💬｜general-chat

[velvet fossil]

okay thanks

first things first

where's the upscaler?

are you kidding me? it takes credits?

that is getting old

[green warren]

https://replicate.com/xinntao/realesrgan
https://replicate.com/cjwbw/sd-x2-latent-upscaler

These are the models that DreamStudio upscaling is based on, you can try them for free here

[velvet fossil]

okay

what about importing images?

[green warren]

There's a box you can click to upload any image you want but only one at a time

[velvet fossil]

I see

[green warren]

If you do a lot they will eventually ask you to pay up. AI isn't cheap to run

[velvet fossil]

what about the SD2 Latent upscaler?

same thing?

[proud nova]

Hey there

[velvet fossil]

Hello there

[wild steppe]

https://twitter.com/EMostaque/status/1651328161148174337

Emad (@EMostaque)

Code for IF by @deepfloydai is up, amazing work by the team.

Weights in a couple of days along with blog posts, inference & other tools, key research Qs & more.

Release will be like SD, research => open release after feedback as new model type.

https://t.co/7ziBcmWXXF

Twitter · Apr 26, 2023, 8:51 PM

[peak orchid]

DeepFloyd's IF model is out now!
Though they took the weights offline again? https://huggingface.co/DeepFloyd/IF-I-IF-v1.0

Github link: https://github.com/deep-floyd/IF

[charred mesa]

Holy crap

This is amazing

too bad we'll have to wait a few weeks/months before it gets optimized for low end machines 🤷‍♂️

but still great 👍

16GB vRAM required (or 24GB vRAM if you want 1024x1024) 😬

I'm not so sure how this could run on even 8GB without drastic quality drops

[green warren]

Emad said in the EleutherAI server that the weights will probably be out Friday. They're finishing blog posts first

[cedar lark]

Lets draw some hands:
https://twitter.com/i/broadcasts/1BdGYyOaQOYGX

[summer haven]

hi folks. I been busy working on a custom text to image model and I'm opening it for a small test for feedback, if anyone wants to try it out let me know ill send you the link to my discord where you can try it out via the bot.

[peak orchid]

Also, what is with this license? Am I reading it wrong?

Interfering with content filters built into the code is apparently not allowed at all:

2. All persons obtaining a copy or substantial portion of the Software,
   a modified version of the Software (or substantial portion thereof), or
   a derivative work based upon this Software (or substantial portion thereof)
   must not delete, remove, disable, diminish, or circumvent any inference filters or
   inference filter mechanisms in the Software, or any portion of the Software that
   implements any such filters or filter mechanisms.

Source: https://github.com/deep-floyd/IF/blob/main/LICENSE

[green warren]

Yeah... That's a little concerning but it's not gonna stop people from doing it anyway

[velvet fossil]

hello

I decided to come back, I just needed some space

[pallid sedge]

has anybody tested the new image upscale api released today? esrgan-v1-x2plus is for the esrgan model, what parameter should be used for Stable Diffusion 4x Upscaler ? the documentation doesn't mention anything about the 4x model

[velvet fossil]

hmm

I don't think so

[peak orchid]

Its going to ignite a new wave of drama, and potentially open the gates for Stability AI to be legally forced to go after any project that removes them

[wispy jasper]

What happened? Deep Floyed models were released earlier then pulled.
https://github.com/deep-floyd/IF

[velvet fossil]

hmm

[wild steppe]

@wispy jasper

[velvet fossil]

hello Jae

can we talk please? I want to chat in a VC channel

or are you busy?

[quiet trail]

Hi. Anyone who knows how to get the new upscaler to run under auto1111 ?
https://huggingface.co/stabilityai/stable-diffusion-x4-upscaler

[velvet fossil]

I might try that

but not today

look I don't want to say this but I want to actually talk with my voice on this matter please

if that's fine by you Jae

if you want to

I want to apologize for how I lost control earlier when I was talking in the stream chat

where do I go to share my ideas?

[frosty saffron]

i think the huggingface demo is missing/down

for deepfloyd IF

[glad acorn]

TheLastBen colab is broken just for me or anyone else having issue lately?

[polar snow]

RuntimeError: "LayerNormKernelImpl" not implemented for 'Half'
wat do
this is StableLM in JupyterLab with the notebook file
i tried installing visual studio and cuda toolkit, no help
had same prob with SD solved with command line param --no-half but can't see how to do that with StableLM.

[novel narwhal]

i love this

ive been having a problem with training lora

i made a 50 image 5000 step one a few days ago

now i couldnt do that again

but after just spamming and retrying it works after like

5 hours of sitting here

[polar snow]

maybe test your system with OCCT and Furmark to make sure your memory/gpu are stable

[novel narwhal]

a

1 sec gotta google

wait i wont download rn its finally working

for now ill be happy with this lora i hope

if it works well

i guess itll be a test version

so if this works well ill do those tests and try to double the steps and increase image size some time

[novel narwhal]

Update

Doesn’t work well

[ripe heron]

hi guys...can I know it this stable foundation used like midjourney?
how can I generate ai photos?
pls guide me thanks

*stable difussion

[jovial maple]

What's up

[jovial maple]

Download stable diffusion webui auto1111

And yes, your PC must have sufficient gpu

[ripe heron]

ohhh thanks for your suggestions btw, is there any website other than this?

[fervent thunder]

Y'all know any good upscalers?

Need one that works on colab

[topaz lantern]

Hey everyone I'd like to announce 🎤 Nebula is All You Need: A Universal Loss Function for All Deep Learning Workflows [paper]

A versatile and adaptive loss function that works with any model, dataset, or task. Experience the magic of a single loss function that dynamically adapts to your needs, making it easy and efficient to use in any deep learning project.

https://blog.apac.ai/1-loss-function-for-all

[fervent thunder]

its not just stable diffusion, you need a photorealistic model and also a good negative prompt to exclude unasethetic images from the latent space

if you want to generate photos

[dense rover]

Л

[fervent thunder]

What even is this thing

[fervent thunder]

Not really, mj based on discord, can generate images and...pretty much it's capabilities ends there.
SD you can install locally or using cloud service and...go from there, there's too much to explain.

But yea...you'll need to install webui if you want comfortably with it, models of a style you want to generate images...
That's minimal kit for a start.

You can use /faq command
(It doesn't work for me again for some reason, so I can't give you answer to how to start 😄 )

[potent spire]

https://www.reddit.com/r/StableDiffusion/comments/1302eby/made_a_stable_diffusion_tool_for_maya/

Thats interesting

[novel narwhal]

Thought about upgrading from 3060, any of u have a recommendation?

[broken smelt]

4090 for sure

[novel narwhal]

budget under / around 1k

[wicked patrol]

3090

[novel narwhal]

yup probably

[fringe current]

question, is installing or adding model or extension all have the same process?

[opaque salmon]

hello, this happened yesterday and it was recorded but I missed the link to youtube and can't find via search. Can someone help me please? Thank you!
https://twitter.com/StableDiffusion/status/1650572265229258752

[novel narwhal]

is tehre anyyonne i can ask to train a lora for me

since i gotta wait for a 3090

[fierce scroll]

would you say that prompts for ai art generators are zero shot prompts?

[uncut flax]

Hello, I got a problem,

for an art project, I am trying to generate an image similar to this photo, but even with a variety of prompts and Lora, I am not able to generate for example lying pig breastfeeding small pigs. Any ideas? Maybe some explicit model? (the image should be a drawing of this scene but generated by ai) I am using stable diffusion.

i cant post image here so it like lying pig breastfeeding small piglets.. I belive u

❤️

[pastel dune]

Anyone might possibly know why my local ip isnt working with stable diffusion? I think i installed everything correctly

[fast cape]

Anyone have a thought when I do PNG Info in Automatic1111 on a fresh new install it no longer adds a override for the model hash as it did before?

all other settings transfer over find when you hit the send to txttoimg button 😦

[broken smelt]

Check your a1111 settings, I remember seeing model hash as toggleable setting of the PNG info in there

[wraith night]

what is the windows feature called that would allow sharing of models across two applications? Trying to get vlad and a1111 and maybe easy looking at the same spot

[pastel dune]

does anyone know how to turn on realtime progress? seems that progressbar is no longer an option in settings

did it get removed?

[fast lake]

Guys, I have a long switching models in the stable, sometimes takes up to 50 seconds, although writes that the approximate time of 2-3 seconds, and also sometimes starts the web version through the console, it seems to freeze and that's all

[alpine bear]

you running on hard drive or ssd?

[peak orchid]

Looks like the EU has opted to make it easier for groups to sue model creators who used copyrighted content: https://archive.is/6lwYp

Under the new provisions being added to the EU’s AI bill, developers of generative AI models will have to publish a “sufficiently detailed summary” of the copyright materials they used as part of their creation, the draft says. The latest generative AI models are trained to create their own content—such as marketing copy, photorealistic images or pop songs—by ingesting billions of existing texts, images, videos or music clips.

The new provisions could give a boost to publishing executives and other creative workers who have been trying to determine the extent to which their content has been used to train generative AI tools, and how they can be compensated.

[vestal dew]

Maybe all these legal issues are why SD has been stagnant

[copper crystal]

That's not the law's purpose. You've got to realize that the EU is much less litigative as America. They've already determined that using images to train a model is fair use and i'm pretty sure that's part of the act that's coming. I'm not from the EU so the details elude me there, but UK and EU both have determined ai models to be transformative creations.

The purpose is so that analysts can determine what the knowledge base and ground truth's of a model is.

America's copyright system is one busted ass disgusting monster that the rest of the world sort of works AROUND. most countries don't work like it does

[rocky root]

start

/start

[odd geyser]

Hello everyone ! Just been absorbed by the SD ecosystem to days ago ! What a wonderful thing ! I'm planning to use it to do some video clips in the rock paper scissors fashion (although in lower quality I guess since I don't have the filming equipment). So far I've finetuned deliberate to my face and I did try to render a 3s 30fps clip on my local machine (which took 45mn) so today I'm trying to use runpod to accelerate the rendering. As 2mn paper guy would say, what a time to be alive !

[glass creek]

i'm in, new noob, studying hard

[broken gull]

does anyone know about alpha3D? it claims to have a AI that converts images to 3D models I have uploaded one to see how itgoes but I am actually kinda scared in a way that these are just overworked people making these models..

there is no further information on the site about AI

[fervent thunder]

çllo,gfv,lvfrjkcde,xs

mjytyygygiyigyig

hehehheheheh

jipfhdfhdhyik0sripjzsrzsj

[fervent thunder]

?

[novel pine]

hello what is this error please: RuntimeError: "log_vml_cpu" not implemented for 'Half'

[delicate tiger]

does the new controlnet v1.1 work with the old 700mb models? or do you need the new 1GB+ ones

[novel pine]

fixed it by adding --precision full --no-half --skip-torch-cuda-test

[warm junco]

no you would need the 1.5 or the pruned 600mb models from here:
https://civitai.com/models/38784/controlnet-11-models

[delicate tiger]

so can i delete the old ones?

[merry kelp]

hi all. where can i get some help to automatic1111? Its randomly freeze the ui. when i refresh my ui browser page its working again.

