🤝｜tech-support

[ornate elk]

Flux upscaling with zluda seems not possible for now xD

[cosmic furnace]

It cant be that forge & flux so bad optimised as of now that 2x less vram is used more properly? Ive read alot of ppl said hires fx is still not a thing and as u said tiled scaling is what shud use to net get vram errors

[ornate elk]

Swarm with zluda? I doubt since your comfyui didn't worked

[cosmic furnace]

Its just site says he uses txt2img and + hi res fix . But that info might be wrong? That guy is an perfectionist maniac and he run 10 times a detailer and alot of inpainting

[spare crow]

Was this guide I followed

[cosmic furnace]

Wat o.o . Dont tell me it works on zluda not as on nvidia? 😦

Pretty sure flux is so bad optimised that it wont work on nvidia also . =\

Zluda emulates not the latest cuda but pretty new enough version -_-

[lucid yoke]

upscaling flux works pretty well on my 3070TI (8gb vram) just takes a bit

[spare crow]

I'm not sure if I should be using different upscale methods or not but just wanted a quick easy method to see if it worked this time, do you normally use the 2x it says under the image?

[lucid yoke]

sometimes but i prefer to refine it

[cosmic furnace]

Really? like upscaling up to 1920-2560 & all that? o.0

[lucid yoke]

yeah lol

[ashen gate]

is it possible to samplers with Flux1D other than Euler-Simple? I can't seem to get it to work.

[cosmic furnace]

well dang... then zluda or either rocm needs updating... or indeed flux itself not optimized...

ppl recommend tiled upscaling with ultra sd upscaler for flux tho

and dang it took 12 minutes for an 960-1280 to become 1920-2560

which is outrageous

[ornate elk]

For nvidia its easier

[ornate elk]

Zluda still gets updates

[ornate elk]

Yes some might work but idk about comfyui samplers

[ashen gate]

im using Forge not Comfy

[ornate elk]

Oh then yes there are flux sampler at the bottom

And simple might work too

[spare crow]

is amd just not loved ? whats going on there? how come its so much more complicated for amd?

[cosmic furnace]

sadge

https://tenor.com/view/sad-frog-mood-rain-pepe-gif-22350498

[balmy monolith]

is there a way to determine if the t5xx text encoder has censored your text when encoding it into tokens?

[cosmic furnace]

and there i got and thought amd is not an outcast anymore...

this is very specific,narrow and dumb moment there

[spare crow]

Am i better off just taking the vram hit and using nvidia then? or is it just ive had bad luck setting it up and amd sucks more on some uis more than others?

[cosmic furnace]

with only upscaling in hires fix +flux not wotrkin as on nvidia...weird...

[lucid yoke]

The ai models are trained/made with cuda (nvidia) in mind. Amd had to stimulate it

Like emulating xbox games on PC. Its gonna be slower, glitchy and take effort to run properly

[spare crow]

That makes sense

I feel like there would be some clever way out there of using 2x the same gpu and tricking the pc into thinking there is only 1 with twice the vram xD super speed generations

[lucid yoke]

So while Nvidia can just run it. Amd has to simulate it's working the same way thus taking more resources

[lucid yoke]

Sadly its simply not possible. Or enterprise cards would be useless

[spare crow]

Truee, id just stack 10 1080tis for a fraction of the cost

[balmy monolith]

isn't there a new software layer codec or something that makes the underlying hard link to the cuda architecture irrelevant?

the pytorch or what ever talks to that instead of nvidia cuda's and that layer talks to cuda or amd instead

meaning that nvidia would have the monopoly on LLM's anymore

i heard about it a little while ago, but am clueless as to it's details

[lucid yoke]

Don't

Its a scam

[balmy monolith]

thanks, i'll block them

[shut geyser]

someone knows why hyperflux lora 8 steps gives this result?

[ruby yoke]

wrong vae

[shut geyser]

ae safetensors

[ruby yoke]

you either have the wrong VAE for the model you are using, or you are using the wrong sampler/scheduler - and since you're using hyperflux, and you apparently downloaded the hyperflux vae, then start changing your samplers and schedulers

[shut geyser]

im on zluda amd

would that change things?

[ruby yoke]

did you follow @ornate elk 's AMD gpu guides?

[shut geyser]

to install comfyui? yeah comfyui works nodes and stuff, i use normal workflows all the time

giving decent results

[ruby yoke]

if you followed his guides for your gpu, then that should nto be a consideration.

[cosmic furnace]

and up to now everything runs fine unlike hires fx? mb theres some other hires fix or upscalers that run fine? hm

[ornate elk]

SD upscale script

But Hires fix in auto1111 and forge runs fine

[cosmic furnace]

mostly ppl recomend use ultimate sd upscaler on flux( tiled upscale ) not without A Reason pretty sure y)

as u said long time ago best one upscale is tiled upscale

also pretty popular is that + control net tiled upscale combined

to get best result

so even nvidia users seem to avoid usual hires fix

at least i tried googling & diggint it yesterday and id say hires fix is least popular method with flux

not a flux speeds id say is an spectacular on amd as for now. at least on forge... everyghing runs smooth & fast

only flux its slow & errors sometimes...

[cosmic furnace]

whuht ;O

ah u meant without flux?

[slow adder]

boys i need help, first time seeing this:
i have vs2022
trying to run hunyuan

[cosmic furnace]

ive tried usual upscaling on flux (img2img) just straight up . And my 7900xt 20 gb vram . cudnt get higher than 1536-1536 . its just FLUX eatinh up shttons og vram on both nvidia and amd pretty sure...cuz bad optimization & slow. i have a feeling cuz its free Flux dev type (not paid pro)

they sabotage optimization & vram usage there on purpose so ppl pay ...

and indeed xcept 4090 users i dont see ppl havr ANY good upscale - speed resul;t with flux

correct me if im wrong there -_-

[ornate elk]

Nope its just slow but I guess because the GPUs dont support FP8 nativly

If that would be the case its much faster

Rtx5000 series will support fp4 native

[cosmic furnace]

also

ive had alot of problem yesterday trying out fp16

and it turned out

as from comments...somehow sometimes . vram can transition into ram

and if u dont have 48-64 ram u cant go fp16

even if u have 20gb vram

ive had everything clogged up & turning up slowly cuz 32gb ram...

yet another doubtful flux optimization & turns out if u have 64 gb ram instead of 32 ram matters

not sure when & how often but it seems sometimes alot of ram helps... as if not enough vram can transer to ram >_>

[ornate elk]

yep more ram helps performance wise

it was like that with sdxl and 16gb compared to 32gb ram too

[timid stratus]

Hi this may sound stupid but could anyone help me with formatting a 128gb sd card to FAT32 on a chromebook

I'm trying to get it to work on my 3ds but it keeps saying it cannot be displayed

[ornate elk]

formatting the drive on a ds or ona a pc?

[timid stratus]

On a chromebook...

Could anyone help me out

BTW I know you can format by clicking on the file but it doesn't seem to be working

[ornate elk]

maybe the sd card is locked?

[timid stratus]

How would you check that

No I don't think so I can move files on to the sd card from my chromebook

[ruby yoke]

you're not out of space?

[timid stratus]

Nope

[misty marten]

I'm trying to generate a video with Hunyuan in Comfyui, but this error appears "Invalid or unsupported data type Float8_e4m3fn", does anyone know if there is a solution? I have an rx 6600 8gb

[ornate elk]

windows or linux?

[misty marten]

windows

[ornate elk]

you would need an other text encoder or model type

with zluda?

anyway huanyun wont work with 8gb vram
as it doesnt work with my 24gb vram either on amd zluda xD

[misty marten]

oh that's tough

[ornate elk]

but ltx video could work

it uses much less vram

[misty marten]

Is there another video generator that works well with AMD and 8GB?

[misty marten]

I'll do some research on this

[ornate elk]

here are my workflows for LTX-Video (GGUF)

[misty marten]

does he need comfyui?

