🤝｜tech-support

[ornate elk]

set a square resolution

its for the tile size

if your on 1.5 set it to 512x512 or 768x768. For sdxl choose 1014x1024

[cosmic furnace]

and if i wanna go from 1024-768 to 1536-1152 ?🤔

[cosmic furnace]

Doesnt matter if i upscale or not it seems =\

about 2-3 generation on same seed with variations & same resolution and i get this error

[ornate elk]

with or without upscaling?

sdxl or 1.5 models?

make sure you use tiled vae when upscaling

[cosmic furnace]

where do i do this tiled vae?)

https://civitai.com/models/317902?modelVersionId=469115 trying this pony model further... probably shudnt use 1.5 loras? o.0 but only 2 of them fixing legs & pose ...

hmmm

either random or loras was causing errors 🫠

[ornate elk]

Yes never use 1.5 loras with pony models

Pony is sdxl based

[cosmic furnace]

y hes said partil compability with sdxl lora his especial model and pony is need even pony special loras i guess 😵‍💫

how do i enable tiled vae?)

this? o.0

[ornate elk]

Yes

Install it, reload the ui or restart. And then just enable the Tiled VAE.
Leave the settings of it on default

[cosmic furnace]

mb smthing else useful i shud enable-install?)

[ornate elk]

!after detailer for easy face enhancing
Boorutag autocomplete for anime and pony tags

[cosmic furnace]

this ?

https://github.com/DominikDoom/a1111-sd-webui-tagcomplete

hm

[ornate elk]

Yes these two

[cosmic furnace]

y found it all) thx

like this enough or

need to be like this?)

oh okay)

[ornate elk]

Yep it even makes image generation faster

As the VAE step takes less time

For sdxl with upscaling its perfect

[cosmic furnace]

btw from all this i shud pick fastest?)

[ornate elk]

Automatic

Don't use sdp

[cosmic furnace]

But auto always slowest doggettx -_- no ?

[ornate elk]

SDP and amd still not working great together. SDP can cause crashes and slowdowns even with zluda

Leave it on automatic

So its dogetexx yes

[cosmic furnace]

u tested them all out? 😦

oh well speed difference seems almost 0 or 1 sec anyway >_>

mb sub quadro? 🙂

[ornate elk]

Not needed its for directml and low vram GPUs

Only SDP could speed up the generation if it would work correctly with AMD, rn its bugged

I'm off now, gn

[cosmic furnace]

mmkay 🙂 thx, gn )

[stiff prism]

Anyone know why there even being a Lora folder causes my Pony Diffusion (Pony Realism) generations to fail due to a CUDA OOM error? It's super weird, if I have the Lora folder on another drive and don't plug in that drive until the program loads, everything works properly but if I have it pointing to a Lora folder on my C drive, it seems to be able to allocate sufficient memory and get a "Nonetype isn't iterable error" also SDXL/Pony Lora don't even show up, it just acts like they don't exist but 1.5 Lora do show up.

I'm using Forge with a 6 GB 2060 so I know it's underpowered but so long as that Lora folder doesn't exist so far as it knows, it generates fro the base model just fine, I just can't use Lora

[fresh phoenix]

Hi,
would anyone help interpret the following error from stability AI's searchAndReplace endpoint (https://platform.stability.ai/docs/api-reference#tag/Edit/paths/~1v2beta~1stable-image~1edit~1outpaint/post)
"{"errors":["finish-reason: required","seed: required"],"name":"internal_shape"}"
The same image/prompt input works majority of times but we see the above error sometimes. Since, we are omitting the seed, expected it to be internally set/generated as per the specs.
Any ideas?

[umbral fox]

tagger does not appear on Stable Diffusion.
How can I display it?

Stable Diffusion ver.1.8.0
python: 3.10.10

There was a problem with tagger ver.1.6.0, so I installed the fork version (https://github.com/picobyte/stable-diffusion-webui-wd14-tagger), but it was not displayed.

[fair oxide]

@ornate elk Are you good with SDXL Pony? I had a question about a good way to get amazing quality, during the txt2img process
My specs are pretty good:
4080 Pny
32 GB DDR5 4800mhz
i9-13900k

just wondering what your methods for getting good image outputs are

[ripe bronze]

Hi, using comfui. I started to use some SDXL checkpoints. When I am using some SDXL checkpoints the resulting image only shows the faces in the frame. But if I change it back to a non-SDXL it shows the full body. Do you know why? I tried prompt "full body in frame" but still the same problem.

[wheat helm]

is there a way to get deforum to end on the INIT image too instead of only starting on it?

[warm otter]

edit: I'm dumb XD (it's there)
Hey everyone - json api spec of stability ai is not up2date I assume?
I'm missing control (sketch, structure) parts :/ (https://platform.stability.ai/docs/api-reference#tag/Control)

api spec is at the top here: https://platform.stability.ai/docs/api-reference

[drifting pendant]

Hello, I have tried to follow you guide, seems to work but when I try to generate something, nothing happens :/
I've followed the tuto, step by step and it was very clear ! 😉

[ornate elk]

your on the final step 🙂
Noteworthy: The first image generation can take from 15 up to 40 minutes. First time only.

just let it load until you get an image

as long as the generate button is grayed out after pressing it, and you have no error in cmd, its fine

[drifting pendant]

Oh, ok, sorry then :/
I relaunch it right now, thanks again for your time 😉

[ornate elk]

no problem ^^

let me know if it worked

[onyx rivet]

rocBLAS error: Cannot read /home/meezopc/stable-diffusion-webui/venv/lib/python3.10/site-packages/torch/lib/rocblas/library/TensileLibrary.dat: No such file or directory for GPU arch : gfx1032
List of available TensileLibrary Files :
"/home/meezopc/stable-diffusion-webui/venv/lib/python3.10/site-packages/torch/lib/rocblas/library/TensileLibrary_lazy_gfx1030.dat"
"/home/meezopc/stable-diffusion-webui/venv/lib/python3.10/site-packages/torch/lib/rocblas/library/TensileLibrary_lazy_gfx900.dat"
"/home/meezopc/stable-diffusion-webui/venv/lib/python3.10/site-packages/torch/lib/rocblas/library/TensileLibrary_lazy_gfx906.dat"
"/home/meezopc/stable-diffusion-webui/venv/lib/python3.10/site-packages/torch/lib/rocblas/library/TensileLibrary_lazy_gfx908.dat"
"/home/meezopc/stable-diffusion-webui/venv/lib/python3.10/site-packages/torch/lib/rocblas/library/TensileLibrary_lazy_gfx90a.dat"
"/home/meezopc/stable-diffusion-webui/venv/lib/python3.10/site-packages/torch/lib/rocblas/library/TensileLibrary_lazy_gfx1100.dat"
./webui.sh: line 297: 8809 Aborted (core dumped) "${python_cmd}" -u "${LAUNCH_SCRIPT}" "$@"

TensileLibrary_lazy_gfx1032.dat

where do i get TensileLibrary_lazy_gfx1032.dat

[timid notch]

i was trying to download some plugin and edit this command a few days ago, but now it comes with some problem. i am wondering if there is a way to set this software to original version without delete any models i already installed?

[ornate elk]

Here:
https://github.com/brknsoul/ROCmLibs/

[onyx rivet]

ok

[onyx rivet]

oh btw i need it for linux

[ornate elk]

Oh then idk

[onyx rivet]

does this work on linux

[ornate elk]

Idk, but these files work on windows with zluda and rocm

[onyx rivet]

sadly for some reason it seems like nobody knows how to build libs for gfx1032

[cosmic furnace]

It seems that using any lora with this model eventually causes error o.0

about 3-5 times if 1024-768 and fewer times if res higher

and author says pony loras shud work with it. Ive only used one style lora hm

no errors in command =\

[near acorn]

anyone here use Kritta with ai generation?

