet as per one suggested fix. They had also said that python might’ve been the issue, but wasn't certain.

In addition, OpenPose sometimes works, as does some depth models. OpenPose only works with the skeleton models and no processor set.

[ornate elk]

Do you have the right controlnet models?

Also make sure your webui is whitelisted in any browser adblocker

[blazing mantle]

I have the correct models. That much I do know. How would I go about checking on the latter?

[ornate elk]

In your browser, do you use an adblocker ?

[blazing mantle]

As far as I'm aware, no. Nothing I've downloaded or installed.

[ornate elk]

Okay, what does the cmd shows when you click on install at the controlnet extension?

[blazing mantle]

Is there a chunk of the cmd you're looking for in particular? Big blob of text here

[ornate elk]

The whole cmd log would be good

[blazing mantle]

The only thing above line with Error is the usual stuff when I first launch SD.

[ornate elk]

Okay can you show it too? Just to make sure

And then try to delete the tmp folder inside the Stable-diffusion-webui folder

And try install it again

[ornate elk]

Okay looks normal, but you should also use xformers for better performance

[blazing mantle]

If you have a link for that, by all means. Also, I'm not seeing a tmp folder.

nevermind. Found it. Deleted

[ornate elk]

Can you show me how you tried to install the Extension?

Did you found it in the Extension list or did you copied the URL of it?

[blazing mantle]

I'm an idiot and have been trying the wrong link. Let me try the right link this time...

Stupid GIF popped up. But yeah. Copying URL and pasting

[ornate elk]

If you click on the Extensions tab, click on Available, then click on "Load from" then in the list search for sd-webui-controlnet

And then click install

[blazing mantle]

Just says installed

[ornate elk]

Then restart

And check if you have it in the txt2img section

[blazing mantle]

I mean, I couldn't click on install as it said installed in that section.

[ornate elk]

You can delete the Extension again and then install it correctly if you want to be sure

[blazing mantle]

I'll delete and try again

Download the models again too, I'm assuming?

[ornate elk]

Use these from here:
https://civitai.com/models/38784/controlnet-11-models

And put them with their config file into the models/controlnet folder

[blazing mantle]

Did that. Same error as before when trying to load preview.

Here's what pops up in cmd when I try:

I know almost nothing about computers or any of this stuff, really, but I feel like Python might be causing an issue somewhere?

Or my own incompetence. Either or. I have a habit of somehow doing something so wrong that it leaves people speechless.

That bad, huh?

[ornate elk]

Http errors

Like connection errors

You may need to change your DNS setting of your PC

Or disable a VPN if used

[old parrot]

Any new/updated plugins to save state and remember last parameters used on Stable Forge?

The plugin I used on A1111 didn't work on Forge last I checked

[blazing mantle]

Not using a VPN, so how and what would I have to do with my DNS settings?

[ornate elk]

It seems that python was installed trough the windows store and not trough the official site

[blazing mantle]

Yeah, both guides said I could do either or

[native latch]

Hello everyone, I need to do a small task of merging two images. I have to merge the image of a baby with an image of their grandparents. Is this possible? How do I do it? Thank you in advance, best regards.

[blazing mantle]

So, delete Python from my computer, reinstall with the version from the official site?

[ornate elk]

Yes

The 3.10.11 64bit version

Then delete the venv folder

And relaunch the webui-user.bat

[severe sand]

I can't seem to get this fixed no matter what I do

[blazing mantle]

Now cmd says it can't find Python despite me having just installed it

Nevermind. Forgot to select the PATH option

[blazing mantle]

The error persists

[ornate elk]

Okay then you need to try change the DNS server

[onyx solar]

Do you have tiling turned on in your settings tab?

[sinful halo]

Rx 5700 xt, bcs of this ? i tried to switch models

[ornate elk]

Did you used the rocm test library or the normal one?

[sinful halo]

the one on the guide

[ornate elk]

Because the 5700xt would need the test zip

[severe sand]

I can tell you that after it's done reinstalling for the 69th time

[severe sand]

anyhow, i never touched the settings. should i?

I mean, not before actually getting it to work

[sinful halo]

i used this one, for who has an X on HIP SDK

[ornate elk]

Okay thats the wrong one.

