💬｜general-chat

[still glacier]

Not really the place for paid stuff as we can't monitor what s going on in DMs and such. And we do to the scale of this server there is a bunch of bots ready to prey on users. I would rather recommend fiver on any other platforms with some sort of insurances for paid work.

[atomic mortar]

looking for work people hate me

[still glacier]

For instance /\ (EDIT : woops they re gone... strange 🤔) .... new account, no messages, have claimed pretty much every role available in this server, that s a very sus account. (But please prove me wrong)

[brittle slate]

[atomic mortar]

another one damn, bots be abound

[exotic maple]

hei

[atomic mortar]

hai

[fervent thunder]

finna gonna haw

[fast cargo]

I have a 1660 super but it's not mine.... I borrowed it. Anyway, it's only 6gb and probably won't be able to do much?

[atomic mortar]

SD1.5 will be slow

[fast cargo]

I figured that. What is the better card to pick for it? 3090 or 5070 ti?

[atomic mortar]

more vram the better i say

3090 is showing its age

[undone dock]

does we can generate image in this server of stable diffusion?

how?

can you explain the general approach to do that?

[still glacier]

No they can't because they are scammer trying to lure you into their DMs

To generate stuff in this server you'll have to use the #artisan-1 / 2 / 3 channels. But it's a paid service, check #artisan-faq for more informations. Most people here generate their pictures/audio/videos locally or using trusted online services.

[twin moth]

Hello

[fast cargo]

Yeah, I thought maybe a used 5070 Ti - get at least 16gb and warranty left - although I don't know how that applies to used gpus nowadays. But, 3090s will have no warranty and could be mining cards - plus, I read that the 50 series has a lot of AI features now

[atomic mortar]

Used? They just released

Yeah blackwell is a nice architecture now that more programs are updating

[fast cargo]

Maybe I should buy it new? Depends on how much $ I'll have

There's quite a few that go on the used market. Dunno why

[atomic mortar]

It really depends, are you gonna use it to have fun and use it for other stuff or just ai

[fast cargo]

Sellers need money. Or upgrading? No idea

I have some friends who are gamers but I don't think I have time to play games. Gonna use it for AI/ ComfyUI and video editing

[atomic mortar]

If just ai, i personally cant recommend it

[fast cargo]

Why?

I'm reluctant to go with a 3090 because of what I mentioned

It's good for the 24gb but....

There's no other cards out there that would be as good unless I can come up with an additional $500 for a used 4090

[atomic mortar]

And if you just use it for editing and ai, its probably not worth the value and can go with a cloud service instead

[fast cargo]

Oh? Then what card to get?

How much is the could service? Is that to make up for the lack of vram?

[atomic mortar]

Cloud services have upto 90gb+

Just pay per hour really but that's not my field

[drowsy void]

Looking for an expert in realistic AI character creation (same face across multiple images). Need very natural, imperfect, real-life phone selfie style (not model, not studio). 20-50 images. Paid work. Please show examples of same character.

Looking for high quality work, budget flexible.

[fiery ore]

You can send a PM
I would love to show my recent project

[twin moth]

Comfy UI is the tools I am using in generation

[crystal edge]

what the best ai currently for artistic generation with a lot of details and creativity

[finite magnet]

pretty much all of them are excellent, the creativity comes from you directly with prompts

nano banana 2 pro is certainly the best, i use that to generate 30-40 consistent images for training lora’s on my characters

[finite magnet]

It’s easy, definitely learn how to do it yourself

[twin moth]

Hello

[wanton garnet]

Do anybody know how to train a lora for a custom Z-Image-Turbo model? I want to train a lora for Z epic realism model but Osteris AI Toolkit will only accept diffusers format and they dont have that available anywhere online. I tried making my lora using the orginial ZiT model but it doesnt work well with the custom models I want to use....

[junior urchin]

Did you ask osteris?

[wanton garnet]

I could'nt find any socials for them

[void forum]

Considering wanting to retrain some of my pics in a new embedding file. How can I do that on a 5000 series Nvidia card? With the card I have now, I can no longer use Automatic1111

[unique ravine]

Any comfy megaminds here?

[hazy spear]

i need help to find this lora: 2old_Anne_Droid_3.4 https://civitai.com/models/697146/2oldannedroid34 https://civarchive.com/models/696802?modelVersionId=779744 i cant find where to download it. if anyone got a copy please share

[atomic mortar]

Onetrainer?

[atomic mortar]

A1111 is severely outdated and i recommend you forget that software
In #🤝｜tech-support pins, theres a install guide for forgeneo (similar UI to a1111)
Or swarmUI (even more modern )

[atomic mortar]

And embedding, i think OneTrainer or koyaSS(more difficult imo) can do it

But embeddings are not needed anymore really since SD 1.5

[shut otter]

Anyone know how can I use a custom checkpoint such as illustrious for WAN image to video in comfyui? Would love an example workflow

[atomic mortar]

yes, you first make a image using illustrious and then use the output image as a input

if you want to do it on one go

1. its slow
2. you need more vram to load both models or you need to load > unload > load

[still glacier]

#✍🏼｜rules-and-tos
rule 2, No NSFW Content

[atomic mortar]

He's asking me on how to make deepfakes of his step sister

[untold thunder]

anyone getting this error in a1111 using reactor? LoadLibrary failed with error 126

[shut otter]

I dont mind generating it seperately. I just want wan to retain the style of illustrious model when i convert the image to video

[sick juniper]

Any good recommendations for anime models? I use stability matrix

[mild forum]

can anyone help me what is the best image to prompt generator

[mild forum]

yes please help

everything nsfw and sfw

[atomic mortar]

bot account probably, becareful of scams

and payment requests & downloads

[mild forum]

ohh

ok man

but i need help

[atomic mortar]

you said prompt generator

like

you want random words?

[mild forum]

no i mean image to prompt

i want generate same image with my loras face

[atomic mortar]

you dont have the metadata anymore?

or you mean image to image?

[mild forum]

no i want to generate someone elses image

imaginge a image i found from instagram

i want the same image with my lora

same outfit same pose same expresion same bg

[atomic mortar]

ah you just want to use flux klein 9b for that

[mild forum]

but in place of the person i want my lora

[atomic mortar]

"replace the man in image 1 with the man in image 2. keep the expressions, clothes and background the same"

no lora needed for that one just a second image

[mild forum]

in stable diffusion ?

[atomic mortar]

in a webui that supports flux klein 9b

[unique sage]

When I want something more detailed should I use detailed "thing" or ultra-detailed "thing"?
if ultra-detailed, is it ultra-detailed or ultra detailed?

Fraud warning
This site may be hacked or fraudulent. Giving it passwords or personal information could put you at risk for identity theft or financial fraud.
ok...

[atomic mortar]

Was it a bot that spoke to you

Never follow random links

The trusted links in this discord are in the pinned messages of #🤝｜tech-support for installs and civitAI (google it) for prompts and models
If your more experienced there are more but with these you get most of your stuff done fine and safely

[unique sage]

ye i already talked to CS1o, but sadly he's got no admin rights, and so this bot has been roaming this server for 5weeks+ now

[twin moth]

Image for prompt generator?

[mild forum]

yes got any good ones?

[twin moth]

Wdym?
Is it image to image you're talking about ?

[mild forum]

no man look i find a image on instagram i want to recreate exact same image with my lora

i want a ai which can write prompt for me to recreate the exact same image but with my lora

[atomic mortar]

what you want is impossible

[pseudo bough]

Does anybody here have an Intel GPU? I'm working on translating pie torch into opencl with Intel compatibility if anyone wants to try it

It's not anywhere near done yet but I'm working on it

Pytorch

Voice text

[void forum]

Just wondering, it is a small problem but is there a way to disable zooming on img2img on Forge Neo? Reason why is because when I scroll down and yet if highlighting the picture with my mouse, i will stop scrolling and either zoom in/out of the image

[pseudo bough]

I'm sick of being stuck on cuda

Random question, if you're a senior full stack and AI engineer, do you have an Intel GPU I can experiment with you with my py torch for Intel

[still glacier]

yeah those are scammers, don't fall for their dms

[dim matrix]

hi, nice to meet you.
i am hamada and from japan.
i am looking for someone to collaborate with me.
if u interested in my suggestion, plz send me DM.

[pseudo bough]

Haha dang sophisticated I would never fall for the dm crap though

[junior ember]

Intel has a developer cloud where you can rent gpus. Idk how publicly available it is tho

[pseudo bough]

Thanks, I'll try to get on that

Imagine if I can get py torch to work that new GPU would have a whole new f'in meaning

[junior urchin]

probably still gonna be slower than amd's

[high ruin]

What are the best models for image upscaling lower quality images and mid quality image??

[twin moth]

💯

[wise heart]

You can try Wan 2.2

[sick juniper]

So as I am getting used to using SD, what would you recommend for ssd drive, because its used 300 gb on my hard drive

[still glacier]

It would make loading and switching models faster but that s about it.
Unless you re using any kind of offload / partial load optons such as --medvarm --lowvram then it would help a lot.

[still glacier]

oh a new bot

[sick juniper]

Gotcha thank for the info, I will be looking into upgrading my nvme ssd, its 2tb at the moment, going to see if i can get a affordable 4tb drive, since the prices skyrocketed.

[still glacier]

yeah not a good time to upgrade and prices aint gonna decrease for a while probably

more power to you

[sick juniper]

I haven't tired this method yet but will it make a difference if I put the models on a 4tb external ssd?

[still glacier]

depends how you connect the ssd.

Is the ssd using a bad quality m.1 / sata -> usb adapter

is the usb an usb2, usb3, superspeed... it can get really confusing.

[sick juniper]

Its usb 3

I have to check again for the specs for it, currently at work at the moment

[still glacier]

worth a try then

[icy wadi]

Hi everyone, I'm looking for AI image creator. We can discuss payment. If you're interested, contact me

[still glacier]

not really the best place for that. Beware of scammers in DMs. I would rather redirect you to fivers and other websites with insurance / protection.

[icy wadi]

Thanks

[fiery ore]

You can send a PM
I would love to show you my recent project and also create a sample if you don't mind

[mortal haven]

@icy wadi I can help you with that and we can also work through Fiverr

[compact pendant]

Are there anyone has SD v1.5 quantized

[atomic mortar]

That's a wild concept, its probably on huggingface but why??