[ornate elk]

yes works only in comfyui

[misty marten]

damn, I'll have to install it again

I installed comfyui using Conda, is there a better way to install it on Windows with AMD?

i hate Conda

[cosmic furnace]

If this is only amd problem with upscaler. Why then everyone advice & use only tiled upscaler(ultimate sd upscaler). So i guess its more like with all gpu's . Probab with adm worse. But who knows? I mean if in the end its same bad speed.
I didnt tried to go lower than 16 gb vram in the forge. Also didnt tried - possible but not reccomended run without stopping generation when vram is used at max and goes beyond that so it wont be very slow. Btw when trying fp16 ive had generation even when it said -8gb vram available. So i guess it is transitioned at ram at some point...

and holy shit... 22gb size models are all fully loaded into vram for real? Theres very few gpu's that have 22 or higher... Dont believe it works like that -_-

I didnt tried to lower less than 16gb used vram in forge or setting up continue generating even if vram is maxed... Maybe turned out same as nvidia and had very long upscaling in the end? Its just i dont feel like upscaling to 1920-2560 is worth 12 mins or more .... Even if its flux... Dontcha know

Some ppl tried going upper a lil bit by 128-128 or 256-256 every time with low denoise so its faster. Like 0.25-0.3 denoise. Also was thinking about it. Somewhat work but kinda blah.
Topaz is probably good way to upscale if u dont have anything needs to be fixed in image =\

[timid stratus]

Anyone think they can help me with the sd card situation

[lucid yoke]

What why use conda at all? Cant you install it like directly through git/unpack a zip & update

Unless you got some extra versions of python etc running

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

The best way i know of for amd

[timid stratus]

Ahmeds a scammed btw

Scammer*

[cosmic furnace]

@ornate elk #🤝｜tech-support message Cs1o some comments or recommendations bout this ?) Shud i try comfy or continue even if vram at limit? Id prefer tho not to waste too many time on high details & upscaling >_> .

[ornate elk]

yes with my AMD setup guide

it uses normal python

not conda stuff

you can find my AMD Install Guides here:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides

oh it got already linked by euro xD nvm

[ornate elk]

have you updated your comfyui instance?

[ornate elk]

i wouldnt just use upscaling with flux. thats the solution :/

for me flux isnt worth the time and resource it takes

sdxl performs much better and makes much better Anime images xD

[lucid yoke]

Wish we got a bot what would just link that message when we do like !amd or !install lol

Would save much time

[ornate elk]

true!

[cosmic furnace]

true xD just trained sdxl that understands promt better... I havent tried comfy at all cuz dont like nodes... But my friend says it eats up alot less vram on any gpu and overall better optimized... -_-

I wonder why noone made flux faster, we have alot of things for free & it all was initially. Also if pro version alot faster & have they made alot of profit like this. Recently illustrous become pay to use. I wonder how itll affect it. Alot of ppl say noob is same & even better cuz trained. & hope noone will buy illustroues now >_>

[ornate elk]

it has nearly the same performance as forge and auto but its not better optimised

yea and illustrious merged models are better than the original anyway

[lucid yoke]

Illustrious was poorly communicated. Its situation it a bit complex

[cosmic furnace]

They say huge chineese company just made an offer they coudnt refuse.

I hope they wont get that money back 😦 . This all is an sabotage to SD community -_-

[lucid yoke]

But iirc since i read it this morning and its midnight now. They needed funds and the other releases will be open source again. I will look up the source sec

[lucid yoke]

https://www.reddit.com/r/StableDiffusion/s/8Ghifus1Iy

[cosmic furnace]

For example tho . Space King creators rejected all corporations offers to buy them . By the hints it video i can suggest that it was even Netflix probably. And they choose not to sell them out... Sadge Illustrous creators cudnt. Probably not enough donates or they dont care >_>

[flat compass]

why does my generator does nothing

[lucid yoke]

It cost them 180k and future releases will eventually be open source as they release bit by bit. Yes the money stings but they deserve to get a little back in their investment and let's not forget. All merges of the base model (what we all use anyways) are free

[lucid yoke]

What ui is this?

[cosmic furnace]

Oh wait. Theyr gonna be free after all ? Or partially paid and partially free? o.0

[flat compass]

Stabble diffusion Next

[lucid yoke]

The Reddit post explains it better

[cosmic furnace]

Wow. Good news everyone?)

hmmm

[cosmic furnace]

Seems like sd next. Yeah i wonder how much popular it is xD. Pretty sure forge & comfy is ahead >_>

[lucid yoke]

Hmm I'm seeing that your trying to run it off your cpu? What gpu do you have?

[flat compass]

rx6600

[lucid yoke]

Do you have zluda or directML?

[flat compass]

im not sure

how can i check

[lucid yoke]

Then you probably don't. I recommend following the guide in the pinned messages from cs1o and use web-ui forge or swarmUI for amd tutorial

[cosmic furnace]

if yur not sure u dont have zluda

xD

[ornate elk]

sdnext is a bit complex. would recommend for AMD to follow my install guides

[flat compass]

can i have a link

[cosmic furnace]

zluda is very specific & needs effort to install. Not easy but worth it

[ornate elk]

sure:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides
for your card follow Auto1111 or forge webui both with ZLUDA

[cosmic furnace]

Do i need to update zluda separately when using forge btw? Not sure . Lil bit sadge not uses latest cuda. But latest enough id say

[ornate elk]

nope, delete the .zluda folder and relaunch

[cosmic furnace]

Also i dont sure but speeds of 7900xt feels like somewhere between 4070 Ti and 4080 which id say good considering pricing >_>

[cosmic furnace]

Delete the on in forge folder or the one ive had to use for zluda? o.0

[ornate elk]

the one in forge

[cosmic furnace]

Although idsay flux is giving good enough stable results. So u need very few inpaint or even none. Before simple uspcaling with 0.1-0.2 denoise or even just Topaz straight ahead...

Not sure why this method of using ultra upsclaer + control net is popular. Probably because it add missing details where needed or fixing... Ppl say just go resolution of tile that will make it 6 or less tiles. To go faster

Is it true that sd 3.5 doesnt catch up to flux yet? And why its only Comfy currently -_-

I was hoping that sd 3.5 just gonna run over flux... By all means. Not yet it seems 😦

y? Friend that cud barely test it online said it doesnt catch up yet. -_- havent tried cuz... Comfy 😒

[lucid yoke]

3. got a poor launch due some licence issue so people flocked to flux

[flat compass]

help, the guide ask me to download 3 different files, and then tells me to rename and moves "folders" but doesnt tell me which

[lapis wedge]

heyo! havent touched STAB DIFF. in a few months, is this new or outdated?

asked around n told me to ping you@ornate elk

inviting me to random discord servers eh? really helpful. 💀

[fair oxide]

@lapis wedge wait for @ornate elk he is afk atm probably

[lapis wedge]

ive followed some tuts. went int cmd and typed git pull

seems to be upto date

[fair oxide]

ahh ok cool

i only use comfyui so didnt have the answer for you

[lapis wedge]

whats comfyui?

[fair oxide]

its another UI for stable diffusion and more

https://github.com/Comfy-Org/desktop?tab=readme-ov-file#download

its node based and very versatile with all the latest updates built into it

[lapis wedge]

is it easy to switch over from A1111 to this? how does it work?

[fair oxide]

you would have to take it slow when shifting from A1111 cause the concept of node based UI and A1111 is quite different but in the long run you will grow to love Comfy for what it can deliver

[lapis wedge]

doesnt they both deliver same results?'

[fair oxide]

persoanlly i started with a1111 and took me more than 1 year to finally move to comfyi and never looking back

[fair oxide]

yes but approach is different, comfyui being more versatile in terms of functions and features and also what it can do beyond stable diffusion

[lapis wedge]

i see a desktop download for this, is it just an app that has all the files and none of the file scroll n edit hassle?

[fair oxide]

the link is their official github, the windows download package include all you need to get started, except for the models and stuff for stable diffusion

i run it inside linux environment which is set up as WSL on windows 11, like a container for safety reasons

[lapis wedge]

is the download and setup fast?

[fair oxide]

you can try the windows installer instead of the zip

takes about 5 mins if you have good net

[fair oxide]

few places for comfyui workflows if you decide to learn using it...
https://comfyanonymous.github.io/ComfyUI_examples/
https://comfyworkflows.com/
https://openart.ai/workflows/home
https://www.youtube.com/playlist?list=PLH1tkjphTlWUTApzX-Hmw_WykUpG13eza

[ornate elk]

python can be updated to 3.10.11 64bit, torch is old so delete the venv folder.