[sinful halo]

/-/

[ornate elk]

You need the one which ends with test.zip from here
https://github.com/brknsoul/ROCmLibs/

[severe sand]

Similar results with tiling on

yaay................

[onyx solar]

It should be off, which is why I asked

[sinful halo]

same error + another on (the one below) when i try to switch models
man, if i bother you, you can leave me, i can accept my budget amd card fate /-/

this happens with newly downloaded models, and old ones too

[ornate elk]

What's in your webui-user.bat?

[sinful halo]

@echo off

set PYTHON=C:\Users\DITlieD\AppData\Local\Programs\Python\Python310\python.exe
set GIT=C:\Program Files\Git\cmd\git.exe
set VENV_DIR=
set COMMANDLINE_ARGS=--use-zluda --medvram-sdxl --update-check --skip-ort

call webui.bat

[blazing mantle]

What do I need to change, and what does it need to be changed to?

[severe sand]

it is off
not working

[ornate elk]

Does image generation work?

Only errors out when switching models?

[sinful halo]

oh, a curious thing, i tried sd next, i used web user to use args on the webui from sd next and it says i dont have hip sdk installed

[ornate elk]

Oh then please screenshot your path settings in system variables

[ornate elk]

Okay looks correct

Have you replaced the library folder?

With the test.zip folder and renamed it to library?

[sinful halo]

yes

i started again to see if i can at least generate a image but now another error

[ornate elk]

Remove --debug

Or whatever you added xD

[sinful halo]

it wasnt there before wtf

trying to generate a image

[ornate elk]

Okay

Then replace the rocm library folder

With the test zip one from github

[sinful halo]

Idk man, I tried to replace it like 10 times now, I'll try a clean installation of direct ml, reinstall hip sdk and replace the library folder again in the weekend after I come back from work, thanks for everything until now

[ornate elk]

Okay, no Problem, let me know how it goes 🙂

[severe sand]

I think you gotta replace the link in the pinned message

[blazing mantle]

@ornate elk so what do I have to do with my DNS settings exactly?

[severe sand]

i need openai to release a super agi asap so that it can fix sd for me

[ornate elk]

I will update my guide when I'm back home

[ornate elk]

You can try to change your DNS setting from your PC to 8.8.8.8 or 1.1.1.1 and then try again.
You need to search a tutorial for how to change the DNS

[blazing mantle]

Still didn't fix the error

[ornate elk]

Are you located in Asia or russia or Iran ?

Some governments can block connections

[blazing mantle]

Nope. United States.

[ornate elk]

Can you show me the error from cmd?

[blazing mantle]

error message is in the second image at the bottom

[ornate elk]

Maybe your antivirus program is blocking it?

Or do you have any third party antivirus program?

[blazing mantle]

Shouldn't be blocking it. Never got any notices or anything. Would usually tell me if it blocked something or if I had downloaded something unsafe. I could poke around with it and maybe temporarily disable it, but I don't think it's the cause.

[ornate elk]

The issue is something with the connection between you and the github site of the controlnet extension

Win error means something on your system

So either a proxy, VPN, adblocker, pihole, antivirus software

[blazing mantle]

It'd have to be ESET then but I just temporarily disabled protection and no changes. I have been having some issues going to Github recently though. I got a connection issue error message while trying to get the ControlNet models.

[ornate elk]

Okay thats strange too

[ornate elk]

It seems that the error occurs when it tried to connect to this site
https://huggingface.co/lllyasviel/Annotators/tree/main

There the Extension normaly downloads the body_pose_model.pth

And the other annotators depending on the selectet controlnet model

[blazing mantle]

Come to think of it, the very first guide I followed had me do something with that site, but the other one didn't. Maybe that site might be the root if these issues and I downloaded stuff that might be causing problems.

Actually, that's where the ControlNet models I downloaded are located.

[ornate elk]

The controlnet models should go into stable-diffusion-webui/models/controlnet

I'm off now its late here. Can look at it tomorrow again

[blazing mantle]

I too will look into it I the morning. Pretty sure the models are in that folder, but I'd have to look again.