[warm junco]

You can but dont need to

The still work

[warm junco]

Hey come to #🤝｜tech-support and try an other Browser

[delicate tiger]

But if they dont work with v1.1 they’re just taking up space

[dense tinsel]

Hello guys,We are integrating stabble diffusion XL with the api Access. We have one issue we don't get. Once we type the prompt on the front end, it goes well on the backend variable but when it's Time to generate, there's a api security blocking the output to the UI. I don't get it. It's it just on the api demo version ? Or once we pay this limitation is removed

[opaque needle]

Does anyone know what is the workflow for these funny anime filters that are applied on existing dance videos ?

I can't stop seeing such video everyday on Reddit but haven't found any workflow for it never actually tried to do anything except images

[warm junco]

Thats done with the mov2mov or video2video extension

[spare plume]

Has this already been voted? How will this work? I mean sure, now they have to disclose what their model is trained on, but how will this be used against them if training a model on copyrighted content is faire use?

Also, @peak orchid, could you please send me some sources about using images to train a model being fair use in the EU please?

[copper sand]

Hello, I wanted to know where I can edit videos? I'm new to discord and I don't know if it's done here or on some show?

[oblique garnet]

Hey fam! Is there anyone in here that's pretty well versed in DeForum and wouldn't mind me picking their brains a bit? I'm using it well but I have a couple questions that I need to figure out

[opaque needle]

Thank you ! I'm going to take a look at it !

[tulip yoke]

whats currently the best vae for anime

[alpine bear]

Kl-f8-anime2

[tulip yoke]

i tried it, and it looks disgusting

[gentle otter]

is there an online service for stable diffusion where you can install add ons like control net and embeddings and stuff? my gfx card cant do it

[tulip yoke]

L

[gentle otter]

what

[tulip yoke]

L

cant u read

[warm junco]

Whats your gpu?

[gentle otter]

amd radeon 7700

[gusty oasis]

Ey yo what was that announcement

[timid bloom]

abouy floyd but its strange

they deleted it after 1 min

[warm junco]

RX 7700 ?

[gentle otter]

doesnt say rx anywhere, "amd radeon hd 7700 series"

[warm junco]

Ah okay yea that wont work

[gentle otter]

yea it refuses to run

so i was hoping there was some kind of remote options

[latent ice]

where did the Vicuna announcement part go?

[slender vault]

@karmic brook what happened to the stablelm and vicuna announcment :/

[mild ridge]

the join button for the event is not working

[latent ice]

?

[hardy current]

anyone know is there a way to generate PNGs with transparent background?

[tulip yoke]

rx vega 3

into your mouth

[spare plume]

Guys, do you think DeepFloyd is getting us closer to the end of graphical designers? If it can make text and follow prompts way better, then what are the cons?

I don't get how such a powerful tool can come with no issues

[plucky shoal]

@hardy current i think it is not posible, but you can try "constant green background" and it should provide key color somehow possible turn to alpha in some program like Krita photoshop gimp....

[hardy current]

AI won't replace you, what will replace you are the people who use AI

[plucky shoal]

i dont know if there are some such plugins

[spare plume]

Well, is this bringing us closer to the beginning of prompt designing industry then, to put it this way?

[hardy current]

is it because SD was not trained with PNG?

[plucky shoal]

i dont think it is related. But saw this question before and it doesnt export with alpha

[alpine bear]

Then you're doing something wrong and its a skill issue. It is the best vae for anime gens currently - has been for months. L

Paired with an anime-centric model like AbyssOrangeMix, it should be fine. You can also try the noise seed delta of 31337.

[hardy current]

if we could generate PNGs with transparent background it would be so much easier to do things like changing the background

[plucky shoal]

try that tag i posted, if it helped in term of constant green background
If it works properly you can change backgrounds ex post as well.

[mortal drum]

Any channels in this server dedicated to discussing AI video techniques?

[peak orchid]

In the EU, one of the ways training is considered "fair use" is under the TDM exception and introduced in the Directive on Copyright in the Digital Single Market I believe

[spare plume]

Thank you very much. I'm tired of seeing too many misinformation from the Anti-AI side. At least now I can look at the articles, read them myself, and send them relevant things.

[thin ledge]

I know I may annoy someone asking this. But I used stable diffusion and one day it stopped working. I am unclear/Unsure where or what is the new download and where to get it. I tried the resources page but I only seem to find an online source. I want to install it on my PC again

[warm junco]

Here is the sd-webui from automatic1111:
https://github.com/AUTOMATIC1111/stable-diffusion-webui

[fervent thunder]

Any one can change dress in photo?

[honest prism]

hey guys how do i get an api key? Any tutorials for integrating with react app?

[tawdry summit]

If StableVicuna compares at all to gpt3, that's huge.

I would otherwise give it at least a few more years before it would become open.

[fair nest]

StableVicuna has 1536 characters limit in the demo. pretty good

[patent atlas]

Someone have this issue with SD via Collab where you disconnect when switching checkpoints?

[ember crag]

yo guys i have a weird problem

my images are automatically upscaling without my input

the image size has been creeping up from 1024 to 1689 for some reason

i am streaming right now (link in my bio) if anybody oculd come give me a hand i would be indebted ;_;

[hazy wind]

Where can I use the new SD upscaler for free without having to download anything?

[fervent thunder]

How to prevent announcements from pinging me =.=
I disabled all notifications and shit...

[dawn mulch]

calling StableVicuna "Open Source" is quite misleading. anything based on LLaMA is not really true "Open Source" since it cannot legally be used by anyone who hasn't received access to the LLaMA weights by Meta

so it definitely cannot be used commercially in any way

[wispy jasper]

Can't seem to download the jupyter notebook from IF, chrome keeps deleting lol

[storm shard]

It's even worse for me, it just doesn't answer my quetions.

[torpid surge]

i tried to get it to give itself a nickname, refused, just kept saying it can't do that. which would be okay if it was good at supplying information

maybe in certain areas it's useful

i get that it's a first try

[storm shard]

I asked if it could do some python code. It says it can so I asked for simple merge and order table code and all I got was
import pandas

[torpid surge]

what's import pandas?

i like the idea of an open source language model but making the claim it's almost as good as the best one out there invites strong criticism

[storm shard]

import pandas as pd is how you start python

[torpid surge]

ah

the personality is a huge part of what makes these tools magic for people

[storm shard]

But I wanted the code to do the sorting, just to test it.

[torpid surge]

since they're just remixing stuff

[storm shard]

I don't like it's personality, there's none.
It didn't respond to my question when I asked to let me know if more info is needed. And also nothing after I said "Hello"?

[torpid surge]

is the language model they're using not trained on much creative writing?

[storm shard]

It said it could do python...

[torpid surge]

lol it still has the misplaced confidence part

just no creative brain

all dunning kruger

[storm shard]

No, it should be able to write code https://stability.ai/blog/stablevicuna-open-source-rlhf-chatbot

[torpid surge]

well maybe it just needs tweaking for it to give you code

[storm shard]

At this moment it doesn't respond on anything I ask

[high magnet]

Anyone know how much VRAM is required for inference with Vicuna?

[forest crow]

Can anyone tell me if stable diffusion local would work on a GTX 1080ti?

[old steeple]

I'm still using SD 1.4. Anything change much?

[fervent thunder]

does it take credits/cost me money when i run stable difussion on my own machine ?

[high magnet]

no

just your energy bill

[crimson dagger]

is there anyway i can view winning entries of POW 27?

[karmic brook]

#🏆｜winner-gallery and #1093352157048217682

[fierce musk]

if stable vicuna delta is CC licence and Llama base model is NC then what is the licence for stable vicuna working model and to what extent it is both open source and business friendly???🤷‍♂️ 🤔

[errant yacht]

Is there a difference list between StableVicuna and StableLM-Alpha?

[mighty wolf]

Exactly, @fierce musk not stated, so isn't ?

[frail wolf]

are there any good models based on sd 2.1 yet? seems like the best models are all based on sd 1.5

[fervent thunder]

yep, just 1.5 afaik

[hollow atlas]

hi guys

I am new here and have no clue how to make things working. I went through #1072220168534642768 but can you guys assist and tell me where to begin ?

[hexed hatch]

if you just need the service and not to download it, https://beta.dreamstudio.ai/generate

[lethal sparrow]

that ai that got posted in announcement kinda bad

[nova zodiac]

Artius v2.1 is probably your best bet, followed by rMada merge, Perpetual Diffusion (sun version), and Digital Diffusion

[frail wolf]

I've tried 5 different ones now and artius was consistently the best out of all of them

[near falcon]

is it possible to use chatbot for games without need to download 6gig models and stuff ??

[frail wolf]

only if its hosted online, can call it from an api within the game

[nova zodiac]

I can recommend trying "RCNZ Orangutan with a Plan" as well, but I think Artius is still better

[frail wolf]

whats the diff between hi res fix and just upscaling the picture later?

[near falcon]

is it possible for multiple users to call simulatnieouly if calling from model hosted online?? Also is there ant hosting service that does this ??

[frail wolf]

look at skin ai, midjourney, huggingface spaces, etc. already done by a lot of places

tbh if youre asking these questions youre probably way out of your depth to attempt such an endeavor..

[nova zodiac]

hi res fix does the img2img as part of the workflow

[frail wolf]

is there any benefit to this? cause it seems to take more resources/time

[plucky shoal]

https://github.com/AUTOMATIC1111/stable-diffusion-webui/discussions/6509
here is detailed answer. And i see its benefits

[fervent thunder]

Anyone got an access to the GPT-4 api already? I'm willing to buy it.

[errant yacht]

Which?

[errant yacht]

I've been using gpt-4

[fervent thunder]

Do you have an access to it on the OpenAI Playground?

[errant yacht]

Playground?

[fervent thunder]

Yes

[errant yacht]

I go here https://chat.openai.com/?model=gpt-4

Idk what the playground is

[fervent thunder]

hi, I'm new to deforum, need help, just installed it and generate the prompt that come with it without changing any setting and get this error message: Error: expected str, bytes or os.PathLike object, not NoneType.

[silk eagle]

is there a discord for sharing celeb deepfakes

[dusty harness]

how this was made!

https://www.youtube.com/shorts/zfDnsVyzxEI

which library is thisssssssssss!!!!!?

[copper crystal]

no. it's quite illegal and discord has always removed servers involving non consensual intimate media (to put it lightly)

[sinful elk]

is this the right place to look for help, been trying to get auto1111 to install for a few days now but with no luck.
i keep getting these error messages

[silk eagle]

#🤝｜tech-support

[silk eagle]

thts gay af but ig discord's gotta look after their asses

come to think of it I havent been to 4chan since SD dropped. maybe they can help

[agile swan]

are there any models that are good at depicting blurry lights/bokeh effect as seen in the night?

[peak portal]

that's going to be very subjective. this is from the Ultimate Realistic Prompt Model (URPM): https://tripleback.net/public/discord/1682702807.6626532277cf962b36fc97a8df4b94d7c45c9f.png

[fierce musk]

@karmic brook Again, pls can I get the answer? if stable vicuna delta is CC licence and Llama base model is NC then what is the licence for your stable vicuna working model and to what extent it is both open source and business friendly?

