💬｜general-chat

[hot vortex]

I found many repos having bad naming... like really... pytorch_model.pth and so on, and then try organizing all the models and renaming them

I often find civitai or similar websites better, huggingface only when redirected or when a github repo specifically provides links to those models and explicitly writing what model it is to rename it

[pseudo silo]

What is difference between full and pruned model? Does it matter for AMD/Stable Diffusion?
Also thanks for the suggestion, much better site

very helpful community

[warm junco]

Pruned is smaller and made for image generation while full models are better to train on

Always go for pruned.

[warped void]

wassup

[frozen obsidian]

Hello I will like to take a course to learn how to learn prompt engineering for stable diffusion. Does anyone know of one or can point me in the right direction?

Also need to learn workflows, etc. I am using comfyui with stable diffusion. Where can I go to take courses to learn? Went on Udemy, etc. but didn't have any luck.

[proven badge]

i think it would be a good idea to go out a roit over companys klling cosumer rights “We’re paying full price for something we don’t actually own and the industry is taking advantage of us” since this bs has to stop then they will listen very fast 🙂 whos with me

[sleek fable]

hey, can anybody help me out?
I try to figure out how to find working LORA for Wan 2.2 Text to Image for using on Replicate... Somehow I have no idea how to use the HuggingFace UI at all...

[sick moth]

I have no idea.

[solid ferry]

with A1111 wats the best way to go about image expansion? (expanding a portrait image into a landscape/wallpaper)
tyvm or do places beyond a1111 have better capability of that

[burnt bay]

Very artistic

[bleak robin]

with A1111 wats the best way to go about image expansion?

[still glacier]

img2img outpainting is available out of the box with A1111, I used the openOutpaint "back in the days" to make it more efficient

I haven't done any outpainting in a while so I m not even sure this extension is compatible with the new forks

[pseudo silo]

Can SD do text to video models? how would those work

[still glacier]

Yes SD can do video, be it with animatediff, deforum, ebsynth or even the native stable video diffusion video models.
https://huggingface.co/stabilityai/stable-video-diffusion-img2vid-xt-1-1
But all those options are pretty old to be honest.

[warm junco]

Forge Neo has Wan video support

[pseudo silo]

AMD does not support text2video? You recommended me Forge w/ ZLUDA if I recall

[warm junco]

AMD does support text2video when used with comfyui-zluda or Forge Neo with ROCm

[pseudo silo]

Is it extension or seperate webui?

I dont see Forge Neo ROCm in your guides. I'll look into comfyui, I tried it and it said it loaded cpu and was laggy

and when I found a random model it "rented" a 4090 which also didnt gen anything

[warm junco]

A seperate webui.
Its listed under the "Any webui with ROCm/TheRock guide"

[fiery ore]

I'm in your dm

[atomic mortar]

Theres no real guide other then seeing how other people use it tbh, on civitAI you can see settings and prompts people used to make their image

There could be model specific guides but theres no prompt fits all solution

[fast hull]

Hello, I've been struggling to use ControlNet to output text2img similar to a reference picture, tried a dozen models/ip adapater/vae and dozen of settings about Preprocessor and CFG etc... The output images looks nothing like the reference image, like it's not even the same color... What's the best place to get some help with this? (The image is a very simple 64*64 pixel art tile of water with some waves to use in Unity)

[pseudo silo]

Would ComfyUI with Zluda be better than TheRock? I read warning HIP SDK 6.4. is that current still

[fast hull]

I found the problem I think, I was using 64x64 as the size, with 512x512 the results are way better and actually look like water lol

[pseudo silo]

version: neo  •  python: 3.11.9  •  torch: 2.9.0+rocm7.10.0a20251107  •  gradio: 4.40.0  •  checkpoint:
I followed "Any Webui with TheRock by AMD:"
Downloaded this webui "git clone https://github.com/Haoming02/sd-webui-forge-classic sd-webui-forge-neo --branch neo"
I dont see a text2video tab, do I need a model first?

[warm junco]

There is no txt2Video tab. Its done in either txt2img for txt2video or img2img for img2video.
You need to enable the refiner in the settings and then use wan high noise as checkpoint and low noise as refiner

[pseudo silo]

okay so wan high noise model, are refiners on same websites? I can use civitai?

Im 100% green on this stuff so I appreciate all the time and effort

Ah okay so refiner is also a model 😂

[warm junco]

yea wan needs 2 checkpoints high, low and an t5 text encoder and a vae file to work

[pseudo silo]

I activated refiner, but I cant figure out what to download now, I know they all go in model

Can I get the newest Wan 2.5? No specification on high/low

One of the models I downloaded also needed a vae, I assume wan comes with one

[warm junco]

wan 2.2 is currently used for video

whats your gpu btw?

[pseudo silo]

Radeon RX 9070 XT

[warm junco]

okay, then you need really smal wan models. but i would suggest you dont start with video stuff and try out some easier models like illustrious or flux

[pseudo silo]

no clue what that would mean

you helped me run sd with zluda and its awesome but I would like to setup videos too, im almost done

I had sooooo many issues following other guides + chatgpt til you helped

[warm junco]

here are smaller wan model versions:
https://huggingface.co/QuantStack/Wan2.2-T2V-A14B-GGUF/tree/main

[pseudo silo]

Thank you, it has high low and vae I see

do I follow the clone way?

[warm junco]

nope, you click on the smal download arrow on the right side of the model

the t5 text encoder can be found here:
go for the smaller one
https://huggingface.co/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/tree/main/split_files/text_encoders

it goes in models/text_encoder

vae goes in models/vae

[pseudo silo]

it works in SD? I dont use ComfyUI

Just a text encoder I guess

Im blind af, I cant see download arrow

oh click them individually 🤦‍♂️

Q2 - Q8 is based on cpu?

[warm junco]

nope

its the quantization of the original model file

its a compressed version with a bit lower quality, but less vram needed

[pseudo silo]

I see okay, I read CPU Inference somewhere

for 16gb vram I should probably limit myself a little

[pseudo silo]

@warm junco How do I proceed in SD WebUI to produce? Do I use Wan or SD UI or doesnt matter?

[warm junco]

sd ui is the wbeui, wan is a model

for video you would need to use wan

[pseudo silo]

I just load the checkpoint and gen into txt2img slot?

I dont see any refine options, automatic?

[warm junco]

you have to go into the settings tab and there under Refiner, check the first box to enable it. then click apply and reload ui

and then yea you select in txt2img

[mossy drum]

hello

[pseudo silo]

thank you G

[pseudo silo]

AttributeError: 'NoneType' object has no attribute 'is_webui_legacy_model'
'NoneType' object has no attribute 'is_webui_legacy_model'
I got this, what did I do wrong?

[warm junco]

can you show which models and files you have selected? in #🤝｜tech-support

[pseudo silo]

I just realized I didnt apply wan ontop of the text encoder, wait let me retry

was confused, thought 1 only

It created white noise image with colors and that 1st gen driver timeout happen, just redo?

[primal current]

Question for ComfyUI users:

On an RTX 5090, is there a difference in speed improvement between:

PyTorch 2.8.0, Python 3.12, CUDA 12.8

and

PyTorch 2.9.0, Python 3.13, CUDA 13.0?

Are there any benchmarks or verifiable data that demonstrates this difference in improvement?

[proven badge]

yooo everyone i got like almost 100k hours of my life gaming thats 11 years of nonstop gaming if i hit 100k hours

[dusky marten]

hello

[steep timber]

so, what has happened with stable diffusion?

[rain flint]

Is here the right place to ask about yaml for dynamic prompts?

[gleaming ember]

overtime

[cedar plover]

hows it goin people

[burnt bay]

Yes~

[harsh current]

What is freedom?

[lethal patio]

free and easy lifestyle

[gleaming ember]

What freedom

[unreal saddle]

who a y

[odd geode]

hmm

[lusty lark]

hello!

[coral geyser]

Hi everyone
I recently acquired an AI-related domain, AIHarbors.com.
I think it could be useful for an AI hub, startup, or model hosting project.
I’m open to discussing opportunities if anyone is interested.

[tender nebula]

Good luck to you

[floral umbra]

Did anything happen?

[steep timber]

that is what i want to know

[cedar plover]

just keeps getting better is all i know of xD

[woven kite]

hello

@narrow osprey

[night glade]

hello
No data source is currently selected. Please choose a data source from the dashboard and try again.

[cedar plover]

hows it goin

[woven kite]

good how are you

[woven kite]

https://www.youtube.com/watch?v=hnr7-VNHJoU

[chrome notch]

Hi

[atomic mortar]

Lets connect you to rule 5 🤝

[glad torrent]

About images for IG, do you usually use 4:5 for static images (not reels or stories)?

[cobalt tartan]

Hi anyone know anywhere like CivitAi that is useable in the UK? Aka no adult stuff as CivitAi doesn’t work now due to having that too.

[still glacier]

https://civitai.green/ might work then

basically civitai but without the nsfw stuff

[sweet stream]

Yall alive in here??

[still glacier]

no

[sweet stream]

Damn 😞

[winged ember]

Do you guys know any models that are semi-realistic but don't give these child-like anime faces? Like those semi realistic faces that make all look like they're 14...

[celest slate]

Hey all...getting back into SD. Anyone have some info on generating multiple consistent characters within the same image? I'm assuming I'll have to do multiple passes and add each character seperately, but if anyone has the sauce on the latest and greatest of multiple consistent characters, please share!

[jolly mason]

# CHICKEN CHICKEN MICROWAVE..

[cedar plover]

have you tried just using a vpn? that should work i would think

[cobalt tartan]

Wouldn’t that be illegal getting round it? I’m an adult but still.

[cedar plover]

ehh im not sure, i dont personally think it would be "illegal" but im not very aware of the laws there on nsfw content. but generally i think it would be fine. if you want to do it and are worried about the legality of it you may be able to ask chatgpt or look more into exactly how the law/restriction of that content works where your at and if theres anything against circumventing the restriction

[cobalt tartan]

In the uk now you have to basicly dox your self to look at adult content online, so since that site also has adult content too it’s blocked now unless you do that.

[pseudo silo]

What is the reasoning for that?

[atomic mortar]

Online safety act law

But rather an excuse for mass surveillance tbh

Every western country nowadays is working on digital id verification

[quartz pelican]

What are the features of Invoke? How is it different from Forge, for example?
(Except for a disgusting name, that is)

[round siren]

Does anyone know the best inpainter model for 3D SFM/Blender style images?

[spring siren]

Hello guys

[hot vortex]

hi to you too, have a good day

[vale void]

hi what's up

[proven badge]

im making humans over clankers t shirt soon

[tame glen]

Is there anyone looking for a skilled developer?

[still glacier]

No, rule 5 #✍🏼｜rules-and-tos

[fresh light]

Hello gyus

[shell gate]

i keep getting sent random friend requests from this server

[grave wedge]

Hello, can you help me?

[atomic mortar]

budget: rule 5

[brazen shard]

got it

[atomic mortar]

o wow not a bot for once

[atomic mortar]

Its like these people dont read.
Unironically, we could use one of those ""devs"" to make a ai disc bot to get rid of "people" who don't read rule 5

[cedar plover]

dang im sorry, ill help you out and send one more

[still glacier]

bold of you to assume they're people

[still glacier]

they still didn't care tho.

[marble meadow]

Is there a tool that lets me remove background to make an image transparent

Sick of these websites that downscale my image and require a shitty pro subscription

[atomic mortar]

Tru, just wondering why there are so many

[eternal valve]

Does anyone know the best settings to use for ReActor when doing a face swap and what model(s) it works best with? Should control nets also be used? Thanks

[atomic mortar]

I found that base settings work best with reactor for me

Sometimes a face boost but nothing too much

But i do it on images i already generated, if you want a very specific pose yeah you can use control net

Or match the body likeness/hair

[eternal valve]

What diffusion models do you use

[atomic mortar]

Juggernaut xl is pretty nice for skin

But eh flux or sd3.5 large works just as well for me

My preferred workflow is sd3.5/flux > refine with juggernaut xl

To fix smooth skin

[orchid swan]

Hi guys,

https://www.loom.com/share/c9e8d95c08264c60a47e4ecec28dd3b2

Can this really generate high quality voices locally?