[lapis wedge]

delete venv folder?

[ornate elk]

yep needed when updating python or when you want to get the latest torch + xformers

xformers needs --xformers in the webui-user.bat

the venv folder is inside the stable-diffusion-webui folder

[cosmic furnace]

@ornate elk u did compared some of nvidia's to amd gpu's y?)

I was so happy running powa amd that felt alot better price wise... also with SD... but this latest doubtful upscale fails felt a lil bit... morale drop 😦

if its longer than tiled up with 12 mins than blah ofc... shud ask probab how long it took with flux & upscaler fix

[ornate elk]

nope

[misty marten]

I followed the guide, I had this NUMPY error, I installed Comfyui-Manager, but in the guide it asks to click on manager, but where does this manager appear? in Webui? because I can't find it

and this appers in cmd "Prestartup times for custom nodes:
0.3 seconds (PRESTARTUP FAILED): C:\ALL\SD-Zluda\ComfyUI\custom_nodes\ComfyUI-Manager"

[ornate elk]

The manager should be located in the webui as button on the top right

How did you installed it?

Normaly you just have to git clone it into the comfyui custom node folder.
Don't download it as zip manually

Also feel free to provide a full Cmd log

[misty marten]

I followed the guide, installed it via git inside the ComfyUI/custom_nodes folder

[ornate elk]

Then something else might be not working correctly

[ornate elk]

Can you share the full Cmd log?

Or comfyui/users/Comfyui.Log

[ornate elk]

Okay try to delete the comfyui manager again and then reinstall it

@misty marten you can also install numpy this way:

Go into the main folder of where the comfyui-bat is located and click in the explorer bar (not searchbar)
There Type cmd and hit enter and then run these two commands one by one:

venv\scripts\activate

pip install numpy==1.26.4

Then relaunch.

Its very late here I'm off now. Can help tomorrow again

[fair oxide]

Where do I fine these?

I already know how to put them into stable diffusion but I don't know where to download them from

[gray geode]

I’ve sent you a guideline @fair oxide add up and follow the process

[fair oxide]

Hello
That account is limited permanently since for some reason discord will limit my account and then ask for my ID every time my ip address changes (which is alot)

Uh ik that's technically against the rules but im not giving discord my id

And also I've never done anything bad here

[misty marten]

this worked, i can generate images now, but the Manager menu still doesn't appear

[fair oxide]

Oh that's probably a scam

KEK

I just realized

[flat compass]

help, the guide ask me to download 3 different files, and then tells me to rename and moves "folders" but doesnt tell me which

[teal bloom]

Hello, is the a character limit to a prompt with 3.5 large turbo?

[vale veldt]

Anyone know why my Forge install is missing the Loras tab??

[worthy pebble]

Hello, I don't understand how to use this discord to generate a picture with stable diffusion?

[ornate elk]

The gfx files go into the library

[flat compass]

i fixed it already!

and thanks you very much!

[ornate elk]

No problem 🙂

[shut geyser]

where to put lava captioner files?

for this

[trail badger]

Hello, I have a problem. No matter what model I write, it doesn't matter, sometimes I only get grayscale photos.

i use flux model

[odd sequoia]

ValueError: Failed to recognize model type!

[lucid yoke]

Thats a lora your using i think, you want to use a checkpoint!

Also i recommend using prompts that match the checkpoint your using mostly

[odd sequoia]

ouh, let me try later, im still download the model right now

[hallow lodge]

hihi

I'm using Forge WebUI for SD1.5 and SDXL. today i want to try Flux on my poor pc(8gb vram 32gb system ram)

I downloaded the GGUF file from https://huggingface.co/ostris/Flex.1-alpha

I also get the clip_L, ae.safetensor, and then this. t5-v1_1-xxl-encoder-Q4_K_M.gguf

Then i selected all 3 in forge webui VAE menu. Then pressed generate on a simple prompt

I'm encountering below error when it starts to run>

TypeError: linear(): argument 'weight' (position 2) must be Tensor, not NoneType

Did i do something wrong?

[ornate elk]

Hey, what's your GPU?

[hallow lodge]

3070 8GB

maybe i ran into OOM is it?

haha

[odd sequoia]

is this okay? this is the result, model with sdxl 1.0 with artstyle lora. a bit weird, but this is the first time my stable diffussion working.

[lucid yoke]

Hmm the base 1.0 sdxl is not the prettiest

I recommend either a illustrious model if you insist on sdxl

Or a anime specific model

[odd sequoia]

wouldn't sdxl be better for 4gb vram?

[odd sequoia]

yes, i use anime style to draw, i want to use ai to correct my colors

[lucid yoke]

I recommend using a anime specific model then

Illustrious is also sdxl

[odd sequoia]

okay, thankyou for help me!

[ornate elk]

Yes you would need the Q4 gguf but its not fun to use flux with 8gb vram

So you should focus on sdxl

[hallow lodge]

i see

ok. i will try to learn on sdxl and sd1.5

perhaps one day i can use flux.. 😦

it looks so nice

[tawny saddle]

🍍

[ruby yoke]

an ip address that changes like that is a red flag to discord - that's what you see happen with bots - so discord itself thinks you are suspicious

[fair oxide]

Don't most ip addresses change

[lucid yoke]

not that often nio

[fair oxide]

Oh

Mine changes every few hours so I need to use a vpn to trick it into thinking I still have the same ip

[lucid yoke]

one question. why...?

[ruby yoke]

is that your isp or are you running something to cloak your ip?

[fair oxide]

My isp

They change my ip very often

[ruby yoke]

see if you can get a static IP from them

[ruby yoke]

yeah. you need a dedicated IP.

[fair oxide]

https://tenor.com/view/cat-gif-25799218

[scarlet igloo]

Why did I apply a mask to the image for local redrawing, but the generated image didn't produce the desired result? After generating the image, the area I wanted to redraw turned out black

[brittle lantern]

@ornate elk i did zluda comfy ui the auto installer linked in ur git tutorials i did everything but i keep getting this

[ornate elk]

ohh not this error

[brittle lantern]

what should i do?

[ornate elk]

whats your GPU?

[brittle lantern]

6600 xt (im that person again XD im back with more problems lmao)

[ornate elk]

okay

what issues you got before? sry to much people i helped setting stuff up xD

[brittle lantern]

a while back had some common issues with forge saying im low on vram when it shouldnt

and i think other cmd issues

[ornate elk]

ah okay

[brittle lantern]

prolly check my pings on u and u may remember me XD

but yh i wanna learn comfy

so i treid installing but

gave me this

[ornate elk]

wouldnt recommend using comfyui with 8gb vram, but i can try help you fixing it

[brittle lantern]

ill assume auto1111 is better then?

or another UI?

[ornate elk]

Forge or auto1111 as you can control the vram usage much better

[brittle lantern]

ive been away for a while idk how many new ui or updates that have happened

[ornate elk]

whats your rocm (hip sdk) and zluda version?

[brittle lantern]

i can try forge again but i just hope it doesnt give me the low vram issue again and making it really slow when it shouldnt

[brittle lantern]

6.2.4 and zluda is 3.8.8

so all recent ones

[ornate elk]

have you replaced the dll files already?

[brittle lantern]

with the 6600 xt optimised one

rocblas?

[ornate elk]

no i mean the zluda files into venv / torch

[brittle lantern]

i have not actually wait how do i do that?

[ornate elk]

the guide states it:

[brittle lantern]

donet hat now

do i try now

works thanks

which would be better forge or auto1111 for my system?

[ornate elk]

good question

both are very similar in performance

Forge has additional Flux and v-pred model support

Auto1111 has the better extension support (but both have near the same ones available)

[brittle lantern]

Oh I use those V-Pred models a lot

I think I’ll go forge

[fossil stream]

@ornate elk a friend of mine wants to install automatic111 with zluda on linux, and I was wondering if you had a linux install guide

[wintry dawn]

Hello. I was sent here from the Mikmumpitz group and was doing this video. https://www.youtube.com/watch?v=gHI6PjTkBF4

It will not work for me because i have a 5800 AMD card. The part at about 5 min where he installs missing nodes is not working on amd. Please advise. i have looked at github and other sources here but am lost at where to start or go next in my use case.