[cosmic furnace]

it seems my friend with 1060 nvidia have kinda similar issue... after 7-8 generations sd froze he says. Either recent updates has ruined smthing or this model? o.0

is there no erorr log txt file? -_-

512-768 20 generations no error o.0

seems the higher resolution is the faster error or resolution matters dunno

probably the model have some issues -_-

30+ images with no errors on 512-768 o.o

probably its this author who recommended higher than 768 or 1024 to do only via upscaling, not sure tho dont see any of it on model page 🤔

40+ images =\

[obsidian carbon]

Allo! I am trying to learn Control Net and AnimateDiff and when I have everything set up and ready to go, the generation barely starts and I get this error

RuntimeError: CUDA error: device-side assert triggered
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 have an NVidia 1070 ti, 8GB Vram and 32 DDR4 ram. I am using SD 1.5 and my cmd args are: --xformers --medvram-sdxl --no-half-vae --disable-model-loading-ram-optimization

[cosmic furnace]

hm this lora didnt appear in ui if its in Auto folder instead of the one specified in file. Is this okay or it wasnt working? o.0

yeah it wasnt working 3-4 rd time even 512-768 is error =\

[near acorn]

anyone here use Kritta with ai generation?

[ornate elk]

Remove --disable-model-ram-optimization
Then set animatediff batch size to 4 and test

[obsidian carbon]

Did that, I get this

RuntimeError: CUDA error: device-side assert triggered
Compile with TORCH_USE_CUDA_DSA to enable device-side assertions.

Animate dif is being strange

[timid notch]

the workflow got some error when ksampler working, but it didnt stop, wondering ~does this error matter?

[near acorn]

anyone here use Kritta with ai generation?

[alpine rune]

for some reason iam not allowed to download stablediffusion no clue why

[cosmic furnace]

parent control

[obsidian carbon]

I’m still having issues with this. Tried several things and no luck so far.

[obsidian carbon]

Never mind I am an absolute tool. I forgot to set the batch to 4. Thank you @ornate elk for the support, life saver

[late tree]

Not exactly tech support for SD, but is anyone else getting "Application Errors" from CivitAi?

[ruby yoke]

which model are you trying to render with?

[late tree]

I'm not, I was just trying to look at Loras and the page keeps crashing saying "Application error: a client-side exception has occurred" but it happens on two different machines and on different browsers

When it happens it shows a sign in button even though I'm signed in.

[ruby yoke]

i'm not having any issues - but i'm also not signed in.

@jagged plover #artisan-3 message in case you missed my response

[ruby yoke]

civit just posted on their discord they're rolling a fix out for the issues

[azure nimbus]

Is the API down again?

Last time, I only noticed SD3 went down. Now, it seems like SDXL and SD Core might also be down.

[jagged plover]

Image Core API appears to be working fine

[azure nimbus]

SDXL just came through for me now

[rare cloud]

Me when no paid subscription

[late tree]

Cool.

[willow kettle]

Might be a niche question but has anyone gotten ControlNet inpaint_only+lama working on amd? Most of it works great but apparently ControlNet inpaint_only+lama needs cuFFT which seems not to be in lshqqytiger's zluda release

[ruby yoke]

you might ask PurzBeats on twitter

[cobalt latch]

Can Stable Artisan be used by directly messaging the bot privately? If so, how? Also, when I singed up and subscribed to Stable Artisan it subcsribed me to Stable Assistant instead. Is this what is supposed to happen?

[jagged plover]

Yes, artisan subscription is assistant on the inside.
For using it outside this discord, not yet but soon™️ (probably a week or so)

[stable kindle]

Does anyone here have any experience with using ControlNet with SDXL? I'm trying to upscale using tile but it keeps generating a new image on each tile instead of just upscaling the whole thing coherently, and this is while using Ultimate SD upscale. Would really appreciate any help.

[stable kindle]

keeps tiling like this, no clue why

[ruby yoke]

https://twitter.com/PurzBeats is the guy you wanna ask comfyUI questions of

Purz.xyz (@PurzBeats) on X

🟣 Artist/VJ 🟣 Educator 📽️ Creative Exploration
🟣 Blender, ComfyUI, AI Animation Specialist
🔗 https://t.co/mbzejbxMrw

Twitter

[stable kindle]

is he in this server? I dont have x

[ruby yoke]

i dont' know if he's on here or not. he has a discord but you need to connect with him on twitter i think

[stable kindle]

@serene valley I think this is him, I guess I'll shoot a DM. Thanks!

[ruby yoke]

yup. that's him

[cosmic furnace]

any pony model when used with lora generates about 3-4 images & get error after . On zluda

[cosmic furnace]

Fixed vae problem on sd next. No errors there soooo ... gonna stay there 🤔

[wet flame]

anoyne using forge ui?

my tiled option disappeared

[ornate elk]

I don't have this issue, what's the error?

[ornate elk]

Yep it doesn't work. But inpaint harmonious works

[ornate elk]

You need to remove every tag that describes the person or environment. Only leave quality tags. Then it should work

[wet flame]

ok so after test A111 is best for making any kind of seamless textures

forge ui is not at all good for it so if you are making textures use A111

[terse lichen]

does comfyui generation speed also depend on your gpu and cpu processing power or it depends ?

[ornate elk]

yes it depends on your gpu

as for every webui

cpu doesnt affect the generation speed (only if some extensions require it)

[drifting pendant]

Alrighty, tested and working like a charm ! 😉

[solemn atlas]

Hey guys, what does that mean

[normal field]

Hi guys, I installed stable diffusion A11111 but the interface doesn't load properly. I have tried uninstall then reinstall, clear cache, trying on different browser but problem persists 😦 Maybe someone knows what to do?

[drifting pendant]

Does anyone know a website or a documentation for optimising the sampling method by models ?
Sometimes I have very wierd results and I have to push the samples steps like 100+, sometimes 150 @ 1024x1024 😭

I run a 6800XT

[brittle lantern]

How's the 6800 xt handling SD and do u use windows or Linux also do u ever run out if vram on 6800 xt?

[brittle lantern]

Guys should I dual boot linux for SD on AMD GPU for ROCm performance

[wet flame]

comfyui saying

how do fix this issue

man full day gone fighting with AI set ups

[brittle lantern]

I had trying to get Zluda work with my 6600 xt just for it to just not work

[drifting pendant]

Well, it's kinda slow compared to Nvidia stuff, I'm planning to buy a 3090... 🙃
But I'm not complaining, It works, I just want to know how to optimize all this stuff 😉
I'm on windows, tried linux, Never make* it work... Dont know why really.
Understand better all this "complicated" world

[wet flame]

i hate nvidia for being so greedy

for what reason their gpu are so expensive for low vrams

i mean only for AI nvdia but for rest amd is still better

[brittle lantern]

I was considering cuz it's £600 but then I'll sacrifice gaming performance compared to AMD offering new for £600

[wet flame]

if i had no AI biz i will never buy nvidia lol

amd gpu n cpu better on price vs performance

[brittle lantern]

I don't wanna either I'm considering getting a cloud based pc yk for gaming just for SD but it only has 3070 which is eh

Also £18 a month is even more eh

[drifting pendant]

Yeah I know but the pricing is ridiculous on the last gen NVIDIA cards

[wet flame]

you can get 2x vram on amd and 2x speeds on cpus for way cheaper lol

[brittle lantern]

Yep but I don't think it's worth personally just cuz better ai support and 24gb of vram for lack of gaming performance compared to AMD

[wet flame]

anyways new amd gpu n cpu coming out with full AI support

anyways some one tell me how to solve

[brittle lantern]

And hopefully the dons behind some of these will do optimisation

For it

If not fuck it

(Not buying nvidia just fuck it)

[wet flame]

nvidia is like adobe lol

giving out expensive goods to fanboys

and fanboys jumping with them blinded lol

[brittle lantern]

I swear gimp apart from AI stuff basically does the same shi

[drifting pendant]

Yeah that's absurd

[brittle lantern]

And RT if u don't fully care for insane RT performance go AMD but if u do go Nvidia

And I'm not talking about buying a 4060 or 3060 cuz atp ur getting shi RT anyways

(Mention FG and DLSS ima cry ur running 1080p on those cards 8/10 times)

[wet flame]

does comfy really need cuda enabled?

i am running a111 for something webforge for most stuff and foocus for characters

[brittle lantern]

Iirc ROCm works fine Zluda if it doesn't wanna absolutely kill itself works but I think Cuda works best

[wet flame]

i can't start comfy

[brittle lantern]

It's a funny name comfy UI

[wet flame]

says error

[brittle lantern]

It's the polar opposite

[drifting pendant]

Yeah I'm in IT, I know hardware stuff 😋
And I totally agree with you !