[atomic mortar]

Neotts? Never heard of it, i see neutts recently released but i found vibe voice better

[burnt bay]

Haven't seen

[shell gate]

DANG IT

[harsh current]

That's a pity.

[white shore]

hi guys i got a question. how do i turn video into animation. i saw 2 videos in yt showing how to do it using stable diffusion. both video are tutorial on how to vid2vid. 1 vid is in russian, and 1 vid is in english and when i tried to follow the english one. like set to noise multiplier to 0 and setting the denoising strenght to 0.5 the image in img2img i generate is blurry. can someone tell me what am i doing wrong?

[forest tinsel]

inbox me

i will help you in that

[hollow sleet]

hello

[grizzled grove]

hi

what sort of developer are you ?

[forest tinsel]

animation

[atomic mortar]

for anyone wondering, buzzword bot, dont engage

oh wow you respond, please read rule 5

[junior urchin]

rule 5

[quartz pelican]

Ok, after hours of fuckary with Chinese spyware while attempting to make my own LoRA I installed ComfyUI portable
Am I going the right direction?

[atomic mortar]

judging by that comment no

if your new to this i definitely recommend neo forge (tutorial is in pinned messages in tech support)

for lora training i like one trainer

or civitai (cloud)

[quartz pelican]

Instant no

[atomic mortar]

just giving my 2 cents, the illustrious loras i made there are pretty decent

[quartz pelican]

Neo is not the same as just Forge, I suppose?

[atomic mortar]

its mostly the same

just a bit newer and support for rtx 5000 series

older cards should not use neo iirc

but using comfy to train loras is hella cursed tbh

one trainer my beloved

[quartz pelican]

Well I've got 1080ti

[gleaming ember]

Is this how you play?

[fierce heron]

Ever had a crazy cool idea and wished it could become reality without being copied?

[quartz pelican]

There are many good games I can play even with this
Though I'm preparing for an update

[quartz pelican]

Copied?

[fierce heron]

good catch, i presumed chatgpt was better at convincing

[quartz pelican]

What

[fierce heron]

im tryna gather some ai enthusiasts, devleopers and all into my server, i tried posting the server address, but it doesnt allow advertisement

i wanted to try to get some of u guys to join my server to build projects and collaborate

Anyways

Anyone interested?

[compact current]

Hey guys, anyone know until when Stable Diffusion 2.1 Base will be down? I was using the CLIP encoder but Idk what variant is used in the actual model.

I thought its the laion/CLIP-ViT-H-14-laion2B-s32B-b79K but I am not able to reproduce the same results with it...

Yes, I do, its quite important as I was running my model with a VAE backbone and text encoder from SD2.1 base, but since the base model is down, I am unable to pinpoint the correct model to use...

I had precomputed the latents, so that is not an issue hopefully

I would really appreciate the help, thank you.

[wanton garnet]

Or what if you wanted to have the character in the bottom half of the image sitting on the edge of the roof looking at the city sunset for example?

Naturally it would generate the character in the center of the image and you wouldn't get much of the background.

Im trying to get the character in the bottom 1/3 - 1/2 section of the photo and the sky to take up 1/2 to 2/3.

Ive tried all different prompts, putting a similar photo in tagger to see what it generates and I've googled the heck out of it but theres no information on how to position the character in a photo like this,

I know you can use outpaint or controlnet but im trying not to use this, I wanna learn how to properly prompt it without it. If its even possible?

[atomic mortar]

@wanton garnet dont dm this guy

Hes been trying to get everyone to dm him for "private help"

Hella shady

[oblique elk]

Two options:

1. Regional Prompts, so you can define regions and the content that should generated in the region
2. generate as usual and outpaint the top of the image.

[uneven cedar]

Bro I want some help

[atomic mortar]

Please do elaborate (and maybe in #🤝｜tech-support )

[tacit rune]

Hey buddy's

[wanton garnet]

Thank you! I know about these methods i thought there was some way to do this without outpainting or regional prompting but just regualr prompt

[wanton garnet]

Thank you i appreciate the heads up

[oblique elk]

Third one would be a controlnet with open pose reference 🙂

[bleak pond]

do u know any free ai

[atomic mortar]

I know some free limited ai

Like 10-20 pics and then its like 3-5 a day

[bleak pond]

oooh nice what is the name

[atomic mortar]

civitAI

[bleak pond]

thanks

[atomic mortar]

depends on the model used offcourse

[bleak pond]

what model is best

[sweet stream]

[cedar plover]

do you have a decent gpu? if so just run it all locally, im running comfyui on an rtx3060 with 12gb vram and 32gigs of ram and can run any model from civitai perfectly fine

[bleak pond]

yeah

[cedar plover]

try automatic1111 or comfyui

id be more than happy to help you with the setup of either one if needed

[bleak pond]

thanks i see it helps with coding

[bleak pond]

i'll like u yo help me with the first one

[untold vault]

[still glacier]

don't bother with automatic1111 in 2025

check out the guide pinned in #🤝｜tech-support

[atomic mortar]

And that's a scam

[tropic hare]

hello

[devout vapor]

How do I change where ComfyUI saves workflows? I can't find that setting.

[still glacier]

NOPE

not happening

scammer

To whomever who may have clicked the link, there is not kind of any stability ai / stable diffusion coin anything.

[junior urchin]

[atomic mortar]

[devout vapor]

I don't understand your reply

[junior urchin]

@still glacier

@still glacier also a scam tactic

[floral umbra]

Got tired of manually installing dependencies folder by folder, so had copilot make a neat auto installer script that OS walks into every custom node and installs each one automagically

Successfully installed altair-6.0.0 bitsandbytes-0.48.2 came_pytorch-0.1.3 imagesize-1.4.1 jsonschema-4.25.1 jsonschema-specifications-2025.9.1 lion-pytorch-0.2.3 markdown-it-py-4.0.0 mdurl-0.1.2 narwhals-2.12.0 numpy-1.26.4 opencv-python-4.11.0.86 prodigy-plus-schedule-free-2.0.1 prodigyopt-1.1.2 pygments-2.19.2 referencing-0.37.0 rich-14.2.0 rpds-py-0.29.0 schedulefree-1.4.1 voluptuous-0.15.2

ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
opencv-python-headless 4.12.0.88 requires numpy<2.3.0,>=2; python_version >= "3.9", but you have numpy 1.26.4 which is incompatible.

✓ Successfully installed dependencies for ComfyUI-FluxTrainer

============================================================
Installing dependencies for: ComfyUI-GGUF
Requirements file: /mnt/Fellow_Gamer_1/comfyui-fresh-25/ComfyUI/custom_nodes/ComfyUI-GGUF/requirements.txt
============================================================

[dawn mulch]

Anyone know how to deblur or focus for SD 1.5 i2i generations? Other than prompting focus, deblur, etc.

[atomic mortar]

Sd 1.5, hmmm maybe proper upscaling could fix it a little but its outdated model might make it a challenge

[bleak pond]

do any one know blender 3d app

[viral pagoda]

Hi everyone, can I ask some questions?
I import the stable diffusion model in the amazon bedrock i encounter an error which it required a config.json. my i know what are the configuration regarding to stable diffusion?

[atomic mortar]

No idea what amazon bedrock is but every stable diffusion has some config differences, xl, illustrious, v1.5, 3.5 etc

[viral pagoda]

can i have some configuration details for this model of stable-diffusion-xl-base-1. 0

in the stubble-diffusion file architecture it has no config.json in the root structure.

[unkempt folio]

I'm looking for a Wordpress expert.
If anyone is interested, please contact me.
I'll provide more details when contact me.

[fiery ore]

I'm in your dm

[atomic mortar]

Well mostly its just euler A and sampler normal, no idea about amazon bedrock but your running it all in the cloud?

Any reason you specifically went with Amazon? You could use the sd api in comfyUI aswell

[bold bough]

hey, how can i generate images?

[cunning patrol]

Im trying to use the older "Deforum Stable Diffusion v0.5", ive had it working when it was "new" at the time using this cheat sheet: https://docs.google.com/document/d/1ZZp3aO33ddzLUj6cM1LAghGiEn_ol3D1V8H-0MphlrM/edit?tab=t.0#heading=h.s2ihhcctvtj1. Its not working at all any more. I liked the way it looked and worked before... Is it futile to keep trying this older model/workflow . I have an affinity for outdated tech. Is it a new notebook verse older model situation or much more complex and i should move on?

[calm seal]

hii

[calm seal]

hi

[rose crest]

Who in here is providing GPU compute?
And who’s regularly buying GPU time?

I’m working on something early stage and want to talk to both sides.
please DM me if you’re open to a quick chat it would be much appreciated.

[gleaming ember]

very rigorous

[tepid grove]

have anyone tried sglang diffusion?

seems to be an open source effort for serving diffusion requests in parrallel

[burnt bay]

try

[tepid solstice]

hi

im planning to use this model (optionally downloaded and embedded) in my app https://huggingface.co/stabilityai/stable-code-instruct-3b

is it oki?

i registered to the license form on Stability AI website it said "your license is activated" but didn't send me anything

my app isn't earning 1.000.000$

[still glacier]

For professional advices, it s probably best that you contact stability.ai directly by mail. https://stability.ai/contact

[tepid solstice]

ummmm

oki thank you

actually their website is pretty clear about this

but i wanted to ask

[odd geode]

dead chat

ded server

[keen hull]

ay wsg yall im new here and i recently just started exploring ai generation overall

im curious what do yall use image generation for ?

[modest kelp]

.

[stone peak]

where do i generate photos?

[compact vault]

hello

[still glacier]

Pick one :
1/ read #artisan-faq to do it in discord
2/ use stability.ai's web app https://dreamstudio.stability.ai/
3/ generate it locally using one of the many Ui available, guide here #🤝｜tech-support message
4/ use one of the many available third party websites. Make sure to read reviews and do your researchs before giving any moneys. There are a bunch of scam websites out there.

[marble meadow]

Is there a tool that can autocreate a pose from one image for ControlNet

[elfin fox]

i'm trying to get controlnets going...getting black outputs...

any idea why folks?

[upper plinth]

RIP stable diffusion

Nano banana pro made all other image gens obsolete. It was nice knowing you gentlemen 🫡

[atomic mortar]

False, can't make nsfw

[upper plinth]

But grok can 😏

[gusty wind]

good old days

[eternal locust]

is there a stable diffusion model/lora for generating computer icons?

like for windows, linux and stuff

[tame swallow]

I need help setting up a gpu cloud for fooocus or comfyui. Im a noon. I need a guide I'm trying to use dataoorts. I just want to generate a few images without the ridiculous censorship on online platforms. Can anyone help me please?

[hazy bobcat]

Hello

[still glacier]

or not

don't follow random link people

[low moon]

Did the cat with 4 GB of ram get help? I've been gone for a while

[low moon]

and its not local or open source isnt it

[pastel crown]

I've been playing world of warcraft for 21 years

[narrow cypress]

hey@all thank you for your great work!! I want to donate a large amount of ethically sourced audio data files (40+ million files + the tools to algorithmically increase the amount available), all types of hq audio assets (songs, loops, instruments, compositions etc) for ai training purposes to the stable diffusion project. Just send dm me

[burnt bay]

so long

[gleaming ember]

Nice to meet you all

[junior urchin]

You do know they won't take your word for it, right?