[alpine bear]

#1010934719455707218

[fierce charm]

hi all, quick question: when using tile-resample in Controlnet1.1 in Img2Img, what is the benefit for the downsampling setting?

[fierce musk]

Sent. Since you're here, do you use stable vicuna model?

[signal dome]

thanks for your sd tips for getting me installed anything loved ur tutorials

[forest crow]

Can anyone tell me if stable diffusion local would work on a GTX 1080ti?

[silk eagle]

is that the real olivio

damn thats kool

[plucky shoal]

@forest crow yes

[copper crystal]

it loves cuda

[plucky shoal]

and it loves VRAM

[fierce musk]

The question is why it's not stated, but on the other hand it was @karmic brook who posted the announcement, wasn't it? So, my intuition is this: The announcement states expressis verbis that stable vicuna chatbot is open-source. That was undoubtedly about the model and not about the inference software, because only the models can be "trained" or "finetuned" in nn. Next, we know that vicuna delta doesn't work alone at all , so in order to make it working at all, the weights must be merged, and you have to use very specific weights, not any weights. So it's clearly about stable vicuna merged. The trouble is, the aforementioned Llama weights are not open source and they're nowhere to be found afaik, not in public. So, since fruit is SAI stuff, so she definitely knows what she's talking about. This was on an official discord server in an official announcement channel. To the conclusion, I don't think that SAI guys would ever lie to us, or trick us, or gaslight, in any way. Consequently, I'm 100% sure they're competent employees, so there must've been a mistake made somewhere or a kinda sorda misunderstanding. right? 🤷‍♂️

[sharp crown]

Hello all. Could someone please help me figure out a problem I'm having installing SD 1.5. I keep getting this error when running webui-user.bat ERROR: No matching distribution found for torch==1.13.1+cu117

[broken smelt]

I'm guessing this is for automatic1111? Have you installed git and python? If I recall correctly, the git clone should install those two things for you.

[fair nest]

Hello everyone, how is your Saturday? a perfect day to generate images c:

[heady pewter]

Installation on local drive
https://youtu.be/zsezZexAH1Q

[daring girder]

I keep getting this error for stable vicuna gptq:

    1. (arg0: torch.Tensor, arg1: torch.Tensor, arg2: torch.Tensor, arg3: torch.Tensor, arg4: torch.Tensor, arg5: int) -> None

Invoked with: tensor([[ 0.0096, -0.0412,  0.2666,  ..., -0.0141,  0.0032,  0.0063],
        [-0.0015, -0.0194,  0.0394,  ...,  0.0240, -0.0155,  0.0186],
        [ 0.0080, -0.0338, -0.0230,  ..., -0.0223, -0.0049,  0.0009],
        ...,
        [ 0.0219, -0.0059,  0.0252,  ..., -0.0189,  0.0675,  0.0023],
        [ 0.0080, -0.0338, -0.0230,  ..., -0.0223, -0.0049,  0.0009],
        [ 0.0033, -0.0362, -0.0326,  ...,  0.0259,  0.0222, -0.0013]],
       device='cuda:0'), tensor([[ 2121707368,   457542826,  -945195331,  ..., -2073450857,
          1792684905,  1735107715],
[-1987283047, -1743210410,  -711497642,  ..., -1435925116,
          1169652587, -1401772731],
        ```

        [-1822058328, -2057637321,  1764202098,  ...,  -110598587,
           562530637,  -611808531],
        ...,
        [ 1732233627,   693523906, -1782213273,  ..., -1462978455,
           874408505,  -663053667],
        [-1755736713,  2054633383,  1519753877,  ..., -1380239175,
          1882756410, -1450744471],
        [  965107305,  2140358600, -1701992110,  ...,  1971290564,
          1807125381, -1487234918]], device='cuda:0', dtype=torch.int32), tensor([[0., 0., 0.,  ..., 0., 0., 0.],
        [0., 0., 0.,  ..., 0., 0., 0.],
        [0., 0., 0.,  ..., 0., 0., 0.],
        ...,
        [0., 0., 0.,  ..., 0., 0., 0.],
        [0., 0., 0.,  ..., 0., 0., 0.],
        [0., 0., 0.,  ..., 0., 0., 0.]], device='cuda:0'), tensor([[0.0319, 0.0156, 0.0291,  ..., 0.0322, 0.0178, 0.0310],
        [0.0107, 0.0082, 0.0084,  ..., 0.0220, 0.0157, 0.0173],
        [0.0074, 0.0070, 0.0075,  ..., 0.0152, 0.0155, 0.0113],
        ...,
        [0.0058, 0.0047, 0.0043,  ..., 0.0141, 0.0084, 0.0072],
        [0.0057, 0.0043, 0.0045,  ..., 0.0119, 0.0087, 0.0069],
        [0.0052, 0.0047, 0.0048,  ..., 0.0102, 0.0086, 0.0066]],
       device='cuda:0'), tensor([[ 1719170664,  1484158581,  2004248422,  ...,  1720083575,
          1987601783,  1986492246],
        [-2005505945,  2003326567,  1719027540,  ...,  1182169223,
          1466398345,  1449555559],
        [ 1432773974,  1987470965,  2004252006,  ...,  1986549079,
          1466328966,  2018867303],
        ...,
        [ 1431660134,  1734768229,  1735882358,  ...,  1718060902,
         -2041218697,  1450665831],
        [ 1702328407,  1735882375, -2041157769,  ...,  2020050536,
          1484289894,  2004252022],
        [ 1987532151,  1986418311,  1485276792,  ...,  1736861542,
          2004244071,  1751611239]], device='cuda:0', dtype=torch.int32), tensor([38,  7,  1,  ..., 21,  8, 12], device='cuda:0', dtype=torch.int32)```

[daring girder]

I have no idea, it can load the model, but actually generating output fails

using the webui thing

4090

okay

it says triton is not, but it doesn't work on windows

not a question, but if triton is needed, I couldn't figure out how to install it

[fierce musk]

that's moderation ticket line

[ember crag]

Can I make text in SD

[hushed quarry]

XL is decent at it by itself, controlnet/t2i-adapters are good options for the other models (1.5/2.1) right now

[ember crag]

what is XL

[hushed quarry]

SDXL is a new SD model that is still actively in training but can be used in its early stages on dreamstudio, clipdrop, and pickapic. Its a larger more flexible base model coming 🙂

[ember crag]

awesome. also, what tool can i use to take part of an existing image and "extend it" beyond its original borders, or image

[tulip goblet]

hey are there any tools I can use to generate prompts for me

I'm bad at generating detailed prompts

my friend wants me to generate a picture of a capybara standing on a rainbow in the sky

weird

but

i just need to figure out how to get a detailed prompt for it

I just got stable diffusion setup on my computer

so I'm trying to generate them myself

lexica just seems to be finding images

ill try umprompt

i think my prompt is just too wird

weird

can't find anything for it lol

[ember crag]

can you teach me about outpainting?

[ember crag]

Huh, so I'm trying to install openoutpaint. it says I need the API flag enabled. I have --api enabled in my webui start batch file and it doesn't seem to work?

[plucky shoal]

not sure it will be easy as this, but havent you it --api-enabled?

[ember crag]

set COMMANDLINE_ARGS=--xformers --no-half-vae --share --port 7860 --api

[ember crag]

https://i.imgur.com/g223eIo.png

https://i.imgur.com/Kxv1YyM.png

[silk eagle]

"big squooshy boobies" legit creates the best looking boobs in Realsitic Vision 2.0

you can thank me later

"mombod" is also insane at realistic bodies

[harsh scaffold]

Guys can someone help me in summarizing cnn/daily mail dataset

i was wondering whether you could
fine tune bart on cnn/daily mail using only google colab.

i am aware that an already fine-tuned model of bart exists, but i want to reproduce the results/

[knotty blade]

@granite iron Hi man .

[fervent thunder]

I need someone to create me some images with some specific socks, I only have 2 pictures of those socks, I pay whatever it takes. I will pass you the two images first and after that you send me a watermarked image showing that you can do it.

[silk eagle]

socks? lmfao

I'll do it. if the price is right 😏

Hey do you play mincraft 1.20. If so can u recommend any setup for optimal graphics at 4k 60fps

[fervent thunder]

That's more of an #🌶｜off-topic question

[silk eagle]

good lord its not even out yet and all they're doing is adding a bunch of camels

what a joke of a once amazing game turned crap

[fervent thunder]

I can offer 50$, i need the socks in models

[silk eagle]

let me see the socks

[neat condor]

if anyone has a dataset and wants a quick lora/lycoris training on my 3090 for free - let me know, currently queue is open.

[silk eagle]

nah I called dibs G

[neat condor]

u do u brah, i'm offering training to those who already have dataset and no proper PC, giving back to the community and all. didn't even read the convo that happened before in the chat lel.

[agile swan]

does upgrading to python 3.10.9 have any affect on the A1111 experience? Because i had downloaded a previous version of python, as it was mentioned in the SD installation guide.

[fervent thunder]

#🏞｜general-with-images message

[polar coyote]

Anyone here good at making comic covers? I'd be willing to negotiate financial compensation if possible.

[hardy current]

any good APIs that remove image backgrounds?

[broken smelt]

There should be an a1111 extension that implements meta's SAM model that does that really well

https://github.com/continue-revolution/sd-webui-segment-anything

Here it is

[candid marlin]

Hi, I hope you are all well.

I have strong expertise and experience in working for Frontend with React.js, Next.js, Angular, Vue, TailwindCSS, Material UI as well as AI, ML, NLP backend with Python and Django, Node.js, and, used MySQL, Postgres, MongoDB so far.

I have experience in AWS, GCP/Firebase, Heroku cloud services as well.

Please let me know if you are looking for Full Stack Web, AI, ML Engineer on your end.
Looking forward to hearing from you soon.

Regards!
Portfolio: https://0xjonatham.xyz

[silk eagle]

dont upgrade

[silk eagle]

no python? nobitches.jpg

oh yeah u did mention python

can u help a niqqa out for free. theres an improvement I wanna make to a1111 but the code is chinese to me

[exotic kindle]

Hi , help with my problem, I did everything according to the video from youtube https://www.youtube.com/watch?v=ycQJDJ-qNI8&t=4s and I get this error when I start webui-user:

D:\SD\stable-diffusion-webui>git pull
Already up to date.
Creating venv in directory D:\SD\stable-diffusion-webui\venv using python "C:\Users─рэш¤ы№\AppData\Local\Programs\Python\Python310\python.exe"
Unable to create venv in directory "D:\SD\stable-diffusion-webui\venv"

exit code: 3

stderr:
The system cannot find the specified path.

[plucky shoal]

Better ask in #🤝｜tech-support i think if problem isnt in рэш¤ы№ in some special characters? @exotic kindle

[exotic kindle]

this program displays Russian letters in this way, username, I'll try to contact tech-support

[plucky shoal]

i know it is azbuka, just some are special. In tech there are experts

[rough sable]

Guys, I was trying to make an art about programming, like those programming shirts with puns and stuff, do you have any tips on what to put?

[plucky shoal]

biggest problem will be text and numbers, should have be added ex post

[fervent thunder]

One message removed from a suspended account.