[wet flame]

torch not compiled with cuda

can i not disable that check?

[brittle lantern]

So am I it's my job

[drifting pendant]

You're on windows ?

[wet flame]

yeah

[brittle lantern]

U can but then it runs on CPU

[open fox]

Comfyiu does also work with the apple silicon macs so i am pretty sure it work without cuda

[wet flame]

but if i enable cuda what will happen to my rest of the AI softwares like fooces forge and A111

they are running all good just not stupid comfy

i hate comfy but krita needs comfy to use AI asistance

[brittle lantern]

Yh but those a chips laptop specific complete polar opposite from laptops on windows and desktop

I would know I run it on macbook pro m2 pro and windows

[drifting pendant]

Try this one from @ornate elk #🤝｜tech-support message

[open fox]

I just wanted to point out that comfyui does not need cuda to work.

[brittle lantern]

I mean I did mention it

[drifting pendant]

He is a boss

[brittle lantern]

Yh I'm not gonna get 3090 for £600 GBP

I play more than AI I think most do

But I prefer having the gaming performance for £540

and AV1 encoder

Tho my problem on macbook is I have 16gb of ram and not 32gb so as soon I do SDXL with high res it's GG

[wet flame]

can you use comfy without cuda?

i got an nvida 3050

[brittle lantern]

Yes u can it seems more of a error

[brittle lantern]

3050 hav 8gb of vram or u got the scam 6gb one

[wet flame]

well i am running A111 right now

scam 4 gigis one LOL

on asus rog strix laptop

[solemn atlas]

what can i do 😅

[brittle lantern]

Ah that's a GG

4gb one is fucked as it is

[wet flame]

it did not even come with numpad lol

gotta make money wtih what i got

[brittle lantern]

[wet flame]

i mainly use it for making game assets and game dev

AI is amazing helper for it

[brittle lantern]

60% on a laptop 101

[brittle lantern]

It is

[wet flame]

i don't have to deal attitudes of folks much anymore

and those unwanted characteristics and over charging for a doodle

man before artists be ripping off folks

[brittle lantern]

Indeed it is

[wet flame]

1 stickman 100 usd OMG

[brittle lantern]

Honestly sometimes the prices are justified if they are really good

[wet flame]

if they are yeah but not stickmens

[brittle lantern]

I've paid some people for them and it's real nice

[brittle lantern]

Oh yh not that

[wet flame]

ye i seen lot of rip offs

[brittle lantern]

Alr cya I'm off

[wet flame]

screw this ima do krita thing later

i am making some icons based on my doodles 400 on batch xD

[brittle lantern]

Nvm I'm back

[drifting pendant]

That's fire

[brittle lantern]

I would know what's the best option Auto1111 on macbook or draw things app

[wet flame]

maybe we should move to general chat before we get hammered by mods

[brittle lantern]

Dw I've alr hammered mods before

But it is also somewhat tech support

[wet flame]

yeah it says to uninstall pytorch to use comfy ha

i ain;t doing that right now

[brittle lantern]

To much effort

[wet flame]

who made this comfy crap

who in the world likes comfy and why i have to clue

i hate it

[brittle lantern]

Idk why I looked at ur pfp and I keep seeing Charles Leclerc

[wet flame]

whose that

i hate node based crap

[brittle lantern]

F1 driver

[wet flame]

comfy is for unemployed rich kids

who has time unlimited time to figure 10101010101 nodes

[brittle lantern]

Bros raging cuz it ain't working knowing well that u don't mean that 💀💀💀

[wet flame]

nah i mean it

foocus is right now the best tool in my opinion for SD

[brittle lantern]

Ah so u can't actually fully prompt or do settings properly (and that's not me saying focus is bad)

Icl focus on macbook is tragic

[wet flame]

nah i don't like sphagghetis

example unity mechanim hate it

or unreal blueprint hate that too

[brittle lantern]

"Transfering to GPU" takes another 23 hrs

[brittle lantern]

Ah so just use Auto 1111

[wet flame]

yeah using that

[brittle lantern]

That's not node based

[ornate elk]

What's your GPU?

[wet flame]

i use all 3 😄

besides comfy

[brittle lantern]

Also imo node based things are my fav

[wet flame]

but krita why you need it lol

[solemn atlas]

can someone help please

[brittle lantern]

Answer @ornate elk question

[wet flame]

oh i got a darn 4 gigs 3050 gpu sad face

[ornate elk]

What are you trying to do?

[wet flame]

got ripped off when buying the laptop

[solemn atlas]

I wanted to generate something

but then this came

[brittle lantern]

Should've done some research

[solemn atlas]

i tried to add here something but it didnt worked

[wet flame]

well iget nice results like this

using foocus

[ornate elk]

Then I would use Auto1111, as you can better tweak it to not run out of vram

[brittle lantern]

Cuz no half then putting command argument after doesn't work that well

[brittle lantern]

I would actually ask u what's the best optimisation for auto1111 on silicon mac

[ornate elk]

What are you trying to do ? Don't add --no-half but add --xformers --no-half-vae

If your on nvidia

[ornate elk]

In the pinned messages there is an script for MAC that sets the best performance settings

Checkout the github of it for more information. There are some tweaks depending on your RAM

[solemn atlas]

it was like that before and then it popped up with what i send before

[brittle lantern]

M2 pro 16gb ram I should use this "export COMMANDLINE_ARGS="--skip-torch-cuda-test --opt-sub-quad-attention --upcast-sampling --no-half-vae --medvram-sdxl --use-cpu interrogate"

[ornate elk]

Oh wait silicon mac. Thats rip. Py torch removed the support of it.
I think the script does the settings needed for it to work too

[brittle lantern]

So can I download that version of auto1111 in ur pin?

[ornate elk]

If you can't explain or screenshot your txt2img settings idk how I can help xD

The error is not enough information

[solemn atlas]

yes I'm on Nvidia

[wet flame]

why light theme its so bright cmon join the dark gang

[brittle lantern]

So what does these commands do what do they mean?

[solemn atlas]

sorry here it is

[brittle lantern]

Let people use what they like

I personally don't mind light on some applications

[wet flame]

unity is to make you pay to use dark theme lol

[brittle lantern]

I actually have my phone set to logjt mode from 11 am to 9 pm

[wet flame]

lot of people actually bought it for just to have dark theme

[brittle lantern]

From 9 pm to 11 am it's dark theme

[brittle lantern]

Honestly waste of money imo

[brittle lantern]

^ I feel like I'm kinda being annoying but no one will see it

[ornate elk]

It makes it run on your CPU

[brittle lantern]

M2 pro doesn't actually have a dedicated gpu does it?

[ornate elk]

Yep. It always runs on the M2 chip

Thats why it needs these commands. Macs don't have cuda

[brittle lantern]

How fast is m2 pro for AI just as is not compared to anything

[ornate elk]

Idk

[naive shadow]

Link?

[brittle lantern]

Hmm I also wonder if like those dedicated apps for SD like draw things for mac is better than auto1111 mac install

[brittle lantern]

?

@ornate elk what's ur fav model to use for realism?

Ngl also feel the 16gb ram is gonna just drop me out of vram error once SD3 comes out lmao

[willow kettle]

Why do you recommend medvram-sdxl specifically here?

[drifting pendant]

How you do that !

[wet flame]

check out the _cheyenne model
https://civitai.com/models/198051/cheyenne

[drifting pendant]

I never had a result like that on my 6800XT I think I am, verry dumb, like, verry dumb

[drifting pendant]

Thanks mate

[wet flame]

i use nvidia

[brittle lantern]

Does hyper sdxl take more or less vram?

[wet flame]

3050 4 gigs the lowest requirement lol

i need to buy new gpu on desktop i got no cash haha

[drifting pendant]

Holy crap !

[brittle lantern]

I'm surprised it's even 4gb I expected 6gb being lowest requirements

[wet flame]

well its working making things with AI assited artwork and been selling too so its fine

https://tenor.com/view/april-fools-joke-dog-its-fine-this-is-not-gif-1750056094467610487

[brittle lantern]

Ah where u selling it on?