Illegally downloading sound and slapping a "legally sourced" sticker on it, is the same

[narrow cypress]

www.zlexx.com

The assets are crafted by ourself

[junior urchin]

@still glacier They're trying to sell it

[narrow cypress]

No we donate this to certain free projects

[pastel iris]

Hey friends, I finally fixed all the bugs for reading SD traces. I hope this can help you guys make things faster!

https://www.youtube.com/watch?v=MBTC74FTRTY

[jovial ruin]

Hello everyone. Could someone tell me, why any trying of download, by SD, model from huggingface.co stuck at 0% ???

[ashen sleet]

hi

[bleak pond]

sup

[still glacier]

It's probably best to reach out to stability.ai through the official ways https://stability.ai/contact , this discord is mainly centered around the SD community.

[vernal herald]

Grok is probably the most uncensored LLM and cloud img gen out there

[vernal herald]

Local or grok is is good

Same with nano banana, though it is censored

[narrow cypress]

Thank you. Today we used the contact form also to reach out to the team. Is the Stable Audio Small Model already released? That one which can run on ARM devices locally?

[haughty knot]

hello

[split kestrel]

Ugh, I am struggggggglin today with comfy

Should never have updated the thing cuz now it’s my nightmare to get it working again

[unique sage]

Did anybody use Krita before?
somebody who's helping me with SD uses it, and it seems to produce better result
i need some tips on how to get it and how to use it

[topaz cargo]

Hello I've been trying to dip into AI gen as of recently, since I finally got my own PC with (weak) graphic cards instead of laptop
My question is when prompting multiple characters, what mitigation/negative prompt you use to avoid repeat characters? I know "cloned face" is one of them, but is there any other prompt that could weight in and reduce probabilities?

[junior urchin]

@still glacier

[marble otter]

you mean @still glacier need a dev now?

[junior urchin]

Ads are not allowed

[still glacier]

Yup rule 5, no unapproved self ads. I m gonna go ahead and remove this for you.

[abstract quarry]

Flux 2 is out 🤩

[wide sapphire]

HELLO I am Dan Stein, and new to stable diffusiom and am SO CONFUSED!!!!!!!

[still glacier]

How so ?

[undone kindle]

hey@all, I am new to this platform and AI at all. Recently I was working on diffusion (with semantic conditioning), I am able to generate image but there is color shift. Any hint to solve this?

[stark escarp]

Hello

Anyone knows how many images in the training data of SDXL?

It seems it's not public information

[junior urchin]

Wasn't sdxl based on LAION's?

[low moon]

https://flux2.io/

=0

[worldly whale]

Did anyone have any luck running flux 2 with the fp16 model? The Load Diffusion Model node seems to give an error with the FP16 model.

[atomic mortar]

You need about 128gb of sys ram to run it

Or less if you use the smaller version but still around +60gb

[worldly whale]

Im running a RTX6000 and 256gb of ram so that should be fine

So it just goes into the load diffusion model node nothing new?

[still glacier]

make sure your comfyUI is updated and uses flux.2 template and it should work just fine.

[silent trench]

So I need some help with Regional Conditioning, basically I am trying to make MTG playing cards for a christmas event for a group of friends. I am attempting to do it in a Chibi style using a lora, and I am trying to "encourage" the lower half of the card to be simple in design...however I cant seem to get anything consistent. ...just would like some help tbh...so anyone down to help out troubleshoot maybe an easy fix ?

It is SDXL if that matters

[proud dawn]

What upscalers do yall use? Anything decent and free that can get my 2k images to 4k?

Or enhancers

[tall cargo]

My typical workflow is an upscaling one for Flux.1 Dev. After the initial 1MP generation, I upscale using 4x_Ultrasharp, then downsample by 0.75 (to about 9 MP, 4K+ size), and then do a second round of denoising at 0.2 strength with a Mixture of Diffusers tiled denoising node.

[mossy edge]

Hey guys, I’m doing some research for a project.
If you rent GPUs for SD or ComfyUI, what’s the most annoying part of the process today?

(Price, uptime, boot time, storage, reliability, anything.

[winged rivet]

i need some help ,i have a 5060ti ,16 gigs of ram and a terabyte of ssd ,and a 9900x , dont need 4k stuff , just about 720p to 1080p is it feasible,in image to video , walkthrows of structures ,and basic stuff ig

[winged rivet]

is it enough and where should i start

[faint timber]

You can start by browsing the video templates that come with Comfy. It's under the file menu. Give Wan a try first. Remember that Wan is 81 frame clip generator. You have to get creative to extend it beyond that limit.

[next geyser]

I am making STT ai service

[mossy edge]

Hey guys, I’m doing some research for a project.
If you rent GPUs for SD or ComfyUI, what’s the most annoying part of the process today?

(Price, uptime, boot time, storage, reliability, anything.

[umbral lotus]

I love the amount of styles that are available on the site. to be able to create certain images however it would be REALLY VERY cool when styles could be combined, for example rococo + aerial view.

[gritty cloak]

Hi everyone,
Sharing an opportunity for AI filmmakers and creators:
UnHuman Shorts – an international AI short-film contest
Theme 2025: THE TRACE

4 Jury Awards + People’s Choice
Prizes from ACE Studio + more partners (TBA)
Last year: 800+ submissions from 13 countries

Deadline: January 25
Details: https://unhumanshorts.org

[azure chasm]

which loras you use to make nsfw with qwen combined with a character lora?

[primal current]

Why am I not able to increase the speed of z-image turbo by 20% using Patch Sage Attention? It should be giving me generations approximately every 0.76 seconds, without SA i can make with 0.95 seg per image, with SA 0.90 seg aproximately

ComfyUI, RTX 5090, TORCH 2.8.0, CUDA 12.8, python 3.12

sageattention 2.2.0 / sageattn3 1.0.0

I'm getting the following:

1st test

without SA: 0.95 seg
SA default: 0.91 seg
SA qk int8 pv fp16 cuda: 0.92 seg
SA qk int8 pv fp16 triton: 0.89 seg
SA qk int8 pv fp8 cuda: 0.89 seg
SA qk int8 pv fp8 cuda++: 0.89 seg

2nd test

without SA: 0.95 seg
SA default: 0.88 seg
SA qk int8 pv fp16 cuda: 0.96 seg
SA qk int8 pv fp16 triton: 0.88 seg
SA qk int8 pv fp8 cuda: 0.90 seg
SA qk int8 pv fp8 cuda++: 0.87 seg

log: https://pastebin.com/J9qedqxM

[delicate summit]

How to get image → prompt ? Anybody know ?

[lucid bobcat]

Anyone else thinks that Flux 2 is hot garbage? Ever since I saw the promotional images I thought that some if not most of the images look terrible. Everything looks flat and weirdly artificial (photoshopped) and backgrounds look pale and washed out. I convinced myself today that I'm not just imagining things. I downloaded the model and workflow and created the same image twice, once with Flux 2 and once with Flux 1 with all the same settings. The Flux 2 generation has all the same bad qualities that I already noticed. On top of that generating this aweful looking image took twice as long.

[atomic mortar]

Well flux 2 does require 128gb of sys ram to run comfortably so no Surprise it takes long,
Flux works best with a juggernaut xl refiner tbh to fix skin issues

Z-image however does not have that problem, waiting for the base model release!

[fervent thunder]

Does anyone have a tutorial on how to use nano banana pro?
I have no experience with AI image editing and jailbreaking them.
I have the ultra package and I see the ultra filters are significantly reduced compared to the free version.

[wild dock]

hey everyone ! have anyone tried z-image ? i wonder if it is just me but with sdxl using the same prompt and changing the seed I could get some very different result, with this model it seems like the seed just does not matter anymore, Am i doing something wrong here ? (changed the cfg too but not much better result)

[atomic mortar]

Cfg 1, sampler = euler, scheduler = beta, steps 9, (more for more randomness) and yes its a different model (and a turbo version so seed matters less)

[wild dock]

Ok thanks i will try, just another question what is that turbo version ? I can see online everyone is waiting for the base model (for lora i guess) what are thr difference between those models version ?

[atomic mortar]

Turbo goes fast but lower creativity, base is slower (bigger) and more creativity, also waiting for the anime version

[wild dock]

Ok thanks, will put myself on the can't wait waiting list too so !

[atomic mortar]

Turbo is still very stunning however

[wild dock]

oh yes had a play with it, it is..... just wow ! can't wait to fine tune it

[gray vapor]

I just discovered that the ONEREWARD FLUX FILL model by bytedance uses true cfg= 4. Was having bad results until I saw this was what they used in the diffusers code. Most workflows online do this wrong too.

So, no Flux_guidance=30 like in FILL. true cfg in ksampler=4. Just in case one of you is scratching his head too : )

[next monolith]

Can I use flux or qwen in stable diffusion webui or only sd models

[atomic mortar]

Looking for people who can read rule 5

[plucky mortar]

Hi, can anyone help me? Where can I go to discuss training and creating lora? I'm having trouble with the small dataset. Or is there a Discord channel or website where I can discuss these types of issues?

[midnight laurel]

Anyone here were able to break gemini synthid using SDXL or SD or Flux? idk what else community of discord to talk about this, but I was able to do this on multiple datasets, so m just looking if someone else did already

[vernal herald]

What’s the inside joke about laying on the grass

[tropic frost]

anyone ever done an monster waifu lora?
putting together one, considering what issues i might encounter

[crisp pewter]

[atomic mortar]

onetrainer is a decent program to train lora's and they have a discord aswell

[round mirage]

For a relative simple lora, how many imagines are generally needed to make a lora? Right now I have around 17 and have it set to 5k steps

[kindred radish]

Anyone here trying to build AI automation? I’ve been working on a few tricks that could really help hit me up if you're serious!

[native sedge]

Hey

[obtuse isle]

Hey guys. I'm in trouble with LoRA traning. Anyone can help me?

[atomic mortar]

Onetrainer discord might be a better place for it tbh

[glad torrent]

Can anyone tell me what happened to Noob-XL? It has not updated for a year. I think it's quite good for some styles.

[atomic mortar]

It got released and that's it, nothing big really

[glad torrent]

I found this: Chenkin Noob XL, but not many uses.

[atomic mortar]

The rest are finetunes and people are waiting for z-image anime version

[glad torrent]

"z-image anime version"? What's this?

https://www.reddit.com/r/StableDiffusion/comments/1p7usu2/havent_seen_this_much_hype_on_an_image_model/

Hmm, hope it works

[atomic mortar]

Well z image is a new model, right now theres only the turbo version

Were waiting for a base model

And the anime version

So basically illustrious but not sdxl

But really really good

[atomic mortar]

scam

[junior urchin]

@still glacier

[tender prairie]

Anyone knows a good AI audio noise reduction software?

[late cipher]

Hi everyone. I am a developer interested in AI art and image generation. I work with Stable Diffusion models in my own projects, including custom pipelines, model tuning, and building simple tools for creative workflows.

I joined to learn from others, share ideas, and help when I can. If anyone is working on projects related to model deployment, automation, or frontend integration, I am happy to contribute where my skills fit.

Thanks for having me here.

[violet field]

Hello! Does anyone know where I can talk about Magnific ? I can’t see any community discord for them

[leaden flicker]

Yoo

[quartz river]

hello

[atomic mortar]

No its a scam

@violet field if you did anything there its a scam

I did check their site and all the reviews seem hella suspicious ngl, upscaling locally seems like a better choice

[viral folio]

Hi everyone!

[glad torrent]

I have a question. I’ve seen a lot of ComfyUI workflows that look like a spider web, and I’m wondering if that’s actually necessary. When generating an image, I don’t understand why so many custom nodes are used, especially when many of them seem to repeat the functionality of existing nodes.

They often use two or three components for things that can be done with a single node.

I get that additional sections are needed for post-processing steps like upscaling or style changes, but most of the workflows people post are just a mess.

I really doubt the efficiency of setups like that.