[teal bloom]

I'm having an issue with Stability's 3.5 Large Turbo API. Frankly, the images that it's generating through the API are complete dog shit. However if I use the same exact prompt in the 3.5 Large Turbo space on Hugging Face, the generation is pretty good. As far as I know the HF space isn't doing anything special to it's generations. Why is there this discrepancy? How can I fix my API usage?

[tawny saddle]

i...

I think I did it

it's not confirmed just yet tho

hopefully it'll make a picture

[wintry dawn]

Anyone here using a AMD card and this stuff work?

[tawny saddle]

nvm didnt work

i did git checkout cfa6e40e

it worked, it started, was back 1.7, loaded fast, was great, nu errors

just when u clicked generate, it just, never ever ever moves

sad, really is, 1,7 was awsome, took 1:20 to make a thing

just cant do it, i tried downloaded the 1.7 for amd from the repo, and then doing git checkout cfa6e40e, that didnt work for somereason, it said that it wasnt a repo, i presume bc, it wasnt an "installed" already thing like the venv or wut ever

i literally would need sum1, to toss me they old 1.7 version, thats already installed and used

:V

[deft hemlock]

ok so i got some simple doubts

the symlink

i installed reforge

this is the models folder path

F:\ai\reforge\stable-diffusion-webui-reForge\models

my a1111 models path is

F:\ai\sd.webui\webui\models

any my reforge webui-user.bat file is

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS= --always-low-vram --skip-torch-cuda-test --disable-nan-check

@REM Uncomment following code to reference an existing A1111 checkout.
@REM set A1111_HOME=Your A1111 checkout dir
@REM
@REM set VENV_DIR=%A1111_HOME%/venv
@REM set COMMANDLINE_ARGS=%COMMANDLINE_ARGS% ^
@REM --ckpt-dir %A1111_HOME%/models/Stable-diffusion ^
@REM --hypernetwork-dir %A1111_HOME%/models/hypernetworks ^
@REM --embeddings-dir %A1111_HOME%/embeddings ^
@REM --lora-dir %A1111_HOME%/models/Lora

call webui.bat

is there anything wrong ?

maybe a wrong character or something

cuz i ctrl c +ctrl v the @ rem from the forge and pasted it into reforge

im a big dumass i know

[fair oxide]

Why do her eyes look weird and how do I fix

[tawny saddle]

i mean

they look oki to me

just looks like they up to sum

prolly gonna try to mess wit ur stable when u not lookin

xd

[ornate elk]

You can use the adetailer extension for auto fixing them

[ornate elk]

Yes, first link of the pinned messages here, are the guides

[fair oxide]

Idk what that is

[ornate elk]

An extension for facefix

[fair oxide]

I don't know what face fix is either sorry

Is it a lora?

[ornate elk]

If you use Auto1111 or Forge webui, Go into extensions tab, click available, click load from,
then you get a list. There search adetailer and install it. Then reload ui

[fair oxide]

Ok

I do

[ornate elk]

Then enable it and it will fix the eyes

[fair oxide]

Thanks that worked

Also

The interrupt button almost never works

So I just close the cmd thingy amd reopen it instead

Is there a fix for this?

[ornate elk]

Double click it

Should be faster as one click

[tawny saddle]

if this dunt work then, its just, super natural at this point

its so close.. just cmon move to at least 1 D;

just dont understand screams

like, other than a missing a commit hash, its works ;o
why it no move ;o

.0.

bruu we got a five percentttt

if this works, that means id have done it, i got back to 1.7.0 succesfully, next challenge, get 1.7.0 for direct-ml

difficultly impossible

like, if i cant get direct ml, and this works, then thats fine, im happy with a version of automactic1111 that actually follows my prompts correctly, slow but listens, or fast and not accurate, i choose slow u-u

besides, maybe threadripper is am4, might can just aquire that at some point

wonder if theres a way to like, modify cpu mode version, to be an direct ml version, cant be that much coding involved

seems to be working good nows :33

good as in, it made a picture, could use some adjusting, but it seems to start generating now as i press the button

feels like i frankinstined this version of stable xd

[wintry dawn]

@ornate elkThank You

[warped canyon]

why is my kohya ss not launching the webui atomtically

[deft hemlock]

take a screenshot of ur cmd

[deft hemlock]

its still loading

wait for it

[warped canyon]

no

[deft hemlock]

damn

[warped canyon]

i launched that like 30 minutes ago, i just needed to put the link on the browser

[deft hemlock]

i got hit with the "no"

[warped canyon]

sry

it launches pretty fast actually

but it doesnt open automatically

[deft hemlock]

does the link work?

[warped canyon]

yes

[deft hemlock]

im lost like u

im not even one of the smart guys

im trying to symlink

[ornate elk]

What do you want to do?

[deft hemlock]

link it

idk the correct format

i fixed it

yippe

turns out they were backwards

[solemn atlas]

hello, what does that mean?

[zinc arrow]

Hey, quick question
whats that one model that released a while back that generates in like 1 step, i think it had L in its name

[ornate elk]

Hey it was SDXL Lightning

[zinc arrow]

Thanks

Also has it had any changes recently, or since it came out in general?

[vocal burrow]

There are also "Turbo" models which are kinda similar

[zinc arrow]

im downloading realvis5 lightning rn
Is it any good?

is turbo "better" than lightning?

[sterile fractal]

When i use vae without "sdxl_vae"vae it keep giving me like those worst image, please help me to fix it 🙏

[lucid yoke]

are you using Vae made for SDXL?

if not thats the problem 🙏

[sterile fractal]

im newbie of this im using "ebaramfcgponymix_v12" for base model

i been try to change other base model still same result

only when i using sdxl_vae give me normal image

[lucid yoke]

why are you changing the vae?

Pony is a SDXL model so you need a SDXL vae

try looking for a different vae on civitAI with these filters if your using the pony model

[sterile fractal]

i see

how i know other vae for what model

[lucid yoke]

you generally dont change the vae much

just the checkpoint

[sterile fractal]

actually I want to ask why all base model that call checkpoint

like the base model that i download from civitai then it should be use at the top left coner right

[lucid yoke]

it comepletely depends on the UI you use but yes in forge/a1111 its probably the left most one. the file you place in "Models\Stable-Diffusion"

[sterile fractal]

yap

then whats the checkpoint use for

i understand lora and lycoris

[lucid yoke]

VAE = colors
Checkpoint = image (realistic, anime, etc etc)

very roughly speaking

[sterile fractal]

ahh i see

thanks for replying me buddy >.<

[lucid yoke]

no worries!

[sterile fractal]

btw

i still have a ridiculous question

I want to recreate someone's style of image

[lucid yoke]

IP adapter could help but generally youd get better results with a lora

[sterile fractal]

i been finding a lot of lora

however all my image look so weird

[lucid yoke]

is the lora compatible with the model your using?

a SD1.5 lora is NOT compatible with a SDXL one

[sterile fractal]

is this

[lucid yoke]

oh your using base

yeah also explains it

go on CivitAI > Models > and use the filter button for ILLUSTRIOUS for anime and SDXL for realistic most likely

[sterile fractal]

uhm

all sdxl right

my mind gonna crash now

[ruby yoke]

SDXL base is a very good model and doesn't need a lora - or a fine tune - if you learn how to prompt it correctly

in fact, the only base models that really do need loras or fine tunes are sd 1.5 and sd 1.4

[sterile fractal]

wait is the base model == checkpoint???