[wet flame]

unity itch things like that

[brittle lantern]

I see I'm more of poster selling fan

[wet flame]

yeah i am embracing Ai in my workflow

i can do it but why not make it faster with AI

its like the new photobashing

[brittle lantern]

U may have a issue with some copyright and other stuff along u mention its made from ai otherwise u will end up having a problem

[wet flame]

artist awlays keep all the reference and other artist art to bash from anyways

nah its based on my doodles 😄

[brittle lantern]

I mean the model u use more than what ur basing it off

[brittle lantern]

My reaction to that

[wet flame]

its true from ancient times lol

you can deny all you want it is what it is xD

[brittle lantern]

Every new gen rn is looking at this like

[wet flame]

the copywright folks stole from someone at some point and made it copy wrighted so no one else can steal xD

oh idk new gens m old lol

idk even what they categorize or whatever

[brittle lantern]

I'm 19

[wet flame]

enjoy 😉

[brittle lantern]

"Max Verstappen is the F1 grand prix winner"

[wet flame]

alright gnight have fun. time for some sleep

[brittle lantern]

Gn my fren

[ornate elk]

Depends on the GPU and vram

The first tweaks are already done by following my guide

Then you you can improve performance by using smaller models, faster samplers, extensions like tiled vae

[ornate elk]

JuggernautXL V9 and realistic vision xl are great for that

For 1.5 you can try absolute reality

Or realistic vision

Or juggernaut reborn

[ornate elk]

SD works with 2gb vram too

But its tricky xD and not recommended

[brittle lantern]

If I was doing high res on my macbook I should just do SD1.5 as with sdxl models high res dies cuz 16gb of ram

[brittle lantern]

How about Juggernaut X the new one?

[ornate elk]

Why using fooocus with an 4gb vram GPU?
You have to wait ages for images I guess.
Because fooocus only uses sdxl models.

[ornate elk]

Its worse than v9 :/

[brittle lantern]

Yh idk why focus on macbook takes like 7 years to transfer to gpu lmao

[ornate elk]

Yep 1.5 2gb models will work good

[brittle lantern]

2gb as in storage size

Or vram usage?

"Vram" 💀

[brittle lantern]

Anime models better on SDXL or SD 1.5

[wet flame]

sd 1.5 is really good btw

[brittle lantern]

Ik

[ornate elk]

Versatile art styles are better with 1.5 models. But the quality of sdxl is better

[wet flame]

sometimes no need of upgrade is needed when you don't need

i use different UI like foocus or etc for different use cases

[brittle lantern]

Quality as in Res or like the actual yk image in terms of eyes and stuff

[wet flame]

example textures in A111

forge ui for lot of sd 1.5 stuff

[ornate elk]

The quality comes from the resolution. You can have good quality in 1.5 when using upacale

[wet flame]

foocus for getting those refence pictures to make art from

like make ur 3D models based on them

[brittle lantern]

Oh yh and the pin auto 1111 for mac u mentioned about pytorch not supporting silicon but can i still download that and use the commands? @ornate elk

[wet flame]

and comfy at last the worst one to use krita AI help build in

like making icons or texture bases frrom my doodles

then clean up iterate etc and repeat

[brittle lantern]

And I can use bad hands lora for anime on 1.5

How about realism images does bad hands work on that I forgot if it's anime only

[wet flame]

wth krita choose comfy idk they should have choosen foge ui instead

[brittle lantern]

Krita is?

[wet flame]

yeah they use comfy

[brittle lantern]

I mean what is it sorry I didn't specify

[wet flame]

its better than adobe firefly

[brittle lantern]

Icl firefly 2 was cool when it came out

[wet flame]

https://www.youtube.com/watch?v=Ly6USRwTHe0&feature=youtu.be

[brittle lantern]

I like it's podcast

[wet flame]

here that what it is

[brittle lantern]

Anything that I can run native like adobe podcast?

[ornate elk]

You can try yes

[wet flame]

besides using stupid comfy there is complete freedom

choose ur models make ur own do what u like

and put in on use with krita

[ornate elk]

Bad hands lora doesn't help much

[wet flame]

you make ur own image dataset and use that

[brittle lantern]

I see so Krita is like photoshop I can't watch videos rn I'm outside on data

[wet flame]

krita is best digital painting app

its a drawing app first editing later

PS is editing first drawing later

[brittle lantern]

How would I fix hands and not be difficult and if not how do I prompt it to not have the composition not include hands

[wet flame]

if you wanna draw digitally krita is best raster program

and free too

[brittle lantern]

Me that has apple pencil (is it available on ios?)

[wet flame]

idk man i don't like those fancy brands like apple nvdiai n such lol

i use things that are functional and on budget

[brittle lantern]

Fair enough I have s23 ultra, macbook pro and m1 ipad pro

[wet flame]

i just got a motorola nice battery

[ornate elk]

Upscaling help, also prompts like
"Hands behind back" or "hands behind head" can help xD

[wet flame]

ur fancy stuff w/o the power is garbage lol

[ornate elk]

Then you can try DrawThings app on them to compare with Auto1111

[wet flame]

battery is king xD

[brittle lantern]

I put that at the end or close to start or do I use "(example prompt:1.2)"

[wet flame]

btw upscaling means what? it will upscale my image output chosen?

[ornate elk]

Before the quality tags

[wet flame]

say i choose 1k res

[brittle lantern]

I'm surprised android doesn't have any good SD stuff

[wet flame]

how do i use upscale here to create a better image?

[brittle lantern]

2× upscale 512×512 to 1024×1024

[wet flame]

yeah i don't want 2k

but i just wanna make the image output quality better

say i got a tiny 128 image as input

[ornate elk]

In auto1111 its done with the hires fix or the SD upscale script

[brittle lantern]

What's does steps and noise do on upscaler

Ik the upscale itself what it does

Most do ofc

[wet flame]

anyone using sd3?

[brittle lantern]

Not out only the API

[wet flame]

is it any better some goodies for amd users?

btw i heard that sd folks getting bankrupt?

[drifting pendant]

I have a 6800XT 16Gb, but I found it slow, really 😭

[ornate elk]

Denois is "how much should the output differ from the input image"
Like how strong should the upscale effect go over the image
The steps are like the sampling steps, like how often should the upscale do something to the image

[wet flame]

set it up properly maybe?

i saw videos people doing nice with amd gpu particular ones

but they got like 16 gb+

[brittle lantern]

So in most cases use 0.2 to 0.35 and 1.5 to 2.2+ upscale and like 10 to 15 steps?

[ornate elk]

Can you try a 512x512, DPM++ 2m, 30steps, image without hiresfix and make sure you use an 1.5 image.
Then check the cmd for the it/s

[wet flame]

don't fret amd going ham on their next gpu and cpu

[ornate elk]

0.3-0.5 works best and yes 10-15 steps.
I mostly do 0.4-05 and 10 steps

[wet flame]

i am sure they gonna make nvidia go crazy again

[brittle lantern]

They ain't making high end next gen

[wet flame]

like they did few years back with their cpus

[brittle lantern]

Focusing on mid range

[wet flame]

nvidia was crying lol

[brittle lantern]

They said they want GPUs to be under 1k

[glad kraken]

Sorry for talk about another topic,I can not use . I type “/“ and tap “dream” and type prompt but it tells me “You do not have authority”.I hear everyone can use few times with no credit . Do I have to pay at first time?

[wet flame]

sd is free though

where u paying things for cloud service?

[ornate elk]

I think you need to link an account first

[wet flame]

if ur using online make an account

[brittle lantern]

Well depends people that don't have pc will pay for MJ or don't have requirements for it also really ez to prompt on it and get good results

I have a sub to MJ and I use SD also mainly when I can't have access to pc or laptop

[wet flame]

i used MJ for one month

its cool but SD is cooler

[brittle lantern]

Amd draw things is not the fastest on m1 ipad

[brittle lantern]

I will say it's very good quality wise and hands and eyes are pretty fucking good

Like all AI it does now and then mess up like Niji beta

[wet flame]

but u can't get all kind of characters 😉

[brittle lantern]

That wink tells me stuff xD

[wet flame]

in games u need all types good ones and bad ones

[drifting pendant]

I search for this, but don't what to do, because I'm new in this

[brittle lantern]

I've made some non family fren stuff with SD 💀

[drifting pendant]

Alright, thanks !