One message removed from a suspended account.

[surreal perch]

FAQ: How do I generate images? Is there a bot on the server?

Currently, there is no bot on the server that generates images. However, there are plenty of other ways such as the official https://beta.dreamstudio.ai/ website or by running Stable Diffusion locally using your own hardware! Check out #1080946152318443610 for more details! You can also stop by #1025467151206854736 for any issues you experience while using DreamStudio or #🤝｜tech-support for any problems you encounter while installing it locally!

[spiral turtle]

Hi fam! i dont know if i'm in the right chan but i saw that there is a stable diffusion for music on tiktok is this true or not at all ?

[full zinc]

Anyone got any recomendations on where to begin my AI endouvour

[broken smelt]

The discord has a pretty nice FAQ

I started a month ago and all I really needed as a complete newbie at the time was simply the GitHub of automatic1111. It has installation instructions and everything. Once you install that, you're good to go. The advanced stuff can come later. Seeing what prompts work, what each slider does, etc helps you learn a bunch firsthand.

[full zinc]

thanks a million man, appreciated

[tepid dune]

hello

[boreal narwhal]

3

[fervent thunder]

How do you make Ai art in the group?

[surreal perch]

FAQ: How do I generate images? Is there a bot on the server?

Currently, there is no bot on the server that generates images. However, there are plenty of other ways such as the official https://beta.dreamstudio.ai/ website or by running Stable Diffusion locally using your own hardware! Check out #1080946152318443610 for more details! You can also stop by #1025467151206854736 for any issues you experience while using DreamStudio or #🤝｜tech-support for any problems you encounter while installing it locally!

[plucky shoal]

you mean this? @fervent thunder

[quartz walrus]

same

question]

[tough nimbus]

Does anyone suffer the problem 'lose is Nan' when training SD model using dreambooth

based on sd2.1

[next haven]

"Error: 'CUDA out of memory. Tried to allocate 128.00 MiB (GPU 0; 6.00 GiB total capacity; 5.06 GiB already allocated; 0 bytes free; 5.24 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF'. Check your schedules/ init values please. Also make sure you don't have a backwards slash in any of your PATHs - use / instead of . Full error message is in your terminal/ cli."
Tell me, I don't understand what to do. Except buying a new video card. What does "try setting max_split_size_mb" mean Where to install? What to install? And where is "See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF" ?

[plucky shoal]

try put --medvram or even --lowvram here
set COMMANDLINE_ARGS= -- xformers --medvram in your webui-user.bat <-

[next haven]

Thanks I'll try.

[mellow stratus]

Where did the news channel go?

[tough nimbus]

I have a serious question

What are you doing at that time during that minute or two while you're waiting for the stable diffusion to open and the hour or two while you're training the model

[digital cave]

Hi All, I have problems with xformers and I do not want to go into a wrong chat! can I just go into technical support or is there an extra channel for this ?

[warm junco]

Hey, your right with #🤝｜tech-support

[digital cave]

Thank you 🙂

[torn wadi]

hi

anyone here?

i need help

is deforumation still good, because it stopped working for me today?

[fervent thunder]

I pay $50 for someone who can create a model that can generate images of these socks in models. #🏞｜general-with-images message

[oak holly]

guys I made some models with control net, and then pasted them togther into one picture in photoshop and after I did some inpainting and img2img I get a really low res pic with not much detail,what can I do to make it super detailed?

by 2?

can I dm you so I can send you some examples?

I just can't learn how to do it

thx

[bleak matrix]

Good morning, everyone! I hope you are all well today

Thanks! Still in hospital--it'll take time to recover but it's all good.

What's everyone up to?

[silk eagle]

In the hospital why?

[fringe current]

anybody knows why i cant upscale? it says RuntimeError: "bernoulli_scalar_cpu_" not implemented for 'Half'

[runic igloo]

Guys, hello everyone. I installed Deforum on my laptop. Iron is almost the same. 32gb ram, 3090 graphics card. But on the computer the video is rendered for 30 minutes, and on the laptop for 3 hours. With the same parameters 1 in 1. Could it be that he uses an integrated video card, and not the main 3090?

[sonic reef]

theres a code you can run to check if cuda is being used or not

if it is, keep in mind a laptop 3090 will never be as fast as a pc 3090

[runic igloo]

Dude, how can I check this? Tell me please)

[sonic reef]

I made a debugger a while ago. let me find it

[warm junco]

you can check in the taskmanager if its uses vram and cuda of your gpu

[sonic reef]

def check_cuda():
if torch.cuda.is_available():
print("CUDA is available.\n\n")
else:
print("CUDA is not available.\n\n")

ugh well its just a snippet lol

but yeh CS's method would also work in theory

[rare smelt]

Uh there is guy with name original maks on Instagram he is using 3d deforum right? I maybe wrong can someone tell me? I Just playing with 2D for now 😁😁

[copper crystal]

original._.maks << not really 3d. the transforms that are done on each next frame use 3d math, but it's not like he's working in a 3d environment. It is still 2d work in that sense and it shouldn't be hard for you to get there. You just gotta knowhow to work those 3d frame transform equations.

it's like skewing the previous frame before it starts diffusing it, if that makes sense

[vestal dirge]

after the update it cannot find my GPU even with reinstalled Cuda drivers, tips?

[rare smelt]

Yeah I need to work on it. rotations x,y,z and kinda confusing but trying... Thank you

[sudden nest]

Can anyone using vladmandic stable diffusion webui help me with smth in dms?

[alpine wagon]

Hey everyone, I am new here and trying to figure out something. I have concept art image, and I would like to generate the views (Front, Side, Top) out of that single image, I tried the prompts (Concept Art, Front View, Side view, Top view, Rock Golem, Flame Head, Drawing) in DreamStudio, there is something I may be missing?

Thanks, any advice will be appreciated!

[sly lodge]

DOes anyone know what techniques i could use to change each of these faces over the course of time, kinda like Deforum, but much more directed/sutrctured (I'm somewhat of a noob):

Oh I guess i cannot post images in here?

Is there another channel I can ask this sort of question?

[tulip yoke]

Laptop 3090 doesnt exist

Only 3080ti

[scenic compass]

hey i'm using thelastben's google colab and keep getting this "Bad LoRA layer name." error. anyone knows what this might be? It only happens with a style LoRA that i've beent rying to train

[bright axle]

hi

knock knock

[copper crystal]

whose there?!

shit i let that one linger way too long

[pulsar vector]

Hello everyone! Does anyone have the link to DeepFloyd IF on Discord? Tried searching Google, cannot find it. Thanks in advance.

[karmic brook]

Check #1072229057707659404 for the link

[surreal perch]

FAQ: I'm new here! How do I get started?

Welcome! Start by heading over to #1072220168534642768 to get yourself situated and help find the channels you are looking for! Please make sure you review our #✍🏼｜rules-and-tos and feel free to assign yourself some #👥｜roles as well! Answer any questions your may have at our #1072229020520947753. There are many ways of accessing Stable Diffusion, take a look at #1080946152318443610 to start your journey!

[rotund silo]

Did you get a solution to this commander??🤠

[sonic reef]

Is using the batch img2img still the best option for doing img2img on a video?

Or is transforming to gif and using the gif2gif better

[warm junco]

There is also video2video or mov2mov extension

[sonic reef]

thank uu

[undone marsh]

Does anyone know if there is a way where given an image with several characters you can generate a new image with the same characters but with another position each of them? Appreciate

[sonic reef]

latent coupler

[undone marsh]

Are you reffering to me?

I'm looking at it. It seems it doesn't make what I'm looking for

[sonic reef]

yh it does

u do a new prompt and specify where u want the characters

one on the left and one on the right for example

it only works well once in a blue but its better than nothing

[undone marsh]

Yes, my needs is that it makes the pose that I want each character from the first image

[warm junco]

You would need ControlNet Extension for making poses

[fervent thunder]

https://www.reddit.com/r/StableDiffusion/comments/134oqt9/average_stable_diffusion_experience/?utm_source=share&utm_medium=ios_app&utm_name=ioscss&utm_content=1&utm_term=1

[undone marsh]

Let's say four character on the image are sitting on a sofa and I want that two stand up in the next image but preserving the same style and characters of the image. Is that even possible?

[undone marsh]

Is it for me?

Ok, that wasn't funny

[sonic reef]

bruh

you want six characters in a single image

unrealistic

[sonic reef]

lmaoo

[undone marsh]

It is for an animation. Just want to see if it is possible to change the pose for a new image generation without changing the characters and the styles

[sonic reef]

well that depends. do you have a lora/dreambooth of the characters? or are they popular enough to come up in google search

I recommend doing stuff the hard way and training ur characters via dreambooth. then u can shape them consistently in any way shape or form

in theory it is possible although six in one image? cant guarantee it's going to work

[undone marsh]

The characters are done by designers. For the animation project I have the task to create different frames with the characters changing their pose

[sonic reef]

oic. then make a dreambooth of ur characters

ima post on general-with-images what I was able to do with two custom characters that aren't in google images

with a lot of hard work, let alone animation lol

[undone marsh]

Ok, I see

It is an important project because I could save time to the team

But I guess it isn't possible. Thought also about controlnet from Huggingface

But it doesn't work by giving an image and from there changing the different characters poses

[sonic reef]

controlnet is everything

If ur not using controlnet ur not using true power

U would have to use it in conjunction with either latent coupler or with inpainting

Inpainting is also really good for achieving what u are trying to do

[chilly torrent]

is it possible to train this model for cybersecurity best practices and also to have internet acess to stay upto date.

[sonic reef]

the lesbian kiss I showed you was a combination of latent coupler, inpainting, controlnet and upscaling

[undone marsh]

Can we have a chat?

Would like to know more

[sonic reef]

yeh

although im telling u

what ur trying to do seems very far fetched

[undone marsh]

You mean is pretty difficult?

[sonic reef]

yeah. but if u show examples maybe I can tell u for sure

[undone marsh]

Let me check

[sonic reef]

thats actually me on an average tuesday night. except for the fact he didnt write any negatives like a n00b

[undone marsh]

I can't upload images here

Can I send them to you directly?

[undone marsh]

Sent you a DM

[tropic matrix]

hey guys! How do I check what Stable Diffusion version I am using?

and if I for example install the latest version, can I still use older models like 1.5 or so?

[warm junco]

The Programm itself isnt the SD Version.
The models are either based on 1.5 or 2.1
The sd webui can load all

[tropic matrix]

ok so I guess new features introduced in 2.1 won't work with 1.5 based models?

or somehow sd webui interface will react, reload and change when moving between these models?

I've noticed latest UI differs a bit from the one used some time ago

[warm junco]

Features of the Programm can work with boath.

[tropic matrix]

awesome, thanks!

[sick onyx]

can someone guide me or guide towards creating my first text inversion?