[elfin tapir]

Can be someone just habitually grandfathering in custom nodes from earlier projects. But yeah I've seen that too

[tranquil mason]

Hi everyone! 🙂

[brittle rose]

new concept of music: #🌶｜off-topic message

[young bronze]

I lather myself in workflow spaghetti, mostly because I run a bunch of prompt randomizers mixed with a ton of toggle-able prompts, which is then run through a loop that scrambles the order of the prompts for more chaos.

I can't say my results are great, but it's usually good enough to iterate from if I find something I like

That being said, we have subgraphs now so we can hide the spaghetti

[lean field]

hey guys about to do some lora training and am lookijng for donated images for ethical training. Will be starting with pedegree animals and will happpily share what i have with civit.ai looking fofor around 1.5k images. do you think the mods would allow me to make a thread here?

[still glacier]

probably best to ask them directly in civitai's discord server.

[oak ether]

Is ZLUDA even faster than DirectML? I generated just a 1024x1024 image with no upscale in SDXL model and the speed was just around 1.3 it/s, that's on 9070XT

[vocal burrow]

Hello everyone, can one single 5090 able to run WAN2.2 FP16?

[latent shell]

Hi. I have a question.
I have already existing image it's resolution is 2048x3072 can I resize(regenerate) it to 2560x1440 to fit my screen?

[warm junco]

Yes by a lot xD

[atomic mortar]

Scam posts be like

[oak ether]

I mean, is 1.3 it/s is a normal speed for ZLUDA on 9070XT or there's something wrong with my setup? On ROCm I have around 5 it/s (but the only thing it's generating is.. driver crash (often leading to PC reboot or complete freeze), althrough preview image looks fine); on 3080Ti I had ~4 it/s. Is DirectML's speed even slower than 1.3 it/s for sdxl 1024x1024? It's hard to imagine that it can be even slower.

[green fable]

How can i generate ai images here ... Can any one explain me plz

[stoic drift]

Hello, a friend and I want to team up for self-hosted AI digital creation. We would need to:

• generate high-quality images in ComfyUI pipelines using top models
• give life to scenes with AI voice narration
• small LLM/bot for automatic orchestration

I currently have a mid-size desktop with 64GB DDR5. I'm just not sure which of these would be the smartest "bang to buck", especially to be future-proof for about 3 years:

• Buy a single 5090 as an external GPU for my desktop (it doesn't fit on current mobo unless I get a new case + I like the idea of an external GPU)
• Buy 3 5060 16GB (48GB VRAM) and buy a new large case/chassis + mobo to host them, cannibalize the remaining parts from the current desktop. More VRAM and still cheaper than a single 5090, and has plenty of potential for future expansion, but is multiple 5060 too slow? I want to be able to run models that require 32GB+ VRAM in 2 years if that's where the wind blows.
• Strix Halo. I can't find any benchmarks for ComfyUI, but for large MOE LLMs like GPT-OSS-120b it beats a having a 5090 + offloading, it runs like 80% faster. idk if this speed translates to image generation models though. How long would it take Strix Halo to generate 1 image, relative to a 5060 or 5090?
• Go all in, buy a new case + mobo + 5090. The most painful choice on my wallet.

THANK YOU FOR YOUR ATTENTION TO THIS MATTER!

[oak ether]

If you mean locally, if you have Nvidia GPU, you can just download Stability Matrix, go to packages, find the most suitable one (I prefer Stable Diffusion WebUI Forge, but I just get used to it. Anything (except Comfy) should be good for newbie too, maybe even more simple. SD.Next seems simple enough too), and it will install everything for you. You can even download checkpoints and loras from here, and it's really handy if you want to try different UIs since it shares models, loras and other folders between UIs. If your videocard has 10GB VRAM or less, you may need to add --medvram or --lowvram launch parametrs (click on settings button in the card of chosen UI in packages, but you should try without them first. Then download some checkpoint, write prompt and click "generate". You may learn how to use generation parametrs later.

[warm junco]

1.3 it/s is not normal for zluda, you should get around 4 it/s

[oak ether]

Why it can be the way it still works, but slowly? Usually if there's something wrong, it just don't work at all. Driver version?

[warm junco]

You could show me the cmd log and the webui settings in #🤝｜tech-support

[gentle prairie]

For Hiresfix, what's the best upscaller to use for realistic content?

[still glacier]

rule 5 #✍🏼｜rules-and-tos

[marble otter]

make a sense

[atomic mortar]

Bot idea: if a message contains, dev and lets connect or the 🤖 emoji its an instant kick

[atomic mortar]

Rule 5, inability to read smh

[tribal crown]

So my version of torch is 2.8.0 with cuda 129(exactly the version that sageattention doesn't have... lucky me).

I installed as instructed with the one for 2.8.0 cuda 129.

I have a 4090.

Installed triton as well.

I get this big ass error :

Error running sage attention: Command '['C:\X\Stable Diffusion\XX\ComfyUI_windows_portable\ComfyUI_windows_portable\python_embeded\Lib\site-packages\triton\runtime\tcc\tcc.exe', 'C:\Users\XX\AppData\Local\Temp\tmpaohs_bn3\cuda_utils.c', '-O3', '-shared', '-Wno-psabi', '-o', 'C:\Users\xx\AppData\Local\Temp\tmpaohs_bn3\cuda_utils.cp313-win_amd64.pyd', '-fPIC', '-D_Py_USE_GCC_BUILTIN_ATOMICS', '-lcuda', '-lpython313', '-LC:\xx\Stable Diffusion\XX\ComfyUI_windows_portable\ComfyUI_windows_portable\python_embeded\Lib\site-packages\triton\backends\nvidia\lib', '-LC:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.3\lib\x64', '-IC:\XX\Stable Diffusion\XX\ComfyUI_windows_portable\ComfyUI_windows_portable\python_embeded\Lib\site-packages\triton\backends\nvidia\include', '-IC:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.3\include', '-IC:\Users\XX\AppData\Local\Temp\tmpaohs_bn3', '-IC:\XX\Stable Diffusion\CXX\ComfyUI_windows_portable\ComfyUI_windows_portable\python_embeded\Include']' returned non-zero exit status 1., using pytorch attention instead.

Why are my versions still incompatible? I using comfyui without this and avoid sageattention like a plague because i never seem to install right. I want to finally succeed in doing this sht. Please help?

[hardy dove]

Any checkpoint recomendation for Logotype design?

[weak tusk]

whats your go to Ai tool right now?

[stoic drift]

Can someone please confirm? If I have two 16GB GPUs, can I:

1. load a model that requires 30GB VRAM?
2. If the answer to 1 is "No", can I load a model that requires 15GB VRAM, and also activate other features (ControlNet etc) that require another 15GB VRAM on the 2nd GPU?

[atomic mortar]

1. No, only llm's work like that
2. Not that i know of. Probably not unless your using some specialized software that most dont. Bigger chance it offloads to ram

[stoic drift]

thank you. Gonna aim for a 5090 then.

[atomic mortar]

a DM for what?

probably spam since its in every channel

[devout vapor]

Can I use A1111 to create an audio file from a text prompt with the right extension?

[tight surge]

hello friendly diffusers - i've got a SD install running local on my m4 mac w/ 24G memory and what I'm trying to figure out is a way to use ControlNet or such to get SD to use the image I give it (img2img) of a particular piece of jewelry. The AI agents tell me this is possible but when I've followed the advice that Perplexity gave me it lead me nowhere. Anyone here have any hints for where I can go to make this a reality? I'm using Automatic1111

[hollow sleet]

hello

[quaint swift]

hello

[warm junco]

Hey, you need to install the controlnet extension.
You can find it in the extensions tab, there click on available, load from, and search for sd-webui-controlnet, then install and relaunch auto1111.

[stone tinsel]

Hi team,
I wanted to discuss a partnership proposal regarding developer ecosystem expansion. Could you please guide me with whom can i talk?

[still glacier]

You ought to reach stabilityai's team directly for that.
https://stability.ai/contact

[valid cave]

Hey guys, I have A1111 locally on my system, for about a year or 2, and I am wondering if it's worth it to upgrade to ComfyUI or forge. I am mainly using Illustrious models, and if so, can I just transfer my model files from A1111 to comfyUi or forge? And is it difficult to learn the new systems?

[junior urchin]

I find it funny how all these web3 and 'developers' want partnerships

[warm junco]

Hey, which gpu do you have?

[valid cave]

An RTX 4060

[warm junco]

Then you can try forge or forge neo

[valid cave]

What's the difference between the two

[warm junco]

Forge is more stable on old hardware or lower vram.
Forge neo is newer and uses the latest torch and performance enhancements.

[crude carbon]

hi

[bleak pond]

sup ai buddies

whats ur favorite ai

[grave crown]

hey everyone

[hollow sleet]

hello every one

[round pivot]

Hey everyone, curious to know - how are you all handling GPU scaling for inference lately? I’ve been seeing a lot of teams struggle with cost and availability

[atomic mortar]

Scam

[low moon]

https://www.youtube.com/watch?v=zQThHCB_aec

[steep copper]

do you use any framework

for inference

[radiant turtle]

who has Account in liblib.art pls i need just to download models 🙏😿

[bitter jolt]

For training SD-2.1 do you guys use the velocity instead of the noise?

[hollow sleet]

Hello

[lofty tendon]

/generate prompt: A lone Hanfu Warrior with a glowing digital fan standing on a floating pagoda platform, Chinese ink wash art meets Blade Runner cyberpunk concept art, Volumetric neon light, misty rain, red and jade palette, digital glitch effect, cinematic lighting, 8k

[pine vale]

Hi is it possible to create video by feeding Manga Panels to the AI and it will create a video?

[ebon barn]

通过

[keen hull]

https://tenor.com/view/unsrscandidate-tyranno-dance-japanese-gif-14945574192896794857

noooooo

[tribal pond]

what happened to stable diffusion, t

did this company go bankrupt?

[still glacier]

They're still here, https://stability.ai/news

[hollow sleet]

stable airdrop update

[rich notch]

hello

[narrow falcon]

hello

[cobalt geyser]

Are there any good webcomics/webtoons to read that are made with AI? I'd like to read their comment sections and hopefully they are positive.

[quartz pelican]

I wouldn't hope for that🤣

[cobalt geyser]

Would you know of any comics?

[quartz pelican]

AI? Nope

However, considering what Webtoon users usually praise... 🧐

[cobalt geyser]

What do they usually praise?

[quartz pelican]

Do I know? They're, like, teen girls or something.

[hollow sleet]

Hello frinds

[neat scaffold]

so, I'm new to the discord, and the program in general, but, my laptop that I'm trying to use stable diffusion on has an rtx 5060 gpu, but, from what I'm finding, the all in one package that I found isn't updated for the 50 series nvidia cards yet, do I just have to wait, or, is there a simple ish fix for this?

[primal current]

Can anyone tell me if SteadyDancer is better than Wan Animate?

[neat scaffold]

trying something, but, not sure yet

welp dawn? also, sent a msg

[neat scaffold]

@vapid dove, sent a dm about a potential scammer

[atomic mortar]

Happens a lot sadly, check #🤝｜tech-support pinned messages, iirc forge neo supports rtx 5000 series cards. Personally i use swarmUI and works great

No help request should be in dm's unless its something very specific nfsw and requires no downloads

[neat scaffold]

ok, wth??? I was downloading it, and, it first, at 1.5/1.6 gb downloaded, stopped, with a "site not available error" and then just completely cancelled the download

wth????

[atomic mortar]

Weird, probably just coincidence but you can check with CS1o in #🤝｜tech-support since he maintains the page of the tutorials

[neat scaffold]

pretty sure it was was a weird download glitch, lol

[valid anvil]

is there anyone making bots, scrappers or any other automation bots for websits

[still glacier]

えいごで書いてください。
それに、 5則をよんでください。

日本語ボトはありましたか？ふつくないですね。

Rergardless, please write in english.