[wet flame]

its working better?

congrats man

[drifting pendant]

Ill tell you when I'm back home

[brittle lantern]

Jony what's ur fav character u made in SD

[ornate elk]

Sure, show me the it/s then I can say if its normal or how you can improve

[brittle lantern]

@ornate elk forge improves times a lot on SD 1.5 or SDXL more? Ik it does both but which one does it affect more

[wet flame]

yes it does tested it

works much faster but no good for seamless btw

[brittle lantern]

Seamless as in?

[wet flame]

so if u want parallax bgs its not for it

means it don;t connect

[brittle lantern]

I just make anime art and as close to realism I can get

[wet flame]

ok no worreis then

yes it is faster and have more tools

[brittle lantern]

I don't do insanely advanced stuff maybe I'll learn and control net soon and just use loras and high res that's it

[ornate elk]

If you have low vram it helps alot with both

[wet flame]

it works very nice on my 4 gigs

[ornate elk]

What are you using rn as GPU?

[wet flame]

my goto AI right now

[brittle lantern]

Ik I got error after it using 11gb of ram on SD saying I ran out of vram on macbook

[wet flame]

unless m making seamless stuff

[brittle lantern]

6600 xt but yk last time zluda does not wanna work even after tut

U helped with that

Remember

[ornate elk]

Yea I know

But others got it to work :/

So idk what's up with your setup

[brittle lantern]

And zluda just said jit gtfo

To me

[brittle lantern]

I'm convinced it's a windows install but I can't say cuz I did a fac reset recently from usb stick

So I'm forced to use macbook and I'm new to macos so that's why I'm asking a lot of SD questions @ornate elk sorry if it's annoying

[ornate elk]

Oh you did a reset ? Have you tried to get it working again?

We can also take a look at it again

[brittle lantern]

That's when I did it after the reset

Same day hour after

[glad kraken]

I see! Thanks!

[ornate elk]

Okay, wanna try again? As there were a few zluda Install fixes done recently

[brittle lantern]

Fine once I get home

Last time tho man that last time like made me triggerd for 2 hrs after

But that's a me problem

When my autism can't fix something it bothers me for a while

[ornate elk]

Yea its annoying if something doesn't work even when it looks like everything is done correctly

[brittle lantern]

Like the first time in 3 yrs I couldn't figure sum issue out

In the sense where everything looks exactly how it should be

At one point I stuck my head in my monitor to see if my eyes had a issue like I'm some esports player 🤣

Cuz normally I need glasses so maybe it was just cuz repeating structure like the files in file Explorer

It's on a row with white background

[ornate elk]

Yea happens when starring at the the bright monitor for a while

[brittle lantern]

It was like 6 am

I actually don't know the time all I knew my room was dark

So I'm guessing

[brittle lantern]

I'm home now I'm just gonna take 5 to 10 min sit down and then u can help me yes?

[river dawn]

How do I create images private and not in open chats?

[brittle lantern]

@ornate elk after ur tut this should be fine?

For mac

The time

Idk why it's blurred it's not like that on gallery

[ornate elk]

Looks okay yes

[brittle lantern]

So this is modified Auto1111 or is this some form of the forge modify?

[ornate elk]

For Mac its normal Auto1111

[brittle lantern]

Ok so this would be better than forge? (idk if u can get forge for mac)

[ornate elk]

The script let's you define if forge or Auto1111 will be installed, on the github its documented

But I would use Auto1111

[brittle lantern]

Yh I just have normal cuz I don't see the whole bunch of addons forge has got

I really wish I bought 32gb ram but bro the price for this was alr expensive

[cosmic furnace]

@ornate elk not sure but it doesnt say o.0

same thing on sd next -_-

[ornate elk]

what yre you trying to do?

[cosmic furnace]

error on next feels similar 🤔

using all kinds of pony with loras that shud work with them

[ornate elk]

and where is the error?

or that does happen?

[cosmic furnace]

nothing specific ?

vae not working in sd again for some reason

in next*

[ornate elk]

oh in next

[cosmic furnace]

it feels same error for both... but next at least gives some error log ,kinda

[ornate elk]

do you use the latest zluda files?

also you should update your python to 3.10.11

[cosmic furnace]

29.04 is latest y ?

[solemn moon]

works with...

[ornate elk]

oh i see you installed it in a wierd folder path

[solemn moon]

oh damn NVIDIA dropped an update for drivers

[cosmic furnace]

yup thats it

and its both zluda folder and in automatic now

how do i fix this in next ?

changing vae doesnt seem any effect

[ornate elk]

is it with pony?

[cosmic furnace]

y

[ornate elk]

try without negative tags

[cosmic furnace]

o.o

[weak fiber]

[cosmic furnace]

nah

[ornate elk]

does it work in auto1111 ?

[cosmic furnace]

yes +)

switching alot vae made result yesterday

but not happening now for some reason

not sure if clicking vae in networks do something

[warm otter]

stability.ai API does not have an inpainting engine for SDXL 1.0? Inpainting does not work for me - it's just inpainting in something extremly different (exactly where the mask is)

[ornate elk]

make sure to lower the denois if possible

[warm otter]

There is no such thing as denoise. You provide init image, mask image, text prompt, cfg, sampler (tried different) :/
these are the results - this is horrible...

[ornate elk]

hmm i see :/

[warm otter]

that means paying now from 0,3credits = 10 times, because only new api seems to work. Thanks for this Price increase! 🙂

guess that's a way to upsell ^^

[near socket]

Um, weird, something I said got flagged here. First time user, can someone perhaps help me with a question or two?

[ornate elk]

sure, whats the question?

[near socket]

Nvm got it solved 😄

You are the good luck charm clearly :3

[thorn plinth]

Im looking for some help getting controlnet to work properly. It shows up in the webui my models are there. I am able to hit preview and see the preview image. I have made sure I have it enabled but it does not seem to have any affect on the generated images

I've been googling all over and nothing works

[ornate elk]

hey, can you show your controlnet settings?

[thorn plinth]

I noticed my settings page looks different than what other ppl have posted

basically there isnt anything there

[ornate elk]

can you show how does it looks in txt2img?

[ornate elk]

okay and what does the cmd shows?

[thorn plinth]

is that readable?

[ornate elk]

yep

controlnet model doesnt work

you need an other one

Make sure you get them from here:
https://civitai.com/models/38784?modelVersionId=44811

[thorn plinth]

I think those are the ones I just downloaded

[ornate elk]

with the config file?

[thorn plinth]

ya

[ornate elk]

and you put them into models/controlnet?

[thorn plinth]

currently they are in \stable-diffusion-webui\extensions\sd-webui-controlnet\models I can move them and test if you would like

[ornate elk]

yea move them, also make sure your webui is updated. check it at the bottom of the browser

[thorn plinth]

its says 1.5?

[ornate elk]

can you screenshot it?

[ornate elk]

ohhh

latest version is 1.9.3 xD

your over a year outdated

also wrong python version

[thorn plinth]

ah lol. I bet that has something to do with it. Do I have to clean install or is there an easy way to update

sounds like I need to fully reinstall and come back

[ornate elk]

best is to uninstall the 3.11 python and then start clean by following my install guide from the pinned messages of this channel.

if your on nvidia here it is: #🤝｜tech-support message

you can move the models over after getting it to run

[thorn plinth]

alright I will give it a try. Thanks a lot for the help

[ornate elk]

no problem 🙂 feel free to ask again if something isnt clear

[cosmic furnace]

tried another promt-lora & everything works (next) 😵‍💫

[fair oxide]

is there any way to download or export all my generated images from DreamStudio?

[cosmic furnace]

and now next generating alot with lora and no errors =\

[stable kindle]

getting this error when trying to inpaint with controlnet openpose: RuntimeError: Sizes of tensors must match except in dimension 2. Expected size 1008 but got size 1032 for tensor number 1 in the list.