[ruby yoke]

for some odd reason this community thinks that base models don't' work and have to have fine tuned checkpoints and loras before they can be used, and that is not the case. however what IS the case is that too many people in this community don't bother to learn how the model they want to use thinks, and so can't actually prompt it to get the results they want

[ruby yoke]

yes. checkpoint really refers to any model

[sterile fractal]

ohh i see

[lucid yoke]

Its good but like you said people need to learn and that's difficult for most

[sterile fractal]

since i change stable diffusion from pixai so i sill have a lot of question even i watched some tutorials ;w;

im english understanding is very bad

[ruby yoke]

it's tedious and time consuming, not difficult - but people are lazy

[ruby yoke]

you're gonna want to run your prompts through this https://www.deepl.com/en/translator before you use them then, as the AIs really only understand english - that's what they were trained on

[sterile fractal]

this like google translate

[zinc arrow]

this is the greatest translator ive ever seen

[ruby yoke]

:) yes it is

[ruby yoke]

this is much much much better than google translate

[lucid yoke]

Far tbh, but the learning curve on some merges is lower bc it does it for them

[ruby yoke]

a merge isn't a base model. that's the entire point. if you're not using sd 1.4 or sd 1.5, you don't need merges or fine tuned checkpoints, or loras, or anything else EXCEPT in very specific situations where you want something that requires information that's not in the base model. you just need to spend enough time talking to that base model to learn how it thinks, and then prompt it so it understands you. but most of the users in this community won't do that. they don't even create their own prompts. they grab whatever loras and models are in the trending list on civit, thinking that's the image they're gonna get when they run it, grab other people's prompts because they worked for that guy so the prompt has to work for them, and then get mad when they get garbage images

[zinc arrow]

Anyone have a simple lightning workflow (realvisv5) for 8gb vram

i dont want upscaling or any of that stuff

[lucid yoke]

im on swarmUI so i got custom nodes but theres NON swarm ones that are in comfy already 👍

[tawny saddle]

i might get comfy ui

jus cuz it looks like fun,. to have a bunch of lil wires going everywhere

[ruby yoke]

you can always hide the wires

[tawny saddle]

wouldnt that make it like

pretty confusing as to wut is connected to wut xd

eh, im interesting in comfyUi (bc) it has all the colourful wire xdd

tho, thats for the wrong reasons to be interested in comfy xd
usually ppl get comfy bc u get more control as to wut u wan from stable, and here i am like, ouu pretty noodles

xD

[sterile fractal]

is Illustrious match with SDXL?

[flat compass]

anyone can help me on why adetailer stop after just one step?

[ornate elk]

Which settings do you use?

[flat compass]

if i disable skip imgtoimg, the end result looks nothing like the original, if i enable it, it ends in 1 step

[lucid yoke]

Illustrious is a specific fine tune of SDXL so loras made for SDXL mostly wont work on it

[ornate elk]

Your using it the wrong way.
Its not an face enhancer for already generated images.
It has to be enabled in txt2img when generating the actual image.

If you want to use it in img2img you have to use the same prompt and settings like for the input image

[tawny saddle]

indeed

[flat compass]

i see, what are good prompts to put in the adetailer box when i generate with txttoimg,

[ornate elk]

Normaly you can leave everything default for adetailer.
But if you want for example change the face expression or the eye color then add it there

Like, green eyes, smiling,

[tawny saddle]

i dare sum1 to launch 2 stables, and then make them both start making an image

[noble idol]

Deal

[tawny saddle]

xdd

[noble idol]

Okay genning

Crashed 👏

[tawny saddle]

XD

[noble idol]

Took 5 minutes, lagged my laptop out for 3 of them

[tawny saddle]

mm, delicous bugchecks

[tawny saddle]

xd

my cpu surprised me when i forgot i had gmod going, in a server of like 8 ppl, when i was making a thing in stable

dunno how it didnt freeze and crash

[noble idol]

My laptop crashes frequently, as long as it turns back on though it can crash as much as it wants 🙏

[tawny saddle]

xd tru yas

i believe the only time my pc bugchecked was for hault code video_memory_managment_internal

was due to stable too

xd

back when i was new to stable, and didnt have the --medram arg set

or really any arg

[sterile fractal]

that mean i cant use SDXL lora with Illustrious base model?

[lucid yoke]

Yes

[sterile fractal]

oh i see

also i dont know why my LyCORIS are wired

like i only downloaded 2 LyCORIS however in the website ui it show me lora at LyCORIS button

[lucid yoke]

You can regard a lora and lycoris as one and the same for most use cases

[sterile fractal]

but i cannot find this LyCORIS in website ui

[tawny saddle]

anyone here good at making custum Ui

and dunt mind making me one where the buttons are like, 3d looking

where like the generate button is kinda like this glossy glass look

o^o

[tawny saddle]

Aero glass theme for automatic1111

imagine u-u

[ornate elk]

This ui is very nice:
https://github.com/anapnoe/stable-diffusion-webui-ux

Also the Lobe Theme Extension for Auto1111 is nice

[livid dust]

is there any ip adaptor suitable for sd3.5 large model for using it on ComfyUI?

[vocal burrow]

oooh free money again 🔨

[sterile fractal]

how to fix when i use Batch function image not stick together?

[keen falcon]

So I'm new at this and I've got what I think is a working SD A1111 Zluda installation as I can make images in XL, SD1.5, illustrious, and Pony, but every time I try to use Flux the preview while generating looks like this and then turns grey when finished. Does anyone have any idea what is going wrong?

[lucid yoke]

Hi! A1111 cant run flux

[keen falcon]

Darn it

Thanks

[lucid yoke]

youd either wanna use comfy or SwarmUI for that. im unsure about Reforge but i remember people saying A1111 cant run it

[keen falcon]

Thank you. I'll try and find a way to install one of those. Can they run the others? Or do I need to have multiple SD installations?

[lucid yoke]

SwarmUI and ComfyUI can run almost everything however i do recommend SwarmUI as it has a nice useable front end but a optional to use comfyUI section aswell

[keen falcon]

Oh, cool. I'll look in to that one then. Thanks again

[lucid yoke]

i do have to warn you though. SwarmUI has more features then A1111 but it doesnt have the same extensions (some of them are built in. some of them not but A1111 is hella outdated anyways)

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

[keen falcon]

I'm there already. 🙂

With not all extensions working, what are you referring to?

Never mind I found it

[broken epoch]

Hey Im using stable diffusion API for my project and Im taking some errors while I was trying to connect txt2img model can someone help me and fix my code.When I use only POST request its fine but when I try to connect GET for Sending a prompt constantly and generating an image will waste a lot of time. Since there are too many products in the database, I use GET to use the previously generated ones. I am trying to get it from the cache using Redis, but this causes some errors. Is there anyone interested in my code who can help?

[fathom sage]

hey guys, I am just starting to create my first animation with comfyui. would somebody be able to help me figure out why I keep getting errors? thanks

[lucid yoke]

can you post the errors? maybe someone will know if they see it

[fathom sage]

for sure, if anyone needs to see my workspace then I can send that too

[rough meteor]

I keep getting the error "Couldnt clone stable diffusion error 128" Can anyone help?

[carmine kayak]

why do my images come out like this? using flux q8 gguf, dualcliploader with t5-q8.gguf encoder and clip, flux vae and realism lora alongside flux redux.
gpu is 1080ti

[tawny saddle]

this my friend

is wut a million m&ms looks like

its just stable telling u, it reallyyy wants m&ms

[carmine kayak]

:/

[tawny saddle]

try disable c0ntr0l net

and add more steps maybe, like round 28 or 30, if results dunt improve, its a question for the one and only cs1o

[carmine kayak]

removing controlnet didnt work :/, does any flux controlnet work with the q8 gguf model?

[lucid yoke]

What UI are you using?

[carmine kayak]

comfyui

[lucid yoke]

its probably a workflow thing but since i dont use comfy sorry fam

[formal ridge]

Do you have your CFG set too high?

[carmine kayak]

cfg 1, am I allowed to link the workflow I used here?

[formal ridge]

I have no idea. I'm brand new here myself.

[lucid yoke]

You can, a workflow is just a .json file

You could convert it to txt if you really want to make it readable or make a screenshot of the workflow

[ornate elk]

Full cmd log pls

[ornate elk]

Make sure your comfyui is updated aswell as the nodes. Then try a simple image generation with flux and the clips + VAE.