Why is everyone so eager to always instantly dm

[coarse chasm]

Who knows how to do nsfw locally on pc? Message me

[still glacier]

1/ no nsfw in this server, rule 2 . #✍🏼｜rules-and-tos
2/ Those accounts who answered were scam accounts. You've been warned.

[fast cargo]

Anyone using a 5070 ti for their gpu? If I combine that with 64gb of ram, is that a good combo?

[junior urchin]

Disabling DMs in this server can save one's mentality

[twin moth]

Hello 🙇‍♂️

[spring niche]

Hello

[quartz shard]

yes

[coarse chasm]

My bad

[sand plank]

Im about to purchase my PC. Will have 5060ti 16gb ,64gb RAM, and 1 or 2tb NVME.

If you have this setup, would please provide an example video what you where able to produce.

Trying to get a realistic idea of what the card can make.

[balmy hemlock]

Hello guy’s ! Anyone know a solid alternative to runpod ( huge shortages of GPU currently ) for run serverless endpoint with ComfYUI ? Thanks

[queen ocean]

hi

[hard haven]

Guys I am a rookie, and I have a question. I am using SD Forge, and for some reason the images generated every time have the exact same face. I am using juggernautXL_ragnarokBy.safetensors and the seed is set to -1, maybe because I was trying to use ControlNet? But I turn it off later, can any expert tell me what could be the problem and how could I fix it? Because I used to be able to generate random faces.

[junior urchin]

@still glacier open ticket scam from @hot pecan

[atomic mortar]

I do have dm requests, sometimes someone asks for something specific or asks for a few specific images and i dont mind those

I do hate the ones who ask for stuff like deepfakes etc

[junior urchin]

Something specific... Like begging for money?

[atomic mortar]

Lmao nah, like instance specific like how to apply a control net in the way they need it or a custom workflow help etc

Rarely do it though

[junior urchin]

Control net for face swap probably then

[atomic mortar]

I'm not helping on those

[twin moth]

hello 😎

[foggy fog]

Hey everyone! I’m Maxime, from France. Glad to be here and excited to join the community. I’m looking forward to chatting with you all, learning, and sharing ideas.

[stoic dust]

Hallo to everyone

[modern magnet]

@visual parrot

@warm junco @still glacier

[junior urchin]

What do you want to spam

[vocal spruce]

built a little tool called PromptForge - tired of staring at a blank prompt box every time I switch models

it's a block-based prompt builder. fill in subject, style, mood, lighting, camera etc. it assembles the right prompt for whatever model you're using. flux, sdxl, midjourney, nano banana, veo 3, wan, elevenlabs and more

weight syntax handled automatically (word:1.4) for sdxl, word::2 for mj, clean text for flux/veo
token budget bar so you know when you're over
drag to reorder blocks (warns you if the order doesn't match what the model wants, with a one-click reset(
preset system so you can save your whole setup and reload it instantly

free, open source (commons clause - free for personal use)

https://github.com/daGonen/promptforge

hope someone finds it useful.

[fast cargo]

really? That's good news. I'd rather get a 5070 ti than a used 3090 that might have been used for mining, doesn't have warrany anymore, is probably losing AI features - I read that Blackwell cards have some optimization features in SD /ComfyUI? Also, a new 5070 Ti isn't that much more than a used 3090

[modern pagoda]

🙃

[atomic mortar]

Scam

[latent wasp]

Hi everyone Im kartik verma from India

[latent wasp]

Im exploring nsfw

Actually im currently exploring this side i dont have much experience in this maybe a quick guide will be helpful

[quartz shard]

5070 ti is faster

but may cause some offloading with smaller vram

I use invoke AI to run simple image models so 5070 ti is faster

blackwell superiority is real

[still glacier]

rule 2, no nsfw here. #✍🏼｜rules-and-tos

[atomic mortar]

Lots of nsfw requests these days

Pick your poison, gooner or spam bots

[still glacier]

why not both

... they're often ... both

[fast cargo]

Good to know, thanks. By offloading - do you mean when the vram is insufficient, it will use system ram?

[verbal ferry]

Hello, I apologize in advance if I write unclearly. I am using a translator. I encountered a phenomenon that is unclear to me: in attempting to mix several characters and styles at the same time, I got a good result. But I discovered that I made a mistake in the prompt. In the Lora weight, I indicated 0,8 instead of 0.8. When I correct it, the neural network produces a blurred, torn, and distorted image.

[cerulean umbra]

Hello!

[verbal ferry]

Is there anyone here from Russia?

[atomic mortar]

There's many, I'm not however but from what i saw in your post, this shouldn't be the case at all

0.8 and 0,8 should basically be the same but it depends on the platform you use, i think you also changed a different setting without knowing

[verbal ferry]

Could you tell me where it would be appropriate to send a screenshot from the console?

[atomic mortar]

#🤝｜tech-support and see if anyone can help.

Do not fall for scams that will try to private message you

[verbal ferry]

thank you.

[verbal ferry]

Would it be appropriate, along with a screenshot of the console, to also send the censored result in 2 versions?

[atomic mortar]

I would make a sfw version of the image

[earnest plaza]

Any India-based LoRA trainers here working on character consistency?

[fervent thunder]

Ayo what is this server dawg

[earnest plaza]

Looking for 4 LoRA trainers to build consistent mythological characters (film project). Paid test. DM portfolio. Mumbai based. Work from office.

[modern pagoda]

Generative AI Art

[high ruin]

I am looking for a decent image upscaler what is the best alternative to Upscayl Topaz Gigapixel and ChaiNNer??

[elfin rover]

Hello @everyone!

Looking to create realistic, high quality AI content for Fanvu, Instagram, OnlyFans or TikTok? I can help with ComfyUI workflows and LoRA training for both images and videos, NSFW or SFW.

📩 DM me to get professional, consistent and natural AI content today!

[atomic mortar]

Scams as usual

[fallen hearth]

Hello @ everyone !

looking forward to create a realistic , content

[ruby vigil]

do you want to create images with videos or lora training?

[unborn hedge]

Every day I look in here and there's a billion scam bots or ads ngl

[sick juniper]

What does lora do?

[twin moth]

Train model for consistent face, skin realism and different angles

[coral drum]

From where do i start with models/lora's?

[atomic mortar]

like starting from 0?

[coral drum]

yeah

[atomic mortar]

hmmm first question: what gpu do you have?

[coral drum]

i tried downloading pony but the images are bad asf

[coral drum]

rtx 4060ti

[atomic mortar]

oh solid nice. yeah Pony models arent the greatest imo due the prompting style

[coral drum]

it worked 2 years ago, although 1-2 min per img

[atomic mortar]

illustrious is more upto date

it really depends on your settings & model after but do you still have a UI installed or got nothing rn?

[coral drum]

i got the UI installed yeah

[atomic mortar]

The? What one? A1111?

[coral drum]

ye

[atomic mortar]

i really dont recommend that one, its out of date

in #🤝｜tech-support pinned messages theres newer ones like Forge, swarm and comfy. Personally i do recommend SwarmUI

[coral drum]

ohhh ok

[atomic mortar]

do you have the 8gb version or the 16gb version gpu?

[coral drum]

shiii

8gb

only

😭

[atomic mortar]

ahh still fine, just gonna take some more time

on my 3070ti it took like 30s per image with a few loras?

[coral drum]

ohh ok

theres a way to make it more efficient isnt there?

[atomic mortar]

hmm yesnt

theres turbo models, less steps and cfg

[coral drum]

but it turns out worse

[atomic mortar]

but ultimately your still bound by vram

[coral drum]

makes sense

[atomic mortar]

for a small model like SDXL quantizing it doesnt make sense

with flux etc it does make more sense to quantize it but

yeah realism

[coral drum]

oh also i will want to generate only anime-style

[atomic mortar]

yeah then illustrious is the way to go really

well if you do switch UI's to forge or swarm
forge: i recommend asking Cs1o for some recommended tips on what to use in #🤝｜tech-support
SwarmUI:
i can help you get started but theres a lot of documentation on the github too and the swarm discord a alot more active

for prompting / loras/settings yeah i can give some tips

[coral drum]

illustrious?

[atomic mortar]

yeah, if i do get automodded for posting links LOL ill just dm it then (civitai only)

https://civitai.com/images/124601387

oh its fine

this uses a popular model nova OrangeXL but "Illustrious" is a finetune of SDXL based on danbooru images

same with pony but without the tagging mistakes

[coral drum]

ohh ok

[atomic mortar]

for most stuff: loras arent really a must unless its a niche character
personally with swarm i use a autocomplete feature to see if i need a lora or not for them

hop into #🏞｜general-with-images ? cant post images here

[eager kraken]

Yeah that’s true. For most cases, LoRAs aren’t really necessary unless you’re trying to recreate a very specific niche character or style. I also like using Swarm’s autocomplete to check if the base model already understands the character before adding a LoRA

[unique ravine]

install studio drivers if you havent

[drowsy crown]

Hoping you all have seen the goonies movie, I just watched it again and they cut the octopus scene! I heard the original scene did not check enough boxes in the "breakdancing octopus" description so at some point they cut it out, even though 'data' references it.. I think we should make an ai cg replacement scene 🙃

[glossy hornet]

Anybody have a good resource for a complete newb to all of this? I’m so interested in learning.

[elfin tapir]

Ask an AI to guide you, around town. Or download comfyui and load from the templates

[atomic mortar]

Asking ai is a horrible way to get lost but :
Pinned messages in #🤝｜tech-support and civitAI to download a model like illustrious/sdxl (for smaller gpus with 6-8gb? Vram) and if you have a better gpu then theres more available

[glossy hornet]

Yea running a 5080 64gb (i was a flight simmer lol) thanks!

[atomic mortar]

ahh similar to me then

honestly the best getting started guide i can recommend is SwarmUI (tech support pins as a link)
but the guide/model support page also mostly works for other UI's like forgeneo
https://github.com/mcmonkeyprojects/SwarmUI/blob/master/docs/Model Support.md @glossy hornet
civitai has a tons of models and images to take a sneak peak from for prompts & settings

big spam&scam

[glossy hornet]

Thanks!

