💬｜general-chat

[round mirage]

Did anyone pass the "exam" thing in SillyTavern? I don't know if I'm missing something stupid or just answer a question wrong but im getting annoyed by it :D

[quartz pelican]

Again, what is considered a good prompt?

[calm basin]

no

well... yes... it won't explode. but you won't iterate on anything useful this century with dev iteration loop that long

[autumn ibex]

Hello, is there anyone who is gonna build smart contract, dApps, NFT and trading bot here?

[faint timber]

@quartz pelican One that produces a result, making you say, "Wow."

[quartz pelican]

Spare me this pseudopoetic nonsense, I was asking about technical part

It's not like any of you is hard to impress anyway

Like, how long good prompts can be? Are there any specific commands that should be present? Any settings considered as standard?

Is it better to write words like separated tags or write full sentances with descriptions?

[oblique elk]

There is no magical rule for good prompts as the different model architecture needs different prompt syntax and weights.
So you will need to be either more specific or at least specify the model architecture we are talking about.

[quartz pelican]

I see. And where can I learn about SD syntax?

Stable Diffusion 1.5, to be exact

Thanks in advance

[oblique elk]

Well SD 1.5 is a model with a keyword based text encoder. Prompt is shorten to about 77 tokens (about 77 words). The order determine the priority so first words will be more important then the last ones. Due to the low resolution (base 512x512 pixel) not many objects can appear.
Most UI support weights so a word “car:1.5” will increase the probability that the object will appear. So boost weights might help.
Next important part is that the keywords must have a good chance the original images where tagged with the words. That’s why some surprising results appear if you use something like “Nikon d850” so a camera brand but it is included in images found in the internet. Mostly portrait or other realistic pictures.
Then you might look at some style keywords to create more interesting appealing images.

[indigo wraith]

hey guys

i need help with stable diffusion

[oblique elk]

So ask your question.

[quartz pelican]

Alright!

[drowsy shard]

Hello, I am looking for a way to do image to image characters swaps in any situations and poses using pre-existing Loras for the characters, I have tried on Automatic 1111 with controlnet and different models but the results are super messy on more complex poses or with two characters. I have considered using ComfyUI but the problem is that I am very limited by my 4gb VRAM, can someone please recomend propper extensions or low VRAM usage workflows for this purpose ?

[gray folio]

@drowsy shard When did you notice that mate 👋

[buoyant stone]

Hello, do you need full stack developer's help now?
If so, please contact me anytime

[stone locust]

I didn't know that because I was stupid enough to see it for the first time...
Thanks!

[primal current]

Is there a Regional Prompt workflow for WAN 2.2?

I want to put two different characters with their loras, I tried with Res4lf and a couple of GitHub projects, but none of them work for me, any help is appreciated!

[unkempt cedar]

guys whats the best model for realisim

[placid hatch]

Does anyone else find that the style of their image outputs changes significantly as they add or remove more and more stuff to it?

[wispy berry]

Hi, is there a way to have a controlnet with SDXL which tries to approximately keep the colours intact? I would like to make clouds which spell words e.g.

[oblique elk]

Why would you need colors for that? If you use for SDXL to create a cloud image then use gimp, krita, photoshop or word,... to add a white text on it, create a greyscale image and use it as an input for controlnet depth you might come close to your goal. But it got nothing to do with the colors?

[wispy berry]

Well, my idea was that the clouds are white while the sky is blue

[abstract quarry]

use img2img in addition to the control net

[limpid ravine]

Am I able to make an image to gif with Stable Diffusion?

I'm looking for a small favor if anyone has the experience, the job is easy, but I don't have the knowledge. 😅

[autumn ibex]

I work with Stable Diffusion to create AI-generated art and also build apps that make it easy for people to use this technology. I use tools like React, Node.js, and Python to connect Stable Diffusion with cloud GPUs so everything runs smoothly. Whether it’s custom images, NFTs, or full AI platforms, I help turn creative ideas into real projects.

If anyone needs my help, please contact me

[jaunty hinge]

Yes, possible but not easy job

haha

[jaunty hinge]

if you use the automatic1111, we can solve this easy....

[limpid ravine]

And will the transparent background be preserved?

[jaunty hinge]

sure

[limpid ravine]

Are you able to do this?

[narrow kite]

is it possbile to use wan 2.2 to generate images? i can only find out how to generate videos

[light elk]

Hello! I'm a passionate and performance-oriented developer.
I’m a Full Stack Engineer with over 8 years of experience building scalable website, cloud-native software solutions across industries like healthcare, fintech, e-commerce, gaming, AI and energy.

[quartz pelican]

Do people still bite on adverts like this?

[tacit karma]

hello

[quartz pelican]

Hi!

[sharp meteor]

Hi, can anyone brief me on how to use Wan 2.2? Is that something I can use within A1111?

[warm junco]

Hey, wan only works on comfyui

[sharp meteor]

Thanks, I'm figuring it out as we speak

[soft hound]

Is there a moderator or administrator I can contact here?

[naive scarab]

Hello guys I'm new here! Hello from Brazil!

[rain yarrow]

how can you make more objects appear?

[limpid ravine]

Where do I report people here?

Receiving scam DMs claiming to be mods

[still glacier]

scammer, don't follow that link

[still glacier]

feel free to dm me or anyone with "Stable Staff" or "Community Guide" role

[oblique elk]

Only a few options if you are sticking with the SD1.5 models. Upscale the image and use inpainting to add some more elements. Upscale with tile upscale to add more details during the upscaling process. Try regional prompting but again if a normal object like a scoccer ball is at least 100x100 pixel to keep some details and the general resolution of the model is 512x512 not much for the ai to fit on the base resolution.

[rain yarrow]

Thanks for your elaborate answer on the possibilities 🙂

[visual pier]

Since I made the switch SDXL is such an improvement :D

I was sticking to 1.5 for the longest time

[autumn sage]

@grand bolt Hiya, sent you a quick question in DMs, can you check it out when you get a sec? Thanks in advance!

[runic patrol]

hey everyone. I hope we are all doing rwell today

[clear maple]

hello

[inner hare]

hello

[lavish dagger]

hey guys! quick question, how can i make these type of ai generated videos ?
https://www.instagram.com/reel/DNgXb2wInol/?utm_source=ig_web_copy_link&igsh=MzRlODBiNWFlZA==

[old lichen]

How to get a poor 3D visual
Which model would you recommend?

[brave tinsel]

Hello everyone

Anyone wanna discuss ideas about making money from ai models?

[quartz pelican]

Ah, web3 NFT NLP LLM enhanced creativity workflow game changer? So based and techbropilled

[quartz pelican]

Chat is this rizz?

[still glacier]

scam* (and cleaned, buh bye)

[buoyant stone]

gm

[brave tinsel]

How is your ai model doing?

[floral umbra]

Be a wordsmith, learn how to prompt, Get loras/train loras on characters, style, camera movment, and have lots and tonnes of lots of vram if you plan to make something that is also high res.

@warm junco Hoi, where does talk/discussion about training loras go?

[wanton roost]

dudes

the rtx cards make me want to kms. theyre so expensive. meanwhile the rx 9070 xt cards are super affordable in my country.... in chance that in the future we ll have proper support for stable difussion on amd cards?

[warm junco]

Yes AMD is working on it currently

[wanton roost]

then i suppose i could do a smart boy move and get an amd one instead. and gamble that amd will do a good job

[young tinsel]

hi

i have quastion

i need help

i downloaded flux dev krea f16

and i dont know how to run it

please help

[abstract quarry]

you need to install a tool for using it. There are plenty to choose from

swarmui, invokeai, forge, comfyui, ...

[ionic fox]

o7

[quartz pelican]

Right, my very own original brand new AI model for optimized creative workflow and producing stunning imagery that evokes our deepest feelings

[brave tinsel]

Mane cmon mane

[quartz pelican]

Don't you understand sarcasm?
Beep boop?

[brave tinsel]

No

[quartz pelican]

Sure

[thick olive]

Hey guys, we are running an AI Image challenge to test our new tool. It's a fully hosted service and we have the SD model hosted as a starting point. Where is the best place (if it's even allowed) to post about this? And can we also post it in the general reddit?

I just want to check first and avoid being banned 😛

[still glacier]

#1092446741984444416 or here probably. Just don't overdo it. I can't speak for reddit.

[humble grove]

Hello 👋

[fervent thunder]

Yo.

[light elk]

I’m Tobi who has 8+ yrs building scalable apps & clean designs for startups and tech teams.
I handle design → dev → deploy so you don’t need multiple hires.

[civic fern]

hi here

[muted stag]

Is anyone else having triton language missing for ksampler advanced, I have Triton-windows installed but it's still complaining

Trying to get I2V workflow working for wan 2.2

[floral umbra]

For wan lora training, was it 3 or 5 sec per clip that was recommended?

[floral umbra]

Double check in CLI logs if there's any mismatch in dependency versions, or other missing dependencies.

[radiant frigate]

I'm wondering whether the option that we could generate images in discord using SD still exists or not

[still glacier]

now those are the #artisan-faq channels

[autumn ibex]

💯 Hello, is there anyone looking for professional blockchain developer here ?
If you wanna build smart contract, DeFi, NFT projects or trading bot, I can help you. 💯

[floral umbra]

NFT's is a meme lol.

[somber orchid]

did civ ai just go down?

[somber orchid]

yes

my issues has been fixed

[lucid bobcat]

Someone ban this guy please.

[rustic citrus]

I need a mod please @vapid dove

Yea

Nah, I gotta speak to him

For Mai

[frank hollow]

hello everyone

is there any option to pay with cryptocurrencies in platform.stability.ai ?

[slender halo]

Hell yea

[still glacier]

spoiler : they'll probably try to scam in dms.

[quartz pelican]

Like a good half of this server

[still glacier]

working on it

[quasi swan]

Hey friends, I recently found a little tool that can make images look clearer.
Here’s a comparison I made with it, and I think it turned out pretty good. You might like it too. Would you be interested in trying it out?

I’ve shared the results in /with-images. feel free to check them out if you’re interested!

[nova kelp]

Anyone know some solid anime inpainting models for SDXL or SD 1.5? Thanks!

[maiden hatch]

Has anyone been successful in making Stable diffusion work with ZLUDA on a 9060 XT?

[still glacier]

it's a scam, don't follow

[meager fulcrum]

how do i add upscalers to my stable diffusion? i dont have \models\ESRGAN\ i have models\RealESRGAN but i put 4x_NMKD-Superscale-SP_178000_G.pth in there and it still doesn't show up in my upscaler

[warm junco]

Create the models/ESRGAN folder

[meager fulcrum]

oh, thanks

[warm junco]

np!

[meager fulcrum]

i downloaded the same loras and checkpoint of a civitai post i saw and did the png info thing and copied it to txt2img but my images are completely different, would you have any ideas why? their fairly realistic while mine looks completely deformed

[warm junco]

you need to click the copy prompt button on the images there.
That will copy even the hidden settings. Then paste it into the prompt field and press the blue/white arrow to apply it

[meager fulcrum]

it still looks awful and nothing alike :(

[warm junco]

can you show boath images in #📝｜prompting-help ?

with settings and all

[meager fulcrum]

sorry if im bothering, im now getting OutOfMemoryError: CUDA out of memory. Tried to allocate 54.00 MiB. GPU errors (on forge), i have 16gb ram, is that not enough? i tried adding --medvram but i still get the error

[lucid bobcat]

You need VRAM not RAM.

[meager fulcrum]

i have rtx 3090

[lucid bobcat]

Haven't used forge for a while. Maybe you need to enable partial loading. When it happens during decoding you need to enable tiled vae decoding.