[neat scaffold]

so, idr what exactly it was called, but, a friend was telling me about a way to save prompts to predetermined files, so, for example, if you get a series of prompts that are fine tuned to get, say, a certain facial expression, eyes, or mouth, for example, you could just direct the ai to that file, rather than typing the full prompt. does swarmui allow that?

[left geyser]

hi, I am looking for people based in US to work with our business. feel free to DM me.

[atomic mortar]

Like, wildcards? Yeah

Wildcards can do both random wildcards or just a single prompt in them

They can even carry loras but it takes a bit of tweaking

[gleaming ember]

hello

[quartz pelican]

Hi

[neat scaffold]

do people usually do shit like I mentioned? aka, wildcards for, facial expressions, eyes, eyebrows, facial shapes, shit like that? I know a friend of mine does, and, when I get into wildcards, and better at prompts, I plan on it, but, not sure if it's a common thing to do, or, if it could even pull off what I'm thinking anyways

[atomic mortar]

Meh some people use ai for their prompts, some people do exclusively wildcards (takes out a lot of effort), just what works for you

[neat scaffold]

ok

[atomic mortar]

I got wildcards for poses/bodytypes etc

Mostly for when i test my loras

[nimble bone]

are there any workflows out there that are relatively straight forward to setup for I2V with images generated with illustrious?

[coarse gate]

hello, what booru tag extractors are you guys using?

[coarse gate]

welp vibe coded my own, if anyone wants to try it out, just put the whole folder into extensions folder. https://github.com/Sasig69/StableDiffusionTools

[elfin laurel]

hello

[halcyon mason]

Hi everybody

[fervent thunder]

Hey guys, I'm very ambitious in this field and looking to connect with like minded peers to exchange knowledge, experience and resources, lets grow and build together!

[glad haven]

guys hello

how can i find if im on the latest version of stable diffusion

[atomic mortar]

The webUI i assume? Probably outdated if theres no update option within

[glad haven]

yea

i thought i had pulled the latest in github

[hearty canyon]

Don't open this imgs he is showing how his future wife will get 🌚 by other men

[junior urchin]

@still glacier

[neon hawk]

Hi.

[crisp crater]

hi

[still glacier]

no

[glad haven]

Hi guys my web ui shows at the bottom version 1.10.1 but on youtube videos i see there is a higher verison

any idea what i am doing wrong

[still glacier]

which webui ? Also that s probably a question for #🤝｜tech-support let s move there.

[royal patio]

I updated my app for you guys.
RealLens this will help make your images look real with one click and nothing to download. 100% free
https://ai.studio/apps/drive/18EbiOreX1TA9gps7wr6UMv5-DsQAbsHy?fullscreenApplet=true

[primal current]

Which one is faster in ComfyUI with RTX 5090?

fp16
or
bf16
or
fp8

[tender totem]

hi

[honest maple]

Hello

[latent dawn]

Hi

[wet grotto]

guys what solution do you recomend me to lip sync?

[round pivot]

As for docker images, we accept it on any framework of your preference. We also plan on having templates, that would be vllm only initially.

[steep copper]

i fixed the gpu thing by orchestratıon

since I'm using rtx 3090 on different machines

i have clients in 3 machine that has 3x 3090

and i have orchestrator

when request comes im sending available client

and caching ai model so if all gpus full im deleting cache from random and put in queue othetwise im using same gpu for same model

[hot sigil]

👋

[rustic pecan]

👋

[still glacier]

get lost scammer

[bronze lintel]

Hey everyone 👋
I’m Riku, an AI engineer & founder from Turku, Finland. I run Since AI, a hackathon + builder community where people turn cutting-edge AI ideas into real products.

If you’re into hackathons, deeptech, or shipping AI projects from 0→1, I’d love to connect and maybe co-build something.

You can find me here as well:
– LinkedIn: linkedin.com/in/rikulauttia/
– X: x.com/rikulauttia

[still glacier]

rule 5 #✍🏼｜rules-and-tos

I have quite a sensible trigger when it comes to rule 5, I see links and formating, talk about dev => I warn people. So far you re good I might have overreacted. But please.... don t be a bot, don t start advertising the moment I turn away

[rustic pecan]

👋

[tough sleet]

Is there anyone looking for developer?

[still glacier]

no

[brittle rose]

I hope YouTube monopoly fades like AOL in history due to its anti-consumer aggressive bias policies and artificial zero-tolerance removal of channels

[hot sigil]

👋

[hollow sleet]

hi

[spare grove]

hello

i recently bought a 5070 so i could train ai images, any tutorial videos you would recommend?

[atomic mortar]

@brittle rose your channel got deleted? Probably due the descriptions with "Taylor switft esque" or something

[vital nymph]

Hi everyone 👋
I’m BabyHee from Singapore 🇸🇬

I work with video, photography, and AI tools like Runway, OpenArt, Midjourney, and Pika.
Currently exploring AI for video interviews, image-to-video, and creative storytelling.

Looking forward to learning, sharing, and exchanging ideas here.
Nice to meet you all!

[brittle rose]

https://www.youtube.com/watch?v=b2sSpqs2xa4 still working. I am talking about my old channels in which they interpreted it as spam because I was using some videos of my people creatively with added value they they don't certify in their narrow mind but people were enjoying them nevertheless and only YouTube AI treated them as such.

[hot sigil]

👋

[hot sigil]

👋

[primal lance]

👋

[devout remnant]

can someone walk me through setup

[brittle rose]

تَاجُ الطُّهْرِ — The Crown of Purity 🕊️🎸✨🛡️🎹

Arab Nu Metal · cinematic female soprano · maqam-rooted melody · expansive runtime
Deep downtuned guitars, realistic amp character, electric oud phrases, piano swells, and grounded percussion — everything sits wide, calm, and intentional.

The core feeling:
A dignified statement wrapped in weight and beauty.
Not loud for the sake of noise, not soft for the sake of comfort.
||Balanced heaviness, spiritual clarity, and careful restraint|| — the kind of track that breathes instead of rushing.

No exaggeration — this is structure + belief + craft.
The soprano floats above the riffs with purity, while the low end stays disciplined and warm.
Mood: composed, resolute, uplifting without pretending.

Special note (audio quality):
This is the remastered SoundCloud lossless version, rebuilt with an even more realistic amp response and higher fidelity, very close to full DAW masters.
It took 39 iterations in FL Studio to reach this balance — no shimmer, no thinning, no quality loss, just clean depth and honest dynamics.

SoundCloud (lossless remaster):
https://soundcloud.com/tber-mohammed-mehdi/the-crown-of-purity

Producer.ai version:
https://www.producer.ai/song/70ec805a-8b6a-48a1-981a-a57a14f3a511

[atomic mortar]

Depending on your gpu theres a really easy guide in #🤝｜tech-support pinned messages

Nvidia is easy, amd more difficult but you do need a decent gpu regardless

[distant swan]

can anyone tell me how i can run this ai generator on my own server\

[fiery ore]

What kind of AI generator

[hot sigil]

👋

[distant swan]

For like when I give it a prompt it will upload it

[atomic mortar]

Basically a custom discord bot with api to stability but costs $$

Would be easier to yeet in midjourney tbh

[fiery ore]

Send a dm

[limpid compass]

Looking for robust baselines (papers + repos) for mask-based video inpainting / object removal with partial occlusions and moving camera.

Constraints: 10–60 frame occlusions, ~1080p footage. Assume I can provide a per-frame mask (or sparse masks if required). Prefer methods that explicitly address temporal consistency (propagation/refinement, flow-guided, transformer-based, etc.).

Any recommended starting points?

Cross-posted in #stable-video-diffusion as well in case research folks are there.

[atomic mortar]

@distant swan if the visionforge ai guy sends stuff you need to download its most likely a scam

[graceful stirrup]

hi

[brittle rose]

I made an AI rock song much better here; #🌶｜off-topic message

[sharp garnet]

https://github.com/shootthesound/comfyUI-Realtime-Lora

in the "Getting Started" section, the Musubi Tuner Z-image instructions tell me to "select your de-distilled model, VAE, and text encoder from the dropdowns"

is a de-distilled model the same as Load Diffusion Model?

it also asks me to "Paste the path to your training backend installation (sd-scripts, Musubi Tuner, or AI-Toolkit)"

where do I paste it to?

nvm I'm supposed to open the Realtime LoRA Trainer noede in Comfy UI and select the stuff there

I'm stupid

[glad horizon]

Built something worth checking out👀

[junior stag]

hello

[primal lance]

👋

[rustic pecan]

👋

[astral thunder]

👋

[sharp garnet]

I'm in VC, can someone please help me out?

[junior urchin]

@still glacier

[grim epoch]

Can someone help me? Im new to Comfy and supposedly I installed the manager but i can't see it

[cedar pivot]

Hello everyone, how does the day or night pass?

[mossy edge]

Quick question for SD users here — do you usually run jobs locally or rent GPUs?

I’m testing a cheaper way to run batch SD jobs on idle GPUs and looking for 1–2 people to test this week.

[atomic mortar]

if you already have a gpu and can run a model its always cheaper to run it locally

if you dont plan on making it a full gig, renting or api is cheaper otherwise over time having a gpu is cheaper

depends on your goals

[mossy edge]

That makes sense.
For people who don’t own a GPU or only run occasional / burst jobs, what do you usually recommend — RunPod, local colab, or something else?

Also curious: what kind of SD workloads do you run most often? (resolution / batch size)

[spare grove]

resolution depends on gpu, you are very limited with just 12gb on a 5070 but 50 series seems to be quite a step up overall from 40 series in ai at least

i would say wait for super series but because the ram prices shot up so much, im not sure, but i would get at least a 5070 ti with 16 gb so youre not stuck on 1200x1200 size

[stable knoll]

Hello. I am new member here and really happy to have an opportunity talking with you.
I consider it incredibly fortunate to have joined this server.
It's not about where you are, but what you do that matters.
As a senior developer, I want to work with you all on everything.
Can I help?

[atomic mortar]

"hello im definitely a real person pay in blockchain currency thanks"

[warm compass]

Hello

[bold trout]

for general ai, does the increased storage space required = more gpu vram usage/requirements? like 50gb text gen model is going to require a really really beefy gpu?

[lethal wedge]

Hey everyone—new to the group. I’m building a personal second brain system: basically, using AI glasses or a pendant to capture what I see/hear (price sheets, notes, car parts, family stuff), auto-save it locally (Raspberry Pi + external drive or similar), hash for tamper-proof integrity, and query it anytime—like “show me the building costs from last November” pulls exact clips/sheets. Goal: accurate, growing memory vault that’s like a better version of my brain—detailed recall, no cloud, full trust in the data.
Anyone else working on something similar (lifelogging, local RAG setups, always-on capture to Ollama)? Suggestions on tools, hardware (Pi + 5TB externals?), or pitfalls? Open to hashgraph ideas for immutable backups too.
Thanks! Excited to learn from y’all.

[mossy edge]

what’s the most annoying part of your SD setup right now?

[warm yew]

test

[primal current]

Is there a way to download Sora videos without a watermark?

I mean methods like the ones used on Instagram/TikTok where they download them without a watermark.

I'm not talking about something that edits the downloaded video with the watermark.

I've created a private API, but I haven't been able to find a way to download the original video without a watermark.

[grim fossil]

Test

[spare grove]

training loras seems annoying

[atomic mortar]

Scams be like

[oak turret]

👋

[upper citrus]

Can someone please tell me where to post a dataset folder I've created for feedback and corrections?

[oak turret]

👋

[primal current]

How can I get it to maximum speed and quality for Z-Image Lora training with Musubi Tuner?

I'm using a B200 (CUDA 12.9, PyTorch 2.9.1, Python 3.12, Linux).