[proven badge]

Oh no, somebody’s getting fired at Anthropic someone leaked the source code for Claude AI and the person who leaked it changed the source code to be working in Python, which is actually pretty smart because they can’t actually get a lawsuit

[long owl]

does **comfyui ** has server like this ?

[atomic mortar]

Id check the comfyUI github

[long owl]

tysm

[onyx escarp]

Yeah

[onyx escarp]

Check comfy.org you're going to see it in their footer

[glacial whale]

Dont suppose anyone is around who would be interested in waffling about their experiences getting SD working in python and or why or how worth comfyui is, ideally in VC 🤷‍♂️

[ivory jungle]

waffling?

[glacial whale]

Yeah I know some people get passionate about the stuff they do, like to share, often dont have people to share with and I need to work with SD in python for a project (WITH a whole bunch of widely different ais) Last time I touched SD I kept it 100% gui and remember it taking me like 6 hours to set it all up. Just not trying for that kinda headache, dont know much about it either so would be nice to pick up the lingo a bit and see what cool stuff people are working on

[brave widget]

any manga workflow comfyui with IP-adapter?

[glacial whale]

Right now im just testing the waters with diffusers

[ivory jungle]

wait you got stable diffusion to work?

im stuck on trying to find a repo that keeps comign back as invalid

[glacial whale]

Yeah ikr, i even got it to output imgs and then rapidly realised that this is NOT going to help with 2.5d isometric buildings (for a game I worked on breifly ages ago) I mean I used vector grahpics for everything so I was just like its not worth the headache to get it to actually work lol

[ivory jungle]

are you a bot?

[glacial whale]

urm, lol people tell I sound like ai alot but no

Probs all the fuckin rittlin im taking

xD

and all the talking to chatgpt

also what sorta llm asks to switch to VC, lol

[shell yarrow]

Hey everyone
I’m an AI & Full-Stack Engineer focused on building production-ready AI systems, not just prototypes. Most of my work is around connecting LLMs with real infrastructure — APIs, databases, tools, and business logic — so AI can actually run reliably in real workflows.
I usually work with things like:
LLM systems & orchestration (DSPy, LangChain, AutoGen, CrewAI, ReAct), RAG pipelines with vector databases and custom retrieval
Multi-agent systems with planning and tool use, Multimodal AI (Whisper, CLIP, YOLOv8, TTS), AI image / video generation pipelines
Backend & full-stack (FastAPI, NestJS, Next.js, React), Automation & integrations (n8n, Zapier, Make, custom APIs)

I enjoy working on real AI products that ship to production, especially systems that combine multiple models, tools, and automation.
Not here to promote anything — just happy to share ideas, help out, and collaborate with other builders in the community 🤝

[quartz shard]

Invoke better

Invoke is goated

[atomic mortar]

Lmao

[quartz shard]

How to add text?

It's just shitty with my models

[atomic mortar]

Here's the neat part, you don't
Most text after 3 words is a mess

Generate text and fix it in Photoshop

[proven badge]

Interesting this must’ve been their plan there whole entire time I told u all for years this would happen..
read it and there’s one thing in it that is concerning age verification for operating systems like your phone or your computer everything

[vague crow]

i have RTX 3050 4GB GPU, what can I try guys?

[atomic mortar]

Copium and sd1.5 im afraid

Sdxl will be slow too by offloading to ram

[dusty lodge]

hello

[atomic mortar]

Hii

[velvet nacelle]

Hello

[copper finch]

What methods are available to generate pixel images?
The size is approximately 16x16

[prisma horizon]

with comfy workflow optimization, low step and quantozation, you can gen 1020x1024 in a min or 2, even at a bigger res one more or 2 minutes

[proven shard]

I have some idea how to make consistent characters in Z-Image Turbo using character sheets. But how can I make consistent sets? E.g., a professor standing in his office doorway, and then sitting behind the desk in the same office.

[unique sage]

anyone here knowledgable about making loras? i need some help

[silent sonnet]

Yo are any of your guys interested in quantitive trading

[cursive thorn]

How much does it usually cost you to rent a 4090 for an hour?

[atomic mortar]

I recommend checking on gpu renting sites as it varies
I'm not too involved in it but runprod/vast ai has some

But i cant guarantee how good they are

[reef flicker]

Hi everyone! 👋

I'm working on a project called AIA Festival – a workshop event for AI video creators (screenings, hackathons, meetups between creators and tool makers).

Before moving forward, I really need your input to shape the concept.

👉 1-minute survey: https://forms.gle/N7h9bE7TNWbnXhcz7

Thank you so much for your help! 🙏

[burnt geyser]

hello humans

whats up

i wanna try and run some local AI image gen on my M5 Pro

how would i go about that

[prisma horizon]

ask grok, but looks like mac it's pretty good to launch big models

[atomic mortar]

Im not entirely sure about mac but in #🤝｜tech-support theres a guide for UI installations

[prisma horizon]

from what grok says, it sound pretty promising to launch big stuff

[warm junco]

Try out DrawThings, its for MAC and IOS
https://drawthings.ai/

[burnt geyser]

nice

[vernal nebula]

@vapid dove

u there?

[twin moth]

Hello here 🤠

[pseudo bough]

Is anybody interested in a super easy lora trainer for LTX 2.3

I'm working on it right now. I've got it to where you upload a video. It cuts a video on proper chunks and then captions it automatically and then runs the text encoding and then does the training all in one program

No more captioning no more slicing of videos for training

Right now I'm confirming everything is working all the way up to text encoding. I haven't got to the training part yet because I've been debugging it for like 10 hours at least

Actually started making a program previously to this where it takes video and slices it up in the chunks for Gemma 4 to watch and write up an article on the video and then make a highlight video out of it and then I realized I could use that same tech to do Auto captions. Then I have a whole lora thing that I'm working on

[vapid dove]

whats up?

[fast cargo]

Can I ask about hardware here?

[low moon]

Yes doggie but the answers will be very similar. 24+ VRAM you are good, 12-24 ok I guess. anything less than 12 ouch

[remote mantle]

anyone know pls

[atomic mortar]

Know what?

[vernal nebula]

id like to report someone

can u check dms?

@still glacier

do u have ban permissions?

[brittle rose]

@vapid dove

delete

[reef citrus]

Can stable diffusion do 3D yet?

[still glacier]

been capable of it for a while https://stability.ai/stable-3d

[candid salmon]

Does anyone need a 3d artist ?

[reef citrus]

thanks, its mostly for footwear i am curious about. I have used the free huggingface a couple of times and love it's ability to accurately capture detail, but it would be amazing to be able to feed it multiple angles. Vizcom is interesting too and you can feed it multiple angles, but it mostly ignores them i think and focuses on one.

[sly eagle]

Can stable diffusion makes animation yet?

[atomic mortar]

theres other companies that released video models. stable video from years ago isnt the best atm

[vapid dove]

yeah, DM me

[tame imp]

Hey all! I’m a founder scaling an AI/ML startup—would love to connect with recruiters and folks in the hiring space.

[atomic mortar]

No recruiters here other then scammers

[marsh bramble]

G

[knotty prairie]

hello people of the world

hope you're having a great day. If not: Please have one, ok?

😄

[west kindle]

Looking for workflow for AI-generated comics (multiple panels, consistent characters across scenes, same style).

Ideally with some automation (prompt → generate → evaluate → iterate).

Has anyone here built something similar or experimented with this?

[tame imp]

hi guys

[shell yarrow]

Hey everyone
I’m an AI & Full Stack Engineer focused on building production ready AI systems, not just prototypes. Most of my work is around connecting LLMs with real infrastructure APIs, databases, tools, and business logic so AI can actually run reliably in real workflows.
I usually work with things like:
LLM systems & orchestration (DSPy, LangChain, AutoGen, CrewAI, ReAct), RAG pipelines with vector databases and custom retrieval
Multi-agent systems with planning and tool use, Multimodal AI (Whisper, CLIP, YOLOv8, TTS), AI image / video generation pipelines
Backend & full-stack (FastAPI, NestJS, Next.js, React), Automation & integrations (n8n, Zapier, Make, custom APIs)

[real rivet]

hello

anyone good at building sdxl loras? need some tips as I'm trying to build some loras locally

[atomic mortar]

I recommend joining the one trainer discord for that one

Its on their GitHub

[real rivet]

oh thank you, where can i find the invite link?

[atomic mortar]

I'll dm you it instead, im in it too

[sly eagle]

I know but I wanted something personal you know those are so expensive hahaha

[vivid badger]

I made the project which can generate the AI short movie from prompt or image.

I used the KIE.ai.

[untold thunder]

which model/workflow has the best character consistency and least identity drift for i2v? I heard about some flows maybe using ip-adapter to keep injecting reference image, but dont know if that is a thing. I experience identity drift even in 5 second clips and it's ruining to authenticity. What do you recommend?

[vivid badger]

I can feel your pain on the identity drift—it’s the biggest vibe-killer for I2V right now! 5 seconds is usually where most models start to 'hallucinate' a new person.

[ancient mauve]

does anyone here use qwen with forge?

[atomic mortar]

Expensive 🥀 also no advertising

[wind crystal]

why？

[atomic mortar]

The #✍🏼｜rules-and-tos rules #5

[wind crystal]

im sory

[fervent thunder]

Wsp

[remote bay]

Hi guys, I'm currently learning ComfyUI, FLUX, and other models, and would love your support along the way. I'll definitely support others when I'm capable. Thanks.

[timid sable]

hello..

[atomic mortar]

Hi

[ashen berry]

🚀 Hey folks! I’m Stiven, launching NeuroBot Labs — we’re building an AI agent that digests business info and actually gets stuff done.

Looking for AI engineers in the US who love LLMs and building real products, and recruiters who can find top AI talent.

Early-stage, fast-moving, high-impact — if you like messy 0→1 projects, DM me or tell me what you’re working on. Let’s make something amazing together! 💡

[grand fossil]

test

[atomic mortar]

Testing

[grand fossil]

yo

can i ask u some basic questions

does this work

oh it doesnt :/

sad

r
r

:q

:wq

[atomic mortar]

You can

I was busy ish

[vestal chasm]

🔒 Message hidden
-# Discord now requires ID verification in order to see certain messages.