[sonic reef]

Step 1. Don't. Go for Lori or dreambooth

[sick onyx]

can you elaborate? i realy just want to start toying with that not spending too much gpu time on a testrun

[sonic reef]

then go for lora

TI is outdated and not nearly as powerful

[broken gull]

are there any extensions that generate textures? with depth map normal map and what not? ik you have the depth map extension which I have installed but I would like to have it combined

[sonic reef]

look up aitrepreneur's lora tut, it works on a1111

[sonic reef]

sadly not but if u find one let me know. sadly SD's achilles heel are skin textures

[sick onyx]

@sonic reefwatching aitrepreneur right now 😄

[sonic reef]

use his LoRA tut if u dont want dreambooth, although personally I'm a dreambooth freak

[broken gull]

you can easily generate depth maps now but donno about normal maps

[sonic reef]

Oh I thought you mean textures on the image

controlnet?

[broken gull]

wdym? no just basic textures that you can use for 3D models

https://cpetry.github.io/NormalMap-Online/

you have this but it doesn't seem to allow to upload images

nvm

it does allow it

[cedar lark]

Lets draw hands and make AI go crazy: https://www.youtube.com/live/wWjgJtp4YfA?feature=share

[sick onyx]

@sonic reefwhat the difference in gpu time for each of these approaches?

[copper crystal]

doing that with stable diffusion won't have consistency. you'd be better off rigging up a 3d model of the character and using that to create new frames

[grand pecan]

Hey guys! Does anyone know which settings is this model using?
https://replicate.com/stability-ai/stable-diffusion-inpainting

I've been trying to replicate (heh) it for a few hours, but I never got even close to it's quality :/ The closest I got was by using stable-inpainting-512-v2-0 with DDIM sampler, but it still wasn't even close

[vivid crest]

I had installed python 3.10.9, but just see a note today on A1111 repo that Newer version of Python does not support torch - does this apply to 3.10.9 also, or is it fine as it is still 3.10.x? can someone throw light?

[warm junco]

Python 3.10.6 - 3.10.11 works with SD.
But not python 3.11

[vivid crest]

thanks, thats what I thought. Just wanted to check on any performance implications - 3.10.6 vs 3.10.9

[warm junco]

3.10.9 is better for compatibility for extensions and tools like kohya

[vivid crest]

great, thanks!

[undone marsh]

Can I DM you?

[distant isle]

guys can someone help me to install stabble diffusion? im trying to do it has 3 weeks

[copper crystal]

afk right now. wouldn't be much help. i can tell you though, that the capability you want isn't stable diffusion.

i'm also a big believer of doing help in public so that others may benefit from it or correct me where i'm wrong in my advice

[undone marsh]

I thought using Automatic1111 with Dreambooth fine tuning with the characters and from OpenPose to get all the different gestures. It isn't som realistic image it is cartoons like The Simpsons

[distant isle]

im actually needing some plublic help here then..

[undone marsh]

Thanks again, for your support

[distant isle]

i need to learn stabble diffusion becouse im having this job demand , and i cant even install it

[copper crystal]

dream booth may help you but you wont' always get great results still. i don't know much about DB training but from what i understand is you want good regularization photos as part of that process.

[undone marsh]

I just need the characters to have different gestures. I guess the best approach is to do it individually each character or maybe with latent couple and select each character. The animators would do the animation

[copper crystal]

i would probably explore inpainting. do a rough painting of the new gestured keyframe / keypose and then inpaint that so it blends and isn't roughed out. a dreambooth model for this would likely just create entirely different versions of the character, not a consistent next pose.

[undone marsh]

I thought training Dreambooth with for example 40 images of the same character with the same outlook would preserve the outlook. Clothes, hair,... The character have always the same outlook. Think of The Simpsons as an example

[copper crystal]

i got into dreambooth training once captions and regularization images were introduced. I've been doing it that way the whole time. People keep telling me it's possible doing it wiht just an image set, but i don't know how. i resize and crop each image myself, clip interrogate them all for captions, then go through and hand edit all those cpations for higher quality. then i use all that to generate 100s of regularization class images before i start training

i kind of stopped using dreambooths too as loras create the same quality with that process

[indigo ermine]

VOX released a feature about Stelfie and his process using SD. Very impressive. https://youtu.be/K0ldxCh3cnI

[celest forge]

Dont know if this is the right place to put this:I have the dreaded “Torch is not able to use GPU” issue when I run stable diffusion cmd. Followed by press any key to continue. Is the site down for downloading the certain file to get GPU to communicate?

I noticed that when I delete the VENV file and re-run stable diffusion. It give a lot of errors when downloading.

I have tried it on my desktop and laptop now. So it must be a source problem?

This is the link I recon it tries to download but fails because it’s down: https://download.pytorch.org/who/cu118

Maybe it’s used for the communication of the GPU CUDA cores and the stable diffusion software? This is my understanding (correct me if I am wrong)

[broken gull]

#📣｜announcements message

I don't have access to any of these channels

[feral cairn]

The channel links in #📣｜announcements all show No access

[signal dome]

ping

[broken gull]

heya flex

[signal dome]

i am early member yet no access

[feral cairn]

oh, we get to watch the channels get opened

[timid bloom]

ehm

yeah

access dont work for now

[broken gull]

I am not a early member but a beta user

[feral cairn]

think we've gotta wait a minute because they set up the announcement before the neccessary channels

[signal dome]

agh i remember coming sd around 2600 members

that time

[timid bloom]

yeah

good old times 😂

[broken gull]

idk anymore tbh lots has happened

[sick onyx]

is it possible to use several different text inversions in a single prompt?

[uneven heath]

Oh

[frigid gull]

I don't have permission for showdown or pantheon

[fervent thunder]

i got mentioned

https://tenor.com/view/rachet-cat-attack-death-gif-22457670

but im not early membr

[mossy onyx]

Hi

[karmic brook]

all channel perms should be fixed. Thanks for your patience

TLDR: bot gens- only for /earlymember and /stablesociety
#🏅｜pantheon and #1100484581037195384 - for everyone

[winter pike]

i wish they would have included beta members too damnn

[karmic brook]

slowly but surely. . .

[coral brook]

Does Stable Diffusion Automatic 1111 work with AMD? RX 6800?

[alpine bear]

There's a version of a1111 specifically for amd cards

[crude stream]

Hey person reading this, I haven't used SD for quite some time but wanna try some things with it now, what is the most up to date version?

[white quail]

Hey guys, is sd 1.4ckpt safe?

[broken smelt]

probably but if not then a lot of people are compromised and we just dont know it 🤷‍♂️

[storm shard]

I guess beta testers from the second/third wave aren't considered as early members? 😢

[sonic reef]

there are no unsafe ckpt's reported as of yet

[shadow sundial]

how to know which version i have? i have radeon graphics card too but i dont know if i have a1111 for amd users

[alpine bear]

https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Install-and-Run-on-AMD-GPUs

[shadow birch]

Hello everyone, is there really a possibility that we will no longer be able to use Stable Diffusion through Google Colab?

[sonic reef]

someone said they blocked gradio

and yet some ppl here had the audacity to kiss google's ass for it lamo.

unfortunately, free image generators are on the way out probably due to legal issues

[copper crystal]

the problem isn't that all ckpt's are unsafe. its that someone could pin a bunch of executable data onto one as a payload, and then another process could find and use that binary data

[sonic reef]

but noone has

Out of the hundreds of thousands of civitai downloads not one has reported a virus

[copper crystal]

yeah you just gotta be smart about where you're getting them from. major sources are fine

[swift roost]

Can modern AI models be run effectively on CPU using Intel Ivy Bridge Xeon processors or would the machine need a CUDA GPU? Wondering what I would need for exploring AI on my own hardware.

[copper crystal]

a jpg could have binary data pinned to it for that matter. people used to payload dump using discord. your temp folder would have malicious scripts and a game crack would find those and run em

would be an easy enough attack to launch. takes smart deployment and timing. it's an obvious attack once it's launched though.

[sonic reef]

no

it can run it, just not efficiently. I would say minimum 10 minutes per 512x512 trolol

or maybe 5?

as opposed to 2 seconds on a 3090

[swift roost]

Multi-thread? Mainly looking into running SD, LLaMA on the blade hardware (3x BL460c G8 in single chassis) I am getting from a friend soon. If not, could stick to having it on a desktop with a GPU installed.

[sonic reef]

yeh I have 13th gen and no amount of threads will ever match cuda

cpu is always a bad idea when it comes to AI

[swift roost]

Makes sense, GPUs are great for that sort of thing

[warm junco]

the amd version is called stable-diffusion-webui-directml

[shadow birch]

but it works fine...

[undone marsh]

Hi everyone! Is it possible to save the images generated with OpenPose on Automatic1111 in a APNG file?

[sonic reef]

is there a discord dedicated to AI porn

[swift roost]

How about text-based? Which I would use the most on servers anyway.

[sonic reef]

if ur talking about llm or tts then the same is true

anything that uses tensors as its processing unit is fucked when it comes to cpu

[swift roost]

ah, ok

[white quail]

Do I have to download the file such as (full model fp16) when downloading models from civitai?

[copper crystal]

There are but they tend to be restrictive since discord itself doesn't allow portrayals if real people at all. Nor should they

[sonic reef]

dont necessarily need to be deepfakes

[warm junco]

Yes but dont ask here for that pls

[white quail]

Should I download the Pruned Model or the Full Model fp16?

[static roost]

У меня ничо не работает , в интернете написано что бля якобы тут картинки в видео можно превратить , хуйня ваш искусственный интеллект

[placid comet]

huh

[sonic reef]

chill out putin

hook me up with a hot russian babe

[vestal dew]

SD moved quickly from new and hot to stagnant to dead.

[lofty zephyr]

That's a bit hyperbolic isn't it

[vestal dew]

nope

Look at even the chatter in channels like this one. Used to be a lot of people talking, asking questions. Now...no more

[amber aurora]

what's the new hot thing that replaced it?

midjourney? I think thats really lacking in its ability to control it

[fathom marten]

Hey

Just figured out how to get inpaint to work with automatic's webui (stable-diffusion-webui-directml) and AMD (rx570 8gb here)
Symptom: Inpaint just work and do nothing, same pic as original. No error.

Just wondering if the solution is known, can't find it anywhere else

anyway, the solution is, somewhere, in :
COMMANDLINE_ARGS= --no-half --precision full --no-half-vae --opt-sub-quad-attention --opt-split-attention-v1

pin this 🥲

[finite cloak]

SD community spread out* rather. Conversations are now split between tons of varying communities (here, r/SD, civitai, SAIL, idek how many more big ones)

and in terms of development upgrades... I mean, 👀 DeepFloyd a few days ago, SDXL expanding beta access literally earlier today??

[copper crystal]

winter vs spring

it'll surge again

[weak wharf]

Canada needs to get its act together.

https://www.cbc.ca/news/canada/montreal/ai-child-abuse-images-1.6823808

Look at the link I just posted, it is absolutely possible she is an AI, look at the dimensions of the photo, check for exif info, on all of the images things of that nature.

[tacit fable]

Hi all, I am currently having a Stable Diffusion problem right now; I really like this image Stable Diffusion made but I want a less lewd version, so I need to find away to cover up the girl's chest without simply making it look all messed up

[broken smelt]

ur gonna wanna stay away from prompting things like large breasts then, cause the ai is gonna wanna draw in some cleavage lol

[sick onyx]

for training lora, is it better to have super varied pictures, poses, and what not or should i try to somewhat go for pics that are going to resemble the scene the lora will be used in anway?