[cosmic furnace]

Yet again pony with vae in it and not working =\

and its not switching to any models that are inlist from alternative path . although theyr in list

[solemn atlas]

hello, can someone help me please

[ornate elk]

in img2img or txt2img?

[ornate elk]

when do you get that error?

also txt2img or img2img?

[stable kindle]

img to image, i fixed it, seems like it was a resolution issue and it wanted a width divisible by 64

now I'm stuck trying to get good hands using openpose, I've followed some tutorials but no matter what I try it generates dismembered blobs instead of hands even with a perfect "preprocessor" skeleton

so if anyone has alot of experience with openpose or controlnet lmk

[cosmic furnace]

1 generation with loras and it errors me out =\

maybe i shud try no medvram or dunno

removed medvram -sdxl

and errors stopped ...

[thorn plinth]

trying to install the proper version of python but I cant get past this error. Anyone have experience with this

[ornate elk]

try removing everything of python and then install it again as administrator

[thorn plinth]

I've tried that as well as removing appdata folders and registry entries. Doesn't seem to fix it

[ornate elk]

and you downloaded the 3.10.11 64 bit installer ?

[thorn plinth]

ya

[ornate elk]

okay

try to install 3.10.6 again and then uninstall it again

seems like something didnt got removed

[thorn plinth]

its saying it cant uninstall the core interpreter if that means anything to you

[near socket]

I'm very new to all this as said above. I got the dream function working but it only produces images. How do I get video?

[ornate elk]

do you have an third party anitvirus programm? if so turn that off when install python,
then install python again

[drifting pendant]

When I push le sliders I god this error too, IDK why

[drifting pendant]

Alright I'm @ 3.01 it/s :p

With some Undervolt / Overclock

[drifting pendant]

I got the error : "Connexion errored out."

[cosmic furnace]

@ornate elk Isnt it werid that medvram -sdxl . Gives me errors ? when witha loras and kinda fast 2 to 3 generations & boom

And it doesnt seem like it affect speed 🤨 . Very slightly and probably not at all 🤔

[ornate elk]

you can try remove --medvram.sdxl as you have 24gb

[thorn plinth]

I finally gave up and decided to test it with with python 3.11 since non of the 3.10's would install. It appeared to be working properly now though. We will see if I run into compatibility issues at some point. Control nets working though so I am happy lol. Thanks a ton for your assistance

[cosmic furnace]

yeah i removed and it works fine. But isnt it weird? o.0 that it doesnt wanna work flawlessly with it -_-. Pretty sure ive had some issues when medvram on usual directml non zluda

[ornate elk]

Alright, good that it works now

[drifting pendant]

@ornate elk Gotta new wierd problem, sometimes I loose access to my folders, in the drive dedicated to SD, like I launch Helldivers 2, cannot access the folder, I have to reboot my rig to regain access, same, sometimes while I try to generate heavy Images like 1024x1024 150, w/o the hi res filter, I cannot explain this my drives are fully tested and clean

PS : It's a 512 NVME SSD, dedicated to all the AI stuff, especially SD

[ornate elk]

Is it an external drive?

[drifting pendant]

Nope, Internal NVME, Samsung 970 pro

[ornate elk]

Did you checked it with Crystaldiskinfo ?
How does the error looks like?

[drifting pendant]

Yep and no error, SMART is OK too

[ornate elk]

Hmm

[drifting pendant]

https://tenor.com/view/i-aint-do-nothin-j-alphonse-nicholson-john-robinson-self-made-inspired-by-the-life-of-madam-cj-walker-didnt-do-nothing-gif-16892174

My SSD be like

I've just disactivated indexing

[ornate elk]

Windows 10 or 11 ?

[drifting pendant]

10 22H2

Up to date too

[ornate elk]

Open up a cmd as admin and run
SFC /scannow

That will fix corrupt system files

[drifting pendant]

I'm doin' it but, I don't trust it x)

Since it's brand new drive I've Windows on x)

[ornate elk]

Okay, how many ssds do you have?

[cosmic furnace]

do i need to reboot to animate diff to work? o.0

[drifting pendant]

2, One Crucial 2Tb and the 512 Gb Samsung for the AI stuff

[cosmic furnace]

reloaded ui and seems do nothing

[ornate elk]

And your OS is one the 512? With stable-diffusion ?

[drifting pendant]

Nope

On the decided drive

Is that the problem ?

Duh hell ?

D:\StableDiffusuion\SD-Zluda\stable-diffusion-webui-directml
In that specific order

[ornate elk]

nope not a problem

[drifting pendant]

My windows install is on the 2Tb sorry

But the only setting that change today is that in a discord voice chat RN

OK, does the SDXL turbo models cose any troubles ?

Never had one before this 🙃

[ornate elk]

How much RAM do you have?

[drifting pendant]

32Gb

[ornate elk]

Okay that's enough

[drifting pendant]

That's a weird one *problem

[cosmic furnace]

installed animediff extension +motion module - & it shud work y ?

[ornate elk]

it should work with auto1111, i use it

[cosmic furnace]

the simpliest way-simpliest giff 🤔

but we got directml ?

or whuht

[ornate elk]

do you get an windows error saying it cant aceess the file path?

or how do you get the error

[ornate elk]

?

[cosmic furnace]

will it work with the one zluda by your guide?)

[ornate elk]

yes

i use it with zluda

[cosmic furnace]

erm... i click and nothing happens... well usual img generation 😒

[drifting pendant]

After trying generating a image

[ornate elk]

Check the cmd

[drifting pendant]

Not a specific path, but all the folders containing files, like I can reach the folder, but the any folder containing a file is not "openable" (I don't have the word for that, sorry)

[ornate elk]

Can you screenshot the error?

[cosmic furnace]

https://tenor.com/view/stellarnushu-gif-10511910975370734967

but thats a lie? erm ....

[drifting pendant]

Can a Lora do this ?

[drifting pendant]

I'm trying to reproduce it, maybe it's a model error

I can reproduce it RN with this one https://civitai.com/models/43977/leosams-helloworld-xl

[cosmic furnace]

ill try 15 motion module then dunno

[drifting pendant]

1024x1024 / DPM++ M2 / 150 Sampling steps

The message

I've not screened in time, brb for a retry, need to reboot my rig... Again...

That was fast :

[cosmic furnace]

Same error with 15 =\

[ornate elk]

What error do you get?

[cosmic furnace]

no motion module detected 😦

downloading theese 2 modules now >_>

[drifting pendant]

The "error message" x)

I suspect it's I'm in a too high setting like the 150 Sampling steps

It can happen in the middle of the render and at the end, this time it was at the end

[ornate elk]

Don't use 150 steps, its not needed

Also whats your GPU?

[drifting pendant]

An AMD 6800XT

I've got the best results with this setting so far

[cosmic furnace]

all motion modules same error o.0

probab this one left but im not sure thats even module -_-

[ornate elk]

150 steps dont improve quality

it just changes stuff 150 times

[drifting pendant]

Okok

I cannot find any solutions, want to give up RN

[cosmic furnace]

it works with sdxl or not? o.0

[ornate elk]

Only if you tweak it with my guide. But I have no time RN to explain how

[cosmic furnace]

🫠

[drifting pendant]

@ornate elk I'm starting to loose hope with this rig, for real, I've tried Linux and a lot a of things, never worked... maybe 50 times on Windows with your install and done... that's it... I'm just gonna grind gym till the and of times...

[cosmic furnace]

dont tell me i get this error cuz pony model -__-

[ornate elk]

So does the error only comes when using sdxl models.

?

Make sure your amd driver is updated

[drifting pendant]

When tring to genrerate anything, 1.5, SXDL, Turbo

[cosmic furnace]

guide pinned? mb i shud try myself for now>? or too complicated ? 😒

[ornate elk]

Nope

[drifting pendant]

I've erased all my models, I'm re downloading them RN

[ornate elk]

Here is the guide to use the sdxl animatediff motionmodule:
the problem is that the webui blocks .ckpt files as security measure.
To fix that:
i copied the motionmodule into the models/stable-diffusion folder.
then i editet the webui-user.bat and added --disable-safe-unpickle
then i went into the Checkpoint merger tab. there i loaded the motionmodule as A.
then selectet "No Interpolation" and on copy config from set it to "None"
then selectet checkpoint format Safetensor. Then press Merge.