Oh heck nah

[still glacier]

not a thing (yet)

[untold thunder]

but i see clips longer than that that have perfect consistancy and they dont even use a character lora, so how is it done?

[vivid badger]

ahh, as my experience, to generate long AI movie, we have to pay $100~$150 for api.

So the short AI movie is prefered.

[fierce prism]

can someone assist me with an image im trying to generate please? dm me if you can

[deft topaz]

hello

I'm looking for a long term partnership

[sonic shard]

is there no longer a bot I can use to generate stuff for free?

[atomic mortar]

Stuff costs money, civit ai gives you some free daily credits to generate you can use to make stuff. You can also save it up and earn more by posting the gens. No nsfw tho

[remote bay]

Hey, just wanted to know if someone know Comfy and is in Jaipur, India? I just started out a couple of days ago, I can't afford a PC and before renting GPU online (cloud) I want to learn (practically by experimenting), so as too I'm not wasting money there. Anyone in the Jaipur, please let me know and if anyone has a solution/suggestion as to how I should I proceed please recommend your opinions. I'm very grateful to all of you. Thanks.

[signal glade]

Hi I m AI/ML Engineer and new to this server!

[lean moth]

where do y'all store the characters you create?

[still glacier]

beware of dm scammers

[slow canyon]

hey guys, i was looking for an llm under 24gb vram, what i need is that i'm seeing images or videos that i really like, but when i try to recreate similar things to them i mostly fail, so i need a model-llm(hopefully can work on kobold) that will give me the prompts for me to create that videos and images, thanks in advance

[onyx escarp]

Hello @deft topaz we can connect

[inland bear]

Hey guys, i need some help with my prompts to enchance my Ai Model can someone help me out?

[atomic mortar]

beware of dm scams, any help offered outside the discord should be treated as suspicious!

[atomic mortar]

vague question, shoot

[brittle slate]

I made an image laudry a while ago, qwen3-vl:8b did exactly that. For the image part you want a model with a good prompt following and large token count. I also tried the 32b variant, but that is not really needed if you work on the prompt for the vision model. The larger model just obeys orders far better like do not comment, do not use titles

[inland bear]

Someone want privat chat? i think its a litte to deep to discuss here. need help with expicit nsfw promting..

Just feel free to add me!

[foggy ice]

The thing that everyone does but no one admits to... 😆

Anyone know any good resources for training models? I used to play with SD, but havent in a long while. Now Im getting back into it, but have great hardare to use..

[tawny portal]

[valid jackal]

Anyone using AI image generation in their video projects? What tool?
Post:
Curious what tools editors are actually using for AI generated visuals — backgrounds, thumbnails, concept frames etc.
Is the complexity of tools like ComfyUI putting you off? Or have you found something simpler that still gives good results?

[upper steeple]

hey im new here and im interested learning ai can anyone helpout there how to make out scrpits real quick and accurate

[atomic mortar]

Be wary of dm scams

[upper steeple]

let me learn from the scratch

[upper steeple]

will take care of that

[peak cradle]

hay everyone
hope you're having a great generating time lol

i need the best free anime models can you make me a list with the links or the names and i will try to find them

thanks everyone

i am using stablr diffusion to generate photos right now and all of the anime filter that i have found on citivia is bad my prompting went over 200 char and still bad sharpness bad shapes and bad eyes alsi the negative prompt are full also lol

[inland bear]

https://civitaiarchive.com/

[peak cradle]

arigatoooo

[peak cradle]

ValueError: Failed to recognize model type!

tried 3 diff models

[elfin tapir]

Older vanilla stable diffusion approach will use vanilla CLIP, out of the box that only supports 77 tokens(150-200 chars), but comfyui or whatever you use likely does some modifications and stitching to still get the whole prompt to the mode, but in a way that can cause issues.

Something newer, say the Anima-preview, or flux, or whatever, will instead of CLIP use an actual LLM bolted in there, so their prompt understanding becomes magnitudes better and supports much longer prompt. But even if their prompt understanding is better the model still need to be properly trained and tuned to produce good image output.

You said you want anime, I recommend to use comfyui, and get the "Anima-preview" t2i(text to image) template, zero setup, much better prompt adherence. A bit of jankiness as it's still a preview model

[inland bear]

you can just download form there

[last summit]

Hey—quick question for anyone here 👋

Is anyone comfortable talking with clients in English and joining calls?

I’m currently working on freelance projects and looking for someone to team up with long-term. I handle all the technical work (development, delivery, etc.), but I could really use someone reliable on the communication side—joining meetings, talking things through with clients, that kind of thing.

You don’t need to code at all—just be clear, dependable, and good at conversations.

I’m serious about building something consistent over time, not just a one-off thing.

If you think this might suit you (or you’re even just a bit curious), feel free to DM me. Happy to explain more and see if it’s a good fit 🙂

[onyx escarp]

Hey am interested

[bronze apex]

Is there any place in here to commission artists/ai experts

If you are good with realism (U can use AI or art) and want to do a photo edit commission de- ai ing a realism image then shoot me a DM

[desert dagger]

most of us are.

[dense osprey]

can anyone guide how to ad the image generation bot to my server??

yes please

[atomic mortar]

Dont fall for scams in dm

We dont have such a bot here, theres only #artisan-faq

[still glacier]

Likely a scam and maybe even one of those very dangerous one where they fly you to a callcenter, confiscate your ID and then force you to scam others to reclaim your freedom.

[atomic mortar]

Why not talk to people here

Interesting reaction of being called out you delete the message @elfin rover

[still glacier]

yeah @elfin rover if you could stop trying to bait people into dm that would be great. Thanks

[shadow ingot]

Hello guys, what do people use to expand an image nowadays in forge?
Openoutpaint is still in use or are there other options?

[compact perch]

Sorry for being unrelated to image generation, but what is the word on when gemma 4 will be released locally? I think i saw it already been released but on huggingface everything has the tag "preparing for release"

Google is shit teir for answering this question.

[atomic mortar]

oh gemma 4 the LLM is already released

[brittle slate]

26B-A4B-Q6_K

[fervent thunder]

https://nexus-evolve-worlds.base44.app

I have made a new Pokémon style game

[bronze apex]

Can someone give me an actual discord for commissioning AI/Artists. I refuse to believe there is so few people interested in making money from AI editing

[atomic mortar]

Market is over saturated and its too easy to run yourself ngl

Even with cloud tools its cheaper&safer to buy a subscription then to hire someone

What kinda edit do you need?

[atomic mortar]

I love it when i forget to swipe reply, see message above

[bronze apex]

Thats so f****** stupid. The fact that its cheaper should mean 100x more people are willingly to photo edit especially when being allowed to use AI. This is the first time in my life I resonated with boomers saying "Kids these days dont want to work anymore"

This has nothing to do with the economy its just pure laziness

[atomic mortar]

Well the problem with "everyone can do it" is that well the rates of doing it isnt worth while to charge for

[atomic mortar]

Like ai images go for 1-3cents per image

Just not worth the effort to commission a image, better provide a setup or something as its pricier etc

[bronze apex]

Not commissioned images. Do u mean like some ai character genned or something?

I refuse to believe people are doing high detail precision edits for 1 cent an image

[atomic mortar]

Yeah that kinda stuff

Meh you make a lora, get a base prompt going and then use a bunch of wildcards

And you wait

Thats it really

[bronze apex]

Oh im commissioning something a bit different from that. I hope people dont think thats what i mean...

[atomic mortar]

You said edits right? Well if it's inconvenient to mention here feel free to dm but if i do anything for someone its either 1. Free edits/images
2. Known github site links referrals
3. Links to huggingface/civitai

The day i ask for payment is the day i hate my hobby

If you need a lora i got like 3k buzz on civit ai I'm not using so shoot i suppose

[bronze apex]

Well even if u do it for free I'll pay u anyway if it's good. Not sure what u mean by 2. And 3. Though

We'll see I'll dm

[atomic mortar]

Never payment, its against my morals

[long ravine]

سلام

[atomic mortar]

Hii mair

Salam

That's hyper specific, not that i know of no

[long ravine]

Wait lol

Wrong chat 😭

[atomic mortar]

[long ravine]

I was gonna ask in a Islamic server

[atomic mortar]

Got a laugh out of me at least

[long ravine]

Sorry

But as Pakistani

[atomic mortar]

No its fine lol

[long ravine]

We have to deal with a lot them

I use to be in server which was fill of there silly stories like there one sage married a dog recarnation of his partner

[atomic mortar]

There's one ultra religious person here but it's not really the place to discuss these kinds of things tbh

[long ravine]

Yeah that's why i deleted my msg

Or 🧃 will ban me

[atomic mortar]

Rip

[untold thunder]

what approaches are there to "perfectly" generate images of a given character in img2img workflows? i want to generate images for a lora using a character that has already been created.

[onyx escarp]

Use good base images, keep img2img denoise low so the face doesn’t change, and use something like IPAdapter to hold identity. Then train your LoRA with a simple trigger word and keep looping better generations back into the dataset. @untold thunder

[foggy ice]

Man, getting this to work on my AMD SoC is a real treat... lol

[atomic mortar]

Oh amd is annoying, if you didn't have any luck try the guides in #🤝｜tech-support pinned messages

[foggy ice]

Thanks! I switched back to my Nvidia for now... lol

[finite mica]

Anyone with diffusion? Need a small help

[atomic mortar]

Technical questions: #🤝｜tech-support
Prompt questions? Just drop em i suppose

[finite mica]

I have 2 10 sec video and I've tried online AI for face swapping and nothing works

So I was wondering if anyone could me help me out and swap faces n giv me

[atomic mortar]

To clarify:
You want face from video 1 to video 2?

[finite mica]

Face from image to video

The thing is, I created a video online already since I used ai to animate it, the persons face has changed quite a lot and it doesnt look like him anymore. So I just have another picture of him and just put that face on this video, but in the video there's 2 people. Only want the guys face changed

[twin moth]

You can actually get a faceswap video workflow 👌

[finite mica]

I dont have a pc/laptop would that still suffice?