[warm junco]

You need a smaller flux dev model version

Dont use the 23gb file

Use an fp8 version

[lucid bobcat]

That's not true. I can use a 23gb model even with my 8gb graphics card without getting OOM.

[meager fulcrum]

i dont see that option anywhere,, also should i have both clip.I, t5xxl_fp16 VAE and text encoder set on?

[lucid bobcat]

You need those things but you can keep clip and t5 on the cpu.

[meager fulcrum]

i gave up and am trying comfyui now...

[robust cave]

Hello
How are you doing?
I am a passionate developer, so far attended various kinds of projects.
so if you have some recommendations or looking for extra developer, I'd love to collaborate together. 😇

[lucid bobcat]

Hello scammer. No we are not interested.

[autumn ibex]

hello, is there anyone looking fr,

[lucid bobcat]

Yet another scammer.

[molten sage]

Hi! I'm trying to build my own Pony checkpoint and I want it to have a Lightning version as well, but I don't even know how to start researching about it, so I was wondering, do youguys know any sources of information for me to learn how to create Lightning Checkpoints.

[lucid bobcat]

Why would start with the hard stuff? Train a LoRA first, maybe it does the job.

[robust cave]

you are scammer or bot

[lucid bobcat]

Hello scammer. Every single post from you on this server has been "looking for dev?". You've probably spent your whole day spamming on other servers too. Just leave.

[quartz pelican]

tbh you ALL should just start a battle royale to see who is the scammiest scammer

[plucky fractal]

is it okay to ask questions in this channel?

[abstract quarry]

I don't think anyone trains lightning checkpoints on his own..What they call lightning checkpoint is just the lightning lora merged into a trained checkpoint. It's a shortcut for users who don't want to use an extra lora.

[fringe epoch]

😂

[earnest vine]

do you guys have any recommendation for a semi realistic checkpoint (illustrous not flux) which is also free

[fervent thunder]

Does anyone know if Ace Step supports a 5090 and Miniconda 3.13? Because nothing is working. And no, I'm not at all technical with coding and installations. I've been getting better as I've been screwed over by about 20 generators over the last month, but I'm still not technical

[mossy owl]

Hello guys any config. recommendations for a GTX 960M?

[still glacier]

oof apart from don t even bother with models other than SD1.5 and its derivates. not much

get pruned fp16 models whenever possible.

it will run and it will be better than cpu only... but still it will be slow.

and go with Forge for your client

[warm junco]

Illustrij

[earnest vine]

thanks bro

is there any loras you'd suggest or any one would work fine

cuz some models need certain loras

[molten sage]

Oh, gotcha! I'll research how to merge the Lora then. Thanks for pointing that out ✨

[warm junco]

The dark lora for Illustrious can help getting more vibrant or darker images. Its nice

Upscaling and adetailer can improve the quality.

[floral umbra]

Sucks so bad that 3090 can only just barely do 2 seconds worth of video, or 33 frames worth of clips for training loras

[fervent thunder]

Speak for yourself. The 5090 laptop's architecture is so bad, 99% of AI generators can't even be installed. It's literally the PS3 of GPUs. I barely got AudioX installed and it can only do 10 seconds of sound/music

[lucid bobcat]

5090 laptop just sounds wrong lol.

[lavish current]

I'm tired of mods deleting my images on other servers

None of it is even offensive

[iron laurel]

I’m a Senior Web & AI Developer with over 8 years of experience building scalable, high-performance solutions across a variety of domains, including E-commerce, SaaS platforms, AI automation, and CRM systems.
With a deep understanding of both frontend and backend technologies, I specialize in delivering end-to-end solutions that are robust, user-centric, and results-driven. From architecting AI-driven workflows to launching full-featured web apps, I take pride in turning complex requirements into seamless digital products.
I'm passionate about innovation, clean code, and delivering real value to businesses through technology.
Let’s connect—I'd love to learn more about your project and how I can help bring your vision to life.

[fervent thunder]

In my own defense, I simply don't have the physical space for a desktop

And actually, it's been really good

Minus the crap architecture for some things

Then again, it might actually be the developers who screwed up as well as most AI generators haven't been updated in 1-2+ years

[sharp gull]

https://www.tiktok.com/@life_laps_official/video/7541170435963014431

Do you guys think this is made with midjourney or WAN?

[primal current]

Qwen Image Edit + ControlNet Openpose es posible?

Basically, I want specific poses for my character, which I have uploaded to Qwen Image Edit, to replicate the poses I could input with OpenPose ControlNet. Is that possible?

I know they can both be used separately, I've tried to make them work together but no luck.

[desert dagger]

there aren't any controlnets trained for qwen

[primal current]

exist

https://www.modelscope.cn/models/DiffSynth-Studio/Qwen-Image-In-Context-Control-Union

https://huggingface.co/Comfy-Org/Qwen-Image_ComfyUI/tree/main/split_files

https://huggingface.co/Comfy-Org/Qwen-Image-DiffSynth-ControlNets/tree/main

https://w www.reddit.com/r/StableDiffusion/comments/1mwr430/qwen_image_union_diffsynth_loras/?utm_source=share&utm_medium=web3x&utm_name=web3xcss&utm_term=1&utm_content=share_button

https://drive.google.com/drive/folders/1vhipvq2LXId5Ed7QkDpozEsu8cShPGYe?usp=sharing

[desert dagger]

those aren't open pose

[primal current]

https://drive.google.com/file/d/1DCBXvLtF5zF_51x0F72O91FUbAQU2FKB/view?usp=drive_link
works with qwen_image_union_diffsynth_lora.safetensors

[desert dagger]

that's still not open pose

[primal current]

If anyone knows what I'm talking about, let me know!

[coral turret]

hello everyone am having a hard time trying to find a lora 2500 people has download it before .. looking specifically for this version (FFM_threesome_girl_sandwich.safetensors) 217MB for pony

[runic patrol]

@warm junco i am so sorry for pinging, yuou have been the most amazing person and have helped me. My friend wants to get into ai generating. But they have an amd card and they are on windows. I heard "rocm" might be a thing. Or is there another method? If rocm is the method, is it easy to install. Once again.. sorr fo rpinging

[warm junco]

hey, amd is no problem, i made guides for it. Which gpu is it?

[runic patrol]

Thank you so much!

she uses a 9700 xt

[warm junco]

okay, and do you use auto1111 or forge?

[runic patrol]

I will have her install forge. It isn't me. I use an Nvidia card.

[warm junco]

alright

[runic patrol]

she is also on windows, is that an issue?

didn't mean to make that sound rude

[warm junco]

then she needs to follow my "Forge with Zluda" setup guide from here:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides

[warm junco]

no issue

[runic patrol]

Thank you so much!

[warm junco]

np, let me know if you have any questions about the setup in #🤝｜tech-support

[runic patrol]

Forge with Zulda. Got it. And thank you for beign amazing!

Also, lovely to talk to a fellow one piece fan./ I think thats luffy anyway!

[warm junco]

yea! One Piece is best 😄

[runic patrol]

It really is. Its my favourite anime. And its also the anime i watch whenever i want to cry. There are countless moments that make me cry without effort.

Are they fine installin' stability matrix by the way?

[warm junco]

True! It has so many epic and emotional moments

[warm junco]

No

Dont use it with AMD

It creates problems that can easily be avoided xD

[runic patrol]

Okay! just letting her know

she was grumbly a year ago when she switched to amd, she enjoys doing the ai things but she didn't know when she switched to amd that she couldn't do as much as the nvidia cards at the time

[warm junco]

Yea true in the beginning it was very rough for amd

But the future with Rocm looks promising and zluda is a good alternative meanwhile

[warm junco]

9070xt or 7900xt ?

[runic patrol]

9070xt

https://tenor.com/view/one-piece-anime-one-piece-giants-giants-elbaf-gif-2198175696127755340

you did that to the scammer!

[velvet beacon]

I need help

All bote for los bunkers

So important 🙏

[warm junco]

No

[runic patrol]

"Interrupted with signal 2 in <frame at 0x00000232DF2CF250, file 'C:\ZD-Zluda\stable-diffusion-webui-amdgpu-forge\venv\lib\site-packages\torch\nn\modules\module.py', line 1762, code _call_impl>" she got an error

[warm junco]

Complete cmd log and please in #🤝｜tech-support not here

[runic patrol]

My bad!

[burnt cairn]

hi

I was blocked from stable audio, how can I resolve this?

[lavish ice]

Yo, can I get someone to ban @quasi depot They are a moron who is actively trying to scam me with some stupid Steam Con.

I have screenshots saved.

[floral umbra]

Oof, when you realize after 30 min the reason training fails from resume point was because resume somehow managed to use “ and not " One is "fancy squiggly" and the other is straight

[lucid bobcat]

Does anybody know how to get rid of the small amount of leftover noise that Flux produces? When zooming in, materials like cloth etc have more noise than actual texture. I'm especially comparing it to 4o that produces exceptionally clean images. I'm not sure if that's a weakness of Flux or something else.

[old lichen]

efb7f0faf7a4
How can I find a model using this code?

[warm junco]

normaly paste that into the civitai searchbar. or google

[crimson quest]

Does anyone know how to get SDXL on mac?

[still glacier]

cf pinned messages in #🤝｜tech-support

[inner sedge]

Is there a way to programmatically run GGUFF files for image generation? I know I can do diffusers pipelines but I still have to download the full models off huggingface.

[true shuttle]

Which gguf model you mean?

And you have to download model from huggingface even if you are using programmatic method.

[inner sedge]

I'm trying to run a WAN2.2 model I downloaded. I've used diffusers to make a SDXL command line app before and I'm trying something new

[inner sedge]

sorry I meant the code examples I've seen need both the GGUF and the "normal" model safetensors

[true shuttle]

Cool!

[true shuttle]

Makes sense.

Could you please share soem screenshot for that?

[inner sedge]

this is where I got it from. He replaces the transformer part of the model with his GGUF model https://github.com/huggingface/diffusers/issues/11878

looks hacky as hell but tbh, any information about this stuff is gold. Everyone appears to just be using ComfyUI

[true shuttle]

Just a sec.
Let me check.

[inner sedge]

localai seems a bust. But even then it's too heavy for my needs

nobody asked. But here's my SDXL command line https://github.com/luckielordie/sdxl
It's made to keep the aspect ratio as close as possible to your target resolution, but things get weird with extreme ones.

[austere cobalt]

does anyone have a good TTS module/workflow?

[dreamy relic]

No one seems to use stable diffusion anymore

Is sd3s tos still so bad that we still don’t have it on popular platforms ?

[abstract quarry]

I mean, stable diffusion is a brand, nothing more. The people behind it left the company long ago and made Flux, which I would argue is much better and more popular than Flux. We now have many new alternatives such as Wan, Highdream, or Gwen Image - the latter two based on sd3/flux architecture. I don't have the feeling that the stability ai company who owns the stable diffusion brand has much interest in doing something new

[still glacier]

No and it has been like that for more than a year
https://newsletter.civitai.com/p/stable-diffusion-3-unbanned-civitai-discover-new-ways-earn-buzz-cash-check-early-access-features-civ

[lethal igloo]

Hi everyone. First step here.
I have a question about controlnet. Nothing too technical I believe. Where shall I ask it?

[oblique elk]

Maybe #📝｜prompting-help suites best

[lethal igloo]

I will try there then 🙂

[coral hollow]

Hey all, I need some guidance on the best way to approach something pretty simple. I have an image of a figure. I want to turn the image around and see it from the back. I have stablediffusion. When I use deepai photo editor, it correctly flips the image, makes some mistakes but it is generally the same figure from the back instead of the front. I can touch this up in photo shop but it does save a lot of time. With stable diffusion it generates variants of the image, but really doesnt simply view it from the back. I did some searching and see controlnets recommended but will that just generate images which are similar, whereas I need exact. ... Not sure where to ask this.