After that you copy the new motion module (that is now a .safetensor file) from models/stable-diffusion back to the folder from AnimateDiff
Importan. Close the webui and remove the --disable-safe-unpickle from the webui-user.bat again.
then save and relaunch.

[cosmic furnace]

seems harder than i thought itd be 😦

i guess most animated giffs out there are 1.5 then 🤔

[drifting pendant]

Already checked, up to date

Maybe it is a VRAM problem, tried to render the same image as someone and monitoring my vram, I've rached 16080 Mb (maybe more), so maybe the soft try to allocate more than possible and just crash, without any message or anything ?

[wind thunder]

hello i installed Mov2mov from a url but i dont see the extension even after the restart,check,apply

[ornate elk]

Yes make sure you don't run any other programms nor Wallpaper engine in the background, also activate tiled vae

[ornate elk]

here is the fix for mov2mov:
https://github.com/Scholar01/sd-webui-mov2mov/issues/146#issuecomment-2054157728

[ornate elk]

yes

[wind thunder]

yes thanks it worked

but i have another problem why when i load a video it doesn't show up like when i drag a video in the mov2mov tab

[wind thunder]

@ornate elk do you maybe know why ?

[ornate elk]

nope

please contact the support directly with their support mail adress

[wind thunder]

okay it work but is it normal that it take so much time like many hours

for a 9s video

is my graphic card this much old

lol

[ornate elk]

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

[vivid flume]

My DirectML is suddenly not registering all my Vram, it was reading all 20gig until a minute ago, and now it only registers 11gigs

[wind thunder]

i forgot to uncheck the movie editor my bad
but still it takes 30 min for a 9s video

[ornate elk]

oh okay,
you need to make sure that you have --xformers --medvram --no-half-vae
in your webui-user.bat at the line COMMANDLINE_ARGS=
to get the best performance

[ornate elk]

oh you are still on directml with a 7900xt ?:O

[vivid flume]

Yeeees.

It was generating great a few minutes ago

[ornate elk]

you should take a look at my AMD Zluda guide, it has much better performance and uses less vram. its the best for AMD gpus right now

[vivid flume]

Link?

[ornate elk]

Here:
#🤝｜tech-support message

im on a 7900XTX and i can easily use sdxl + hires fix without getting out of vram

[vivid flume]

im assuming i should drop the --medvram arg?

[ornate elk]

not on directml :/ directml is bad optimised in vram usage

[vivid flume]

i mean for zluda

[ornate elk]

yes

zluda doesnt need it

[ornate elk]

if you have any questions about the guide, feel free to ask, but as you have a 7000 series gpu you can skip step 4 and 4.5 of the zluda steps, so its a pretty easy guide for you

[vivid flume]

seems much easier than fighting with DirectML

should i get rid of my directml file?

[wind thunder]

it still take a lot of time its normal ?

i did the thing

[ornate elk]

You can, after you moved your models and stuff over.
Also the first image on zluda will take 15-40 minutes.

[ornate elk]

Okay and how long is the video ?

And what resolution

[wind thunder]

9sec

576x1024

[vivid flume]

is this right? i dont wanna mess my stuff up

[ornate elk]

There is a little ` before the C:

Remove that, then its fine

[vivid flume]

omg i almost died

do i need to retart my pc or anything?

[ornate elk]

That resolution is to high then

Thats why it takes so long

[ornate elk]

Nope only if it won't work

@wind thunder also 30fps takes much longer

[vivid flume]

i forgor to save the args 💀

it launched fine tho, i just closed the ui and relaunced

[wind thunder]

ah okay

but this button is shit

how much should i put like 20 fps is okay

[ornate elk]

24 is okay

[vivid flume]

💀

[ornate elk]

Okay then restart the PC.
Make sure you don't have Wallpaper engine running. And then relaunch the webui-user.bat and generate again

[vivid flume]

omg, wallpaper engine is killing my SD lmao

i love my wallpaper

[ornate elk]

It can even crash the PC when used together with SD xD

[vivid flume]

I'm surprised that hasn't happened yet then lmao

[ornate elk]

Yea its random, also it slows down the generation time

On directml more then on zluda

[vivid flume]

I did have some intense lag with my generations on directml

[ornate elk]

Yea that's WE

On directml, SD uses 99% of your GPU vram. And then Wallpaper engine comes in and say "fine, give me the rest" = short freezes or crash

[vivid flume]

https://tenor.com/view/aaaaaugh-gif-26144251

[ornate elk]

Is your driver updated?

[vivid flume]

yuppers

[ornate elk]

Also please show me the full error

The whole cmd log

[vivid flume]

ignore the bottom 2 lines lmao

[ornate elk]

Ohh xD you have an AMD CPU too?

[vivid flume]

yup, taking advantage of that amd expo

[ornate elk]

The webui tried to use the igpu of the cpu instead of your 7900xt

[vivid flume]

💀

[ornate elk]

To fix that you have to disable the igpu in the device manager

It's called Radeon TM graphics

[vivid flume]

ez pz

[ornate elk]

Its Under display adapters

[ornate elk]

Same ^^

[vivid flume]

connection timed out raaaaah

[vivid flume]

6000MHz ram is noice

[ornate elk]

Relaunch it

Then press Generate and look at the cmd.

[gloomy bolt]

I currently have SD1.5, but I want to try out SDXL too, especially pony diffusion. I'm currently using the auto1111 UI for sd1.5, can anyone guide me to install sdxl? Also will it be using the same UI still?

[ornate elk]

Hey, you can just download sdxl and pony models and put them into the models/stable-diffusion folder and then select them in Auto1111

[gloomy bolt]

i thought SDXL was a whole different program

so it's just another checkpoint?

[ornate elk]

Yep its trained on a 1024x1024 base resolution

@vivid flume delete the venv folder and relaunch

Then use and 1.5 model for the first run

[vivid flume]

Ok, restarting desktop as well

Aaaah, I was trying to use my anime checkpoint

[gloomy bolt]

I see, does that also mean it's best used with 512x512 with hires x2 or does it have to be 1024x1024 pre hires

[ornate elk]

No you need to use 1024x1024 native

Pre hires

[gloomy bolt]

oof that requires a lot of vram right

[ornate elk]

I need to go off now its very late. But your not far away from getting it to run.

[ornate elk]

Yep. What's your GPU?

[gloomy bolt]

rtx 4070, 12gb vram

[vivid flume]

oke, i deleted the venv files, threw up some errors when i ran the bat so im rerunning it now

just download errors for torch tho

[ornate elk]

Okay, then make sure you have --xformers in your webui-user.bat

For better performance

[ornate elk]

Oh but thats not good.

[vivid flume]

oh yikes

[gloomy bolt]

mk I should have that already

everything else is practically the same as using for SD1.5 right

[vivid flume]

its downloading everything now tho

[gloomy bolt]

just 1024x1024

[ornate elk]

Without errors?

[vivid flume]

yes, no yellow text to be seen

[ornate elk]

Yes, but sdxl models only work with sdxl Loras

Same goes for pony models but they work with sdxl Lora's too and also are trained on 1024x1024

[gloomy bolt]

mk sounds g

alr thxs, imma try it out later

[ornate elk]

No problem 🙂

[ornate elk]

If it fails again with the same error as before.
Then close the webui.
Open up a cmd and type
Pip cache purge
Then delete the venv folder again.
And relaunch the webui-user.bat

[vivid flume]

oked dokey

go to bed now

u dont get paid enough for this

[ornate elk]

True, I don't get paid xD

Its 4am 🥲

Gn

If it doesn't work we can figure it out tomorrow

[vivid flume]

[vivid flume]

LES GOOOOO

[viral knot]

waifu generated, mission was a success :3

[vivid flume]

depressed miku ftw

[viral knot]

im no Miku expert at all, just wondering, cause i saw some pics and her hair to me is more like light green or something, maybe teal?, yet in the prompt it says light blue, but idk :3

[vivid flume]

she occasionally has teal hair, but its generally a light blue/aqua color

[viral knot]

ah

[gloomy bolt]

So i just got started with sdxl stuff, and I installed some of the loras for sdxl but none of them show up in the lora tab for auto1111

[wet flame]

Pay CSO or donate to him eh

Dude open buy me kofi

He helps so much returning the favour makes sense

[vivid flume]

ah, didnt see he had a kofi, will definitely do

[wet flame]

he does not i think but i suggest him to open one 😄

[sudden hawk]

hello, i have this msg in my webui but I don't understand cause when I check : the file is a safetensor file
"AssertionError: C:\A1111\stable-diffusion-webui\models\Lora\ariesFp16NoEmaNoVae_v32.safetensors is not a safetensors file"
can someone help pls ?