[crystal nacelle]

I need a control net Guru please.. ive manually drawn the proper control net skeleton on Photopea however, the SendToControl net button doesnt work...

control net preview is blank and render result has the old control net image

[scarlet igloo]

stabilityai/sd-vae-ft-ema-original and stabilityai/sd-vae-ft-ema ，What are the differences in hugging faces

Should I choose stability ai/sd vae ft ema for a beginner

[fair oxide]

stable diff completely ignores all my loras no matter the weight

this happens for all of them

[pseudo nebula]

bruh wtf, how do u show the render latency of the gpu

in after burner

[fallen quest]

Maybe you should actually prompt for something and not just call the Lora.

[fair oxide]

I've done that

It still ignores it

[fallen quest]

Is the Lora for the right architecture? E.g. 1.5 model -> 1.5 Lora.

[blazing mantle]

Whenever you see this, I can confirm that the models are in the correct folder as well. Thinking maybe a complete fresh reinstall of everything again with the correct python might do something? Honestly, I have no ideas at this point.

[livid lark]

ignore directory without repeats <--- i get this and it cant find my dataset images, i am tryign to train a sdxl lora, what am i missing?

is there some namingconvention i must adhere to? also my traidnign images are jpg and the reg images are png

[little shoal]

how do you add libxcrypt-compat as a package in stability matrix so i can get my comfyui working

[ornate elk]

You can try to install an other extension, to test if you get errors there too

[ornate elk]

It looks like you have 2 images in photopea. You need to merge them and then try to send it to controlnet

[blazing mantle]

The thing is, only parts of ControlNet didn't work. Some did. Uncanny was fine, OpenPose only worked without selecting a processor and with skeletons, and the Lines one only worked with certain processors.

[ornate elk]

Okay, yea makes sense that openpose only worked without a preprocessor as the error is from downloading the preprocessor

I dont think a complete reinstall would fix it

[blazing mantle]

Others just don't work at all. Like the mapping one, or the IP adapter

[ornate elk]

Do you get there the same error I'm cmd too?

[blazing mantle]

I'm away from my computer for a while so I can't double check, but yes. Trying to get a preview pops up with the same error and the images being generated are not being affected by ControlNet's settings. Just the prompt.

[ornate elk]

Maybe try disable Eset realtime protection then relaunch the webui and try controlnet again

[lunar ridge]

is there an api for stable audio 1.0 and if not when is the api for 2.0 coming? Thank you!

[livid lark]

my koyha _ss only runs 1 epocsh even if i specify more. why?

[livid lark]

whats the best number of steps per image? cna i get away with 10 or less (default is 40)

[tough kettle]

Hello, it is possible to singer with paroles written in the invite ?

[cloud rune]

Hello! I am inpainting an image that has the size of 821x1132 with "resize by" set to 1, so the output should also be 821x1132. However, all my outputs instead have 816x1128. Does anyone know why that is and how i could possibly fix it since i need the output to be the exact same size as the input.

[ornate elk]

Hey, Set the Inpaint Area to mask only, then set the resolution to a square (its not the output resolution)

[uncut jewel]

Hey! I'm facing the issue that when i finish generating an image in ComfyUI, my GPU ressource stay so high (100% 3D and almost 100% in Vram), i can wait for minutes, even an hour and the gpu ressources doesnt decreases even if i don't do any generation or anything inside comfyUI, while when i close the terminal, it then slow completly down to a few %, so it doesnt makes sense at all that it still consumme that much when the command do absolutly nothing. Anyone did ever experienced that issue and know how to solve that ? It started to happened really frequently and slow down my pc really heavily...

[sleek nexus]

Is it better to have a newer, more powerful GPU to run SD or an older GPU that has more vram?

[ornate elk]

What's your gpu?

[ornate elk]

There aren't many older GPUs with a lot of vram.
But for SD it goes:
More vram = higher resolution and more demanding extensions can work together.
Newer GPUs: better performance = faster image generation if the resolution is not to absurdly high

[uncut jewel]

3090

I got like 24Go of Vram

[ornate elk]

Hm strange that the vram gets stuck at full usage with that

Make sure you have the latest nvidia driver

[uncut jewel]

Yea exactly, i doesnt make sense lol

[uncut jewel]

I'm on Nvidia Studio Driver that is up to date

[sinful halo]

hey man, everything is working fine now, thank you so much !! i only have one more question, my SD NEXT install does 9s/it and my direct ml does 1-2s/it, i can fix it or better i keep with direct ml ?