[high sky]

im gettin this error with 4070ti super anyone know how to solve i tried many thing but still cant fix

[ornate elk]

Make sure your GPU driver is updated.
Then dont use the zip file and install it via the install guide from the pinned messages and it will work

[high sky]

okay i will try ty

[ornate elk]

NP, let me know if any step isnt clear

[high sky]

i edit commandline_args --xformers --no-half-vae my card 16gb so dont add any and launch webui-user.bat and get this error

i think i need 3.10.6 version for python right

[formal ridge]

If you're on Windows, just look up Lykos Stability Matrix. It will help you one-click install several different UIs, CivitAI model browser, etc.

[high sky]

im too new for this things lykos stability matrix alternate for automatic1111 right ? i just checked rn

and im on windows yes

[formal ridge]

It's not an alternative, it just makes installing things and downloading / managing models a lot easier.

[high sky]

okay i will check this too

ty sir

[formal ridge]

You're welcome, I hope it helps you out.

[high sky]

first of all i installing python 3.10.6 rn after this i will try

[formal ridge]

I don't even recommend that. LSM will install a single python instance for you that's shared among whichever UIs you install.

[high sky]

then i'm installin lsm ty ^^

[ornate elk]

You installed the wrong python version

Uninstall 3.13,
install 3.10.11 64bit
Delete the venv folder and relaunch the bat

Thats all you need, to fix it

[high sky]

i did what u said and now installing if i get error i will show again big ty for helpin guys ❤️

[ornate elk]

No problem 🙂

[ashen gate]

help. not sure why my Forge is spitting out a Jpg and a PNG of each generation.

[high sky]

im in ty sir ^^

[ornate elk]

Its a setting that gets triggered,

When the resolution is to high.

You can disable it

[ashen gate]

thanks. also various generations keep having a black bar at the top and bottom of the image.

[ornate elk]

Nice, have fun! And for the start go with this model for realism images:
https://civitai.com/models/133005?modelVersionId=782002

The 1.5 ema pruned model can be deleted as its old

[ornate elk]

That can happen if you try to generate at a to high native resolution.
Lower it and try to use upscaling instead

[cunning sapphire]

Hi Stability’s AI tech-supports,

I hate to ask, but I’m really stuck. I’ve been working hard on a project, relying on the free credit, but it’s run out. I can’t buy more because I don’t have access to a card for international payments. It’s frustrating to be so close yet unable to continue. If there’s any way you could help with a little extra credit, I’d be incredibly grateful.

Thank you for your time and consideration.

[ashen gate]

use a local repo and its free

[formal ridge]

ask away

[ruby yoke]

what sampler and scheduler are you using? flux only works well with a few of them

[high sky]

I don't know if this is the right place for this, but do you have any suggestions on which model can do something like this?

[formal ridge]

That's a specific style, but I can't remember the name of it. Sorry =/

[high sky]

anyone got idea for this

[formal ridge]

I've not used flux yet, so I won't be able to help.

[high sky]

i changed vae to ae.safetensors and fixed

i tried 3 different vae this is 4th

[rough meteor]

I fixed it no need lol
It was "Couldnt clone stable diffusion" basically

[brittle lantern]

hi i need some help with auto1111 adetailer when i select mediapipe face mesh eyes only it does not work

it will say nothing detected on image one

despite eyes clearly there

this happens on every gen

[sterile fractal]

how to fix when i use batch function to gern 2 or more images, image not stick together🙏

[noble idol]

[ornate elk]

Batch size or count will just show a collage at first but each image is seperate.
What you want is the X/Y/Z script at the bottom of txt2img

[sterile fractal]

aight thanks

[little bough]

Does anyone know where the prompt that can be recalled from the blue 'load last generated prompt' button is stored? it's not in the temp folder so where would it be?

[formal ridge]

Browser cookies or temporary storage maybe. I recall closing the browser once before and coming back to not being able to load the last prompt.

[little bough]

nah it's not the browser(that's where I thought it would have been), I generate to an external ssd, multiple of them and i switched back to my other disk with a different browser and clicked the last loaded generation and it loaded what came from the other disk so it seems to be storing the data somewhere on my main internal drive where Forge/A1111 is installed. it's not in the gradio temp folder so im curious to know where it is. I know there's a python cache, could it be stored in there?

[formal ridge]

It's been a while since I used either one. Best suggestion I have is to run a file content search on your drive for a keyword from your prompt to find out where it is located.

[little bough]

yeah that's what i plan to do

ah think i found it, 'params' file in forge folder seems to store it

deleted it and tried loading last gen and nothing came up, nice to know

does anyone know if its possible to move the gradio temp folder to another drive? i dont know why it insists on storing some temp files(prompts and with some versions of forge temp images too) despite the fact i have all the folders including temp directed to my external drive

[fair oxide]

i have problem with this

in forge

any solution?XD

help guys

[little bough]

it happens when you're trying to do what/

[fair oxide]

to do image

not working

[little bough]

hmm, seems others are having this issue, is that you in the comment? have you updated to the latest version? https://github.com/lllyasviel/stable-diffusion-webui-forge/issues/2669

[fair oxide]

no its not me

but

i have the same

[little bough]

do you just get the problem with flux or with xl too

[fair oxide]

forge

[little bough]

sorry, corrected it, i meant with just flux or with forge

i was using that very same model in forge yesterday with no issues, i will update my forge and see if i get the same issue

[fair oxide]

hmm its forge i think

idk

Is there any possibility to download the old version?

[little bough]

infact my forge was up to date, generates fine so im guessing you might have some configuration issue somewhere

[fair oxide]

no bro

i install it fresh

doesnt work

maybe when i

download pinokio old version?

i try

[little bough]

did you install with pinokio? i just shared that link because that's what i could find

[fair oxide]

yeah i install with pinokio

[little bough]

hmm

could you just try a regular install or with stability matrix?

[fair oxide]

regular install?

i again install 5 times

[little bough]

yes just buy installing with git

[fair oxide]

pinokio and try it

[fair oxide]

idk how to doit

[little bough]

what, you've tried 5 different installs and got the same issue?

[little bough]

ah its very simple

[fair oxide]

no 5 times the same version

[little bough]

lol ok

[fair oxide]

show me how to do

[little bough]

well if you want a self contained easy way to install, i'd try using stability matrix first, it is like pinokio but a different 'company'

https://github.com/LykosAI/StabilityMatrix

you can install that and then select install forge

[fair oxide]

im here where is download ?

[little bough]

or do it the manual way which is to basically make a folder in whichever drive you wish

[fair oxide]

oh my bad i have it

lets try 😄

[soft nest]

When im training a lora style where the style resembles a cartoon-like style, do i mention cartoon in the caption of the image? eg: "a cartoon girl sitting on a bench" instead of "a girl sitting on a bench"

[tough basin]

https://rom1504.github.io/clip-retrieval/ doesn't seem to be working anymore, is there an alternative?

[fair oxide]

working now

😄

[uneven hull]

hello is anyone online may i get some help please?

[noble idol]

just type out your issue, if someone can solve it theyll let you know

[uneven hull]

i was trying to generate something but when i click the generate button it just say loading after that nothing happen

it doesn't show any error just not generating

[lucid yoke]

What ui do you use? What gpu do you have?. Did it work before?

[noble idol]

[uneven hull]

yes its working completly fine before

[lucid yoke]

Also nothing personal for you proton. but i feel like this should be pinned https://dontasktoask.com/

[uneven hull]

i am trying to reinstall it see if it work

[lucid yoke]

Hmm did you update anything recently?

New gpu? Windows update?

[noble idol]

add any extensions?

[uneven hull]

nope

[noble idol]

i break my stuff all the time adding extensions hahaha

[uneven hull]

i added a new models thats it

[lucid yoke]

What model? In what Webui??

If it's flux in a1111 its not gonna work

@uneven hull will dip i a bit so if you can answer i can help but otherwise i gotta go

[noble idol]

cs1o is here

[uneven hull]

sorry bro, completely newbie here i will try to figure it out myself. sorry for wasting your time man

[noble idol]

he'll fix it real shit

[ornate elk]

Hey, can you show the full cmd log?

[lucid yoke]

No worries, i like to help out whenever i can, but i recommend mentioning your OS(windows/linux) gpu and what your trying to do next time

[uneven hull]

like the console?

its just not running

[ornate elk]

What's the name of the folder where the bat to launch is in?