[abstract quarry]

just use an edit model

• flux kontext
• Qwen Edit
• Highdream Edit
• Omnigen 2

I only tried flux Kontext and Qwen Edit so far. Both should be good enough for your task

[coral hollow]

Thanks, Id prefer to instal everything locally as I have the hardware. So as an example https://www.buildabear.co.uk/dw/image/v2/BBNG_PRD/on/demandware.static/-/Sites-buildabear-master/default/dwa49d338d/29231221227229x20237010.jpg?sw=800&sh=800&sm=fit

I just want to make the back of the image...ie turn it around 180 degrees

[abstract quarry]

Look at my examples: #🏞｜general-with-images message

[coral hollow]

(ive just joined thx!)

Exactly what I need !

[abstract quarry]

Flux Kontext (and all other I mentioned) are local

here is your teddy: #🏞｜general-with-images message

[coral hollow]

(downloading and installing everything now)

Any issues with cuda versions? Ive got A111 running fine.

[abstract quarry]

these are just finetunes of the image models Flux/Qwen/Highdream

so if Flux runs on your hardware, Flux Kontext will run fine as well

[coral hollow]

Whats flux? another version of A1111?

(its been a year since I played with all my AI tools)

[abstract quarry]

A111 is just a tool

Flux, Stable Diffusion, Qwen - these are the models

[coral hollow]

A111 is web front end to Stable diffusion?
Flux etc like SDXl?

[abstract quarry]

yes

[coral hollow]

Huggin face model load into A111?

[abstract quarry]

I would say, Qwen is currently the most powerful model, followed by Flux

Qwen is quite vram hungry, though. Flux is smaller. SDXL is MUCH smaller

[coral hollow]

(first result was a flux website hence the confusion)

[coral hollow]

24gb vram 🙂

[abstract quarry]

enough for all of them

you will have to use a lightning lora for Qwen, though, otherwise it takes 2 minutes to generate a single image

[coral hollow]

https://huggingface.co/black-forest-labs/FLUX.1-Kontext-dev?

[abstract quarry]

but Qwen Edit is better than Flux Kontext most of the time. The only thing Flux Kontext does better in my opinion: it changes less of the image. Qwen Edit is more freely changing the image, sometimes changing too much

while Flux Kontext is really good in only changing the small detail you asked for

(ChatGPT on the other side is the most extreme. it is a master in reinterpreting images, but it is really bad in only changing small details of the image)

[coral hollow]

Are all the models on hugging face?

[abstract quarry]

yes

https://huggingface.co/Qwen/Qwen-Image-Edit

[coral hollow]

What tool do you use for the front end?

(web ui)

[coral hollow]

Yes chatGPT was the first thing I tried and it just made random changes with just a basic prompt

[abstract quarry]

comfyui, but that's the most complicated one 😅

[coral hollow]

Yes , I did want to install that as everyone was suggesting it, but never got around to it

[abstract quarry]

invokeai is quite good

the advantage of comfyui and webui is they have a lot of extensions. Comfyui not only has most extensions but you can also extend it yourself very easily, giving you most flexibility

invokeai has the fewest extensions and new features often appear there quite late. But even without extensions it has a lot of features built in and probably the most powerful ui

[coral hollow]

I'm used to webui, so I ll get it working there and then switch over 🙂

For omni gen2 should I be using the 10gb or 5.87gb model or are they parts of a combined file as they say 0001-and 0002

[abstract quarry]

in huggingface they 0001 - 000x parts are always combined

[coral hollow]

rar? 7zip or does it autocombine?

[abstract quarry]

I don't know, probably latter. I think each file is just a safetensors file

which is a dictionary of string keys and matrices

and if you have multiple files you just merge the dictionaries

[coral hollow]

yeah, they are just safetensors. Im assuming I just need the safetensors and not the rest of the file archive (scheduler, processpr et al.)

[abstract quarry]

probably. Its diffusers format 🙁

diffusers reimplements all the models. So they have their own code and they rename everything making everything incompatible

it depends if your tool can read diffusers models or not

(some tools like invokeai and also some of the webui forks run in diffusers anyways)

[coral hollow]

cant paste screenshots here.. i now realise

[abstract quarry]

thats why I posted the images in general-with-images xD

[coral hollow]

been trying to paste images here for like 10 minutes (thought it was a browser issue)

[calm night]

heelo anyone here is an excpert in image gen ? not anime tho

[true shuttle]

what kind of help you need?

[solemn burrow]

Hello, I started with the image generation less than a week, I would like to find someone veteran who can help me with some stuffs and guide me a bit hehe, excuse my bad English, dm open if someone want to help me

[cunning merlin]

hello, i just downloadet Remacri as an upscaler, but why cant i find it?.. 🙁

[warm junco]

Put it in models/ESRGAN if you use Auto1111 or Forge

[fervent thunder]

I might or might not've just spent 6+ hours screwing around with Invoke AI without eating at all just now. If it isn't a word, well it is

[cunning merlin]

nah, had a different problem, but i figured it out, thanks 🙂

[serene siren]

Hello

[versed marsh]

[magic ridge]

are there any AI agent experts?

[digital wind]

I need to do a interview for a class. If you have time, please respond to this Google form. This questionnaire is about AI-generated images.

https://docs.google.com/forms/d/e/1FAIpQLSfZXNGD1uyDGzXLiVX52XAzYplZ3Fl5nxC_29WlFMIqpfUWLw/viewform?usp=sharing&ouid=110730258411046891797

[north dagger]

hello!

[solemn burrow]

Hello, what happen with civit ai?

[north dagger]

No idea. But, the maintenance mode art is funny.

[round siren]

can i get a sd inpainting workflow that doesnt have so much crap attached to it

[digital briar]

hey yall, im trying to generate a bunch of images in a specific cartoonish style, anyone know what the best model to use is?

I've heard it's Flux Kontext but i'm not sure

The style isn't realistic at all, which is why i'm worried Kontext may not be the right move

[drowsy phoenix]

hi ppl, I dont use SD from around 1 year, how is the image generation this days? I mean for run an instance in my pc... will be power as Chatgpt 5 ?

I have a 3090 24gb

[oblique elk]

Kontext and Qwen Image edit can be used to change situations around a character no matter which style the base image is.

[rich sun]

Hi, Hmu world's leading cyber security and ethical hacking platform

[abstract quarry]

you can do limited style transfer with Flux Kontext or Qwen Edit, but its probably not good enough

probably the better choice is to train a flux lora on the cartoon style

flux can do many art styles if you train it on them - not just realistic

[sonic ivy]

hi, is there a place where we can find unusual / weird/unethical type of LORAs?

[true shuttle]

Of course there is.

It must be google. lol

[sharp jetty]

Good day, everyone. I'm new to this server and to Discord in general! I’d like to hear about your general experiences with sharing AI-generated pictures online. I’ve come across quite a bit of hate, by that, I mean negative or intentionally mean-spirited comments, simply because AI or Stable Diffusion was used, even on dedicated boards such as certain Reddit subs. Have you had similar experiences? I might be a bit too naive when it comes to this whole discourse, but I’m honestly surprised by how harsh some of the sentiments are. 🫥

[runic patrol]

Anyone here familiar with Img2img upscaling? If so, what should my screenlook like. I am looking to use img2img to upscale and add some post processing to it. If its okay to ask

[warm junco]

yea, you can show your img2img screenshot in #📝｜prompting-help

[runic patrol]

Sure!

[pseudo jetty]

quick question for those who have a rtx 5080 how long does an image generation and hires fix take for you, at 720 1280, and image take around 1:30 including hi res? and my old gtx 1080 only took around 2 minutes so i feel like the image generation should be faster. Xformers is on but im not sure what else i could do using a1111

[runic patrol]

i was told the 5080 doesn't have access to xformers yet?

[pseudo jetty]

i mean i have the commandline args set to xformers, whether or not its doing anything is another question

[pseudo jetty]

its mostly highres that takes a while

[inner sedge]

do img2img pipelines do style transfers?

or like, "take the focus of this image and reimagine it like so"

[junior urchin]

Scam

[true shuttle]

Really?

[true shuttle]

But he suggested me to do some c r ypto job.

But I don't know much of it so passed.

[junior urchin]

They just need to insert "(HOW)"

in automod

[autumn sage]

Hey @grand bolt check DMs please 🙂

[ripe flower]

Hey guys been away from the AI image scene for a while, today I saw some funny stuff on IG where ppl were adding AI girlfriends to their pics. I'm wondering what the options are for something like this, I tried with Gemini and it does ok, but not great.

[still glacier]

taking out the trash, don't mind me

[undone solstice]

I'll check my DMs, but I'm not the one you're looking for. The Al image scene is pretty wild these days. What exactly were you looking for with the AI girlfriends?

[ripe flower]

Just something super simple, I posted an image in general with images where I added a girl to an existing real image. This was done with mobile Gemini app from Google, it's actually quite fast. But with Gemini it's hard to get details correct, lots of limitations

Ideally it's a quick and easy workflow, I know this can be accomplished through SD image to image inpainting on my desktop, probably requires some LORAs

[tender nebula]

Absolutely, it’s funny how many people have such a strong and inherent hatred for AI generations

[sharp jetty]

How did you deal with the discouraging feeling? Or did you stay above it? 🧐

[tender nebula]

Personally I don't believe there's any need to deal with this feeling within any means. We're just experiencing the common human behaviour of being emotionally attached to an idea or opinion that isn't inherently true. I think time will tell us how AI technology really dominates society for the foreseeable future, and society will adapt organically

What about you? How do you understand the hate and how do you deal with it?

@sharp jetty

[sharp jetty]

I agree with what you say, there will be an adoption naturally, just as it was with photography or computers in general. I am not sure whether this will happen in our life time 😆 I understand the intimidation factor of the technology and that it can lead to some emotional reactions, especially if it feels to some people like it is taking something away. However, I think this must not be the think if the technolog can be seen as a new form of creating, rather than a form of cheating or undermining other people's efford. While I think I understand the emotional exit point, it is beyond me how people can get so invested that is make them activily harass others (I guess that urge must have been in their system from the very beginning). I feel a bit disappointed, as I also do "classic' drawings and I new that it can be a very time consuming and dedicated efford to learn. Yet, I also see that using for example, Stable Diffusion, can be a very invested project, as it is, as many of us will know here, often more than just "entering a prompt" and there can be so much planning, retouching, setting up, etc. involved in creating an image. I guess I am disappointed when I see this effort disregarded.

[normal parcel]

Hello I am trying to use Stability to create videos from images via API in n8n, image creation works but not video creation, is the API limited to image?

[elder tapir]

hi guys im new to SD, im on a budget and i wanted to try it out using my 6700xt but it's so hard to run with AMD

ive got a good deal on a 3080ti , is that enough for AI images? im not a professional .. just casually tryin to generate anime images

[warm junco]

Hey, you can easily use your 6700xt

You should follow the pinned setup guides of #🤝｜tech-support
There go for AMD Forge with Zluda.

Feel free to ask me if any step isnt clear

[noble widget]

hellow

[elder tapir]

thanks! ill give it try

[elder tapir]

sure!

I'm going to buy the 3080ti, later on too .. but im going to give my 6700 another shot lol

[lost zephyr]

Hey, what's the best channel here to discuss lora training?