[uncut jewel]

Maybe it's coming from an extension, but yea that's pretty weird, i run different workflows on it, but i always have that issue that the vram get stuck at full usage when finishing generating..

[sinful halo]

OH NO, MY SD is using my cpu

my direct ml is using zluda im sure bcs now i can upscale

[ornate elk]

So you got the Directml with zluda to work? Than stay with that

[sleek nexus]

I have an i5 9th gen and according to what I've searched online the most powerful GPU it can (or should) be paired with is an rtx 2070 super but from what I've found it just has 8gb vram whereas something like the 1080 ti has 11gb

[sinful halo]

yeah, better than fryin the little brain matter i have left

thanks again man

[ornate elk]

Haha xD also if you have problems with sdxl models then enable FP8 mode for sdxl in the settings.
That will lower the vram usage of them

[sinful halo]

oKS, understood

[ornate elk]

Your card was the 5700xt right?
So the rocm library test zip was the thing that fixed it

[sleek nexus]

I do want faster generations but higher resolution would also be nice considering my current GPU can't generate higher resolutions without crashing

[ornate elk]

What's your current gpu?

[sleek nexus]

RX 580 8gb

[ornate elk]

Ah well

[sinful halo]

yes, rx 5700 xt, i made a clean install of eveything on my C drive, i followed your zluda guide again, but then i used rocm library test zip, and everything worked

[ornate elk]

Perfect, good to know 🙂

[sinful halo]

🙂

[ornate elk]

Can you try a test for me for (it/s or s/it) speed?
Use a 1.5 2gb Model.
Resolution 512x512
30 steps.
And for the prompt only
Cat in positive and
Blurry in the negative prompt

Without upscaling or anything

[ornate elk]

A 2070 can do highres images and ia also fast.
But 8gb vram is not the best for the future

But I wouldnt buy an 1080ti for its 11gb

[sinful halo]

Yeah sure, just give me a moment

[cloud rune]

i get what i did wrong here now, thanks a lot!

[sinful halo]

sorry for taking so long, i was downloading dreamshaper bcs is 2gb, i used dpm++ 2m karras, cfg on 7 and it took 20 seconds sitting at 1.53it/s

[sleek nexus]

Hmm, would a 3060 suffice then? I found that it has 12gb and from a quick search, pairs well with an i5 9th gen

[timid ibex]

Can't seem to install Automatic1111 using Git, any help on this?

[ornate elk]

Thanks for testing! Thats quite impressive as on Directml without zluda it would be s/it and not it/s

And CPU would be also s/it (much slower)

[ornate elk]

Yes thats a good GPU for SD and for its price

[sinful halo]

so thats good then, nice, everything bcs you helped me man, thx

[sleek nexus]

Alright, should my current GPU stop working I'll know what to switch to

[sinful halo]

got to 1.09it/s at some point, im so happy

[timid ibex]

Any asnwers?

[ornate elk]

Seems like a connection error

[timid ibex]

I think I might need a VPN for this then...

[ornate elk]

If your from Asia/russia/Iran you may need a VPN

[timid ibex]

Yep South East Asia

[ornate elk]

Yep then a VPN will help

[ornate elk]

Yea that's good.
Hires fix is always slower. Best to use 10 or 15 hires steps.

[ornate elk]

Also tiled vae extension can help if you want higher resolution upscale

[sinful halo]

Ok, thanks man, appreciate it !

[elder notch]

hey guys, i don't know if this has been asked before, but when training lora , in captions do i type in what the lora needs to learn? or do i write basic stuff in the images?

[blazing mantle]

While going through the reinstallation process, this series of errors popped up during the initial boot-up. No idea why or if it may be connected to my problem

And now it just won't work at all. Huh.

It worked on the second attempt, but there was something about missing pytorch?

And the error is still there, despite turning off that thing in ESET and even putting the URLs in the allowed section

[ornate elk]

Connection errors again

[blazing mantle]

Well, I got nothing. Even reset router and all that fun stuff.

[ornate elk]

Maybe you should try out using a VPN

Also make sure your connected over Ethernet

[blazing mantle]

I just did ethernet. I'll see if I cant find a VPN to try.

[buoyant pewter]

Continued work on M1 based MacBook Pro. After uninstalling Python and Homebrew, clearing out old stale MacPorts, etc. and reinstalling HomeBrew and Python, setting up for ComfyUI, etc. I have two results.