[uneven hull]

i will just show u the ss

1 sec

ok somehow its working now

i have been trying to generate something just now but its just not running

[ornate elk]

And now it works? What's your GPU?

[uneven hull]

rtx 4070

idk man clueless

[ornate elk]

Okay and which webui do you use?

[uneven hull]

idk i just start using it today

completely newbie here

[ornate elk]

Is the folder called stable-diffusion-webui ?

[uneven hull]

nope

[ornate elk]

Forge, comfyui, sdnext, automatic, Auto1111 ?

Invoke, foocus

[uneven hull]

no idea

[ornate elk]

??

Do you use it localy on your PC?

What's the file called that you launch to start it?

[uneven hull]

its in chinese hahaha

[ornate elk]

Oh, are you also chinese?

[uneven hull]

emm yes

[ornate elk]

Okay then I guess you installed the Chinese version of Automatic1111

[uneven hull]

maybe i should just take a ss of how it looks like

[ornate elk]

Do you have a webui-user.bat in the folder?

[ornate elk]

Can you screenshot how it looks inside the folder of that programm?

[uneven hull]

yep

[ornate elk]

I meant the folder on the PC

[uneven hull]

ok

[ornate elk]

You can click on root directory (in the programm) to get there maybe

[uneven hull]

or u want them in detail?

[ornate elk]

Nopr

Can you right click and edit the webui-user.bat. file?

To screenshot what's inside

[ornate elk]

Behind the set COMMANDLINE_ARGS=
You can add --xformers
That should give it a much better performance when you try to use larger models.
Should look like
set COMMANDLINE_ARGS=--xformers

[uneven hull]

will it break?

[ornate elk]

Should not break but I'm not sure about this specific version of webui

[uneven hull]

my friends got this file online he said that he use it before and its good

then he introduce it to me

i just started today and completely clueless

[ornate elk]

Ah okay xD np

[uneven hull]

thx for the help btw

🤝

[ornate elk]

No problem, feel free to ask for any technical question about the programm.

And yea if the performance is slow with bigger models (6gb large files)
Then try --xformers to speed it up

[uneven hull]

alright thx

[sterile fractal]

im wondreing how the "Add network to prompt" actually work, i mean i can use lora and vae i dont know what it affect..

[vast pulsar]

hey guys, i'm trying to replicate an exact style i achieved with the normal pony checkpoint... i'm not having success porting it to other checkpoints, even the ones still in the pony realm (aka using the same style loras)

any way i can somehow "port" a style?

my only idea is to make an illus lora with my pics... os that doable?

[little bough]

does anyone know the best way or if its even possible to make an application shortcut for the webui.bat for forge/a1111 so you can pin it to your taskbar for quick access?

[lucid yoke]

Yes. And that's probably the way to get a style really. Maybe ip adaptor

[vast pulsar]

Ip adaptor?

[lucid yoke]

Its a bit hard to explain rn but i recommend watching a youtube video for it, style transfer ip adapter

[vast pulsar]

😵‍💫😵‍💫 I guess I'll tey

Try*

[wide imp]

Hey guys, I'm new here, I wondered what's the 402 error when trying to call the API ?

Assuming I have credits and all that

Usually it's some kind of user limit or rate limit thing, but I loaded credits up, and still, can't really use the API

[lucid yoke]

I recommend contacting support or waiting a little if you recently topped up. Since 402 hints at no credits/payment.

Could also a workflow issue if the api is called on incorrectly somehow

[wide imp]

Yeah it was the waiting after top-up situation 🙂 Thank you!

[fathom sage]

can anybody help me figure out why I am getting an error message here?? this is my first workspace ever so i am pretty new. thanks

[lime river]

why would the dmd2 lora be giving me blank images? any guesses?

[formal ridge]

There's a problem with your request (on Windows anyways). .bat files cannot be pinned to the taskbar, even via Shortcuts.

[novel wing]

@ornate elk hey man i took a break from stable diff and it needed to update and sadly its not working properly getting a error of failed to create model quickly. AMD graphics card. hopefully its not a crazy issue. if i remember the last time we chatted i just needed to delete venv.

[ornate elk]

hey, that message can be ignored. its normal for the auto1111 amd fork

but your on an old rocm hip sdk build

[novel wing]

AH

Do i need to just redo everything? in a new format? @ornate elk

[ornate elk]

have you freshly cloned the webui today?

or is it an older version?

[novel wing]

all old

[ornate elk]

ah okay, whats in your webui-user.bat?

[novel wing]

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS= --use-zluda --medvram-sdxl --update-check --skip-ort --api --listen
call webui.bat

[ornate elk]

in the second line (which is empty) add Git pull

[novel wing]

you got it

[formal ridge]

Looking at the left screenshot it looks like you're trying to download a model at the startup. Do you not have a local model installed?

[ornate elk]

and then you need to uninstall all HIP SDK stuff in the system control.
Then download and install HIP SDK 6.2.4 and then reboot the PC.
Then delete the .zluda and the venv folder and relaunch the webui-user.bat

Then your updated and it will be compiling zluda again for the new version

[novel wing]

it should be all stored locally

[novel wing]

@ornate elk is it possible that windows is blocking Gitpull?

[torn lotus]

Is it worth my while to upgrade a 3700x 3060 12gbvram 32 to 64ram?

[formal ridge]

That depends. Are you currently using a 2x16GB kit or a 4x8GB kit? Ryzen CPUs work best if all slots are populated.

[torn lotus]

2x16gb ... would need to upg a mobo for 4x16

[formal ridge]

Then you should be fine. Unless you're running way too many things at once you'll never need that 64GB. I'm running a similar setup as you.

[torn lotus]

I'd like to use flux

[formal ridge]

Then use Flux. The only thing limiting you on FLUX is VRAM. You'll need an nf4 (Q4-0), maybe Q5 if small enough Flux model to fit in that amount of VRAM.

[torn lotus]

is GGUF worthwhile?

[formal ridge]

It seems to work well for a lot of people, so I'd say so. It's just another format, like CKPT or Safetensors.

What actually matters is the quantization.

[torn lotus]

well if it dampens too much the quality result, I'd rather stay on 1.5sdxl

[ornate elk]

You have added git pull at the wrong spot

Its not an commandline_args arg

It has to be alone in an empty row/line

[ornate elk]

Nope

[torn lotus]

what about a sli smaller vram gpu, like 8192

[ornate elk]

32gb is enough for that too

For Flux use the smaller gguf variants

Like Q8 or Q6

[novel wing]

grumbles

[little bough]

yeah i mean is there a way around this to like make an .exe or something that directs to the .bat

[novel wing]

@ornate elk @echo off

set PYTHON=
set GIT=Git pull
set VENV_DIR=
set COMMANDLINE_ARGS= --use-zluda --medvram-sdxl --update-check --skip-ort --api --listen
call webui.bat

[ornate elk]

It has to be just
Git pull

Without anything before

[novel wing]

ah

[torn lotus]

@ornate elk would you advise a sli 12+12?

[novel wing]

@ornate elk that worked got this now XD

[ornate elk]

Delete the venv folder again and relaunch

[formal ridge]

If there is, I'm not aware of it.

[novel wing]

dope

[formal ridge]

You would still be limited by the 12GB VRAM of one model, that's why the H100 / H800 have so much VRAM on them.

Even though the H800 is technically SLI (or a similar form), I doubt nVidia would make that available using lower cost consumer GPUs.

[novel wing]

got it again @ornate elk do i need to also delete the zluda

@ornate elk

[dusky dragon]

Can someone help me with this?

trying to use a lora from civit ai

its saying forbidden when i paste them

[ornate elk]

Do you have any extensions installed?

Check the extensions folder

[novel wing]

@ornate elk

[ornate elk]

Okay good

You may need to reinstall the webui

Ive never saw the numpy error in auto1111 before

[novel wing]

alrighty!

[ornate elk]

You can save your models and outputs

[novel wing]

@ornate elk you got it! do i just follow your link you gave me the first time to install?