[warm junco]

How much is the 3080ti?
Because 12gb vram isnt much

And thats the most important for ai tasks

I wouldn't buy that GPU in 2025.
Better go for something like 5060ti

[elder tapir]

i buying it from a friend $500 aud

he's been using it for almost 3 years

i cant afford more than 600

nothing AI video .. just anime images should be fine right?

[warm junco]

For just images your already fine with a 6700xt with Zluda.
but for 500€ budget I would definitly go for the 5060ti instead of the 3080ti.
The 5060ti is 4 years newer, supports all nvidia features like DLSS 4.
less power consumption.
And 16gb of vram. (Which is important for ai and newer games).
Gaming difference is +10-20fps for the 3080ti, but if you play on FullHD it doesnt matter.

[noble widget]

guys, is there anyone who have even just a small idea of how to generate image like this? or is it impossible in Stable Diffusion?

#🏞｜general-with-images message

[sharp jetty]

This looks very much like something you can create with a checkpoint like pony and maybe some specialized LoRa. What did you try until now?

[noble widget]

none yet i will try it later tonight after i get off work

will let u know the result

[warm junco]

Its possible, go with illustrious based models like prefect illustrious or walnut and Upscaling is the key for the quality.
And you need adetailer for the face fix

[noble widget]

thanks @warm junco will update here on result

[royal patio]

Ai is getting out of control! Soooo anyone else on this new Google Trend? I created an meme instagram post. upload your image... and poof your IG post is made...
https://ai.studio/apps/drive/1nwkCHVk9YtueBc_yv1pFr325IwZTpUPU
no more need for ComfyUi anymore.

[fervent thunder]

Hi everyone! I’m new here and just exploring AI. Honestly, I don’t know much about AI yet, but I’ve been experimenting with some tools and managed to create a few models that look extremely realistic. I was really surprised by the results! I’m here mainly to learn more, connect with others, and also get feedback. If anyone is interested in seeing some of my creations, feel free to reach out!

[worldly hedge]

Anybody using Video Combine node from VHS in their workflows, havent you noticed preview generation becomes very slow, like 10x the usual almost instant, after like 10-20 generations?

[signal solstice]

It’s been a few months since I’ve been in this space. With winter approaching I imagine I’ll be making some more images. Is stable diffusion still getting updates? Is 3.5 the model to look for or SDXL with a lot of Lora’s etc still the way to go. I’m mostly into non realistic art styles.

[rain ivy]

Pls how do I download the stable diffusion model and activate it?

Also is it possible to train the model? If so how pls

[dusk verge]

hack my account

[true shuttle]

????

[solemn burrow]

Hi guys, is it possible to get a generated Pic and then use the hires.fix after the generation is completed?

[lucid trail]

anyone got a recommendation for upscalers for illustrations and graphic designs not realistic images?

[eager lantern]

Man trying to even install and use fooocus is so hard lol. Maybe I'm just dumb.

I'm having issues with Cuda versions and pytorch versions. Would using stability matrix fix these things?

[high ruin]

If your using Chainner which models are best for upscaling people??

[hot sparrow]

Hello 👋

[opaque forge]

yo does anyone do ai content?

[umbral lotus]

i've just discovered the 'infinite' amount of styles you guys from SD added to the online-gui 👀👀👀 Wow i'm speechless 🙏👏🤗😃

[opal coyote]

How do you use upscaler models in swarmui?

i have it in upscaling and its not seeing it, and when i move it to checkpoints it says it needs an architecture ID

[still glacier]

don't follow the link that scammer sent you

[still glacier]

depends of the upscaler

[opal coyote]

4x ultrasharp v2

[still glacier]

then depending of the back backend you've selected you either need some node to use gan models

[opal coyote]

i have both the pth and safetensor

[still glacier]

or put the models somewhere in models/ESRGAN

[opal coyote]

using comfyui backend

[still glacier]

should be the same thing inside, just in a different container. No need for both

[opal coyote]

i have both cause it wasnt working

neither work anyways

Figured it out

[still glacier]

that was fast

[opal coyote]

was looking everywhere for like an hour before i asked here lol

[cunning gyro]

Hello!

Which LLMs do you use with AI Editors like Zed and cli tools like Claude Code? Is there an LLM as good as or better than Gemini 2.5 Pro that is also free or at least very affordable?

On OpenRouter, I see good rankings for Qwen3 Coder, DeepSeek 3.1, and Grok Code Fast. Reviews of Grok Fast were not good, so I am skipping it. And DeepSeek 3.1 is already seeing growth with negative derivative while Qwen 3 is growing. I wonder if there are other models I am missing

I don't need Claude-level coding capabilities; Gemini Pro is enough in my case to ship working code that does what I asked it to do, but obviously, the more powerful, the better, as long as costs are capped at open-source model prices

[narrow kite]

what do you guys find to be the most photorealistic generator with comfy? im looking for more of less a "normal" photo without DOF, something that looks possible to be taken from an iphone

[spiral aspen]

Hello, when using stable diffusion creating depth map. is there a way to save in 16bits instead of 8bits?

[still glacier]

don't

it's a scam

[spiral aspen]

opps!

almost got me

thanks

[still glacier]

and to answer your question people would probably need more informations such as; what program are you using ? what is your workflow for generating depth map ? etc

Because 16 bits depth maps can be done yes.

eg : https://github.com/zhyever/PatchFusion

and this https://github.com/thygate/stable-diffusion-webui-depthmap-script

[spiral aspen]

i am using thygate extension that is how i generate those depth maps but no were to control to 16 bit

[still glacier]

confusing the two repos my bad

I see this thread about it https://github.com/thygate/stable-diffusion-webui-depthmap-script/issues/125

but it's 3:30 am here and I'm gonna head to bed pretty soon

[opal trench]

Hi everyone im new to stable diffusion have been trying to set it up to run as best as possible on my rig using chat gpt and cant lie it has caused me no end of problems I have been able to make some nice pieces of art but ive had more issues with chat gpt pairing the right sdxl VAEs and loras could anyone please help me

So far ive had to start fresh 5 times because chat gpt keeps effing it up

Working on a nvidia before rtx 3050 8gb vram any help much appreciated

[warm junco]

Hey, can you show a full cmd log in the #🤝｜tech-support channel?
Want to see if everything is setup correctly

[whole raven]

Meow, guys!

[real crown]

how do i use a private chat to generate images?

[raven agate]

Hey all, Long time no see 😛

Lot of coding porjects going as usual, mainly comfyUI related, some not.

In any case, figured you guys would appreciate this. Song generated and Edited in Producer AI, and Video tools, I used everything pretty much for Image and video generation. ComfyUI, WanGP, Qwen Edit, Wan Vace Multitalk FusionX, LantentSync, and just a bunch of other tools, some I made, some others made, and an amalgamation of techniques both old and new.

https://youtu.be/BKqp-B2dEpw?si=_uniot7cX08OuX0k

[desert dagger]

@vapid dove

[mild crown]

hello everyone 👋

[runic patrol]

Has anyone here used Noobai? Can i peak at your set up. trying to get mine sorted and i am struggling.

[timid birch]

Looking for long term partnership (US & Europe) Good Opportunity!!! Good Reward!!!

I believe that great results come from trust, reliability, and strong collaboration. That’s why I’m looking for a dedicated partner who is not only professional but also values long-term cooperation.
One of my biggest strengths is being reliable and supportive — I always stand by my partners and do everything I can to help them succeed. In return, I make sure that those who work with me receive better compensation and recognition than most opportunities out there. For me, success only matters when it’s shared.
This will add authenticity, presence, and impact, helping us reach and connect with a wider audience.
I''m looking for a private discussion, if you are interested in.

Enjoy our journey together !

[hearty canyon]

Hope your not gonna present an idea of let's start making ai OF girls

[waxen cobalt]

/

[urban thorn]

Hi I am new to Stable Diffusion, I was given an web address from Alison.com for me to follow the online tutorials, gifthub.com/AUTOMATIC1111/stable-diffusion-webui, this address doesn’t give me access to the interface so I can follow the tutorial, any ideas? thanks in advance

[still glacier]

No idea what tutorial you re talking about, nor what is Alison.com.
But for sure gifthub.com is not what you re looking for anyways, it should be github.com without the f

also automatic1111's stable-diffusion-webui (or A1111 for short) is quite outdated by now you should be looking into "forge" https://github.com/lllyasviel/stable-diffusion-webui-forge

[urban thorn]

Thank you, much appreciated

[high valley]

is runpod.io a privacy friendly solution for running SD ?

[true shuttle]

Of course.
It's definitely the private one.

[high valley]

I was thinking of training a flux model on my personal photos and play around and wanted to make sure my photos aren't stored on their servers

[true shuttle]

Of course your personal photos will be stored on the cloud storage but there will be no one who can access there unless you share your IAM credentials.

[high valley]

so if i were to delete them, would they be deleted permanently ? (i hope it's not like google & other services)

[true shuttle]

Exactly you are it is stored for some period( I"m not sure how long it will be) but after that it will be cleared.

[versed notch]

Hi. do you have paypal account? can u rent me?

[sonic forge]

**👋 Romeo | AI/ML Developer👋 **

Hi, there. I am looking for a paid job or work as a developer with 8 years of experience in AI/ML and WEB development.

Mainly, I focus on Voice AI agent, AI-powered chatbot, Automation, Data Science, Computer Vision and Web Development.

Voice AI agent: Vapi.ai, Retell AI, Twilio, Asterisk, 11labs, etc...
AI Chatbot & NLP: RAG system, Prompt Engineering, STT/TTS, LLM models such as GPT-4.5, GPT-4o, Claude 3-7 Sonnet, Llama-4, Gemini2.5, Mistral, and Mixtral.
Automation: n8n, Zapier, and Make.com, etc...
Model Deployment: Runpod, Replicate, Huggingface, etc...
Program Language and frameworks: Python, FastAPI, Flask, Django, Node.js, React, JavaScript, TypeScripts, Express, Next.js, Nest.js, etc... (Lovable.dev)

🌐 This is my portfolio: https://romeo618.vercel.app/

In addition, I always try to learn new and cutting-edge technologies, and I place great importance on collaboration with team members in development.
If you have any idea or project, plz DM me.
Thanks

[lucid bobcat]

Any devs in here that know the inner workings of ComfyUI? I'm talking about stuff like caching etc. I have a tricky question.

[still glacier]

Might as well ask directly in their discord if it's really ComfyUI specific.

[lucid bobcat]

I tried. The only people replying are scammers.

[half eagle]

Ok inbox me

[primal current]

My images that were generated in 20 seconds went to 50 seconds per image. Has there been any change in ComfyUI? It's literally been 2 hours and out of nowhere this happens, without changing anything.

[analog mortar]

hi im new to sd

[simple widget]

It has a good enough time limit., for you to store it.
You can transfer them in another loc which'll be suggested initially.
(It did that for me at least)

[brisk edge]

hello

[umbral zenith]

Is this the place to ask users here if they know some good partners that offer AI image editing (based on stable diffusion, or not) through API use?

[fervent thunder]

hello

I just came to have a look

sorry

[agile viper]

what vae should i use?

[tall kernel]

Hey everyone 🙂
Before i pay, Can Midjourney help me with making 2d characters into realistic n64 style characters, in 2-3 different poses but still sticking with the same aesthetic and art so I can use many characters and have the same output?

Thanks

And how would I go about doing something like that

Please DM me 🙏🏼💯

[versed notch]

How are you? Do you have paypal ? can u rent me ? I will pay u $100 per month

[meager fulcrum]

is there a wan discord server? im having issues setting up a wan 2.2 workflow..

oh nvm i got it

[next geyser]

hi

[true shuttle]

Halo

[next geyser]

are u bot

[true shuttle]

Kidding?

[true shuttle]

You bot?