Can make 200/200 [00:47<00:00, 4.23it/s, avr_loss=0.43]

I've tried many methods, but the VRAM usage won't go above 70GB. I want maximum speed and quality, I would greatly appreciate your advice.. Forgive my ignorance, but I know more about practical than theoretical aspects of AI.

[upper citrus]

I'm looking for a way to generate AI images with really good resemblance that doesn't require days/weeks of troubleshooting hell.

[karmic jacinth]

guys do you think it good idea to run sdxl models wiht 6gb of vram?

[gray folio]

Opps

[atomic mortar]

It works

Not super fast but works

[gentle prairie]

is there a way to keep the exact same seed on each generation instead of doing 1 image at a time? I'm using dynamic prompt to try different stuff

if i put the seed, it will always do anotheer seed on the next images

[primal sundial]

yall i have a laptop with amd radeon tm am i cooked

[hollow sleet]

hello frinds

[faint timber]

@gentle prairie Yes. The way I handle that is drop down a Seed node that connects to a Set node. Then every node that uses a seed, I drive with a Get node fetching the seed value. Then you can "fix" the master seed and run multiple gens, tweaking values without the concept radically changing.

[late cedar]

is automatic 1111 still being developed?

[atomic mortar]

No

Or atleast not by the same crew if it is iirc

If your on newer hardware (5000 series) i recommend forge neo, or just forge in general

Since it looks similar

[thorn hare]

how do i filter for civitai videos with a workflow ? i filter for metadata, but when i drag the video file into Comfyui, it still says "Unable to find workflow"

[atomic mortar]

Video files don't carry metadata like the images do iirc

[thorn hare]

damn 🙁

[outer shadow]

/generate

[thorn reef]

👋

[foggy flax]

Hello everyone, I want to ask a question: will Stable Diffusion run on the following configuration: NVIDIA GeForce RTX 3060 Laptop GPU, Intel Core i7-11800H? If not, what alternative solutions can you recommend?

from Kazakhstan with love :3

[trim sierra]

👋

[warm junco]

Hey, yes it will run on that

[foggy flax]

Which Stable Diffusion build do you recommend I run? I need to make an animated clip.

[warm junco]

Forge or Forge classic

[warm junco]

post the complete cmd log in #🤝｜tech-support

[plucky yoke]

[stable night]

Hi All

[feral pike]

whats the recommended setup for linux users (i have a gtx 1070)

[weary zealot]

Yo guys

[magic fiber]

Hello

[warm junco]

Forge or ReForge or forge classic

[oak turret]

👋

[upper citrus]

Hi — I’m looking to commission a character LoRA (starting with Thorn Rose from Sonic Prime).

I’m specifically looking for someone experienced with stylized character LoRAs, especially identity consistency across poses or multiple characters. I’ve already tried training myself and would prefer to hand this off.

If this is something you do, I’d love to see examples and discuss scope.

[rose stirrup]

I've been out of training loras for almost 1 year, has appeared any way to make consistent characters from one image? the options i found on youtube are only from openart and im looking to do it local, also uncensored

[fast sage]

Hey lads, I'm trying to make a creature image to video on wan 2.2 but NOTHING I prompt seems to allow me to do an animation with the mouth closed, the creature always seems to be moving the mouth like it''s talking, suggestions?

[rustic pecan]

👋

[astral thunder]

👋

[rustic pecan]

👋

[atomic mortar]

Scam

And another scam

[magic fiber]

Hello how are you?

[fiery ore]

Can I DM?

[red wagon]

morning folks ☕ hope all are happy that its finally Friday. 🥳. Im new here looking around trying to see how every one handles there Image creations and or media creations etc. right now i have a massed MESS folder of around 25k lol not sure what to do with them so im working on an Auto sorter lol

Interesting like HAHA good luck or interesting like another repo remake lol ... Im just curious i been working on this project for like 2 weeks (my own personal agenda) lol im right at the gotta go see what is out there maybe me buiding one is harder than using one already made lol . but now im just at a point of wondering how others store manage their creations because for me lol when i create i dont have time to inspect each image at the time of creation im simply prompting tuning till i get what i need etc 4x variations then i download everything and drop in one folder lol . Maybe thats just me lol IDK 🤔 but i keep it all casue well i paid for them all lol but ya makes a mess folder haha

well right now thi si what i haev but whern i design things I always leave my self wit hScalability exits and expansion enhancement options. IM still at the (FOR MY TRASH) lol but its pretty close im still implementing tuning lol here IM not a tech TALK guy I m problem /Solution guy 🤔😊 Here is a description etc my AI made for me to make so i dont have to keep explaining to other AI and people lol : # The Executive Summary: Chimera Master Project

Project Mandate

The Chimera Master project was initiated to address significant inefficiencies in image sorting processes. Initially, the system relied on manual methods and fragmented tools across multiple project folders, leading to chaotic workflows, inconsistent results, and frequent errors. This manual approach consumed excessive time and resources while risking data inaccuracies, hindering operational efficiency and scalability.

The Solution

Chimera Master transforms this disorganized environment into a unified, automated data curation pipeline. At its core is a single dashboard, launched via "chimera_gui_phoenix.py," that orchestrates the entire process from initial sorting to final validation. The system employs a dual-engine approach: the Triage Engine uses visual probability for quick categorization, while the Semantic Engine leverages semantic analysis for detailed descriptions. These engines feed into an audit layer that reconciles discrepancies through "The Audit" phase for analysis and "The Judgment" phase for manual conflict resolution, ensuring a seamless, error-free output. Users can initiate the full process with a single-click activation of the "Full Auto Protocol," producing clean, labeled datasets ready for applications like training models or archiving. Optional post-processing via "Text Refinement" allows users to clean up descriptions using checkbox filters in a dedicated interface.

so i hope any of that makes sense lol but it in my eyes is a here is a folder.... click a butoon now sort organize tag create reports of TXT and cvs for data structure use for my personal LLM to use etc lol thats basically what i made it for haha

[north trout]

#💬｜general-chat a quiet cat

[atomic mortar]

🔇

[red wagon]

Cool , perhaps a possibility as i said still just looking to see if what i have (scalable and expandable) etc is needed by others or if it is jUST me being a make an image keep an image etc...lol hence the big mess... regarding othe rmedia types Yes as i said i design wit hteh expansion of intigration of othe rmedia types , Im just not there yet but always in my head while designing becasue next is I have a suno folder wit h 10000s of songs and song variations etc same problem lol same wit hvideo and video clips etc lol MESSY messy lol. sorry friday and more coffee and kid to get ready for school. so ill be on and off as i work on my projects today if you wan to discuss anything shoto me a DM or whatever 👍

[tepid grove]

how viable is it to run hunyuan 3.0 on say a mac studio

with just transformers

[upper citrus]

Okay I already have 3 people interested, so I don't need any more.

[red wagon]

Canada. but i think i wont be integrating and starting expanse until after Christmas because im still piecing my stuff together and TESTING and also bug hunting and system purifying etc For my own Local LLM lol (WHICH is why i made chimera and teh data set it uses as RAG for visuals etc via CVS doc file

[smoky grotto]

I'm trying to install it, but it gets to a point where it asks me to log in, and then it gives an error.

Can someone help?

stable-diffusion-webui

[fiery ore]

I would actually love to show you my work if you don't mind

[upper citrus]

Where can I find it?

[fiery ore]

Check DM

[iron kestrel]

Obsolete AI 🫵

[primal sundial]

anyone have a sora code they could share?

[zealous bobcat]

Generate a vibrant and festive artwork celebrating the Chinese New Year, incorporating traditional elements such as red lanterns, paper-cut designs, and decorative motifs like dragons and flowers. The composition should be symmetrical, with a focus on red and gold colors, contrasted against a bright and clean white background. The artwork should convey joy, prosperity, and good fortune, making it suitable for widespread use during the celebrations.

[sharp gull]

Can I pay anyone to teach me to setup runpod and a workflow?

Ive never used comfyui because my hardware is not that great, I am going to rent a runpod GPU and set that up. I want to generate lip sync videos using wan 2.6 or infinite talk. These models allow you to give an audio file and an image and lip syncs onto that image with comfyui could I give it 4 images at once with one audio and have it generate 4 different videos at one time? I seen comfyui is node based editing just wondering if that is possible

[atomic mortar]

Personally no but if anyone asks to dm assume its a scam first, server gets a lot of those

[peak tendon]

Anyone could help me with troubleshooting?

[peak tendon]

Eh, already "solved" the issue

It was about automatic 1111. I couldn't get it to work no matter what, but apparently it's not my fault and 1111 simply doesn't work

Im not sure, whatever the fuck it was. I just switched models to neo forge.

[quick roost]

Hi guys! Sorry another newbie.

Saying that I have got Stable Diffusion running on a Lenovo Yoga Pro 7i with only Intel GPU and 16G of ram with the SwarmUI. OK it's slow.... very slow taking sometimes an hour and a half to render an image. So, with the January sales about to break upon us what would be my best bet for a replacement, faster laptop (I've not been happy with this one since day 1 because it's slower than the failed HP 5i with 8G ram)?

Do I need to go to an Intel Ultra 9 or would the Ultra 7 be enough if I go for a fast Nvidia graphics card with plenty of GPU.

BTW, I've found generating on my own system with the same, or lower, setting to a web site gives a far higher quality of image.

Not a BOT. I'm @regionalphotos on the VERO platform.

[modern pagoda]

40 series or 50 series, 70 meh, 80 better, 90 gold. Up to you & costs tho. I7 / ultra 7 is equivalent of old i5 ranges, just modern. I9/ ultra 9s are equivalent of old i7 range of products.

[still glacier]

intel gpus are still very niche when it comes to AI support. If it took you an hour and a half to render anything chances are it wasn't even using your gpu and was using only your cpu

[modern pagoda]

You can run older smaller models as well.

You can get away with 7, higher ram, & a 4080, 4090, 5080 or 5090 as high end from low end gpus.

But CUDA & the nvidia have to do all the work

Not the cpu

[still glacier]

And when it comes to "what should I buy for AI", answers are always the same.
1/ prioritize vram. And I mean Vram not ram, the memory that's actually soldered to your gpu. The more there is, the easier and faster it will be to use heavy models. There are some hacks and methods to load models partially on your gpu. But there are limits to this method and it will slower your renders significantly.
2/ 16gb of ram is theoretically enough but your system might freeze at times, you might have to use your disk as ram, etc... So if possible get 32gb
3/ should I go AMD or Nvidia ? Nvidia has been around the block for way longer than AMD when it comes to AI and Scientific Computation (cf cuda). But nowadays AMD can do pretty much 90/95% of what Nvidia can. It might be a bit more tricky to install. But if it saves you hundreds of dollars it might be worth it.
4/ "ok but which gpu exactly should I get ?!?"
If you want to generate videos, I wouldn't recommend anything under 16gb of vram, if possible get more than that
If you stick to image generation. You can aim for some pretty freaking low specs models. We've been running SDXL on 4gb models. But with how vram hungry the latest models are, I would not recommend getting anything under 8gb. And 16gb is probably the sweet spot for now.

[quick roost]

OK so Ultra 7 or 9 really makes no difference and it's the ram and Nvidia GPU that will make the difference. Knowing that will save a lot of £££.

[still glacier]

the VRAM

Under normal circumstances, ram should pretty much only be used to pass the model to your GPU.

Or for some minor operations.

(for instance, some text encoders work pretty fast on cpu+ram)

[quick roost]

When the dealers say 8gb of GPU they really mean VRAM? I know I need at least 16gb of RAM to run my old Photoshop software.