[vestal dew]

anyone usee this VLAD? Im trying to figure out what analyze does

[fervent thunder]

hi

when i try to donwload it says pyuthjon wont open

can someone help me

[broken smelt]

have you installed python 3.10.6-3.10.10 and set its path correctly?

[hardy violet]

im also having issues i set the paths but it says error 0900

launch unsuccessful

[fervent thunder]

Yes

[chilly crow]

Stable Vicuna seems to be the best rn

[sly peak]

Hello - I'm a paid member over at DreamStudio - I'm assuming that doesn't mean I get access to for early-member and stable-society groups? If not, how would I do that?

[white quail]

What's dreambooth?

[alpine bear]

Early-member cannot be obtained, as it is specifically a role held by server members who were here when the server was first created.

Stable Society is a role given to those who win a POW or COW contest here in the server.
#⭐｜pow-info
#1087493421209485393

[sly peak]

Got it - thanks!

[brisk brook]

Hello! I just joined in search for new knowledge and to share anything I find while using stable diffusion. My focus is going to be in generating gray scale textures intended for real time vfx. I ll share any progress related and any tips or advice is welcome. 🙂

[white quail]

Is VAE broken? Can't launch stablediffusion with it on it

[simple siren]

Can SD be used to generate AI profile pictures from your real pictures like all those recent websites that offer it for like $50?

[empty fern]

is there a UI for stablediffusion that's not the A1111 webui, or a webui at all, especially not one based on that steaming heap of sh*t called Gradio?

I'm tired of the constant crashes and wild piles of errors

[warm junco]

I tried several uis with gradio and never got an error

Other webuis would be InvokeAI, comfyui and nmkd that dont use gradio

[empty fern]

Maybe I'm just unlucky, gradio tabs constantly crash for me no matter the browser I use

I'll check them out, thx

[warm junco]

Thats very unlucky, i had no problems with Firefox and ublock origin, (disabled for webui)

[empty fern]

ah nmkd is written in C#, blessed

exactly what I was looking for, thanks again

[warm junco]

No problem 🙂

[strong grove]

hello everyone, just out of curiosity, has anyone got this model: Customsmix-05 | hash : 00ed673379 ? I saw amazing results for anime but no results on Google and it was on Civitai before ^^

[median rain]

Hello im french

[left pawn]

Hello, I need a volunteer for an artwork

[umbral vigil]

hii

[nimble matrix]

hello

can some tell me where i can find how to use stable diffusion

[old tinsel]

any suggestions for abstract art models? everywhere i look its just cat girls and anime

[young ivy]

good afternoon

[regal snow]

I'm familiar with quicksettings list (things that appear at top of the UI). What I was wondering is whether there is a native means, or extension, to make a checkbox up top that toggles multiple settings at once?

Specifically now that the new ControlNet inpainting and tiling models are so powerful, I thought it would be neat to have a simple "Inpaint" or "Tile" checkbox to automatically toggle ControlNet, pre-processor, and model.

[coral meteor]

Anyone doing in-person events / meetups re genAI in London?
Or know what discords / twitters to follow to find them?
Also any co-working spaces / groups / hacker houses for GenAI?

[eternal violet]

what's the final fix for the libcudnn error on a fresh install?

lol, right on, understand friend

believe its the same as this one

https://github.com/vladmandic/automatic/discussions/540

even tho i'm not using vlads

think it's just missing cuda libraries but wanted to verify when anyone who might have come across it first

[silver grotto]

Can some people not see the #1100484581037195384 channel?

[eternal violet]

installing the libraries from this guide for Ubuntu currentlyhttps://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html#ubuntu

50% done

python, pip, and nvidia drivers are all working tho

that worked! adding this to my fresh install guide for linux/A1111

[wild steppe]

It should be visible to everyone 👀

[eternal violet]

i can see it and i'm new here

[umbral vigil]

hii

[narrow nimbus]

Hello

where are you from guys?

[clever cradle]

Anyone know where I can get a decently cheap PC with 8 GB VRAM?

I got laughed at for having a 4GB graphics card that only supports 1080p from 2014, so I assume they exist

[crude cove]

Hey everyone!

Is anyone here from Paris? We're organizing a GenAI Paris meetup

[gloomy flower]

Where us the gpt bot

[cobalt cedar]

Q: If I found a spreadsheet bug in the GPU comparison spreadsheet from @weary robinchall (linked in the benchmark survey), where would I post it?

(It seems to charge the price per kwh per kws (kilowatt-second) instead. Meaning it weighs the power price by 3,600x as much as it should.)
Edit: This basically completely changes the GPU ranking to something that makes sense. Didn't make sense to me that the A100 would be most cost effective.

It also seems the fp16 column divides by the TDP of the card instead of its price.

(missing $ sign to lock column)
Link is: https://docs.google.com/spreadsheets/d/1Zlv4UFiciSgmJZncCujuXKHwc4BcxbjbSBg71-SdeNk/edit#gid=0

[vapid tangle]

Is there anywhere I can use SD with clip guidance? It's just nowhere near as good without it. Seems like SD's own website doesn't even use it anymore

[urban fossil]

Anyone have any idea why Automatic1111 only uses 6gb of Vram max when Im running a 500 frame deforum generation with a 3090 gfx card? Figure it should be maxing it out.....

[limpid tusk]

are you using xformers?

[somber cipher]

Hi guys , i dont know if am i right chat here?! need help with installing stable diffusion , python on my PC. everything installed ,seems like ok, but python getting error 9009. reinstalled few times, again error.

[urban fossil]

I am not, updated Auto1111 today to pytorch 2 (?) and was led to believe they are no longer required...

[limpid tusk]

ah then maybe not, you could give it a try, i still have mine enabled after updating

[limpid tusk]

#🤝｜tech-support would get you a better/faster answer

[somber cipher]

thank you so much

[limpid tusk]

np, youll have to restate your issue in that chat

[urban fossil]

Just reinstalled them and is going at 9.8gb now which is obviously better but still feels like a lot is being left on the table....

[limpid tusk]

is it a decrease youve noticed in the last few days or so?

Because i was having performance issues until i disabled the "a1111-sd-webui-lycoris" extension, the may 1st update caused big performance issues for some

[urban fossil]

Only just getting back into it after a while as been too busy so can't really say

[torpid igloo]

where was the thing where you can select the models of characaters? cant find it

[soft vessel]

f

[fervent thunder]

https://civitai.com/ ?

[quasi jacinth]

Hola, alguien tiene el link del Discord de Dot CSV? 🙏

[languid pendant]

I saw a month ago someone making a really horrifying ai genned video of will smith eating spaghetti, and now i saw a relatively good looking movie trailer. What do yall think ai will be able to do in another few months from now? What about years? Are we right on the cusp of ai becoming smarter than we are?

[waxen cave]

thsi is driving me nutz i have been trying to train stable diffusion on a brand of vapepen for last 45 days nothing seems to work

[fallen copper]

ok this is gonna sound stupid, but I just got auto1111 and want to learn how to input myself into a model like revanimated. Is that possible? How would i do that? thanks

[fallen copper]

lora preferably, i just dont know how to use a lora with a model like, lets say, rev animated, dreamshaper, etc.

thank you so much for your. help man!

[gray heath]

hola, soy nuevo en discord en general, alguien me podría dar una guía/resumen para usar la IA porfa?

[kind kraken]

Anyone considers themselves an expert in this Stable diffusion and Civit Stuff? If so, DM me, and I can pay you for a session for some questions I have on this topic.

[tiny flicker]

Hello

[old imp]

.

[fervent hare]

anyone know of an extension that auto tiles images together?

[gloomy stratus]

any mod online? 🙂 would love to know if you would be open for a community partnership and whats are the requirements to be listed as co:creators.

[river eagle]

Imfao

[shadow birch]

well, then colab even ended the free service for those using stable diffusion.

[safe sapphire]

Specifically web uis. Diffusers is okay, unless they've changed their policies!

[shadow birch]

and do you know any good diffuser links to use?

[safe sapphire]

For example, diffusers may be set up in colab like this:

!pip install --pre -U -q diffusers/ xformers accelerate ftfy huggingface_hub transformers triton omegaconf safetensors
from diffusers import StableDiffusionPipeline, UniPCMultistepScheduler
from huggingface_hub import notebook_login
notebook_login()

pipeline = StableDiffusionPipeline.from_pretrained("stabilityai/stable-diffusion-2-1", torch_dtype=torch.float16, scheduler=s)
pipeline.enable_xformers_memory_efficient_attention()
pipeline.to("cuda")

images = pipeline(prompt="whatever", negative_prompt="whatever2", guidance_scale=7.5, num_inference_steps=25, height=768, width=768).images
display(images[0])

[shadow birch]

ok, ty man!

[next fractal]

What are the best sites for using stable diffusion models? I already know prodia , openart , leonardo , randomserd ,pixelai and another one i forgot the name of but most of them have very less models and require ingame currencies... Can someine please give me more sites with some anime models? (Excluding google Collab)

[empty rivet]

Hello guy, I just want to know, for "stable diffusion checkpoint", how can we access to V 2.1, for now we only have 1.5V and that's all :/

[shadow wraith]

is there some kinda "inpainting img2img", similar to the way img2img works but just for the masked area of the input

ill have to look into it, i never really dove too deep with controlnet

[abstract quarry]

don't get that. In auto111 this is the default way if you do inpainting

[shadow wraith]

oh I see, is it the inpaint upload?

[abstract quarry]

just mask the image and select "original content"

[sudden pumice]

hi, guys. Idk if it's off topic here or not. I'm kinda new here but does anybody can explain how to make a smooth transition between the original photo and the animation where the pic starts changing too fast? Is it about keyframes?

[sweet jetty]

Guys I got my final version of chatgpt as a prompt engineer.