[next geyser]

no

how can you watch me immediately

[true shuttle]

Just watching atm.

[next geyser]

lol

ok

I am looking for dify engineer

do you know dify

[true shuttle]

Dify is the low code platform for the chatbot

Isn't it?

[next geyser]

don't you know well

right ?

[true shuttle]

Not really

I've mastered it.

[next geyser]

where do you live

[true shuttle]

You want to generate image using Dify?

[next geyser]

no

business plan doc

for our company

[next geyser]

kidding?

[true shuttle]

Not interested.

[next geyser]

I am looking for dify dev

[desert dagger]

what is dify?

[next geyser]

lol

dify is ai tool

[hearty canyon]

Can someone give me banodoco discord server invite link i can't find it

Even reddit one is not working and website also

[agile dove]

hello, i'm new here

[small ravine]

yo, can anyone help me i'm struggling. i've been trying all kinds of things and i don't understand what's happening. i am trying to make my own model, i trained a lora using kohyass and it just doesn't seem to work. chatgpt is giving me headaches aswell. if someone can kinda teach me then that would be awsome

[orchid cloak]

Hello
Can someone help me to train my Lora(I got images) and help me to get a model from civitAI to combine my lora with model from CivitAi
I will pay the person when is done 🙂
I got stable diffusion but got stuck here

[desert dagger]

Why would you look on the stablity AI discord for people using a tool that stability doesnt provide?

[desert dagger]

first question - what model did you train the lora for. second question - did you specify a trigger word when you trained it?

[desert dagger]

what model do you want to use the lora with?

[small ravine]

just stabilityai/stable-diffusion-xl-base-1.0 and yes in the captions if thats how it works

[desert dagger]

okay so that's SDXL - that means you can only use that lora with SDXL and since you specified a trigger word when you trained it, you HAVE to put that triggerword in your prompt

[small ravine]

thank you but i already know that

ill just try to train a new one

[deft vine]

Idk if its the right place but i use stablediffusion webui locally, and for some reason i tried 5-10 loras and none seem to work, i try putting weightings but it does nothing doesnt change the image at all

[modern pagoda]

Try in #🤝｜tech-support

[grave kayak]

https://github.com/kijai/ComfyUI-segment-anything-2

Hi, I'm looking for a node where i can click on a part of the image and it will segment and create a mask. Basically exactly what this experimental PointsEditor node does. However this hasnt been updated in a long time and I cant get it to work. Does anyone know a node pack or node that can do this and is up to date?

[daring condor]

Hey, if someone could assist me with creating my own lora i would be very happy, i have been trying for 2 days to make my own lora, i have around 150-200 sample images of a experiment person and i have captioned them all, i have downloaded stable diffusion and after many youtube guides and articles i just found out that kohya is awful tool (good, but is not working properly, does not even work for me) and after a lot of trying after downgrading python and around 6 hour of similar download install uninstall remove upgrade downgrade etc im stuck here ready to start punching my monitor.

I would be very happy of anyone who could help me, (btw i have an rtx 3060 with 12gbs of vram)

i am not sure if i should post this on tech support

since im not sure what exactly my question is about

ill write it there too

:P

[true shuttle]

🤣 🤣 🤣 🤣 🤣

[true shuttle]

I used Kohya ss to train my own lora model yesterday!.
Just DM me

[daring condor]

i did it : D

[rustic tusk]

привет, есть какой-то человек говорящий на русском который может мне отвтетить на несколько вопросов по stable diffusion?

[still glacier]

No, this is an english server.

[left quiver]

Hi everyone! I would like to know if Stable Diffusion has something like Runway or Luma's Creative Partner Program. If they do, could someone point me in the right direction for how to apply and what the requirements are? Would really appreciate any insights! 🙏

Thanks in advance! ✨

[still glacier]

I would reach directly to them by mail [e-mail]

can be found in their website's footer, "contact us" section

[manic grove]

Hi, is there any way to have an auto top up for credits or atleast a warning email when credits are low for the API?

[pale holly]

Hi! Im just an ex artist who has a big interest in what Ai art can do. I was told to come say hello! 👋

[fervent thunder]

Hello! I’ve got the whole image generation down but I legit cannot figure out videos can someone help me?

[dapper girder]

Does anyone know the best local image upscaler? I want to upscale to 8-16k

[opal coyote]

Swarmui not finding 90% of my loras anymore, how to fix? it only sees dramatic lighting slider

[queen steeple]

Hello everyone ! I have an existential question : why does comfy and ReForge produce different results for the same prompt and parameters (I know if comes from the way it manages noise and maths, but why ?? Math is math, isn't it ?)
And how can I get more similar results ? I find some gens very disappointing with Comfy.

[still glacier]

Different implementations. I would not bother trying to replicate the same output across comfy and reforge. Plus ComfyUI basically let you create the whole pipeline from lego pieces whereas others follow a (mostly) predefined pipeline.

[queen steeple]

It's not that I want to get the exact same results, but for example, when I use EpicRealismXL, I find the output quality better with ReForge.
It's an issue because its dev already had to step back few weeks for personal reasons, so I would like to be sure to get a similar quality with Comfy

[still glacier]

Would probably need some outputs and settings to compare. Your Comfy graph can probably be improved. (that would be a question for #📝｜prompting-help (I'll be commuting to work pretty soon so I probably won't be able to answer any time soon)

[queen steeple]

Understood, I may ask there then !

[obsidian heart]

Hi everyone! Eager to jump into such a vast and imaginative tool

[brave birch]

how start ?

[desert dagger]

read the information in #artisan-faq

[brave birch]

Thanks

[alpine bear]

Throwback to when we used to do competitions on a monthly basis here
Wish we had access to the old winners channel to see all the awesome artwork

Hope the good ol' gang is alright. Fruit Jae and Kevi. Guiz. Never did get that miku role..

[dusty trellis]

What percentage of your generations come out good and have followed instructions?

[spare gyro]

hi

[lapis swallow]

What is the best way to merge a few styles LORAS with different weights into 1 LORA?

[oblique jay]

Realistically, which model would be best to use in today? Not talking about SD only, but all of the models that are currently available on the market?

Been using Whisk lately just because of the ease-of-use. But I was curious about which would provide me the most granular control of things with the highest quality.

If anyone responds, please @ me 🙂

[quiet spruce]

Hi, sorry if I'm asking in the wrong chat, but I have some questions.
If I want to train a lora character with pictures with a realistic character,what settings or captions should I have or apply to them to get a flexible character that will keep your easy details and can mold on other types of art style and not have problems generating multiple charactere.M I looked on youtube reddit civitai and even chatgpt,but all give me different answers and I don't understand anything.deci
1.Which captions should I keep to add or remove?
2.How do I calculate the epochs and number of repeats relative to my dataset? For example how many repeats and epochs on a dataset of 15,30,60,90 images
3.Which optimizer is recommended for sdxl for characteres?
4.What settings to apply for alpha and dim in training?
5.What other little tricks are there to get a good lora.
Thanks

[rose willow]

🛌 💤

[primal current]

Hi friends, do you know of any open-source AI that creates sound effects? I create my videos with WAN 2.2, but I need to add sound. Open-source is preferable because it shouldn't be censored.

[jovial lotus]

Hello

[still glacier]

saw you there little scammer.... Cleaned and banned.

[fair forge]

hey guys i want to give stable diffusion another shot, what is the best way to start with it?

[still glacier]

look at pinned message in #🤝｜tech-support , install forge, browse civitai for some model you like and install it.

[lavish tulip]

Hello, How can I generate an image with AI here?

[true shuttle]

Sorry but you can't generate image here directly

[lavish tulip]

@true shuttle where then?

[true shuttle]

On your local machine

And this is only the help server

[lavish tulip]

@true shuttle thanks 👍

[true shuttle]

Anytime

[keen ember]

Lincoln Stein is working on a cool search tool called PhotoMapAI that can search your image gallery folder based on image content, clustering the images into groups. Any plans to incorporate this into Invoke?

[still glacier]

ask the invoke folks about it :p

[buoyant quiver]

Hi is this the right place where i can find someone who can replace the face of someone from a youtube video (from an upcoming film) with the face of a friend (male) of mine? It’s just for shits and giggles

If not, can any kind hooman point me in the right direction

[formal imp]

Hey guys, quick question — how do you usually go from img2img to a video?
Do you use some script, extension or maybe another tool for that? Just curious what works best for you.

[trim pecan]

Whats goin on here

[final cipher]

Hello everyone, what model is best used to use txt2image for dnd avatars and such.

I currently have 2.1

[spring finch]

hello

[torn isle]

hey guys, I'm looking for cracked devs building tech products to set up an accountability group with

[karmic zephyr]

🚀 Building a Prompt Portfolio Builder for AI creators — helps turn messy prompts & screenshots into a polished, client-ready portfolio in under 1 hour.

Looking for 5–10 beta testers to try it free & share quick feedback.

👉 Comment or DM if interested!

[faint timber]

@formal imp Personally, I'm still looking for something that works. Currently, we're limited to 5 seconds of reliable output. Most AI videos are stacked up as 5-second clips. Check out FramePack to self-animate stills. Wan, LTX, and Hunyuan are available, but their prompt adherence is severely lacking on all technology fronts.

[true shuttle]

What kind of products do you create?

[next geyser]

[Request Summary]
Project Overview
Our company plans to launch an integrated website that combines "manufacturing, fintech, and energy."
Our goals are two-fold:

• Raising awareness of the efficiency of corporate payments:
• Simplifying domestic payments and overseas remittances and receipts with the ease of a bank transfer. Securely accessing external OwlTing (OwlPay) and JPYC APIs from the UI.
• Factory electricity trading using JPYC: Bridging the gap between online surplus electricity transactions (bilateral, auction, and market-linked UI concepts). The design references the information design of the Digital Grid.

• Our company does not hold its own financial licenses; instead, this is an "intermediary UI" that links and embeds external (OwlTing and JPYC) functions. Fund transfers and other operations are expected to be handled by each service. The design is based on the premise that JPYC is classified as a prepaid WEB3 payment method under Japanese law.

budget : 500＄

[obtuse edge]

Hello everyone...

[primal current]

Hi friends, how can I convert "nsfw_gold_8.5k_final.pth" to safetensors? It's for MMaudio. I've tried everything with claude and gpt, but it won't read it.

That pth is wonderful if I run it alone in MMaudio git, but I'm interested in knowing the process to get ComfyUI to read it.

[tropic frost]

so i started, training loras over on civitai

been getting some decent character loras, but feel like it can be better

i try to scrap together 20+ images for training

does it help much if i resize all images to around same sized?

are 10 epochs enough, or would it be better to go for 20 if i dont have that many base images

[true shuttle]

You should increase your epoch in that case.

Also 20+ images will improve the quality anyway

[manic grove]

Hi, I have attempted to request a Community License and the message after says I should receive something in my inbox but there is still nothing. Could anyone help me with this?

[modern pagoda]

We should give out awards to the communities best artists & users who help others here the most. Nothing more than metaphoric golden banana stickers but as a gesture towards their creativity, dedication, & techniques.

[primal current]

What's the best VACE 2.1 workflow for making a character copy movements from a video? I've tried several, and they come out horrible.

[desert dagger]

to start with, civit's trainer is lousy. luca taco has good trainers on replicate, use those

[abstract quarry]

.pth is not a standardized format, so there is no easy answer. You have to find out what it is

very likely it is a pickled python object, so try

import torch
whateverThatIs = torch.load("nsfw_gold_8.5k_final.pth")
print(whateverThatIs)

in best case it is a dictionary of keys and tensors, then you can directly put that into a safetensors file via the safetensors library

if its a model object, you usually can extract its keys and tensors via

dictionary = dict(whateverThatIs.named_parameters())

even then a tool like comfyui will only be able to read the safetensors file if it knows the model architecture

[fleet raft]

hi all

[wet kettle]

hello, I can build your site perfectly based on my experience.
let's discuss about the requirement of your project with more details.