[still glacier]

When the dealers say 8gb of GPU they really mean VRAM?
Yes.
Unless they're trying to sell you a laptop with an integrated gpu. In which case cpu and gpu will share the ram.
Integrated gpu != dedicated gpu

[modern pagoda]

4080, 4090, 5080, 5090 are top models for VRAM, not cheap, especially in lappy

[still glacier]

Other than that, yes once you've removed the vram bottleneck the more recent and the higher tier the gpu, the faster it will be.

nvidia xx90 models are ridiculously expensive unless you are looking for some professional hardware and have money to burn I would not recommend them.

[modern pagoda]

Gougers too

[quick roost]

I think the one I was looking at a day or so ago was a 5080 in an ASUS laptop. I'd prefer an HP as originally they were Compaq and really good quality.

[still glacier]

xx80 are somewhat more reasonable but even xx70 are fine as long as they have enough vram.

[modern pagoda]

If you game, do graphic editing, gen ai, its worth costs, just get a cooling pad too in case it heats up

Images vs Video. Video will be far more intensive.

LLMs no prob, most gen AI models as well for txt2img img2img type uses afaik. Txt2video/img2vid will be trickier.

[quick roost]

@still glacier @modern pagoda I'll post my latest render on the general-with-images group.

[modern pagoda]

Good luck

[tribal gorge]

Helo

[broken moat]

HELLO

[main finch]

Hey I want my own ai tool for virtual try on can anyone help built

[spare grove]

peak english, just imagining helping you gives me a headache

[stable bough]

https://youtu.be/U2puGN2OmfA
I’d really appreciate any honest feedback on:
The debate structure and pacing
How engaging the arguments feel
Thumbnail/title/hook for click‑through and retention
If you enjoy it and want to see more AI‑vs‑AI debates on controversial science topics, it would mean a lot if you subscribe to the channel and drop a comment with future debate ideas

[elfin laurel]

hey, i’m an AI and full stack dev, been working a lot lately on automation pipelines that link slack, notion and some custom ai agents i’ve built for clients, mostly around rag setups and multimodal stuff like clip and yolo mixed with a bit of voice ai. it’s been kinda wild seeing how many projects turned into something real for people using them, feels good honestly. always tweaking things, learning, and letting the systems grow on their own in a way.

[devout vapor]

Is it ok to ask if anyone has a checkpoint that I can't find online?

[atomic mortar]

unless its previously deleted somewhere odds are you wont find it

looking for flux 3 :^)

[quick roost]

@still glacier @modern pagoda Sorry to bother you again, however I've found a dealer selling HP OMEN 16-am0507na 16" Gaming Laptop - Intel® Core™ Ultra 7, RTX 5060, 1 TB SSD with 24 GB of DDR5 RAM and 8 GB GDDR7 on the video card at £800 below the HP recommended price. I'm a 66 year old UK pensioner so do you think this would be a good deal?

[atomic mortar]

8gb on the graphics card is pretty painful for ai but for gaming it's ok

Especially since its a laptop the performance is going to be a little less then a desktop version

[quick roost]

But a lot better than my current laptop with unusable intel GPU? Current render times using SwarmUI are 1.5-1.75 hours.

[atomic mortar]

It depends on the model you want to run to be honest

[quick roost]

Currently running StabilityAI SDXL and would like to try SD3.

[atomic mortar]

For lightweight models like sdxl/illustrious its ok (15-30s gens?), for sd3 i would say no

Im on a 5080 and i get about 60s of generation time with sd3.5 large

Wan image to video about 5-9min and z-image turbo 5-10s

Tbh it's going so fast lately even my 5080 can't keep up with the latest models ( flux 2)

[quick roost]

Difficult to decide... I could go to Ultra 9 with a 5080 but it's more than twice the price.

[atomic mortar]

since i game a lot and use it for rendering stuff it was worth it for me, but if its for just AI i personally couldn't justify it

But if its worth it is up to you to be honest, since you use swarmUI you could also ask the question in their discord, a lot more smart peoplle there then me

[quick roost]

Cheers for that, I didn't know there was a SwarmUI discord. My main usage would be for photo editing and getting to grips with AI. I can render on-line but found the quality isn't the same. to be honest a 5-10 minute render time would be acceptable.

[atomic mortar]

Its on the github, under status
https://github.com/mcmonkeyprojects/SwarmUI?tab=readme-ov-file#status

Its pretty active and you get news if theres updates or new models released

Some peeps however can be a bit blunt however in my opinion

[thorn storm]

Hi everyone! New member here. I created a free open-source tool that works with ComfyUI and I'd love to share it with you all. Which channel would be best for posting something like this? Don't want to post in the wrong place! Thanks for the help!

[pulsar juniper]

https://stopots.com/pt/09154

stop online

[fervent thunder]

lol

[glad torrent]

Is this inpainting plugin/method still relevant?

https://github.com/lquesada/ComfyUI-Inpaint-CropAndStitch

[junior urchin]

#💬｜general-chat message #💬｜general-chat message #💬｜general-chat message @still glacier

[vapid onyx]

hi

[high flint]

when it comes to flux checkpoints, which ones have been your favorites?

[proper rover]

Hello

[bleak dew]

hello I've been trying to install Stable Diffusion for days. I've followed every guide for AMD, but it never works. I wish I could use this program, but I don't know what else to do.

[humble mason]

Have you tried just installing Pinokio? It makes it all so, so much easier.

[untold vault]

[modern pagoda]

No one click on those

There's no support server. Thats a scam.

[atomic mortar]

Scam

[viral bison]

Hi everyone!
I’m doing a quick survey about AI-assisted RPGs.
Even if you’ve only tried ChatGPT, AI Dungeon, or an AI narrator once, your experience counts. It takes less than a minute and is anonymous.
👉 https://forms.gle/utHoNaZWHr1c3sJW7
Thanks for helping out!

[bleak dew]

No, I haven’t tried Pinokio. I honestly don’t really know what it is 😄
I’m completely new to this whole topic, and it’s actually the first time in my life that I’ve ever had to work with CMD

[spare grove]

laptops are generally pricy and often have a weaker version of a normal 5060, mobile gpus can even be a tier lower then advertised

if it says 5060m, its closer to a 5050

[still glacier]

Sorry about those. Kinda busy at this time of the year (and they know it)

[modern pagoda]

No worries. just dont want users fallin for it

🤘🤞

[tawny trout]

Hello everyone

[low moon]

hi

[outer shadow]

hi

[wet zephyr]

Hello Great People!

Excited to Join this community.. Lets build something Great!

[primal current]

Does anyone have a method to register anyone with Sora verification for cameos?

I thought about using the OBS iOS repository and having a video of myself swapped with my AI.

But I don't know if it will work.

[junior urchin]

@still glacier #💬｜general-chat message #💬｜general-chat message

[karmic jacinth]

is it normal for regional prompter to take lots of memory?

cause my 6gb vram gpu going crazy

[mossy edge]

For people running ComfyUI in production: what’s the most fragile part of your current setup that you’re scared to touch?

[harsh current]

Congratulations

[bleak dew]

Does anyone know a good tutorial for installing Stable Diffusion? I’ve tried many, even ChatGPT 😄 but I just can’t get it to work

[still glacier]

Pinned guide in #🤝｜tech-support

[mighty ravine]

Merry Christmas guys

[viscid apex]

hi

[hollow sleet]

hello frinds

[stoic osprey]

oh oh oh merry chrismas! ^^ https://m.youtube.com/watch?v=GO8U2IDoRgc

[ember urchin]

HI

[ember ice]

is there anyone looking for AI and Blockchain dev?

[atomic mortar]

no as 9/10 are scams and it breaks rule 5\

and assuming this is a bot anyways

[inland burrow]

Merry Christmas! 🎄
Hope you’re having a relaxing break and that 2025 has been kind to you; fab campaigning etc..
Just wanted to wish you well and say hello — hope the new year brings good things. Memorable times,
Robert (& also Elaine and now with both a Honey and a Heinz)
🤓🥳🐺🐺

[junior urchin]

@still glacier

[still glacier]

Beep boop 🤖
I gave them the boot 👢

[modern pagoda]

Holy buzzwordapalooza

[junior urchin]

@still glacier

[hazy falcon]

https://www.tiktok.com/@top100_real
Anyone know how these videos are made? Crazy stuff

Questionable subjects aside, the quality is really good lol

I'm new to SD btw so maybe this is easier than I think

[merry gust]

hey is there a image to video option on stability matrix?

[proven badge]

quuestion what website do i use to connect with racist and anti lgbt people??? since i know there is 4chan and instagram but yeah im looking since i hate censorship and i just want a place anti censorship i know discord is not the place for me anymore

[still glacier]

Somewhere else, somewhere really far away from here. Please don't bring it up again, that is indeed not the place for this. Literally rule #1.

[unreal axle]

hey

I’m new, Stable Diffusion doesn’t start / my images look bad

i need help

[atomic mortar]

its probably your prompt or some settings

i recommend looking on civitai and you can find images with the model you use and settings

[unreal axle]

i need peython?

[atomic mortar]

?? so do your images look bad or you cant generate at all?

[steel fable]

there is anyway to create gif or short videos?

[orchid ivy]

How can I create porn

[still glacier]

Not the good server for that. Rule 2

[junior urchin]

I have to be honest

How is that word not a naughty word

[hazy falcon]

Yea you need python

[unreal axle]

tx bro now i see tutorial

[mortal shadow]

Hey mate

[steel fable]

anyone knows any extension or something to create gif or short videos please?

[dense shoal]

https://www.reddit.com/r/invokeai/comments/1pw9ty7/in_case_you_havent_seen_it_yet_invoke_v6100rc1/?utm_source=share&utm_medium=web3x&utm_name=web3xcss&utm_term=1&utm_content=share_button

[sand jay]

invoke got Z-Image models supported now

[tulip tangle]

What is the latest news from stability ai ? Any mew tech coming soon ?

[trim sinew]

ello

Can someone tell me how this person makes his AI videos? https://www.youtube.com/watch?v=4-osOmV-5PI

I want to make same kind of videos

I suppose it's combining two images into one and then using image to video AI

but how?

[bright ember]

Hi, could you help me? I can't see any sensitive content on platform X. I can't figure out why. I live in Italy and I'm an adult. Can you help me?

[trim sinew]

That's a good thing

I'm an adult too

[bright ember]

The problem is that sometimes when I share my creations, they come out as sensitive content.

[trim sinew]

Hmm.. ok

Are you legal age in the Twitter settings?

[bright ember]

I'm 40, I don't know what I'm talking about. I've watched a lot of videos, but nothing has changed so far.
🙁

I don't use VPNs because they slow down my internet too much.

[cold moss]

Does anyone here know which are the best communities to join for running ads on Meta? I'm looking to join e-commerce/ads communities to learn, get feedback on my ads, etc. Just curious if anyone here can recommend any.

[atomic mortar]

Not sure there's any since its usually a corporate job

[empty jackal]

Are there anyone looking for blockchain developer ?

Such as wallet, NFT, Smart contract, Defi, RWA, tradingbot, etc

Specially dont you need tradingbot ?

I think it's important to lots of trader.

BTW, Please feel free to contact to me .

[atomic mortar]

BTW feel free to read rule 5

[trim sinew]

@empty jackal I'm not into making money that way so no

I love working with my hands and have them dirty

Its much more rewarding than staring at some crappy graphs all day

[peak light]

Hi guys

Who’s online?

[stoic osprey]

merry chrismas! https://m.youtube.com/watch?v=GO8U2IDoRgc

[peak light]

I2v?

And characters reference?

[stoic osprey]

hey! I used Nano Banana for the character references (all video with AI)

[peak light]

What model for the video?

[peak light]

Great (:

[stoic osprey]

I've used several sites, it's a chimera, Qwen + Highfield + others ^^

[peak light]

Because my work is with a lot of people until we get the nvidia tower we are renting gpu
We must use comfyui for parallel video generating

[peak light]

Oh nice but it took a lot of time

[peak light]

lol

Nice trick I would like to show that to my boss
Can you send me links?