[hollow flint]

anyone know of a list of stable diffusion models along with their strengths and weaknesses

[fervent thunder]

https://www.pcgamer.com/researchers-translate-brain-waves-into-scarily-accurate-images-using-stable-diffusion-ai/

Holy shit this is actually insane

[sonic reef]

clickbait alert

[fervent thunder]

i mean just look at the images

cant post here cause of anti img rules

[sonic reef]

idk, im always sketchy about these futuristic "mind reading technology" articles like neuralink that end up being a big ol' nothingburger

time and time again

[fervent thunder]

Yea, just take 2 seconds to look at the image comparisons of what was shown to the subject and what stable diffusion produced.

they are scarily close for something we're literally starting to explore

specifically the images halfway through the article captioned

Shown images (top) and those reconstructed through stable diffusion by reading fMRI scans (bottom). (Image credit: Takagi & Nishimoto)

[sonic reef]

oooh i see now

I thought it was supposed to read your mind to recreate your dream images

that's kool af

[fervent thunder]

yea it's crazy cool

and who knows maybe one day it'll expand to thoughts.

but being able to pretty accurately recreate what you see just through your brainwaves is fucking bonkers

the absolutely unfathomable speed at which ai is evolving is mindblowing

a couple months ago i could install a local chatgpt that wrote crap, now i can run pretty close to 3.5 models locally on my gaming pc. it's insane how fast this is going

same with ai art, i remember using midjourney just a year ago and it produced utter trash in most cases, now it's outputting masterful art in all categories, and SD tracking a bit behind but with waaaay more control

[sweet jetty]

his output:
positiv prompt

award-winning (mountain landscape:1.4) with (Japanese cherry blossoms:1.6) in full bloom, vibrant pink petals, lush green foliage, serene atmosphere, distant misty mountain peaks, crisp blue sky, (golden hour sunlight:1.5) casting soft shadows, stunning bokeh effect, traditional Japanese scenery, reminiscent of Hokusai's art, captured with a high-resolution Canon EOS R5, wide-angle lens, vivid colors, lifelike textures, hyper-realistic detail, picturesque composition, trending on Instagram, raytracing, Unreal Engine 5

negative prompt:

cityscape, urban, polluted, ((winter)), (((desert))), garbage, graffiti, low-resolution, pixelated, oversaturated, dull colors, overexposed, heavily filtered, motion blur, tilted horizon, (((no cherry blossoms))), cropped, bad composition, watermark, artist name, cartoon, 3D, monochrome```

Request: a japanese cherryblossom on a mountain

results in #1070829832151240734 > nature

[fervent thunder]

The real deal will be having a program pull all the metadata of starred images on this discord using the same models and producing a prompt base out of those for quality images.

[signal dome]

ping

[wild steppe]

Join us at our stage event today where we talk about SDXL and experimental models! ⚙️

https://discord.gg/5mXj4HMa?event=1103369195665760276

[fervent thunder]

What's happening, why bot online ?

🐧 or was he always online and I just didn't see it

[wild steppe]

All info is in our announcement, bot is slowly being opened to more members: #📣｜announcements message

[fervent thunder]

Depends on a model, if it's a model you downloaded from https://civitai.com/ , they usually include license to follow.
No idea on base models

[broken smelt]

id read the dreamstudio license in that case

[young ivy]

anyone know what the new "Model Compile Mode" is in vlads auto1111?

[wild steppe]

#1029055412764422214 is open for your questions ^^

We're LIVE! #1005545148211527800 🔴

[split sonnet]

hey.. can anyone tell me ho to create arts like this style?? i dont understand if its disco diffusion or stable diffusion model??

i didnt realise we cant chare images

share*

[austere parcel]

chat test

[wintry stream]

#🏞｜general-with-images 🙂

I mean, I don't get the point of this but w/e lol

Thanks?

[shadow birch]

hello everyone, does anyone here use colab with a paid plan?

[sacred jay]

I do

Anyone here know what the best open source music ai model is right now? I'm hoping to generate music that's like a soundtrack I really like

[mighty swallow]

where can I find a model in wich I can add different attributes to my dog while he is in the same pose

[twilit fern]

Im also interested in this. I used to be a music producer but Ive lost passion for it the last decade . i bang out the occasional EDM club-bangers on my Novation Circuit Tracks , but Im not very serious atm. I just love to experiment with technologies. Hybridize them into something new and fresh

[pastel ice]

can i get some help with my issue?

[shadow birch]

Is it free to use webui stable diffusion in the paid plan?

[sacred jay]

It would require compute credits still, but also heads up that colab has started restricting use of webuis like that

[shadow birch]

I understood. And do you use the paid plan of colab to use Stable Diffusion? my question is if the intermediate plan is enough

[fervent thunder]

is there a video of todays stage ?

i had to leave midway through

[sacred jay]

I'm using it for deforum, so yes, but I'm not sure about the intermediate plans. Unless you're using a custom checkpoint, you could probably get by with the stability platform

[pastel ice]

Why is my webui-user.bat not working

[neat cypress]

first time in the community method i came up with . what you think about it? "How To Find Best Stable Diffusion Generated Images By Using DeepFace AI - DreamBooth / LoRA Training" : https://www.youtube.com/watch?v=343I11mhnXs

thanks

[boreal stump]

where is the best place to ask absolutely newbie questions about prompts?

[shadow wraith]

what are those small round purple marks on some of my results? I've never really thought about them because they dont appear often

[kind kraken]

Hello, if anyone can help me with these, we can discuss and amount and happy to pay.

Sure. Here’s are a bunch of things I need to do, lmk if you can help me with any of these.

1. I have a fitness model pictures in a folder, and I have a lot of it. Highlighting her face, shoulders, legs etc…

Now, I want to create images with different poses with it using stable diffusion, so that the model don’t have to take pictures anymore. Also by using control net with SD, I want to generate images with specific poses that the model hasn’t poses for before.

2. Now I have the models image, I want to change the background of the model. So 1 model with the same pose, different types of background. The background shouldn’t be much different from one another, ie they should be more or less a little variation of each other.

[sonic reef]

check ur dms

[sick onyx]

so is it/s all there is or does batch size influence it aswell?

[alpine bear]

The higher your batch size, the more images are made at once. This uses more vram. While it technically decreases your it/s, your overall it/s can dramatically improve if you have enough vram.

[sick onyx]

iam getting 2s/it with a batch size of 30 on a 4090

[alpine bear]

30x2

You are effectively at 60its

Wait

2s per it

Gotcha

[sick onyx]

yes

[alpine bear]

Yeah I recommend a batch size of 12. Thats what I use on my 4090.

You will hit around 48it/s

[sick onyx]

what it/s do you get?

gradient checkpointing off?

[alpine bear]

I get 4it/s on batch size 12

4x12=48its effective

[sick onyx]

gradient checkpointing on?

[trim sandal]

What's the best way to generate openpose

[zealous sphinx]

is it possible to blend a landscape image with a separate image of a character in stable diffusion?
I know midjourney has a blend option but what about SD?

[warm junco]

They come from a missing vae file

[warm junco]

OpenPose Editor or Posex extension

[tulip yoke]

I get 3its on batch size 8 and 7 on 1

[abstract ice]

meh

[tired condor]

@wide plover

[abstract ice]

how come i dont have the early member or beta member roles? i have been here forever 😛

[sweet nacelle]

can I create 2d assets in stable diffusion for rpg maker games?

[neat condor]

yo have one quick spot open for training on my 3090 - anyone with dataset but not a proper tech that needs training lora/lycon?

[thorny sorrel]

does anyone here have any experience with non-UI stable diffusion?
like, directly from Python code

[hollow hinge]

I type in stuff to get a car, ive tried different models, and i keep getting multiple of the same car? why is this?

https://queenang.net/shx/e01j.png

[sudden ruin]

Because models are trained on square images and youre trying to generate something wider, duplication is a known issues for non square resolutions sadly

[hollow hinge]

Oh okay Thanks

the first one i did square and it had only one

[sudden ruin]

It sometimes helps to describe more of the scene so the AI knows how to fill empty spaces

[hollow hinge]

Yeah i tried to have it in a driveway of a house, i got a field or whatever instead

[sudden ruin]

After some time playing around youll know how to avoid and deal with this

you dont have to make the image exactly square, but being close to square surely helps

[warm junco]

@hollow hinge 1.4, 1.5 model are trained on a resolution of 512x512 and the 2.1 model on 768x768.
Staying closer to these resolutions will give you non duplicates

[hollow hinge]

What ckpt would be good to try?

i have 1.4 1.5 and protogen v22 per the tutorial

[warm junco]

Depends on what you want to do.
Protogen is a good start. Also Dreamshaper v5 can do a lot of variety

You can find a lot of models for different purposes on civitai.com

For photorealistic stuff there is Realistic Vision for example

[sudden ruin]

Im in love with edge of realism and analog madness right now for realistic stuff

Illuminati Diffusion also was great when you still could download it

[safe sapphire]

I use diffusers.

[thorny sorrel]

oh, can you help me out?

pipe = StableDiffusionPipeline.from_pretrained("my_model_PATH", torch_dtype=torch.float16)  ```

is this syntax incorrect?

[safe sapphire]

Looks correct, what doesn't work?

[brazen atlas]

so how where do ug uys sell your ai art please tell me

and anyof u here uses vam? virt a mate?

[tardy schooner]

hey guys - what's the best way to generate some stylized graphics but using someones face - inpainting? (lets say for example if I would want to make my friend as some fantasy character but you can still tell this is him by the face of this character

or I can use whatever model with img2img - cause it's kinda loosing the source face then

but it looks like photo just pasted on image - or I'm just using wrond model maybe

0,75 - sorry propably stupid questions - I try to always look for info myself and with prompt->img I have no problems but this one giving me trouble

nah still looks like cutted out photo pasted on AI art - I'm doing something wrong

Ok I think it's because I'm using regular model instead of inpainting - just inpainting mode

ok downloaded inpaiting and it's better - will expermient further thanks 🙂

[wind kernel]

anyone using controlnet? finding it significantly reduces output quality

[fervent thunder]

https://tenor.com/view/happy-dance-gif-23574866

[simple turtle]

AWS talked about Stability using their Sagemaker platform to train GPT-NeoX models back in Dec 2002 (https://aws.amazon.com/blogs/machine-learning/new-performance-improvements-in-amazon-sagemaker-model-parallel-library/). Is there any code you can release on how to exactly integrate the two? They provide GPT-2 and GPT-J examples ion their amazon-sagemaker-examples repo, but their seem to be nuanced enough differences in architecture that porting those examples to gptneox isn't straightforward. Also, if there is a better channel to be asking this, please lmk.

[abstract quarry]

which one?

[wind kernel]

by mikubill, sd-webui-controlnet 1.1, used with realistic vision from civitai

[abstract quarry]

that's just the ui

[wind kernel]

oh, openpose

[abstract quarry]

worked fine for me..But in the end it's like increasing guidance scale: adding more control to the output might decrease quality

[wind kernel]

played around a bit already, but good to hear its working for others, means i just need to experiment some more

[sweet nacelle]

can I create 2d assets in stable diffusion for rpg maker games?

[abstract quarry]

I would say as more "unusual" your pose is as worse the image. But I got good results so far. It's rather that SD is not good in creating characters from far away in general

[abstract quarry]

I don't think you can create assets fully automated, but you can definitely streamline asset creation

for character faces it's probably easy

for tilesets and animationa is probably more difficult

[wind kernel]

it captures face only poses well, when its full it has trouble with the face

[abstract quarry]

yes, that's normal

use upscaling and img2img or inpainting to fix faces

but that has nothing to do with controlnet - SD is just bad in faces when drawing full body shot. It's just too low resolution

[sonic reef]

youre doing it wrong

check #🏞｜general-with-images to see how it can actually improve quality

[sweet nacelle]

im just into 2d assets that are like the houses in pokemon for an example

[fervent thunder]

When I use extras and Batch from Directory, to upscale many images, some repeated images are created, for example, the first image is created many times.
How can i fix that

[shadow birch]

hello everyone, does anyone have a link to those lighter versions of stable diffusion to download, which work on weaker PCs?

[copper crystal]

open the folder and check out it's contents. you might have multiple copies of the images saved

[warm junco]

you mean an other webui ?