[primal current]

i go to check

[trim pecan]

Whats goin on chat

[true shuttle]

Chat is done here. lol

[strong vigil]

Hello everyone

Can someone help me?

Can I still use stable diffusion without NVIDIA?

If not, how do I use it via Kaggle?

[sweet pivot]

you cannot run the stable diffusion in your machine without NVIDIA.
To run it locally or your own environment, you can use google colab or the other cloud runtime environment such as AWS ML services.
or you can use HuggingFace plateforms.
Kaggle is just only learning plateform.

this is my opinion based on my experience

[trim pecan]

Fr no doubt

[true shuttle]

You can simply do that with kaggle kernel's gpu or tpu

But yeah you know kaggle is not for dedicated service so it will be only used as some experimental purposes.

[true shuttle]

Btw you are an active kaggler?

[oblique elk]

Sure you can. Check pinned message in the #🤝｜tech-support channel. There are some guides on how to install a UI and backend components for AMD Cards as well.

[gleaming bloom]

check out GPU renting platforms. Vast, runpod, simplepod - all are pretty easy to use and price is very attractive

[hollow wind]

hello

[primal current]

My friends, do you know a workflow to change a character from one image to another character that I have in another image?

[strong vigil]

Can someone give me 10$ please?

[iron mica]

Hey, I'm hoping someone would be willing to upscale a class photo for me?

[ruby trout]

hello hello

[true shuttle]

What kind of image do you have?
I'll help you

[rapid laurel]

היי לכולם. יש כאן דוברי עברית?

[still glacier]

not really, this is an english speaking server.

[rapid laurel]

"What is the command for generating images?"

You're welcome in advance to anyone who can help me.

[still glacier]

cf #artisan-faq

[stark hornet]

hiiii

[floral umbra]

@vapid dove 3 compromised accounts above.

It's the mr beast scam

[dapper mauve]

hi ! can anyone tell me what this server is about ?

[modern pagoda]

AI Generative Art

[manic grove]

Hi, I have attempted to request a Community License and the message after says I should receive something in my inbox but there is still nothing. Could anyone help me with this?

I've also got a scale production I'm wondering if there is any way to enable an auto top up feature?

Could any staff/support assist me?

@hallow edge @vapid dove

[still glacier]

It's worth to tag @Community Guides too in those cases.

[still glacier]

Waffles ! It's all about waffles, square ones, circle ones, doesn't matter the topping. Also some of us talk about AI / stable diffusion.

[queen steeple]

Blue ones...

[still glacier]

you stop that

[queen steeple]

Hello people ! I'm looking for docs/tutos/communities to understand how weighting unet layers works (input, middle, output), I have recently tried to use it to get character consistency without a lora but currently it's all luck more than understanding.

[queen steeple]

It's to see who is a web rookie or not !

[nimble wyvern]

Hi

hello, I would like to expand my AI generating skills, where can I go for tutorials? Right now I would like to create a map for my game, I was thinking that first I would create a big terrain map and then add mountains and structures to it later. any hints where to go?

[light talon]

Youtube and reddit. I usually would try to look up loras on civitai, and then check out the images others have created that match what im looking for a experiment from that.

https://youtu.be/FLaics4IT-w

https://civitai.com/models/1114428/cityscapes-maps-and-sketches https://civitai.com/models/1113248/og-vtt-maps

[floral umbra]

A community for us who love messing around with generative "A.I", where you simply input text, and the trained models delivers based on the text input. And all generated content is based on seed. You can either lock seed and finetune one image/video/audio/song, or keep generating more random images based on random seed.

[proven badge]

lets OWO all the ones who does censorship since i almost shoot up a place becuase of censorship and i will if it keeps going.... free charlie screw leftist its time for war is now!!!!!!!!!

[regal ocean]

Can anyone guide me to some resources on how to train an img2img lora?
I want the model to add christmas lights to houses consistently. img2img models give inconsistent results with just prompting and reference images, but they can remove the christmas lights perfectly. I could easily create a dataset by removing the lights from images and the training the lora with the houses with lights as the target and the houses with no lights as the prompts

[winter cloak]

What Node can Route a Model name into Effici. Loader SDXL? (base_chkpt_name)

[light talon]

Im not super expirienced, but im slowly transitioning to SwarmUI from Forge. My project at the moment is about making icons for fictional NFL teams. But i cant seem to figure out inpainting and changing details on some of the images i liked. I tried downloading some comfy workflows but i just seem to get any good results. I have mostly used Flux since i want to add text to my icons (Orlando Wizards for example).

[stuck laurel]

I've tried installing a LoRA into automatic1111 but after setting everything up and reloading the webui the lora is not appearing in the Additional Extensions, any idea why??

[shrewd gust]

Did you hit the refresh icon on the Lora part of the UI itself?

[stuck laurel]

yeah

it just doesn't work idk why

i installed epic realism and it shows , but all the loras i install off civitai dont show

[shrewd gust]

I think also not all Loras work for all models, some dont show up depending on model in use (I could be wrong about this)
Could it be that?
Does it show up if you use a different model?

[stuck laurel]

maybe .. how do i know which model to use for this lora?

i'm really new to this so i dont really know the difference between these things

can i send you in dms the lora i'm trying to use?

[shrewd gust]

Im fairly new to this myself but Im sure not all of the Loras appear on all of the models, its just trial and error I guess
I would say if it doesnt work on the model you really want to use, see if theres another Lora that does the same thing

[stuck laurel]

Photorealistic AI Influencer – Woman877 – Character LoRA [SD1.5 | SDXL | Flux]

this is the title of the lora

i suppose the model it uses is [SD1.5 | SDXL | Flux] ?

[shrewd gust]

That would be logical but you know computer stuff, its not always logical 🤣

Some of them have trigger words too right? Have you checked if there are trigger words you can use in your prompt, and if it works by using those?

Theyre usually listed on the download part on Civitai if thats where you got it from

[stuck laurel]

in the website it says trigger word: woman877

how do i exactly use it?

[shrewd gust]

So if you type that into a prompt that should make it work

[stuck laurel]

can i dm you for a sec?

to show you screenshot

[shrewd gust]

I dont do DMs sorry, I got like 3 friends and thats it 😛

[stuck laurel]

do i type the trigger word , followed by whatever i want and it should make the face always identical on every photo?

[gray surge]

yo

[warm junco]

You have to check the info box of the lora on civitai to see what model type its made for

It will only show up when a compatible model is selected

[fiery ore]

Hello guys

Is anyone into AI influencer here?

[vapid dove]

Hey, shoot me a DM

[manic grove]

Just sent.

[weary zealot]

Greeting everyone I'm new here can someone teach me

I'm new here please

[still glacier]

whare are you looking to learn ?

[runic patrol]

sorry for bothering but which channel can i ask for help regarding a recurring gen issue i am having

[chrome adder]

What's the best i2v out there?

[warm junco]

If its technical then in #🤝｜tech-support
If its about the prompting or getting a specific image then #📝｜prompting-help

[runic patrol]

I am not sure if its either or to be honest. I am not sure what will fix it. I will explain in the post. thank you so much. ❤️

[white night]

🤚 Looking for a long-term U.S. partner — great rewards await. D!!!M if interested!

[desert dagger]

post in the #1092446741984444416 forum

[near halo]

Olá quero gerar um vídeo já tenho o prompt aqui , alguém me envia o link da pagina de colar o prompt ?

[grave stag]

Its been SUCH a long time since I've tried getting back into this and running it I just found it to confusing to use last time.

I don't know if talking about other platforms is even allowed here but is there a way to do image generation like how GPT does it where you talk about the image, it asks you questions and then you generate? Then it asks you more questions if your doing edits or something.

Or is that like way more advanced and not possible at all to run locally even on like a 9070xt? Or even a 5080/5090?

If I can't run it locally is there an alternative to GPT with a free plan that isn't as censored not that I'm making true nsfw but just that GPT is overly strict.

If I'm not allowed to talk about other platform here or something or this breaks rule 2 (maybe I don't think so but it could being less censored then GPT) just delete the comment mods and give me a warning I don't want to get banned here this is just the only dedicated AI server I'm in and idk where to ask? The tech servers I'm in nobody helpful has been responding.

TLDR the way you talk to the LLM about the image for generating and editing is there a way to do that for a less censored model then GPT especially if its not anything paid?

[still glacier]

It would be dishonnest to silence talk about others AI. No probleme here.

With that said I don't think I ve seen anything like you mentioned. And tbh I don't really see the point of it but ok. There are tools to assist you building your prompt for sure. But at some point if you re not drawing nor even writing the prompt / tweaking every extension, where s the fun.

[oblique elk]

I would suggest you simple use a large language model to help you with the prompt building. There are some smaller and less uncensored ones out there. It would not solve complex tasks but simple prompt building is not so hard. Simple look at huggingface for uncensored llms
The prompt based image editing works with either flux kontext or qwen image edit. But they need more runs if you want to change more elements. So “make the dress red” one run, “change the color of the red to teal” second run. Do not try to add complex combination in one prompt.

[molten elk]

Hello guys, I'm trying to find a way to make Paints Undo works but nothing, is there someone that succeed with it that can put an image of mine inside to see the results please ?

[strong vigil]

I am

Guys what AI to use to make videos?

[fiery ore]

That's cool

[strong vigil]

Like this https://youtube.com/shorts/HzSz7YRigZQ?si=Zdhar7vA7luBc1d4

[fiery ore]

How long have you been into it?

[strong vigil]

Just recently

Havent made anything yet because it seems like i need to use SD and have a good GPU which i dont have

[fiery ore]

Which software are you using?

[fiery ore]

Ohh

You can you Google runpod to operate it

[strong vigil]

I dont have money either

Rn im thinking of doing AI Videos

Gotta figure out how to do it

What AI to use

[grave stag]

So if I got like a full, create a cervitaur (deer centaur) in the forest, bla bla I described the outfit and what not and what she is holding and her expression it would let me make whatever I wanted (give or take one or two things I changed in edits later) then if say like her bow was to big in her hands I'd tell it to make that smaller, one edit, next edit I'd change her shirt to a crop top, third edit I'd change the color?

I guess LLM integration of here is the image, here is what I've crated, here is what your asking me to edit, do you want x or x for the specifics of the edit, x it is like x or change the style to x..then I run the edits the way GPT does does not work unless your one if the 3 or so big shots?

Last time I tried to use stable diffusion (and I was on a 4070) I spent like 5-15 hours with chat help (not all at once) trying to make one still of an anime character that could look like from first glance it was or could be a frame from the show and could never get past the uncanny valley what cursed thing do you make that looks somewhat inspired by the girl your trying to recreate. I said if I can't even get one yep that's her alright good enough image I'll never get my refs the exact way I want when I'm more specific with my details let alone edits and gave up. I take it its gotten both better and easier on the learning curve and is usable now for me?

[elder tapir]

im not really good with computer gpu parts, but i have a question, which one performance good overall? I saw two good price for a ROG Strix 3080ti 12GB (Pre-owned) and a Asus TUF Gaming 5070ti 16GB (slightly pricey), I'm planning on using it for Stable diffusion and for gaming as well, which one performs better over all? im noob to all this lol

[faint timber]

You need more VRAM to move forward. Go for the 16Gb card.

[true shuttle]

Exactly!

You should use 5070ti

[torn ferry]

hi

[craggy rune]

is there drop in wheels I can try in a third party mac app?