[twin moth]

Not at all
You'll need a pc to run it

[finite mica]

Fahh 💔

[twin moth]

Or you get who can help you with the face swap

[finite mica]

Yeah thats what I was asking

🥲

[compact perch]

Sorry again for asking here, but what could be the problem when a local LLM refuses to finish a task, gives a short response but multiple "regen" makes it breakthrough and finish the task it's given?

[signal glade]

Hi guys rate my website from my profile, any improvement, recommendation etc?

[atomic mortar]

scam as usual

[untold thunder]

hmm alright! do i put input image into the ipadapter too?

[foggy ice]

Sup everyone?

[finite mica]

anyone who got faceswap?

[foggy ice]

Perhaps.. why?

Is that server unstable diffusion still around?

[finite mica]

Could you help me out? I need to faceswap in a 10 second video

Would be really appreciated

[foggy ice]

Dmed you. Depending on what it is, I may do it.

[void adder]

Have you been resolved?

[finite mica]

Nop

[cloud beacon]

hi

8+ years in AI, and one thing is clear: tools change, but fundamentals win.
From building early machine learning models to deploying scalable AI systems, the biggest lesson I've learned is this—understanding data beats chasing trends.
AI isn’t just about models. It’s about solving real problems, responsibly.
#AI #MachineLearning #DataScience #🤝｜tech-support

[cold moth]

I have a 5080 rn, and upgrading from 16gb to 32gb system RAM (ddr4). I was wondering, would upgrading directly to 64GB make any difference for local models (latest image and video models)? My understanding is that the bottleneck is mostly the VRAM.

[atomic mortar]

Yeah unless you offload it won't have big effect, maybe with like a extreme upscale or a new (unoptimzed) model might that normal ram come into play

[brittle slate]

If you only make images, 32 will suffice. With complex WAN stuff, you want 64. And the MoE LLMs have good use for system RAM for that partial offloading.
If you keep the rig >1y i would go 64 right now.

[faint timber]

@cold moth Can your DD4 motherboard even accept 64GB?

[atomic mortar]

and cpu

[deft topaz]

Hello

[wintry gust]

Hi! everyone , if you wanna pdf guides about soft and hark skills ( D M me) .

[atomic mortar]

?? This + that server tag screams of scams

[wintry gust]

not a scam

it's not pushing anyone to for thiss

if you won't so, just ignore itt

It's a simple guide that's in everyone's life, add some good skills ( it's like my digital product).

[atomic mortar]

👍 then by all means lure people into dm's as your litteral second message

[wintry gust]

Whoever is interested will DM, don’t worry.

[deft topaz]

what do you do for living?

[wintry gust]

side hustles.

[atomic mortar]

@still glacier ill leave judgement with you, someone advertising sidehussles and stuff

[deft topaz]

where are you from?

[wintry gust]

INDIA

[deft topaz]

May I ask something?

[wintry gust]

yes offcourse

yes

[deft topaz]

good

do you have it?

[wintry gust]

not yet !

[mortal scaffold]

hello

[tribal solstice]

Hellooo

[deft topaz]

hi

[signal glade]

Hi

[split jolt]

Who for battefield 1 ?

[quartz shard]

IS IT POSSIBLE to get Illustrious models running on an S25 EDGE

[severe wigeon]

Civitai has a remix feature, How do I do that locally?

[dawn mulch]

Anyone here stil using sd 1.5 or sdxl? I just created a lora creator for it. I would like for people to test it and see if you can make loras with it.
https://github.com/dadadies/ddd_Lora_Creator

Its takes roughly 1 hour to make a lora, on my 6GB VRAM GPU and 16GB laptop.

[shell yarrow]

I’m an AI & Full Stack Engineer focused on building production ready AI systems, not just prototypes. Most of my work is around connecting LLMs with real infrastructure APIs, databases, tools, and business logic so AI can actually run reliably in real workflows.
I usually work with things like:
LLM systems & orchestration (DSPy, LangChain, AutoGen, CrewAI, ReAct), RAG pipelines with vector databases and custom retrieval
Multi-agent systems with planning and tool use, Multimodal AI (Whisper, CLIP, YOLOv8, TTS), AI image / video generation pipelines
Backend & full-stack (FastAPI, NestJS, Next.js, React), Automation & integrations (n8n, Zapier, Make, custom APIs)

[olive wolf]

Hi Everyone

My name is Saketh. I am with a $50M early stage fund based out of bangalore, India primarly investing in advanced manufacturing, DeepTech, ClimateTech and other emerging sectors. Please feel free to reach out to me if you are building something interesting.

https://www.linkedin.com/in/marpusaketh/?skipRedirect=true

[quartz shard]

Which model should I use for pixel art sprite generation? Illustrious fine tunes are bad with the pixel art lora

[atomic mortar]

all of the "pixel art models" really, you could use certain upscaling methods to make it pixel perfect though

well downscale and then upscale

[spare jewel]

usure

[atomic mortar]

Why is that a link

[spare jewel]

Idk

Embed

[still glacier]

Because no pics in #💬｜general-chat

[steady atlas]

Report for player

[desert dagger]

that's why #🏞｜general-with-images exists

[soft ice]

How can I reactivate my account?

[atomic mortar]

Stability account? Just email them. Devs arent here

[soft ice]

Oh okay thanks m8

[uneven pecan]

Anyone else suffering from AI FOMO? 😅 This article strikes close to the target: https://www.linkedin.com/pulse/do-you-any-chance-have-railroad-fever-jonas-angermund-hhxif 🚆

[steady atlas]

https://www.instagram.com/reel/DXJz6tuEjXd/?igsh=b3dsc21neXgxeTJi

Ayush bhai jake dekh isko and kuch inspiration le

[quaint quail]

B

[quartz shard]

which

[signal glade]

Hi

[deft topaz]

Hi uys

guys

[vital olive]

What’s good

[ashen bough]

hi

[signal glade]

Hi everyone, I’m an AI/ML engineer passionate about helping others and collaborating on open-source projects. If you share similar interests, feel free to connect.

[fiery dirge]

wheres content

@warm junco @still glacier

@vapid dove

[vapid dove]

post some. be the change you want to see in the world

[inland bear]

Hey everyone,
My friend trying to get ComfyUI running through Stability Matrix on a new AMD build, but he keep running into a showstopper error. Hoping someone here has experience with AMD GPUs and ComfyUI.
System specs:
GPU: Radeon RX 9070 XT 16GB
CPU: Ryzen 9 9950X3D
RAM: 32GB
OS: Windows 11
The problem:
When trying to run any workflow (even a basic txt2img), I get this error:
text
torch.AcceleratorError: CUDA error: device kernel image is invalid
Search for `hipErrorInvalidImage' in ROCm docs
Device-side assertion tracking was not enabled by user.
Full traceback points to an embedding operation failing inside the CLIP model.
What we've tried so far:
Installed ComfyUI via Stability Matrix (latest version)
Reinstalled dependencies
Checked that ROCm/HIP is properly detected (seems to be)
Our suspicion:
The error looks like ComfyUI or PyTorch is still trying to use CUDA instead of ROCm/HIP, or there's a kernel compatibility issue with the 9070 XT and the current ROCm build.
Does anyone have a working setup with an RX 9070 XT and ComfyUI? Do we need to:
Use a specific PyTorch ROCm version (e.g., 6.2 or nightly)?
Manually force HIP device selection?
Patch the CLIP model code?
Any help or pointers would be massively appreciated. We know AMD support is still maturing, but the 9070 XT has 16GB and great potential for SD.
Thanks in advance!

[atomic mortar]

#🤝｜tech-support .
Cs1o has the most experience
Matrix often has issues i don't recommend that
Try using the guide in the pinned messages within #🤝｜tech-support

[queen ocean]

hi

[hard kite]

hi

[steel ridge]

You are on the right track that error isn’t ComfyUI, it’s a ROCm/PyTorch mismatch with the 9070 XT. Most stable ROCm builds don’t fully support newer AMD cards yet, so you get that “kernel image is invalid” error. Make sure you are actually using a ROCm PyTorch build (not CUDA), and try a newer ROCm or nightly PyTorch. If it still happens, it just means the GPU isn’t fully supported yet nothing to patch in ComfyUI.

[spare jewel]

present

[vast iron]

Hello

[sly timber]

Hello all!

I'm trying to find a way to run comfyui (or something like it, please tell me if any competition exist?) as API... I've found some services that are pretty good but I still wanna know if I'm missing something. I think ComfyOnline and ComfyWorkflows are pretty much dead and do things I want them to do. I have tried Kimara.ai and it's pretty good but not yet sure if best. Comfy Cloud is not for me because API and I fear they do some partner node bs and make everything else worse. I don't wanna run on runpod, too difficult.

So.... Any ideas?

[modern pagoda]

Woo is that a fauxCS1o doppleganger 😅😂

[spare jewel]

nanu nana

[steel ridge]

You are not missing much tbh most of those platforms are either dead or limited.

If you want API without stress, go with something like Replicate or fal, they are simple and actually work.

If you really want ComfyUI style, your best bet is still running it yourself and exposing an API everything else is kinda hit or miss right now. @sly timber

[sand adder]

yo
idk if this is the right channel to ask but do yall think it would be plausible to train a lora on the JJK s3 artstyle?
i have trained loras before but this was almost three years ago and on subjects not style transfers so im not too sure if it would be very doable
and i do undersstand alot of it is up to me and finetuning and what looking for
any insight would be appreciated tho

[atomic mortar]

Why wouldn't it be possible? Maybe someone made it already, check civitai

[steel ridge]

Hey @sand adder, Yeah it’s definitely doable. Style LoRAs like that work fine, you just need a clean, consistent dataset and good tagging. JJK S3 has strong lighting and linework so focus on those in your captions. Expect some trial and error but you will get solid results with proper tuning.

[hollow valve]

How realistic is it to train a Lora model for Ernie Image that consistently generates my handwriting? I tested it with a handwriting prompt without Lora, and it worked pretty well. But my handwriting is really ugly

[sly timber]

I like comfy because it gives more control instead of doing full lottery with prompt seeds. I know comfy so comfy itself is not a problem, just the completely ass updating etc on local. I'm probably gonna go with kimara then for now.