./venv/bin/python main.py
and
./venv/bin/python main.py --cpu

The image shows a 'blue bottle' in the finder window which is the result of running main.py with no options and the Node screen with resulting 'bottles in landscape' is the result of using the '--cpu' option.

The no options speed is about 15 seconds, the '--cpu' speed is about 100 seconds. While that is not as snappy as perhaps I'd like, it's sure better than waiting a week for a card deck to come back from the compter center, with about 200 pages of errors due to JCL card mix ups.

The question I have is there anyone one or list dedicated to Apple and M1/M2, etc. and Stable Diffusion?

[blazing mantle]

Holy shit, it actually worked. I don't know why I need to pretend I'm in Canada, but I'm not going to question it. Thanks a ton for being so patient.

[nova wren]

i have a doubt not sure where to ask it, when doing img2img with atuomatic1111 is there a way to apply the denoising strenght for shape and color separately? or how would it be best if my intention is to use img2img to change the color of an object or character without changing other characteristics?

[ornate elk]

Perfect, no Problem xD

[ornate elk]

Hey, you would need Controlnet for that as you can lock the form of the image with canny or the pose with openpose

[cosmic furnace]

Cs1o?)

[ornate elk]

Hey

[cosmic furnace]

Lil bit not where it shud be probab but what im doing wrong here?

[cosmic furnace]

Hoi)

mb i need to write folder where to put file in the end? doesnt seem to worked either...

i do uesave from-json and a place of json file but...

also doesnt want to convert other way. just edit in temp worked -_-

[ornate elk]

Maybe use " " for the path

[drowsy pagoda]

hey, can someone help me resolve this error? 'RuntimeError: Could not allocate tensor with 9994240 bytes. There is not enough GPU video memory available!'

I have the rx 6700xt gpu which has 12gb of vram so idk why I dont have enough memory to generate ONE 512x512 image

[cosmic furnace]

C:\Users\AMD>"C:\Users\AMD\Desktop\CHEAT-TRNR\Total Conflict-Res\uesave-x86_64-pc-windows-msvc\uesave.exe" from-json "C:\Users\AMD\Desktop\CHEAT-TRNR\Total Conflict-Res\TCR_v60_9.json"

doesnt work like this 😦

error: unexpected argument 'C:\Users\AMD\Desktop\CHEAT-TRNR\Total Conflict-Res\TCR_v60_9.json' found

[ornate elk]

Hey, thats because your using the Directml webui (without zluda) which is very unoptimised in vram management

You should use my install guide for the zluda version which works faster and doesn't have vram problems.
The guide is in the Pinned Messages of this channel

[drowsy pagoda]

I have lshqqytiger's automatic1111 fork installed

idk what zluda is

[nova wren]

i tried using Control net but the generated image was laways wildly different shape base to the original character when it was a person.

And when it was an object (tried to create a truck with a different paintjob) it failed horribl

[ornate elk]

Yes lshqtigers fork can be used with directml or zluda as Backend. He suggest using zluda. My guide covers the steps you need to do to get it work

[drowsy pagoda]

ok, thanks

[ornate elk]

Zluda is an emulation level of nvidias s cuda for AMD cards

Let me know if you have any questions with the Installation

[drowsy pagoda]

btw, how would u say the performance of rx 6700xt is?

what is it analogous to on nvidia's side

[ornate elk]

Zluda is between 2 to 4 times faster than directml

I guess you get like 3-4 it/s

[drowsy pagoda]

no, I mean its performance as a gpu in general

I get less than 2 currently

[ornate elk]

A 6700xt is like a 3070

But with more vram

PS5 performance is a 6700 non xt

[drowsy pagoda]

yea, but Im talking about stable diff performance

[ornate elk]

You said general xD

[drowsy pagoda]

I meant stable diff lol mb

[ornate elk]

Idk the performance of every nvidia GPU

But wait I get you an image

[drowsy pagoda]

yea, ive seen that

it's crazy how even the 3050 has 2x the performance

[ornate elk]

Yea but a 3050 is barely useable for Highres images and is awful slow when using sdxl

So your card won't have a problem with both

When using zluda

[cosmic furnace]

what it means input - output? o.0

i command it to convert file & subcommand to input or output? 🤔