[reef pilot]

I'm trying to use Stable Image API . Why is the image blurry when i using the negative_prompt parameters.

[hardy rapids]

what's a good way to upscale tile seamless patterns, cant use those regular upscalers because it will mess the edges and and they dont tile seamlessly anymore

[fierce hollow]

WTH??
RuntimeError: mat1 and mat2 shapes cannot be multiplied (154x2048 and 768x320)
HELP

[young wharf]

[urban quest]

I am on an M1 with 16GB and tried to use stabilityai/stable-diffusion-x4-upscaler with Python 3.10
``
from PIL import Image
from diffusers import StableDiffusionUpscalePipeline
import torch

image=Image.open("image.png")
model_id = "stabilityai/stable-diffusion-x4-upscaler"
pipeline = StableDiffusionUpscalePipeline.from_pretrained(
model_id, revision="fp16", torch_dtype=torch.float16
).to("mps")

pipeline('xxx', image).images[0].save("image_upscaled.png")
``
It starts up but when it then comes to the 75 iterations of the upscaler it simply stalls. Do I miss the obvious?

[gleaming pawn]

Hey people, I have a noob question : Does anyone has a solution to directly input a PNG image with transparency into inspainting? So the alpha area is targeted by the inpaint.

[ornate elk]

I have made one ^^ linked in my profile bio

[ornate elk]

Hey, your sure the aries is a lora? The name sounds like a model.
Check the file size. If its 2gb or larger its a model and has to be moved

[ornate elk]

Switch to an sdxl model then on the Lora's tab press the refresh button on the right

[sudden hawk]

yes, in my memories when i DL it, it's a model, and I put it in model>stable diffusion, not in loras, that's why I don't understand why I have this error message

[ornate elk]

and your sure its not in loras? recheck to be sure

[cosmic furnace]

Shudnt this model of animatediff work with sdxl without manipulations? mm_sdxl_v10_beta

[ornate elk]

Yes it works with sdxl

But it has to be a .safetensor file

[sturdy forge]

Bah. Still seeing this.
I guess it's either "dig or ticket" if I want to stop getting annoyed by it
Surprised seemingly nobody else has experienced/noted it, though

[ornate elk]

If you want to link all models from an other Webui you can do that by adding --ckpt-dir "Path/to/models/stable-diffusion" --lora-dir "Path/to/models/lora" --controlnet-dir "..." --embeddings-dir "..." to the webui-user.bat, at the line set COMMANDLINE_ARGS= in auto1111

if the models are in auto1111 its easier, then you have to rename the extra_model_paths.yaml.example to extra_model_paths.yaml, then edit it and set the path to auto1111

[sacred cedar]

greetings... ai visual production bots have been removed, anyone know?

[cosmic furnace]

Probably next is messing up because not clean install... What are the most important good cons of Next vs Directml"ish ? Not sure if i shud keep it

Well it has some features-extensions pre installed... Like animdiff -control net, interrogator. Hmgm... Some inbuilt styles? o.0

[ornate elk]

you can keep both and share the models

[gloomy bolt]

Is there anyway to speed up sdxl generations? It's takes around 3 min per gen, and I've seen some comments that removing xformers actually speeds it up?

I mean it is doing a million pixels per gen, but people are doing this faster somehow right

[ornate elk]

Hey what's your GPU?

And what's inside your webui-user.bat?

Xformers only speeds up the generation

[cosmic furnace]

Btw this is weird. Directml needs to have at least 1 model for this to work. Or some time to upload models. Not quite sure but sometimes models are in the list but when i click them nothing happens and i see that model chosen now is the one from start hmgm o.0

[warped canyon]

when training for a specific character, is it ok to use 2 of the same picture on the dataset but one is normal and the other is the zoomed in face (no body)?

[gloomy bolt]

4070 12gb ram

[ornate elk]

You can't use xformers AND opt-sdp-attention.
Only one of them

[ornate elk]

Xformers uses less vram

[gloomy bolt]

hm i see

should i remove --no half vae too

https://www.reddit.com/r/StableDiffusion/comments/14uayos/is_sdxl_supposed_to_be_this_slow/

this is sort of what some people are saying but idk if it applies to me or not

[ornate elk]

If your using anime models or sdxl VAE then keep it

[ornate elk]

Nope, you have enough vram

They would need --medvram-sdxl for 8gb cards

[ornate elk]

How much RAM do you have?

[gloomy bolt]

alr sounds g

[cosmic furnace]

32 g 🤔

[ornate elk]

thats good

[zinc arrow]

whats the fastest ui for inpainting

@ornate elk sry for the ping

UI-UX is really slow and i dont like the ui of forge/a1111

[ornate elk]

for inpainting auto1111 is the easiest
for large image inpainting or outpainting there is also Krita-ai diffusion which works with comfyui as backend

[zinc arrow]

I just want simple things like changing the clothing of a character or changing colors of an object

[ornate elk]

okay, yea still auto would be the easiest

just make sure you set the inpaint area to "only masked" and then set a square resolution

[zinc arrow]

Alr can u suggest any extensions for inpainting

[zinc arrow]

What do those do

[ornate elk]

these are needed for inpainting to work correctly

that limits the area that will change to only the mask area

[zinc arrow]

ah

[ornate elk]

faster generation and better result

[zinc arrow]

What is the difference between inpainting and sketching

[ornate elk]

in sketch you can use colors

[zinc arrow]

What difference does it make

[ornate elk]

but i didnt used it that much

you could paint the tshirt blue

and then it will easier switch the color i guess

[zinc arrow]

Ah ok

[ornate elk]

@zinc arrow there is also this extension:
https://github.com/Uminosachi/sd-webui-inpaint-anything

i didnt tested it but it could be helpful

and then there is also this standalone tool:
https://github.com/Sanster/IOPaint?utm_source=100xgrowth.ai&utm_medium=marketplace&utm_campaign=100xgrowth.ai

[zinc arrow]

Also do i need to download a certain inpainting model

[ornate elk]

normaly not, but it can help if you get wierd smudge around the inpaint area

[zinc arrow]

👍

[cosmic furnace]

installed mov2mov

and tab didnt appear anywhere

hm o.0

made an search and it seems its not working anymore 😒

[ornate elk]

you need to do a fix

to make it work

here is the fix:
https://github.com/Scholar01/sd-webui-mov2mov/issues/146#issuecomment-2054157728

[cosmic furnace]

Btw animatediff with video source gives an video driver error of closing(too long respond?) o.0

nvm it doesnt work anyway o.0 . while yesterday was okay -.-

[ornate elk]

make sure the resolution is not to high

also reduce the batch size can help

[mystic ruin]

my sd control net extension doesnt apply when i click apply and quit

[ornate elk]

your using forge, it already has controlnet build in

[mystic ruin]

oh

[sudden hawk]

https://civitai.com/models/191193/aries-fp16-no-ema-no-vae-safetensors-checkpoint
it's a checkpoint that's why I don't understand

[sudden hawk]

ok I just understand, i had put ariesfp16 in Stable fiffusion AND Loras 🥲 that's why, sorry for bother you

[plucky dock]

Hey all I was wondering if anyone else has been having issues with A1111 inpaint sketch recently? Every time I try to use it I keep getting this error. I have fully uninstalled and reinstalled everything and the problem still persists. Do I need to rollback to a previous version or something? I'm pretty clueless as to why this keeps happening.

[solemn atlas]

Hi, i still have this problem

[ornate elk]

Make sure your on the latest nvidia driver.
Also what image settings did you tried? Resolution, etc loras used?