[hollow valve]

i will try to train it on ostris ai toolkit or fal ai. i collected around 30 samples of my handwriting. if i get some good results i will post it here

[hollow valve]

yeah sure, thank you

[empty beacon]

hi

can some one please tell where i should add my design

@unreal dock hi

hi

can u see my design and i am new to community can u guide me

my design is in #🏞｜general-with-images

no cannot

i don't know how to dm did u mean insta

@unreal dock i have no idea can u guide

shell i send u a personal request

please accept the request

[atomic mortar]

Its a spambot ignore it

[empty beacon]

@atomic mortar are u sure

[atomic mortar]

Why else would a new account, that joined not too long ago instantly try to get you to a DM when everything you wanted can be discussed in #🏞｜general-with-images

Its trying to single you out, maybe get you to install something or even demand payment

It happens a lot here, community's sorta dying

[empty beacon]

@atomic mortar can u guide me i think u are not a bot

[atomic mortar]

Guide in in what way? The image? Personally i think its a bit chaotic

But it works

[empty beacon]

what more i can do as i am a fresher/learner

[steel ridge]

Hey, if you don't mind i can help you with that.

[atomic mortar]

Hmm you can clarify a bit better what your need

Like prompting?

Model? Settings?

[atomic mortar]

Another one of those "lets dm" guys

Crazy

[empty beacon]

@atomic mortar prompting

[atomic mortar]

What a lot of people do is use a model like gemma3, to refine a prompt. What model are you using to make a image?

ZIT? flux? Sd3?

[empty beacon]

gemini pro and gpt

[atomic mortar]

Ah then it sorta does it for you, ask in one prompt to expand your prompt lets say
"Make a prompt for a image that contains a car, dragon, red landscape and has to be in a card or banner" then use that prompt to make said image

[empty beacon]

no i make base design and modify it

i write prompt by my own

[atomic mortar]

Allright then theres not much you can change really other then describe clearly what you want to see in the image. Be precise

Use art styles, (artist names can help), maybe atmospheric lighting or colorschemes

Try to use textures etc

Anyways, don't fall for the people asking to DM here, 9/10 times a scam

[empty beacon]

@atomic mortar thanks

can i modify the dragon image

or is it useless

[atomic mortar]

Sure why not. Im not too familiar with gemini or gpt but they should be able to

Broken english, all roles, brand new account, asking people to DM
All the signs of a scam bot

[empty beacon]

ok thanks can i upload another design

[atomic mortar]

Otherwise, please do show your repo, notebook & drop your streamer url

[atomic mortar]

Sure go ahead, its a open chat afterall #🏞｜general-with-images

reply.

#🏞｜general-with-images @still glacier got a scammer on our hands

[atomic mortar]

Even after that edit its something incomprehensible

Looks like the bot owner took over

[modern pagoda]

"Has anyone really been far even as decided to you even go want to do look more like?”

My official response to everything now 😂🤣

[atomic mortar]

Lmao appearantly its a afrikan insult , explains why he hopes to strike big here

[bitter venture]

ok after some tme wanted download some new models civitai doesnt have hrentai model policy bs wtf alternatves?

wow even here u cant tipe that word

fk is wrong with u all

[atomic mortar]

No nsfw rule but
civitAI.red has nsfw content. .com has turned SFW only

[bitter venture]

r acount separate?

god hate all ths protect chldren internet, i had 2girls 1 cupo as a 12y and guess waht i am alive

[bitter venture]

ty

[atomic mortar]

Nope, they are shared

Gotta login again tho

[bitter venture]

dunno didnt work ccreated new one wth google acount...

[atomic mortar]

I think its the proper url but just in case check the announcement on civitai.com it self

[bitter venture]

i think u r right n civitai.red, but i find it retarded... that why boughted. like why a generator chose a limited version...

mby m just old and there r other sites?

[atomic mortar]

not any i can recommend no

[bitter venture]

i know its a dumb queston but lest say, how would u go about traning a checkpont?

have programmer knowlage not machle learning

[atomic mortar]

your best off joining discords like onetrainer or koyhaSS for that

[grave stag]

If I have an image with someone drawn but the legs and some of one arm are cut off (intentionally by the way) can AI have the context recognition to make it a full body pic? Or is it gonna struggle?

[gentle bramble]

Sup

[twin trellis]

Bup

[wary island]

Hello

[grave stag]

?

[brittle slate]

Yes, it can add an arm and legs, but at best they will look similar. Tattoos or other personal features are missing.

[ruby vigil]

If you did like I can help you with that and it will look very perfect

[gaunt shore]

Hello

[gaunt shore]

Yeah AI can usually fill in the missing parts and turn it into a full body especially if the pose is clear It might struggle a bit with unusual styles or unclear angles but most of the time it does a decent job

[humble canyon]

I'm sorry if this isn't the proper channel but here we go.
I have been using Gemini to generate some images for my roleplay character. What I have been doing is giving a photo of my character to use as the face and a reference photo to use as the pose, outfit, environment etc. It has been working flawlessly but also denying so many stuff for the smallest things. Is that same workflow possible to do in stable diffusion? What I've mostly seen is hyper realistic stuff being done on SD and what I'm chasing is some sort of high quality 3d render aesthetic. I have pretty much no experience with stable diffusion but if you can just give me a brief "yes, use this x method" or "no, you cant" I would really appreciate it.

[steel ridge]

Yes, you can do it in Stable Diffusion.

Use IP-Adapter/FaceID for the character, ControlNet (OpenPose) for pose and img2img with SDXL + 3D/CG LoRAs for the look.

It’s basically the same workflow as Gemini, just with way more control. @humble canyon

[humble canyon]

Thank you.

[torpid tartan]

speaking o f IP-Adapter/ControlNet - Im trying to generate some images of my anime girl character, very specific in look, to create a Lora later. Im using sdxl model in IP-Adapter (special model for NoobAI checkpoint), yet effects are far from exprectiations,, thats clearly not the same character. I've read IP-Adapter works mostly with realistic characters, is it possible to achieve somehow the same way anime girl? I have some workflows from civitai, maybe I'm doing something wrong?

[steel ridge]

IP-Adapter can work with anime but it’s not very stable compared to realistic faces, especially on SDXL.

For better results:

Use ControlNet (lineart or reference)

Keep reference strength moderate

Don’t overcomplicate the prompt

Stick to one checkpoint and settings

With anime characters, consistency usually comes more from ControlNet + good base model than IP-Adapter alone. @torpid tartan

[torpid tartan]

heh, thought so, thank you kindly :>

[steel ridge]

You are welcome

[steel ridge]

If you don’t mind, I can help you fix it and get consistent results. @torpid tartan

[atomic mortar]

reminder, any scams always happen in DM's!

[regal plume]

Hi, I’m Den. Built a thing. It's a fashion social app where real people try on real brands with AI, post their looks, and our engine hunts down the best price and fit across 3M products - from Nike to Farfetch to Shop Simon. It's early access. Some of it works great, some of it is rough. I'd rather ship it to you now and fix it together than wait six months and hope I got it right. Come tell me what sucks.

https://apps.apple.com/us/app/uverest-ai-fashion-social/id6754376796

[atomic mortar]

I kinda wish it followed rule 5

[regal plume]

😞 sorry

[regal plume]

Just wanted to share our new AR model

[atomic mortar]

not the place for it, too bad sonny

Like i get the enthusiasm but you gotta admit.
A unknown person, posting a random app that they have to install on their phone to have a flux Klein/nanobanana workflow run in i hope your secure cloud environment with a unknown business model

Like do you earn comissions? Is it paid? Do you sell userdata? Many unknowns and frankly by not bothering to read the rules just portrays it as shady.

[grave stag]

Well I do have a half dragon that has scales on part of the body and non human hands so?

[grave stag]

Should I DM you the image?

[ruby vigil]

Yeah

[atomic mortar]

Keep in mind the scofield person will ask for payment

But what you want is very doable, no need for dm's really. Its just inpaint with a proper prompt

[ruby vigil]

Yeah sure

[atomic mortar]

I still think they are a bot

Or they use llms to translate either of the two

[plush wagon]

Hello everyone, do you feel like AI filmmaking today is more about visuals than storytelling, or is it just still early?

[atomic mortar]

Still too early for anything coherent

[grave stag]

Then I'd just pay a real artist. Though I should learn how to do it myself for future use. Never did get it right and I've tried setup twice now.

But each time I've tried its been easier then the last so there is that.

[atomic mortar]

If you got some time, im home soon and can show you how i did it

[grave stag]

No visuals can make up for a truly bad story that said AI film making is gonna kill so many jobs we aren't gonna speculate about it because it shouldn't be coming but it is. Not thinking about it might delay it though.

If your talking the YT or tik tok shorts then a bit of both for looks and storytelling no mater how good it is if its just barely better then the early will smith eating spaghetti people aren't gonna watch it has to look closer to human done level even if maybe not on the scale of can't tell if its AI less you go frame by frame.

[grave stag]

I'm slacking off in class so can't, gonna go turn my phone off. I'm defiantly not paying Scofield though. This is all on some where on the slop scale and I don't think people should be paid for it, no offense. Not that I would have any way of knowing if a real artist used AI to make slop and fixed errors and did half the normal work.

Wanna just use my laptop if I can 32gb ram 13th gen i7 (I think) and a 4060 8gb.

Desktop has a 9070xt 16gb a 7700x and 32gb ram but I don't use it as much. Can I get away with just my laptop?

[atomic mortar]

Laptop definitely but 8gb vram will be a limiting factor.

The amd card also is possible but it's a bit more trickery

[grave stag]

I've heard if its same vram nvidia is faster for same quality but if its less vram it won't be as large a model so at best same speed or very slightly faster for worse quality because Nvidia has less vram to load the larger model?

[atomic mortar]

hmm its mostly because AMD uses up more vram for the same task, and some things are just funky / simply wont work due cuda dependency

[grave stag]

So like if its between same vram always go Nvidia less its cheaper for AMD but if it'd same price and AMD has more vram do it.

That's what I've been told.

[atomic mortar]

results otherwise:
no difference

but amd simply uses more vram (unless they changed something recently)

Cs1o might be the better person to ask about this but he often lurks in the #🤝｜tech-support channel

[grave stag]

Well yeah but larger model better. AMD loads larger model with more vram but Nvidia better if same vram. Least I hope. Otherwise I made a very poor choice with my 9079xt even if o paid less then $100 including tax on it if you count the fact that I sold my 4070 12gb for about $500 and was at launch for the 9070xt.