[grand nacelle]

Hello, who should I contact about a possible partnership here?

[viral fog]

is there anyone who can help me generate pics n stuff

im finding it hard to understand the settings of the generator

[still glacier]

Maybe @vapid dove or contact stability.ai through their mail directly https://stability.ai/contact

[vapid dove]

shoot me a DM

[still glacier]

which generator ?

[viral fog]

any really, just the gui settings im confused abt

like what does LR do or cfg or sampling steps or epochs, these are all new words for me😵‍💫

[still glacier]

which gui ?

[viral fog]

A1111

im a newbie so idk whats the regular software people use, i js saw stable diffusion discord server and said why not

[still glacier]

k then I d suggest that you hover your mouse pointer on whatever setting you don t understand

there should be some infobubble 99% of the settings

[viral fog]

ohhh i didnt even know it showed

appreciate it

[still glacier]

Also know, that a1111 hasn t received any update for more than a year. So it s probably best to move to forge instead. It s basically the same UI, you won t get lost

[viral fog]

thanks ill look into that

[still glacier]

There are some install guides in the pinned messages of #🤝｜tech-support

[charred hazel]

hey fellas, is there a channel to ask questions about prompting and config in general? i'm getting good results overall with my prompts but having many many troubles trying to replicate an specific hair style

[still glacier]

#📝｜prompting-help

[charred hazel]

ty ty

[inner barn]

is Stable Difusion really opensource?

[still glacier]

Yes ?
https://huggingface.co/stabilityai
https://github.com/Stability-AI

Just not the training set.

[rotund tide]

hello who can help

[dawn mulch]

SD 1.5 is still kind of the best out there is it It's so much smaller than all the other image generators but it's quite capable. And it eas released 3 to 10 years ago. In that light, all the new image generators in some ways are not even close to sd1.5.

[silver juniper]

Anyone here good at Comfyui? I am new and I am running into an issue where, I upload a new workflow to COmfy, but the missing nodes that I need to install are completely hidden. I can't even see what I need to install, there is just massive gaps of blank space in the workflow. Anyone know how to fix this?

[dawn mulch]

I think you can simply expand the nodes to show what they are, unless you are talking about something else. If you hace comfyui manager, it should tell you what you need to install. But comfyui ui manager isnt all that accurate either.

[silver juniper]

That's the thing, all the nodes I need to install are just not visable. I can't see them on the board and there is nothing to install via Comfy Manager... It's so odd

[patent pike]

Someone here have a RTX 5060 T.I 16GB? How does it compare with RTX 3090 in performance for SD? Can the newer techs be better than more VRAM?

[gritty lava]

hai stable diff peeps, dunt fall for the Wumplus nitro bot scam, dunt add it, spread the word 💜

[knotty bobcat]

Good evening everyone!

[inner barn]

Oh ok thank you

And would ya know under what licence it is by any chance?

[viral fog]

Need some advice guys

Im training a style lora and the safetensor came out to be 2 GB

But apparently thats too heavy to be a lora?

[warm junco]

Yep thats not a lora then

[fiery ore]

The blank gaps in your workflow happen because ComfyUI hides any nodes you don’t have installed. To fix it, you can check the terminal output or the workflow JSON to see which nodes are missing, or install ComfyUI Manager, which makes missing nodes visible and easy to add. Updating ComfyUI also helps if the issue comes from version mismatch.

[viral fog]

Well yea but thats what i dont get. What did i do wrong that my training turned it into a model

[gentle prairie]

does someone else have this issue on civitAI today or is it just me?

i don't need a ticket, just if it's the same for everyone, lol, civitai shows update notification, 25, but it doesn't show anything when i click to see them, was just wondering if anyone was experiencing the same to see if it's on my end or not

not a big issue

[still glacier]

that was most likely a bot / scammer answering you.

[gentle prairie]

oh lol

[still glacier]

as a proof they re now cowardly hiding themselves.

cute

[still glacier]

depends of the model

[grand nacelle]

Thank you. I've sent you a DM, could you please check?🙏

[fickle jacinth]

What are the best options for AI image generation right now? I've tried some free options I've found and Sora was doing the best by a mile in the anime department at least.

[abstract quarry]

Flux and Qwen do a solid job and are much faster than Sora if you have good hardware

yes, Sora has very good text understanding, but it also tends to produce the same styles over and over again. Usually, you recognize immediately that an image is made with ChatGPT

Flux and Qwen have worse text understanding, but for 90% of the use cases they are good enough while being MUCH faster

and with Illustrious or NoobAI you have plenty of different styles

[fickle jacinth]

By text understanding, do you mean that you have to finetune prompts less?

[abstract quarry]

I mean complex prompts that describe a lot of details

text understanding is: Sora > Qwen > Flux > Sdxl/illustrious

if you want to describe every element in your image in detail, Qwen and Flux might miss more and more stuff. They cannot focus on too many details

[fickle jacinth]

I see.

How restrictive are they when it comes to what they can put out? I remember using some generator made by Bing a year or two back that denied most prompts even if they had nothing indecent in them.

[abstract quarry]

I talk about local models

Flux dev/Krea, Qwen, SDXL, Illustrious, Chroma - these are all open and local models

you run them on your own hardware

and there is no rejection or censorship at all

[fickle jacinth]

Oh, I see.

[abstract quarry]

they generate everything they know about. If you want nudity, you usually need finetunes for that (e.g. Illustrious or Chroma), because Flux and Qwen are not trained on that

[fickle jacinth]

I'm not interested in that, I just remember having a bad experience with restrictions at one point.

[abstract quarry]

yeah, I know. Even Sora can be annoying here

I once asked for an image where one character is doing the "thinker" pose and Sora rejected the prompt because of copyright issues 🤦‍♂️

you don't have these issues with local models

[fickle jacinth]

I assume the drawback is that you need decent hardware to run them.

[abstract quarry]

yes and no xD

like Illustrious runs on basically every gaming graphics card

for Qwen you should have a good one. It would also run on older hardware, but then it takes minutes to generate an image

better models need better hardware. But it depends on what you want. Many people here might argue that Illustrious, a very small model, is better for anime and manga than Qwen, a very large model

[silver juniper]

Thank you for the reply. When I check in the manager, There are no missing nodes listed, even when I click "Check missing nodes". Everything is blank. It's very odd because 2 days ago, I could see the missing nodes and they were highlighted in red.

[fickle jacinth]

I see. Thank you for the help, then. I'll look into the ones you mentioned.

[analog pollen]

Hi

[fiery ore]

If missing nodes no longer show in Manager, it’s likely a Manager glitch or the workflow was saved after nodes disappeared. Checking the terminal logs will still reveal what’s missing, and resetting or updating Manager usually fixes the issue.

[silver juniper]

Ok, thanks, I will try this

[fiery ore]

Alright
You can hit me up if you're still facing same issue

[silver juniper]

Much appreciated. Quick question, Im using this through StabilityMatrix, could that be causing an issue?

[proven badge]

who thinks its a good idea's to buy a hat maker and start printing nazi hats shirts and more.. since times have changed and the symbol is not hateful anymore

[proven badge]

you know what im going ye mode with this one

[silver juniper]

OMG, it worked! Thank you so much!

[fiery ore]

I'm glad to hear that

What are you trying to create tho?

[wind harness]

hi folks

[silver juniper]

Everything! I'm new to comfy, just moved over from Automatic1111 and I'm having so much fun with all the possibilities. Right now I'm digging into WAN video. The workflow I uploaded was a "talking head" video

[fiery ore]

Ohh that's nice

Well I actually based on AI model creation which includes create images and videos using ComfyUI.

I can teach you tho since you're new to ComfyUI

[silver juniper]

Does anyone have a good workflow to create images using face as a reference? I am having very little luck with the ones I'm finding.

[knotty bobcat]

Hi. Good morning!

[cobalt silo]

do most people use webui or comfy ui?

[floral oxide]

What’s best way to learn couple latent for forge SDXL?

[proven badge]

question what would happen if 1st Amendment and others just disappears since the leader says they are not improtent and kicks them from the rights and freedoms???

[crystal cosmos]

I have the same issue :/

i can run txt2img, adetailer seems to do nothing

then when I try to detail the image manually in img2img with a mask then i get the error: modules.devices.NansException: A tensor with NaNs was produced in Unet

if I try working around it with --disable-nan-check then I get all black

another checkpoint works fine, only Illustrij gives me that problem so far

running on MacOS M1

[obsidian ingot]

not how that works

[proven badge]

grand theft auto 6 is going to be wild since gta 6 is going to be the liberal and conservative sides of the political spectrum and it going to have tons of left vs right meme's and culture

[obsidian ingot]

Yeah there’s a reason every gta has made headlines

[proven badge]

yes and this time they can havE charlie in the game XD

[warm junco]

reading the github page of latent couple ^^ its not that difficult to use

[knotty bobcat]

、

[terse wedge]

hello everyone

I have an existing Python pipeline in Google Colab using Stable Diffusion 1.5 via the Replicate API.

I need someone to:

• Add LoRA for consistent identity across 4 full-body images per character
• Integrate IP Adapter or reference based control
• Generate multiple unique characters (4 consistent full-body images + 1 avatar each)
• Ensure variation in skin, hair, clothing, and background
• Prevent duplicate faces
• Save metadata to CSV

Everything must run via Replicate API (Colab-based).
Do you have direct experience building Stable Diffusion workflows with LoRA and ControlNet?
If yes, can you handle this as a full project delivery?

kindly send me a dm if you can handle this project perfectlly

[fiery ore]

I'm in your dm

[signal dome]

Hi

[violet prairie]

Check out all thing about VimGolf AI competition: https://vimgolf.netlify.app
All about OpenMule: https://openmule.netlify.app

[orchid cloak]

Hello guys,
I have to find a guy with NSFW for my character and it will be a long tern partnership with payments but i need some samples and i will provide images.

Somebody interested?

[fiery ore]

Can I dm with some samples?

[fiery ore]

I'm in your dm

[humble grove]

Okay

[jagged night]

hi i am new can anybody tell me best web UI for stable diffusion i have intel i5 and nvidia rtx 1650 graphics card

[warm junco]

Hey, Forge webui should be the best for that GPU

You find a setup guide on the first link of the pinned messages in #🤝｜tech-support

[jagged night]

i used it and today i deleted it because local control net is not working

inpaint everithing needs web_ui_controlnet

and forge is not supporting it

@warm junco can you tell me a way to fix it?

[warm junco]

Hmm haven't tried inpaint everything on forge. Is it a controlnet model or an extension itself?

[jagged night]

Sorry it's inpaint anything it's an extension which needs controlnet for it to work properly

Inpaint anything had some issues o but i fixed it with a post from github @warm junco

[warm junco]

yea i just checked github right now and there some fixes in this thread:
https://github.com/Uminosachi/sd-webui-inpaint-anything/issues/154

[jagged night]

@warm junco Yeah this is the post i used to fix inpaint anything but in this post there no fix for controlnet

[warm junco]

what exactly doesnw work with controlnet? do you use the 1.5 controlnet models ?

[jagged night]

@warm junco Forge has given inbuilt controlnet which does not work with inpaint anything

@warm junco i tried using every type of model nothing works

@warm junco if you are using forge Ui try downloading inpaint anything and U will see what the issue is

[warm junco]

will check

[jagged night]

@warm junco inpaint anything has a problem where when you try to segment the image it will be in segment but brush will not appear to paint the specific area

@warm junco https://github.com/Uminosachi/sd-webui-inpaint-anything/issues/163 this is the link from where i copied the code and pasted it in extension folder to fix it after that it worked

it will be biggest code in that post

[warm junco]

so it works now or not? xD

[jagged night]

@warm junco Inpaint anything brush is working but the main part where i should be putting my prompt for changing the image needs controlnet that is not working

[warm junco]

ahh okay now i get it

[echo jackal]

Can someone recommend the best model right now for generating realistic images? I’ve tried EpicRealismXL, but the results haven’t been very impressive. Even after increasing the sampling steps, the outputs still feel lacking. Any tips on how to improve the results would also be greatly appreciated.