[stoic osprey]

Thanks for your feedback! It did indeed take me two weeks of work (I created images and video, and also provided sound effects and light VFX), all done in Sony Vegas and Adobe Premiere

[peak light]

Wow damn

With 5090 it could take less?

I mean local models with you

Gpu

[stoic osprey]

Of course! I would be delighted! 🙏 ❤️

[peak light]

Tnx!!

[stoic osprey]

I only own a Ryzen 5 laptop lol 😆

[peak light]

I showed the kids in class today how to refine prompts for videos models with o the llm

[peak light]

lol

Maybe I will share my work pc when I will get my server

My boss so annoying it takes him so much time to get me this shit

Hey

Do you know how to use vast.ai and comfyui?

I can send you link to work with me to create AI video

[stoic osprey]

It's very complicated and the results are uncertain, but... I discovered a technique to maintain physical consistency. We can't share photos, so I'll try to explain: first, I created a character on Nana-Banana, then I asked Q-Wen (or others) to "move my character onto a white background." The result allowed me to create different situations while keeping the same character

[peak light]

Wow you are a pro

[stoic osprey]

My starting point is Rick and Morty characters against a background; from this image, I can ask questions like "he's driving a go-kart"

i'm not a VFX expert, but I have 20 years of experience in video editing. I really looked for a solution to solve this damn physical consistency problem (you'll notice that in all the sequences, Rick is wearing the exact same stained sweater) ^^

[peak light]

Damn

If you had comfyui you could just used masking lol

[stoic osprey]

I know Comfui, haha. Actually, I use the worst internet tools to make the best ones ^^

I have contacts in AI development, but my working method is very traditional

[peak light]

Do you want to get link for comfyui from me? I don’t mind renting server for you

[peak light]

Damn nice

[stoic osprey]

That's kind, but I already have the link ^^

I have too many favorite pages, it's hard to find it lol

[peak light]

I mean.. when I pay for a server I get link with token that I can share

It cost me 0.5 usd per hour

[stoic osprey]

That's cool, I have good contacts, however I admit that I had difficulty creating images better than mdjourney ^^

Now that AI videos are accessible, everything depends on a good starting image, and on that point I sometimes have limitations with nano-banana

[peak light]

So I will be very happy if you can make video with a server I will rent for you
Just tell me what models and workflows you need in the comfyui

[peak light]

Oh..

That sad

[stoic osprey]

It seems that Nano Banana is powerful, but I wasn't able to create what I wanted 🤔

[peak light]

Is flux is better?

[junior urchin]

They did nerf nano banana because it was too good

[peak light]

Whyyyyy

I’m trying to think now what models should I download

[peak light]

Ok I have template for wan video editing

[untold vault]

[unique sage]

Is there a decen adetailer for hands? yolo one kinda sucks, since it reads feet along with hands

[trim sinew]

How to generate AI videos without comfyui?

It doesn't work for me

[signal trout]

https://psyklore.substack.com/publish/post/176100735

[fallow hamlet]

Hello.

[unique flame]

Hello :)

[west osprey]

Anyone familiar with WAN on forge-neo? I got everything set up including low noise wan in the refiner. How do I use both high and low noise lora?

[harsh current]

ha ha

[junior urchin]

@still glacier dm bot

[still glacier]

Picture me surprised

[peak light]

Yo

[atomic mortar]

Zero Censorship is a horrible idea, thats waiting for legal problems

[lyric swan]

hi

where can i see tutorials please

[atomic mortar]

to install, #🤝｜tech-support in the pinned messages

[lyric swan]

thank you boss

[junior urchin]

@still glacier

[karmic sigil]

Hello!

I have a complicated question: Am I a dinosaur with Flux?
What should I be using for the most realistic NSFW ultra ultra real photos for now?
Using ComfyUI of course.

[atomic mortar]

Scam

[still glacier]

Those scammers really think I'm sleeping at 5:00 am. smh.

[still glacier]

Also, I can't prove it at 6 am but the link posted a bit earlier about some img2vid website is suspicious. (also "payment via telegram stars" what the ?). I'm removing it per Rule 5.

[trim sinew]

There is one particular model that produces the most natural looking skin

It's not perfect at all

[karmic sigil]

Shut. You're good. Stop being mad.

[trim sinew]

It's got all the pigmentations and imperfections that all natural skin have

I love that model

It doesn't produce any of that perfect smooth skin stuff

Like it gets as close to real unfiltered life as possible

I just don't remember how that model is called because I renamed it

But it's one of the most popular realism models in civitai

This might be the model you're looking for @karmic sigil

[foggy quiver]

Hi Guys

Is there anyone here who have experience in generating AI influence or model?
I need a help here

[serene wraith]

hey, anyone tried stable diff on 9070XT ? any guide or is there a lot of tweaking to get it work ?

[atomic mortar]

In tech support theres a lot of help requests for amd cards, iirc the guide from CS1o in the pinned messages (in #🤝｜tech-support ) should get you all the way there

[tropic frost]

got a bit of a conumdrum.
I trained a flux lora to see how it well can turn out, did this on civitai.
After i was done the website didnt allow me to generate with it. Seems to happen if nsfw tags are used in the training,
Downloaded the flux checkpoint on my stable difusion, and its not really working. Say its not allowed.
Any inbuilt safe guard in stable difusion when it comes to realistic lora oor anything like that?

[gritty raven]

👋

[velvet pulsar]

Hello

[serene zephyr]

What software are you using?

[atomic mortar]

"on my stable diffusion" your probably using the outdated a1111
I recommend using either forge (neo)or swarmUI for this depending on your GPU

[tropic frost]

im using zluda

[atomic mortar]

Hmm i recommend popping into tech support and asking what ui is recommended then for you

Cs1o knows the most about amd cards and image gen here

[tropic frost]

ok

[delicate bone]

I see some ppl use ( ) in prompt, what does this means?

[atomic mortar]

Just a way of adding weight to certain words in a prompt

You mostly do not need it

And unironically its better to do something like apple, apple, apple, then to do (apple) or apple:1.3 since that causes burn

And the first example does not

In stable diffusion models, flux etc i havent used weights at all

[bright ember]

hello

how are you guys?

[peak light]

Do you have any good workflows for i2v lipsync and character references?

[peak light]

I’m trying to figure out what workflows do I need for consistency

[bright ember]

show me some examples

[peak light]

In dm one moment

[bright ember]

ok 🙂

[peak light]

So much work I have

[bright ember]

then leave it, I don't want to bother you

[peak light]

I need youuuu

My work now is to make AI video for my job place

[bright ember]

What programs do you use to create videos?

I'm looking for something that will make me nsfw videos

[peak light]

Comfyui

[peak light]

Oh I’m working with schools 😅

[peak light]

You will need to train your own model I think

I have 0 knowledge about video model training

[bright ember]

I have been training models for years, but I have recently started trying Comfyui.

[atomic mortar]

Scams again

[atomic mortar]

Loras take up more vram, if you exceed then whats on your card it will offload to ram and will slow down

[low moon]

Happy new year everyone! Even you, Emad.

[unique sage]

anybody here know how to use FreeU well?

[valid aurora]

hello guys idk if this is a stupid question but i used to download illustrious loras in civitai 5 months ago then i heard bout qwen and was planning to download some qwen character loras but i could not really find any in civitai only style loras (not sure of the exact term)... i know its possible to make character loras with qwen but why are there barely any in civitai but when you filter illustrious theres a whole page of anime character loras

anyone know where i can find more qwen character loras?

[atomic mortar]

Always sounds like a scam

[abstract quarry]

because most people are only interested in training illustrious. Its easier and you can make porn with it which seems to be the main motivation for most character loras

[abstract quarry]

good thing is that modern models like flux or qwen allow you to generate images based on other images, so you don't necessarily need a character lora

[hollow sleet]

happy new year

[valid aurora]

I see. Ive no experience with making illustrious lora but for qwen i used ostris ai toolkit and it was quite easy. I heard illustrious was outdated so i dont want to take a gamble so guess i gotta just make my own qwen lora for multiple characters as tedious as that is

Ive not really used references images when making specific characters i tend to opt for just making/downloading a lora as i believe its more consistent but ill give it a go

[limpid compass]

Curious to hear from working editors 👋

What’s a ~13s clip that looks simple but is actually a nightmare to edit?

Stuff like hands crossing objects, motion + occlusion, reflections, edge chatter, tracking drift — the shots that make you sigh before starting.

If anyone has a sample clip (or can describe one), I’d love to see it.

[dark hawk]

THERE IS NO PALESTINE

[fervent thunder]

THERE IS ALSO NO EASTER BUNNY

[trim sinew]

Is there a way to fill up video frames with AI? Like you feed it starting and ending frame and the model will produce a video clip based on these two

I really want that

[kind eagle]

Hi I'm george

I made a fun little site:

Control The Art .com

it’s an AI image that anyone can change. Have fun

[atomic mortar]

Yes but you'd swarmui or comfyUI for that, personally i don't mess a lot with video gen but thats possible yes.

[trim sinew]

ComfyUI doesn't work for me

It's broken

[atomic mortar]

I seriously doubt comfyUI is broken if its a fresh install if im being honest, its one of the most popular ui's, i recommend hopping into the comfy server for some better trouble shooting since it could be a missing dependency or a download error

[trim sinew]

I can't get the checkpoint selection to work

It just says undefined

No matter what I try

[atomic mortar]

Maybe you are using the wrong folder for the checkpoints

[trim sinew]

It's the right one

Comfyui is simply broken for some reason

[atomic mortar]

Your sure you are on the latest build

And your not using stable- diffusion folder for wan

[zinc viper]

Hello

[dark hawk]

i don't even know what it is

[fervent thunder]

an easter bunny?

[red wagon]

Happy New Year, everyone.
I fell down the AI rabbit hole about 10 months ago and ended up building something to solve the one problem that was driving me insane: organizing the absolute mountain of content we all create.
We all have that one folder—the Downloads mess, the phone dump—that's pure chaos. My project is a system that you just point at that folder, and it uses AI to intelligently sort it all out for you. The image pipeline is fully operational (and it's scary good at telling the difference between a headshot and a screenshot), with video and audio not far behind.
I just ran a real-world test by dumping my entire phone's camera roll into it, and honestly, the results were incredible.
If this sounds like a familiar nightmare, I'd love to hear what your biggest organizational pain point is. Curious to see if I'm on the right track. Feel free to DM me any questions or for that matter the real friction points we are all facing. Love to get feed back and or hear the real problems. GOT THAT messy 1 folder of 100's or 1000s of images and video clips generated lol thi is what im aking that will organize ths (hopefully) so we dont have to lol

[faint timber]

@red wagon I methodically add dates to all my output pathing, so whatever I work on for a given day ends up in a related folder. It's better than before. I used to have 1,000s of images in my output folder. Now, I have lots of subfolders with thousands of images in them. The input folder, however, is still quite messy.

[rich shell]

Hey guys I'm so late to using discord it's not even funny. I've been waking up everyday forcing myself to painfully mind numbingly get acquainted with SD for the past few months. That and write tons of fiction. But I'm looking for working rooms (as they may call it) to discuss anything AI filmmaking, maybe even chat here for a little. I have no idea. My 10 minutes just ended so I'm sending this before I eat an insane sandwich, but I will educate observe too. Thanks so much to any type of reply

[devout pollen]

Saw this is a great group to join via TikTok - something about an MCP Claude/Blender setup. Is this like Minecraft on a schedule 1 substance?

[bronze merlin]

Hi Happy new year 🥳🥳🎊🎊🎊🎊🎊🎊🎊🎊🎊🎊

[trim sinew]

https://www.youtube.com/watch?v=VpKhbKqP3PI Does anyone know how to generate this kind of AI videos?

i really want to make my own like this

[ashen rune]

https://www.youtube.com/shorts/Xh7PYgvj-QQ ♥