[atomic mortar]

For anything AI i say nvidia is still king, but you probably could run flux on the 9070xt though and the laptop would be struggling

for illustrious/sdxl based models. your laptop would work just fine

[grave stag]

I don't know the difference. I do know its always nvidia better if same vram I just assumed if less vram that wasn't the case but maybe I assumed wrong? Especially if only going up by 4gb (when I sold my 4070 for a 9070xt at near cost of 9070xt) but 8gb difference for laptop to desktop that's a bigger jump?

But anyways when I get home in 3ish hours if your still free and its the kinda thing I can setup in just an hour or two I'll 100% try it.

[atomic mortar]

it shouldnt take longer then 30min even

[grave stag]

Maybe its gotten easier last time I spent a few hours and I don't remember if I never set it up or if I did but it never output anything not at best horror movie version of said anime character I wanted.

[humble canyon]

Where can I find some comfyui workflows?

[steel ridge]

Hey @humble canyon, i can help you with that.

[humble canyon]

I mean if you can just direct me to where I can find it that would be perfect, a website or whatever

[atomic mortar]

kimmi is trying to sell/scam they try to get you to DM

what kind of workflow are you looking for more spesifically?

[humble canyon]

oh lol

[humble canyon]

what i'm doing is pretty much face swapping first but that gives me a realistic output as expected. I want to achieve some sort of high quality 3d cgi look if that makes sense. so i'm looking for a workflow that could turn a realistic photo into that sort of style. I can even give pictures for the style reference if need be

[atomic mortar]

hmm honestly you dont need comfyui spesifically for that or a super special workflow. what kind of specs are we working with?

like what GPU

[humble canyon]

I have a laptop 3060

yeah I know : (

[atomic mortar]

8gb or 12gb?

[humble canyon]

I believe 8

[atomic mortar]

hmm with your 8gb your quite limited ngl

using flux klein MIGHT give you a good result

4b is lesser then 9b

could maybe fit?

[humble canyon]

I'm already using flux klein for the face swap so I'll just tweak it a bit then ig 🥲

[atomic mortar]

check #🏞｜general-with-images care to drop an example?

of what you wantr

[humble canyon]

yep sure thing give me one moment

[humble canyon]

is it possible to use a lora with an img2img?

[steel ridge]

Yes, you can use a LoRA with img2img.

[spare jewel]

ghop ghop ghop

[atomic mortar]

yea you can

[humble canyon]

what I don't get is what should my prompt look like in that situation? like i said, i just want to change the style not create or add anything

[atomic mortar]

It depends on the lora trigger phrase if it has one

[shell yarrow]

is there anyone looking for dev?

[humble canyon]

Is it possible to find it out?

The civitai page doesn't say it

[atomic mortar]

Chances are that there isn't one

[upper dune]

Hi

[plush wagon]

Hello everyone, I feel like we’re in a Méliès moment. The tools are here, but the language isn’t defined yet. What is your take on this?

[atomic mortar]

Otherway around, the language exists but the tools are not there yet

[stone dust]

Hey guys, quick question — do you also get a lot of bad outputs like extra fingers or weird hands in anime images?

I’ve been working on a small tool that automatically detects these issues and filters out bad images so you don’t have to check manually.

If anyone wants, I can test it on your images and show results 🙂

[glossy sleet]

Hey everyone, I’m an AI engineer, currently looking for new freelance or contract opportunities.
If anyone is looking for help with AI product development, I’d be happy to connect.

[quasi stag]

helloo

I need some assistance in correcting the prompt I am using for a vton project?

Whould any with the expertise in this field assist me?

It would be a great help

[kind cairn]

I am new to this AI modelling. I would be really happy to use some help to be honest... if you are free 🙂

[atomic mortar]

The ai engineers are spam bots

[atomic mortar]

I recommend popping into #🤝｜tech-support pinned messages for a installation guide if you have a graphics card with more then 8gb of vram

[kind cairn]

installation part is done I want to learn more details when creating images since there is a lot of feautures

[atomic mortar]

Hmm well what gpu do you have? What you can use depends a lot on your hardware

And settings change drastically between different main models

[inland imp]

Hello.
I recently joined this group.
I still have a lot to learn, but I think it's important to have fun.
I'm looking for people I can easily talk to and who I can become friends with.
I'm all for Japanese people.
I'm interested in Japanese culture.
Nice to meet you!

[quasi stag]

Hi

What do you do?

I saw digital artist in your profile

@inland imp could you help with smt?

[inland imp]

@quasi stag
yes

pls send me DM first.

[atomic mortar]

scams are always in dms!!!

[atomic mortar]

dont get scammed by being lured into a dm.

[inland imp]

hey. urotypo!
dom't lie.

[atomic mortar]

wheres the lie

mister brand new account asking people to DM

[inland imp]

[atomic mortar]

Scam

[untold thunder]

what is the best way to join 5 second clips so it looks like a seemless contituation of each other? anyone can show a workflow and eli5 please? because the flow should generate i2v -> take last frame? and put it into a new generation flow, and then join the two and take the last frame in that clip and feed it onwards, and so on?
there must exist well working flows for this?

[kind cairn]

Except the ram everything is perfect the gpu is rtx 5070

Ram is not bad tho it's 16gb

[atomic mortar]

16gb is plenty for image generation, more is always welcome tho

im rocking 16gb aswell

what models do you run or you dont know yet?

[kind cairn]

Yep of course I tried some basic images it was generating really fast but the fingers or eyes were becoming disarster sometimes to be honest.. the only thing I want is to learn specific things about stable diffusion

[atomic mortar]

Well im walking the dog RN but after i got some time to show you some stuff

[kind cairn]

Thanks ❤️ good luck with your dog

[atomic mortar]

What UI do you use btw?

[kind cairn]

Windows

[atomic mortar]

No no I mean what do you use to make images like forge, swarm, comfy etc

[kind cairn]

Automatic 1111 I guess

[grave stag]

I've dabbled with image gen but its not be the most easy (for me been a while maybe it got easier but that's what they said last time and just no) and for one reason or another it just doesn't seem to stick for wanting to make images.

That said I just saw a website that does convincing like really close to real could maybe fool lot of people but so costly. Even if I gotta run the PC overnight can that be done because instead of being cool now my ref pics look a little better or I got a second angle and pic of my girl and stuff that be actual yes I wanna use that and print figures 100%. Or are those models even bigger (even with reference image so no can't do less you got a 3090 or something?)

[atomic mortar]

ah personally i wouldnt use that, especially when your using as 5070

A1111 is pretty outdated

[kind cairn]

So what is ur suggestion

[atomic mortar]

i use swarmUI because a lot of the "addons" are just built in and it supports most models because its built upon comfyUI

[atomic mortar]

what kind of model are you using? realistic or anime?

[vapid vine]

Hello!

im trying to install stable diffusion but i am having issue with the installation of CLIP. I'd really appreciate help 🙂

note: This error originates from a subprocess, and is likely not a problem with pip.
ERROR: Failed to build 'https://github.com/openai/CLIP/archive/d50d76daa670286dd6cacf3bcd80b5e4823fc8e1.zip' when getting requirements to build wheel

[atomic mortar]

your trying to install A1111 its very outdated. check #🤝｜tech-support pinned messages for either forge or neoforge

[vapid vine]

I am a moron (first time trying to install stable diffusion).

[atomic mortar]

nope. its just not very well documented and a lot of videos on youtube are just outdated

[vapid vine]

What is forge or neoforge?

[untold thunder]

what is the best way to join 5 second clips so it looks like a seemless contituation of each other? anyone can show a workflow and eli5 please? because the flow should generate i2v -> take last frame? and put it into a new generation flow, and then join the two and take the last frame in that clip and feed it onwards, and so on?
there must exist well working flows for this?

[steel ridge]

You are already thinking about it the right way. The best approach is exactly what you described: generate a clip, take the last frame, use it as the starting point for the next generation and keep repeating that chain so everything flows naturally.

[steel ridge]

If you want, I can set up a clean ComfyUI workflow for you that automates this whole loop properly. @untold thunder

[untold thunder]

wow thanks

that would be great!

[atomic mortar]

He would ask for payment, probably a scam

The reply is gpt made too

[still glacier]

Please stop trying to bait people into dm (where other people can't monitor what you're advising). If you want to help them, help them in plain sight where others can learn from the exchange too. Otherwise that s prime scammer behavior and you will get the boot for it.

[kind cairn]

Realistic

[minor olive]

hello

[kind cairn]

Hello

[minor olive]

nice to met u

i am full-stack and AI engineer.

[atomic mortar]

well if your using a1111 or if you want something similar forge neo

[minor olive]

and u?

[atomic mortar]

oh another ai engineer scam, let me guess your looking for clients

[atomic mortar]

then you can use adetailer to fix stuff like eyes, but generally if you use older models like SDXL its gonna look gnarly regardless

Z-image has some banger realism without too much effort

[minor olive]

yes

you are right

[atomic mortar]

crazy a manual scam this time, anyways wont work here 👍

[minor olive]

haha fair 😄

not really looking for clients though — more interested in collaborating with people on projects

[kind cairn]

So I should use forge neo instead you say? How to install it over a1111?

[atomic mortar]

exactly the type of scam we see a ton of here

[minor olive]

appreciate the tips btw, I’ll check those out

[atomic mortar]

it looks like A1111 if you want something similar but i personally am not too experienced with forge neo. in #🤝｜tech-support pinned messages theres a few guides

i personally use SwarmUI

[minor olive]

all good, I get why it might look that way

[kind cairn]

And what's swarmUI for?

[atomic mortar]

... image generation

and video

[minor olive]

right

[kind cairn]

No like what kind of images is it specific?