[rugged onyx]

i just installed stable diffusion on my asus tuf f15 for fun and im in genuine awe at the fact that its worked right now wtf

using webui rogue

[warm junco]

webui rogue? never heard of that one

[rugged onyx]

im. an idiot i meant forge

[warm junco]

ah alright xD

[warm junco]

dont forget to add --cuda-stream to the webui-user.bat commandline_args if your gpu is a 30, 40 or 50 series gpu

it can improve the performance

[rugged onyx]

mine is a gtx 1650

which i believe supports CUDA (at least from my experiences torturing it experimenting with other ais)

[warm junco]

Yep it does. Also forge is very good for the vram management of the GTX 16 cards

Even sdxl model can work on it.

[rugged onyx]

0_0 i did not know

wait does that mean it can run other ~7gb models too?

[warm junco]

Sdxl/illustrious ones are around 6GB. If your GPU has 4GB vram they can work but will be slower than 1.5 based models.
Also if you only have 16GB of normal RAM you will need to increase the windows pagefile to not get stutters or freezes when using them.

[outer patio]

Someone with open commisions of generations? :3

Just dm me uwu

More comfy than checkin the chat back uwu

Haha

[rugged onyx]

i have 8gb of normal ram 💔 does that mean i'm cooked

[warm junco]

Hmm you should definitly increase the pagefile then and test

Here is the Guide how to increase it:
https://www.tomshardware.com/news/how-to-manage-virtual-memory-pagefile-windows-10,36929.html
Enable it only for C and make sure its not enabled for any other drive.
Set it to 16000 Min and 24000 Max for C
Then apply and reboot the PC.

Also make sure to have at least 24GB free space on the C drive aswell.

[rugged onyx]

wait should i adjust it to 8000 min and 16000 max since i have only 8 gb?

[warm junco]

No to 16000 min and 24000 max

[rugged onyx]

ahh alright

[warm junco]

It then can use that much SSD to outsource overflowing ram

That should also give a much better performance to the whole system

[rugged onyx]

got it

[jagged night]

@warm junco Did you try it? If it's not working can you tell me any alternative

For segmenting an image

[bleak pond]

hey guys

[blissful bane]

hello

[bleak pond]

was just on this server

[weak sparrow]

hi

[jagged night]

can comfy ui work with i5 and nvidia 1650 card?

[true tangle]

hello

[tiny prawn]

Qwen-edit models without reference images (only prompt) same or better than regular qwen models.

[terse sparrow]

Hey, does anyone know an option or extension in Forge to generate 2 images using the same seed but with different prompts (all in the same generation)?
For example:
Generating 6 images with prompt 1 + ADetailer + Highres fix
I want Forge to also generate 6 more images with prompt 2 + ADetailer + Highres fix
in the same generation, rather than manually waiting for generation, changing prompt, seed, and generating again.
I want to use it for something like before and after.

[fast sage]

Hey lads, want to take a illistration and keep the form of it but make it realistic, suggestions? style transfer isn't working that well.

[oblique elk]

Either image 2 image with a realistic model and try different denoise values. Or use it as input for controlnet depth

[primal current]

Does anyone else experience that for WAN 2.2 Animate, using the Kijai workflow, the color changes during the video? One of my videos changes color after the second 2. I use "color matching," the best option I found with MVGD. It fixes it, but it darkens the image. I've tried changing seeds, samplers, steps, checkpoints, and nothing.

I tried a Chinese workflow that doesn't use the Kijai wrapper, and it doesn't have that problem. The color looks better.

[candid nova]

Good timezone.

[regal palm]

Send me a private message if you are having any issues with your wallet

[silver juniper]

Anyone know what the best WAN 2.2 model would be to use with a 4080 Super - 16gb of VRAM?

[rigid bridge]

what's the best gpu to run this locally?

currently have a 4090 i think, was wondering if 5090 is best or maybe a used v100 or something?

[regal palm]

Hello everyone

[primal current]

Is there a way for ComfyUI to detect the total frame rate? I have several videos of different times. I change it manually, but sometimes I make mistakes or forget.

[fiery ore]

Sure
Can I dm?

[topaz owl]

I am exploring Stable Diffusion as part of a bigger project I am building and I am especially interested in using it for motion and avatars that feel like living networks. I am here to learn, experiment, and share progress. Transparency is important to me so I will be open about my process and mistakes along the way. Excited to connect with others pushing the creative and technical edges of this space.

[normal sun]

Hi everyone 👋,

I’m looking to commission an AI video creator or 3d expert to help me complete a cinematic project titled “The Warrior’s Trial.”

About the Project:

Cinematic 3D game-style video (similar to God of War or Assassin’s Creed cutscenes)

Runtime: ~3–5 minutes

Story: A male and female gladiator fight in a Roman colosseum (script provided)

Output: Scene-by-scene video clips (10–20 s each), later edited into a full short film

[fiery ore]

Can I dm?

[normal sun]

Yes please

[fervent thunder]

Hey everyone, i trained a Lora for a person in ai tools and then I’m using comfy ui for image generation. For some reason it’s not using the Lora what so ever and is basically just generating images as if there is no Lora

I’m really new, so I’m probably doing something wrong

My workflow is correct though, as I followed a tutorial, I just think the Lora may not be

Is there a way to test my Lora in another way?

[oblique elk]

Easiest way would to use a simple python code using the transformer library and use the corresponding model and Lora. You will find code by using google or ChatGPT

[abstract quarry]

check the console for error messages or warnings.
comfyui will not display a popup if a the lora and model do not match

[royal radish]

Hello

[flat cosmos]

hey can someone help me?

[fervent thunder]

Thank you

[jagged night]

Can anyone give me a name of extension which help be segment an image in forge ui i am using latest version which does not support segment anything

[wicked marsh]

Hello Everyone! just joined

[zenith delta]

hello

[regal river]

hello

[jagged night]

Can anyone please answer me my forge ui lora is being detected but not being used in images and there are not error massage

[fiery ore]

Hello guys 👋
I create custom AI-generated content
If you’d like a sample or something made just for you, my DMs are always open.

[jagged night]

Can anyone please answer me my forge ui lora is being detected but not being used in images and there are not error massage

[maiden cradle]

hi can someone help me to fix either error code 9009 or the 123 one ?

i tried to install SD but it's not working

[warm junco]

Hey, post your issue with more info, GPU, and screenshot in #🤝｜tech-support

[formal lotus]

Hi everyone! I'm VD, excited to join this community. I'm passionate about AI, machine learning, generative AI, and prompt engineering, and eager to learn, share, and collaborate with you all. Looking forward to diving into the discussions, and I'm a newbie.

[orchid swan]

Hello guys, I'm new here and I'm experienced in AI and LLM training and web scraping.

If anybody has issue with set up, I can help for free

[primal current]

My people, do you know of any legitimate places in the US that sell the RTX 5090 at the best price? New.

[crimson quest]

Yo can someone make an image for me if I give them the prompt?

[proven badge]

who here does not like racism?

[serene zephyr]

Realistic images?

[dawn mulch]

Like air and water, racism can be healthy if used right. But also like air and water, racism can be unhealthy if used wrong.

[proven badge]

your words that you just used is racist

[still glacier]

let's just .... not do that, ok.

[sleek fable]

hey,
I want to upscale and enhance the quality of old comics. upscalers fail here mostly because the detail is just not there and blurry and inaccurate lines remain as such... Basically I look for a NSFW capable model that I could use. Had good experiences with Flux + NSFW Lora, but it changed too much of the content 🙁

[mild crown]

how is going guys?

[last sierra]

AMD is working on an official way to support AI models not relying on just CUDA conversions?

[warm junco]

yep they call it TheRock and implements native pytorch support with rocm for any ai tasks.
Im updating my setup Guides this weekend for TheRock support

Its faster than Zluda and it works for faster Lora training too, but its still WIP

[last sierra]

Damn! Its fully ready ? Can be used on 7900 xtx aswell?

Ah yeah wipnofc,but works fine atm?

[warm junco]

its still work in progress but it works. I have a 7900xtx aswell and tested auto1111, comfyui, forge/reforge/forgeNeo, OneTrainer and Khoya_ss with it

[last sierra]

Ah heck yeah, so amd gpus are finally usable 🤩 (better usable and faster i reckon?)

[warm junco]

can do an sdxl/illus 832x1216 image, 30 steps, upscaled by 1.5 hires fix 10 steps, in 16 seconds

[last sierra]

Thats massive ngl! Gotta test this out when I get home🤩 ill look out for the guide

[warm junco]

yea its awesome, will update the guide real quick

@last sierra Here is the AMD TheRock Guide:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides#any-webui-with-therock-by-amd
Let me know if it works for you or if you have any questions.

[last sierra]

Thank you 🤩 will have to test it out

[fallow sparrow]

hello looks cool

[crimson quest]

Yuh

Just 1

[rancid basalt]

hi

jsut wondring i have not used stable for like 2 years what version should i download?

[uncut lake]

Hello

[warm junco]

Hey what's your gpu?

[bleak pond]

hi guys

[bleak pond]

What did the traffic light say to the car?

[shadow prism]

hi guys

[cunning verge]

Hi

[merry acorn]

RuntimeError: CUDA error: no kernel image is available for execution on the device CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect. For debugging consider passing CUDA_LAUNCH_BLOCKING=1. Compile with TORCH_USE_CUDA_DSA to enable device-side assertions.

What the problem?

I cant use stable diffusion at all

[warm junco]

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

[sterile sphinx]

any idea how much faster it is?

[warm junco]

with forge and zluda i get 3.40 it/s and hires fix 1.08 it/s getting an upscaled image in 32 seconds
with reforge and TheRock i get 4,7 it/s and hires fix 1.81 it/s getting the image in 20 seconds

[sterile sphinx]

oh damn, that's quite the improvement... I might consider switching

[warm junco]

yea you can also install it and just point the models folder to it for testing

thats what i do with every webui

im currently updating the guides and will include Forge NEO and Forge classic. But need more tesing with them

[rancid basalt]

2060 super, but i have my laptop with npu amd radeon graphics and amd ryzen 7 ai

[warm junco]

Then go with Forge Webui
You find an install guide in the pinned messages in #🤝｜tech-support

[rancid basalt]

On pc or laptop?

[warm junco]

For the laptop you can try setup AMD with Zluda or TheRock

[verbal dome]

any way i can put my image in img2img crop out or mask my image and render the scene around it? i tried all settings by masking my image, or only masking outside my image and no such rendering was possible on flux. Tried inpainting and sketcher.

[desert dagger]

@vapid dove spammer alert - this is what they're doing now to get around the filters on most discords and this guy likely spammed this post in all channels

[modern pagoda]

Nuke em

[abstract quarry]

this is called outpainting.
You need a very high denoise rate for it. It works best with inpainting/outpainting models like Flux Fill

with Flux Fill you can outpaint with 100% denoise. Results are usually quite good without any need for further tricks

[pulsar sphinx]

Hello everyone, could you help me identify a style or Lora that this person uses to generate images?

https://x.com/hypetorig?lang=es

[tall bolt]

coincidence but i joined to ask about outpainting too, do you know why mine looks like this? using flux fill

#🏞｜general-with-images message