[ornate elk]

Yep the link is also in the pinned messages

Automatic1111 with zluda is the Guide

If you want to use Flux, you can install Forge Webui with Zluda

[novel wing]

what would you suggest? whatever you think is better i will do that for sure

[ornate elk]

Both are good, you can also start with Auto1111 again and if you need you can install Forge later

[novel wing]

@ornate elk thanks for all the help i kinda figured id have to restart today but as always i couldn't have done it without you good sir! this what its currently doing!

[tawny saddle]

i think it funi

when some times ppl show their console, for help and they like, forget to hide their smut models xdd

ive done it once u-u

[ornate elk]

looks good, no problem!

[formal ridge]

Why hide your smut models? "let your freak flag fly". Just remember it's a SFW server and don't show images or prompts.

[lucid yoke]

^ smut models are also very capable of sfw

[tawny saddle]

Yas

I also mix pruned with everything xd

bc it adds realism and it also adds like a million more possibilities other than wut the one model could do

[formal ridge]

Yes they are, they're generally better with the human body than SFW models. Details such as eyes and fingers are better

[lucid yoke]

and if people really wanna see they can always just ask in a dm tbh

[tawny saddle]

xd

[azure magnet]

Hi there is someone that could help me?

[lucid yoke]

Hi. it looks like that this is the problem:

unless theres another problem because i dont see a erro

r

[azure magnet]

@ornate elk sorry for tagging could you help me

[azure magnet]

I don't know what should i do

[lucid yoke]

well did you movify that comfyUI or did you get it straight out of the box

[azure magnet]

I don't know i'm just a casual user

[lucid yoke]

.... okay did you add custom nodes or is this the first time you openened it / didnt change anything

[azure magnet]

I have add custom nodes while ago.

[lucid yoke]

hmm well since your probably not un linux we can ignore the first warning.

ThreadpoolCTL

i recommend you visit that link

[formal ridge]

ComfyUI isn't a UI I would ever tell a casual user to try. You should look at SDWebUIForge

[gilded leaf]

Hello i've got a question
I want to try to generate a video on stable diffusion webui
txt2img work super fine, but when i try to make a video with "animateDiff" + selected motion, it just creates a picture what am i missing?

[lucid yoke]

What GPU do you have?

[gilded leaf]

nvidia 1080ti

[azure magnet]

Yes but i'm able to use it but not in fixing "techical stuff"

[gilded leaf]

But am i missing some parameters or it should just work?

[lucid yoke]

hmm its been a long while since i worked with animatediff tbh. hmm you were on A1111 you said?

[azure magnet]

This is easy to fix?

[ornate elk]

motion module only works with v1.2 of animated diff model

[gilded leaf]

Ok sorry (i never got into video stuff so i have no clue) how do i check which version it is?

[ornate elk]

which animatediff model have you selected?

or in the animated diff models folder

[gilded leaf]

AnimateDiff Motion Modules
This one from civitai

sdxl v1.0 beta

[ornate elk]

this one cant use motion modules but should still work, check the cmd log

[gilded leaf]

I'm opening and closing to try again. But meanwhile a question how do i understand if it should work or not?

[ornate elk]

it should show something in the console like a process bar

also you have to set the animated diff batch size to 1 or 2

also you need to have --xformers --medvram-sdxl in the webui-user.bat

[gilded leaf]

Ok yes now it worked thank you! The batch size did the trick when i placed it at 2

[ornate elk]

no problem 🙂

if you get just an gif which switches between 2 images or similar error you have to enable this setting:

[lucid yoke]

did the link have a solution?

[azure magnet]

What?

[lucid yoke]

The link in your errorlog

That basically said "check here for troubleshoot/workarounds"

[azure magnet]

Yes check it, but i don't know what should i do

[ornate elk]

Can't help much with Comfyui.
But it seems you installed custom nodes which rely on conflicting library's.

[azure magnet]

Thanks, couldn't help me even with the second once?

[ornate elk]

Do you get any errors while generating or does it work?

[azure magnet]

No, no error while generating.

Could i upgrade python version?

[ornate elk]

If you got no error then you don't need to.

[azure magnet]

But upgrading isn't better?

[formal ridge]

"if it ain't broke, don't fix it"

[ornate elk]

What's your python version?

[azure magnet]

3.11.8

[ornate elk]

Thats okay

ComfyUI recommends to not get higher than 3.12

So your good with 3.11.8

[azure magnet]

thanks

[spice vale]

Me and some others across forums are having trouble with Stable Diffusion forge

I just got here so I don't know if it's been talked about but is there a fix for it?

[sterile fractal]

Hello! i want to ask how the "Add network to prompt" actually work?

[finite zenith]

Hi, does anyone know if there's a lora tester for SD forge?
I have a bunch of lora trained (20 epochs) and want to test them on a grid to see each training progress

[finite zenith]

Solved

[lucid yoke]

You might wanna give more context

[sterile fractal]

by "add network to prompt"?

[lucid yoke]

By what UI you use lol

[sterile fractal]

a111

[lucid yoke]

https://www.reddit.com/r/StableDiffusion/s/ABZ5qY9b92

Imo its a niche so far as i can read

Cant see it in newer WebUi's

[sterile fractal]

aight

[safe seal]

do i put in every line or all at the same time ?

[ornate elk]

Every line

[safe seal]

Yeah i been trying to install that and instead of my pc my brain shutdown

[sterile fractal]

is guidance = cfg?

[lucid yoke]

yea

[sterile fractal]

okie

[soft nest]

Which type of lora is best? im training cartoon-like style and idk what to pick

There’s soo many 😭

[lucid yoke]

? Are you asking what type of style or type of lora

Lora and lycoris are imo the same but most people make loras

[soft nest]

im using kohya and im training style using lora

The style is pretty much like 2d cartoon style

so i want to know which lora type is the best

like standard, locon, etc.

[sterile fractal]

Why my image looks so simple, even i use all same model and lora

https://civitai.com/images/39367251

i have no idea i test so many times

[ornate elk]

Pony models are sdxl (1024x1024) based so better use a resolution like 768x1024
Also you need to add some negative tags

[sterile fractal]

ohhh i see

cuz i saw some vids on youtube said that 768 is the max for ai🤣

[sterile fractal]

i remember pony no need any negative idk

[ornate elk]

Pony has score tags that you can add for improvements.
In the negative add score_5, score_4, score_3,
For the positive prompt add
Score_9_up, score_8, score_7,

[ornate elk]

Thats not true lol xD

[sterile fractal]

aight man thanks a lot although its still havent hit my expect

[ornate elk]

The image looks good

But the prompt is more focused on quality tags than describing an actual scene

[lucid yoke]

I recommend going on civitAI, images page and filter for PONY, and see how other people make their images (if they share the prompt)

I also recommend checking under the model so you can see what to expect mostly

[sterile fractal]

i got one here

[sterile fractal]

here

totally the same

its still looks like idk how to say

[lucid yoke]

You also have his custom embeddings?

[sterile fractal]

wat custom embeddings?

[lucid yoke]

negative_v2 (negative_v2 Color_Balance_Calibration_0.8) negativeXL_D,

[sterile fractal]

in civitAI

[sterile fractal]

i dont know whats that

[lucid yoke]

Yes, he didnt list them

[sterile fractal]

i just want my images like this style

[lucid yoke]

Hmm well i think he has some stuff that isnt listed on civitai
lora:badanatomy_AutismMix_negative_LORA:1,

In his negative but isnt listed a a resource

So i unfortunately can't help you there, maybe you can dm the person if hes willing to send you those missing resources

[sterile fractal]

"lora:badanatomy_AutismMix_negative_LORA:1" this is the key to make images like that?

[soft nest]

Can someone please help with this?

[lucid yoke]

I recommend following a guide if im being completely honest with you. Most people here dont make their own loras locally

Its trial and error, when in doubt use standard i say

[lucid yoke]

No idea but you need his embeddings and his negative lora to replicate that image

[sterile fractal]

oh

aight

[lucid yoke]

And even then, with euler A its slightly different per machine

[sterile fractal]

i wanna give up🥹 🥹 🥹

i been try to do that a few weeks ago

[lucid yoke]

why that exact style? why not try others/ try your own?

[sterile fractal]

I love it so much💀