[shadow birch]

I remember that I once found the website link with 3 different stable diffusion models

For download

[warm junco]

so you mean models?

not the programm

[shadow birch]

are simpler versions

[copper crystal]

probably smaller models that have been pruned down to 2ish gb

[shadow birch]

no no

yeah

[shadow birch]

👍

[warm junco]

Models like deliberate are 2gb

should work fine on lowvram pcs

on civitai.com you can find pruned models

[shadow birch]

sorry, I expressed myself badly, they are not models, but versions of Stable Diffusion

[warm junco]

okay the normal stable diffusion works on low end pcs, there is an other one for low vram but its old and outdated

[shadow birch]

ah yes, these are the ones i was talking about, thanks. are outdated

[fervent thunder]

#🏞｜general-with-images message
This

[copper crystal]

i think you're talking about models/checkpoints. They all load into the software. there many different implementations of stable diffusion but they generally all load the same models. The files you're seeking are models for stable diffusion.

[copper crystal]

you talking about systems like disco diffusion here?

[silver kettle]

are artist styles back in the new sdxl?

[shadow birch]

I had expressed myself badly. I was referring to simpler, lighter versions of Stable Diffusion that would work on my PC, which is a weak PC.

[peak portal]

this is true. when i render at 3840x2160 it's soooooo high quality but the duplicates are horrendous lmfoa

[short marten]

hi

anyone here who use deforum stable diffusion ?

the google colab notebook

[wild steppe]

plenty! we've got people using deforum in #🎥｜animation & the Peter Gabriel's DiffuseTogether challenge #1097876660101066854 ^^

[short marten]

ok thanks

[thorny sorrel]

uses more than my entire colab RAM

[safe sapphire]

Ensure num_images_per_prompt, height, and width aren't misconfigured.
There are many options that change how memory is used, see
https://github.com/huggingface/diffusers/blob/main/docs/source/en/optimization/fp16.mdx
Though,

• If there is a black screen, retry without xformers
• pipe.enable_attention_slicing(number) the number can be anywhere between 1 and 8

[soft vessel]

How early did you need to join to be a beta member? I was on since early August and I don't have the role.

[thorny sorrel]

Oh no i cant even get there, Its failing to load the model i think(?)

[safe sapphire]

Is it in the diffusers format, .ckpt, or .safetensors?

[thorny sorrel]

ckpt

so like, "openjourney.ckpt"

[safe sapphire]

it is available in the diffusers format. specify "prompthero/openjourney" instead of "my_model_PATH"

alternatively, though it might or might not be possible to use the .ckpt directly, there is at least a program that converts .ckpt to the diffusers format, inside diffusers
like so: !python diffusers/scripts/convert_original_stable_diffusion_to_diffusers.py --original_config_file config.yaml --checkpoint_path my_model_PATH --dump_path model --image_size 512

diffusers automatically downloads it if specified that way...

[brazen atlas]

hi so if i want to use contorlnet to changetthe face mimic to tounge out can i dot that?
f

[quiet trail]

Hi! Say now, a few days after release, did anyone get the new upscaler to run under auto1111 or can point me to an extension ?
Thanks
https://huggingface.co/stabilityai/stable-diffusion-x4-upscaler

[shadow wraith]

why is inpainting leaving a grey spot where the mask is

[hazy sable]

i just finished working on my first extension. I'd really appreciate feedback so I can try to improve it. thx!
https://github.com/Leodotpy/sd-pixel

[fast sage]

This is going to be a stupid question but I'm using AUTOMATIC1111 webui for generating, I've rolled tensorflow and hugging face out before and didn't get as good results, if I wanted to keep the models and LoRA that AUTOMATIC accesses, but move over to cli (linux headless gpu server), what are my options?

• Can I use safteyTesors / LoRA models with something else to produce easy / same results?

• Could you just point me in the direction on how to get started?

[balmy mantle]

Does it have some effect on the output image, in which order the different controlnet processors are set in the configuration?

[floral kestrel]

hello there, im trying to use interogate CLIP with an amd gpu , is this something that causing issues ? I got RuntimeError: Expected tensor for argument #1 'indices' to have one of the following scalar types: Long, Int; but got torch.HalfTensor instead (while checking arguments for embedding)

[young ivy]

you mean the official releases?

[fervent thunder]

hi guys, what section is for help on Stable Diffusion extentions?

[young ivy]

that would depend on the ui

[floral kestrel]

yeah sorry was trying with automatic1111 but i tried CLIP Interrogator 2.1 on hugging face and it dont work neither

[fervent thunder]

how did u install Automatic1111?

i get this: Traceback (most recent call last):
File "C:\Users\Vids\stable-diffusion-webui\sd_civitai_extension\install.py", line 7, in <module>
from launch import run
ModuleNotFoundError: No module named 'launch'

and i cant find a way around

tried to pip install launch and i get this:
The conflict is caused by:
launch 0.1.3 depends on flowdas-meta<1.1 and >=1.0.1
launch 0.1.2 depends on flowdas-meta<1.1 and >=1.0.1
launch 0.1.1 depends on flowdas-meta<1.1 and >=1.0.1
launch 0.1.0 depends on flowdas-meta<1.1 and >=1.0.1

To fix this you could try to:

1. loosen the range of package versions you've specified
2. remove package versions to allow pip attempt to solve the dependency conflict

removed all the lines from requirements.txt but nothig

[floral kestrel]

been a while .. I used tutorials online and lots of research to make it work on amd

[fervent thunder]

do cpu and gpu matter?

mean if team red or blue

[floral kestrel]

cpu not so much i think . Gpu yeah much better with nvidia . Amd is like "problems each steps away" lol

[fervent thunder]

im on a 13900k and a 4090

[floral kestrel]

yeah its easy peasy to run then

[fervent thunder]

my issue is that Stable Diffusion works but i cant manage to install the extentions

like Charturner

[floral kestrel]

I have friend that run on 2800rtx that get more performance than my 6800xt

[fervent thunder]

oh lol

im trying to follow this
https://github.com/civitai/sd_civitai_extension

[floral kestrel]

cuz on amd we dont have cuda and stuff I guess im not a big pro on technical things .. I just try and try until I make it work lol

[fervent thunder]

switch to team green

more expensive but is reliable

and less enginering

[floral kestrel]

cant afford, my card was a gift the price are dumb now , like I could get a card around 500$ but 2K .. will not happen in any life lol

[fervent thunder]

hopefully prices will go back to normal

[floral kestrel]

but tbh installing automatic1111 was pretty easy all my issues was because I have amd gpu. maybe look a youtube tutoriel .. I remember there are some little things to be sure of .. like the version of python you install

[fervent thunder]

i watched all of them

the issue is that a "launch" file is missing

and i tried to pip install it but i get dependency conflict

look at the conflict and are all
launch 0.1.3 depends on flowdas-meta<1.1 and >=1.0.1
launch 0.1.2 depends on flowdas-meta<1.1 and >=1.0.1
launch 0.1.1 depends on flowdas-meta<1.1 and >=1.0.1
launch 0.1.0 depends on flowdas-meta<1.1 and >=1.0.1

and i have no clue how to fix it

[fast sage]

The readme tells you what you need to do.

[fast sage]

Anyone?

[fervent thunder]

@fast sage indeed, thank you. Installing trough Stable Diffusion UI got the job done. Do it from CMD create conflicts. Appreciate 🙂

[waxen trail]

hi, is there any documentation on how "AND" keyword is implemented in Auto1111 and how it impacts images?

[warm junco]

yes here is the wiki section for AND:
https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features#composable-diffusion

[waxen trail]

thanks for the link, but it kinda does not explain what AND aims to do, since i can do the same without it, only i can't use DDIM with AND. Is it only useful for recursive variations?

[fervent thunder]

once I installed the Automatic1111 as from instructions, should not appear in the tabs?

[waxen trail]

AND kinda looks like different syntax for (a : 1)

[warm junco]

You need to copy the ip you get

Open it in Browser

[fervent thunder]

no i no that, im able to use stable diffusion

the tab of Automatic1111 if exist do not appear or if i use the trigger word to use the extention, it doesnt work

[sage musk]

is it possible to run any of the newer LLMs on an AMD gpu? like victurna, stablelm, alpaca or anything really?

[fast sage]

When I train a model in Automatic1111 does it overwrite the prompt for the selected Stable Diffusion checkpoint?

[neat condor]

yo next batch of free training, have next 20 mins free ppl, if anyone has a dataset and needs 3090 for training lora/lycoris dm.

[pastel ice]

guys anyone else's prompt is not working at all? completely ignores the negatives...

whats that

im kinda new to this so i dont know all terms

using the auto11111 web ui

on the negative part

XD

maybe it has to do with impainting

im even using embeddings

alright

nah idk its completely ignoring negatives

how is that even possible

[fervent thunder]

Maybe negative is too long?

longer it is - less weight each thing in it has

[unreal crown]

Anyone know why my image is colorful when rendering but after render it's like colorless

[tawny sleet]

I have the same problem, colours on my renders are muted and dull

[broken smelt]

by colorless do you guys mean like actual grayscale/monochrome?

[hollow hinge]

Batch Size and batch count, which one do i increase if i want to let it generate multiple images at once? Also is there a log somewhere of your prompts, seed e.t.c? or does it only keep a log of image outputs?

[broken smelt]

images you create should have the prompts embedded in them, you can look them up by uploading them in the PNG Info tab

both of them will do what you say, batch size will increase vram used though

you can also generate infinitely if you right click the Generate button instead

[unreal crown]

My colors are not vivid after done rendering:/

[tawny sleet]

Not grayscale/monochrome, but just kinda greyed out

[broken smelt]

could try using a different VAE

https://huggingface.co/stabilityai/sd-vae-ft-mse-original/tree/main

try using the SD 840k vae if you are just using the standard one that a1111 defaults to

[tawny sleet]

Could it be the model? I'm using DosMix

[broken smelt]

thats another possibility

[unreal crown]

For me it wasn't gray scale before but it suddenly started being gray today

[lilac gazelle]

arguing with a gatekeeping artist be like
https://www.reddit.com/r/StreetFighter/comments/134zuyr/comment/jirpgx6/?context=3

[hollow hinge]

WinDows Properties > Details Does not show them

[hollow hinge]

Yeah but i mean if i wanted to set it to make 5 images and let it run for a few minutes, do i increase size, or count

[abstract quarry]

size is how many are computed in parallel

size * count is the total number

[hollow hinge]

Well i just did 5 count and it made 5 images

[broken smelt]

Yeah you can increase size with as much vram usage as you can get away with. I generally just generate forever and cancel it when I get back.

[hollow hinge]

So what does size do?

[abstract quarry]

what I said

[hollow hinge]

well its been sitting on 1, so if i put it to 3 would it be using more vram and going faster?

[abstract quarry]

it would create three times as much images

[broken smelt]

If it's set to two for example, it'll draw the two at the same time

[abstract quarry]

and yes, it costs more vram but it will be faster than just increasing count

[hollow hinge]

so then its basically just batch size times count?

[abstract quarry]

yes

[hollow hinge]

That seems pointless, But alright. Thanks

ill just use count for now

[abstract quarry]

using size will be much faster

but you cannot set it arbitrary high because you will hit oom at some point

[hollow hinge]

Why would size be faster than batch?

[abstract quarry]

because it computes in parallel