Or it can make any kind of images

[atomic mortar]

any? the UI doesnt determine the output. the model does

UI = user interface
model = actually makes the image
you need a UI to use a model

[kind cairn]

Alright thanks I was just curious

[glossy cliff]

https://cdn.discordapp.com/attachments/1447176971892297769/1496815472169975858/20260423_1309_New_Video_simple_compose_01kpwx0ndzet0td72sszqa33e5.gif

[kind cairn]

And any video could be made with stable diffusion or you need something else

[minor olive]

images are flexible depending on the model 👍

for video, you’ll need additional tools

[atomic mortar]

im afraid you learned a lot of wrong terms and what stable diffusion is

[minor olive]

— stable diffusion by itself isn’t really made for full video generation

[atomic mortar]

Stable diffusion is a business, you have SD1.5 SDXL, SD3.5 models for images
you also have flux, z image, anima for images too

you have hunyuan, LTX, WAN2.(and more versions) for video

[kind cairn]

Well u said u are using that uı for making images and videos that's why I wanted to ask sorry

[atomic mortar]

no need to be sorry!

its all good

just pointing out you might have learned some wrong terms and telling you what the right ones are 👍

[kind cairn]

Thankss

[minor olive]

???

is there a japanese?

[gray echo]

Hello 👋

[minor crater]

Henlo

[bright cradle]

chatgpt img 2 is crazy

[boreal pond]

Hello everyone!

[elfin pivot]

So any new breakthroughs in image generation technology ?

saw sd3 dropped but have not seen much talk about it

[atomic mortar]

Z image, Wan, and others have made big strides but nothing beats the good old trippyness off SD1.5

[untold thunder]

thanks for the warning 🙂

[atomic mortar]

rule 5. people not bothering reading the rules are often scams anyways

[deft topaz]

Hello Everyone

I'm looking for a developer with 5~10 years of experience.

[glass bone]

Hello

[atomic mortar]

look on fiverr as theres a ton of developer scams on discord

[plush wagon]

Hello everyone, we’ve been exploring the idea of AI filmmaking while putting together a small AI film festival focused on storytelling. I am curious to see how people approach narrative in this space.

[atomic mortar]

@vapid dove with the increasing amount of spam lately any thoughts of increasing the community guides?

Also second question, how much of the stability staff is even here still? I haven't seen them in ages.

[pale atlas]

Hey guys for training a LORA for an original character, if you want your original character to generally have the same hair style like idk 65%-70% of hte time, when tagging the dataset, would you tag the hair each time or not?

[atomic mortar]

If you want a consistent character, you use a unique identifier anyways so i wouldn't

[pale atlas]

I see. So if i didn't tag the hair, would I be fighting SDXL if i want to change the hairstyle later in a specific photoset?

[atomic mortar]

Not really, it might bleed through sometimes but nothing too breaking imo

[hasty quartz]

Hello, everyone. Is there anyone looking for a Senior developer?

[brittle slate]

[hallow hare]

What the heck is going on in this chat?

So much spam

[atomic mortar]

usuallly

1. official server thats mostly dead
2. only 2 people with moderator perms
3. stability staff is gone mostly

[hallow hare]

[steel fable]

I'm trying to use img to img to recreate an image I want, identical but with a different style, but it always comes out with random results. Any tips to make it work better?

[modern pagoda]

Gimme hammer power i got this 😂

[brisk void]

What image?🤔

[steel fable]

From a game that i play

[jovial monolith]

Hello. I am currently looking for a job. Is anyone here with a current project or a new idea?

[atomic mortar]

wrong place for a job, though theres probably a lot of scammers dming you right now

[uneven plume]

hi

[next siren]

Hey everyone 👋

I’m Akeem, a full-stack developer with a strong interest in AI and generative models.

[pale atlas]

Hey guys, I have a question about dataset tagging for LORAS training for an OC. I know that if I leave a trait untagged (like medium straight black hair), it gets "baked in" to the character's unique trigger word.

However, what if I want a trait to stay consistent ~85% of the time, but leave room for variation the other 15% (e.g., a ponytail)?

If my dataset is mostly straight hair but has 15% ponytails, do I need to tag the hairstyle in every image?

If I train it this way, what happens when I use the character’s trigger word without specifying a hairstyle? Will it default to the majority style, or will the hair become a messy "blend" of both?

[atomic mortar]

didnt you ask this the other day too

if you dont use a auto tagger + one/two extra keywords your doing it wrong

[pale atlas]

Oh yeah sorry, I just wanted to confirm becuase I wasn't sure if i asked correctly that time. Basically tag the hair, but I don't need to keep prompting the hairstyle to get the same general hairstyle if the same hair is used in 80+% of the dataset?

[atomic mortar]

just label EVERYTHING you see in the picture

• a unique tag for just the character

[pale atlas]

Understood haha

[atomic mortar]

scam

[modern pagoda]

Figured you guys might appreciate this brilliant madness

https://youtu.be/TGIvO4eh190?si=nBYdd0EYbZCL3eVS

[vocal stratus]

What kind of Bloodborne pkg do I need is it a specific version?

[atomic mortar]

Not the place, rule 5

[hybrid oxide]

Sorry mate. I didnt know where to post. I am new as I said 😅

I deleted the message

[atomic mortar]

No worries but this server doesn't allow advertising

[hybrid oxide]

So this is mainly for creators and devs instead of infrastructure providers. Check 👍

[atomic mortar]

well devs have mostly vanished too but the dev spambots

[hybrid oxide]

Doesnt sound like a promising future for this discord server 😅🫣

[atomic mortar]

well your not wrong

[waxen fog]

What’s the best base model for pixel art game assets?

[brittle slate]

any models... downscale/color reduce, pixelperfect

[grave stag]

How good is video generation and what can you do with it?

[keen plinth]

hiii

[ancient mauve]

anyone here uses qwen?

[steel ridge]

Hey @ancient mauve, Yes I do. What do you need it for?

[ancient mauve]

do you use it with krita? Im trying to do the typical put the clothes in the reference image into the charcater in the main image with no luck

[steel ridge]

Yes I use it with Krita

[atomic mortar]

dont enter DM's of randoms

they will either scam or ask for money!!!

[steel ridge]

What is meaning of that bro??

[atomic mortar]

just a friendly warning, why. you feel called out?

[steel ridge]

Oh

[ancient mauve]

she literally mp me lmao

[ancient mauve]

do you know this?

[atomic mortar]

hmm not with krita but if you gimme a moment i can check

in a game rn

[ancient mauve]

I want to use qwen for simple stuff like put the hairclip of the reference image into the main image or the clothes, things like that

but it eaither invents stuff or literally copy pastes the thing without even trying to adapt to artstyle or without posing it

[atomic mortar]

hmm care to hop into #🏞｜general-with-images

post some screenshot so i can see whats happening

probably the prompt ngl

flux klein works better for it personally

[ancient mauve]

flux klein is basically the same as qwen?

[atomic mortar]

for editing sure, it works a bit differently but you can still do the " put the clothes of image 2 onto the man of image 1" but not sure how its intergrated in krita

[ancient mauve]

what do you use comfy?

[atomic mortar]

i use swarmUI

basically comfy but with a nice layer

[grave stag]

?

[atomic mortar]

well thats a broad question, its decent and you can generate all kinds of stuff

check on civitai and then under videos

thats what you can do with it

[grave stag]

How is the nsfw side of it?

[atomic mortar]

I recommend checking civitai red

Showing is better then telling

[grave stag]

Does uncensoring work?

If I wanted to like make anime into real stuff?

Or for uncensoring real stuff? That's what is use it for.

[atomic mortar]

Your asking the wrong person but what you want is video editing.
There was a new model that dropped but goodluck running it

Unless you have top of the line hardware

[grave stag]

Is it for some reason harder then making from scratch?

All I have is a 9070xt 32gb of system ram (though I may by another 32 but sadly I'm on DDR5) and a 7700x. Don't need it to be that fast. I just REALLY like the scenes I can find online where people re edit the scenes to be you know instead of the hair and what not.

Though I'm probably asking on the wrong server because if I mention certain words like adult anime (you know the word) it flags me.

[atomic mortar]

Amd makes everything more difficult that's for sure

[atomic mortar]

@vapid dove
Sorry to ping again when your finally online again but im really curious about your thoughts on this

[unique ravine]

Typeshit frfr

[vapid dove]

happy to increase the number of guides. as for the number of stability staff there are quite a few (10+) still kicking around but aren't flagged as staff. They claim its because they don't want to get spammed but they can't hide their digital agoraphobia from me.

[atomic mortar]

Lmao fair enough, more guides would be very welcome to just clean up the spam that flows through here

Just sad to see the massive amounts of people trying to rip people off too by luring them into dm's

[vapid dove]

yeah srsly

[atomic mortar]

Well if applications open up or if you select a few regulars id be very happy

[vapid dove]

https://tenor.com/view/jesus-jesus-christ-jesus-is-our-saviour-yeshua-lord-and-saviour-gif-23033351

you have ascended

[atomic mortar]

Hopefully we can turn the activity around and see a rise in actual users again

[vapid dove]

hopefully, will help when we have some models we can actually talk about again.

everything has to be so hush-hush now

[atomic mortar]

Thats definitely true, still theres a lot of different models still floating about that people can use.
Frankly sd1.5's creativity is still unmatched

I'm gonna be diving into making loras for some of the more dated models again since most people can't run the cutting edge stuff locally

[glacial igloo]

Hey everyone. I had made a management tool and recently uploaded it to GitHub. Where do I go to post it? I'm mostly wanting others to get use out of it other than myself.

[atomic mortar]

what kind of tool?

[glacial igloo]

It's kind of an everything tool for managing Loras, Checkpoints, as well as images. It has Civitai and Danbooru integration. I just didn't want to spam it everywhere or anything, but I finally got around to releasing it after working on it over most of this year.

[atomic mortar]

hmm well theres no real spot for it here but you could try reddit for it.
though dont wouldnt that be better as an extention to a UI rather then a seperate tool?

[glacial igloo]

I mostly wanted to make something outside of an extension, but you're probably right! There's already Lora Manager and Diffusion Toolkit, and this is kind of a morph between them, with some other features on top. I posted it to Reddit recently there but wanted to make sure I went through the avenues I could think of!

[atomic mortar]

hmm the subreddit is actually an seperate group iirc. awesome you made a project and if its open source even better but if anything you could try their discord since its broader then stable-diffusion but im afraid i cant give the approval since i just became a guide

[glacial igloo]

Ohhh okay. My apologies! I had thought this was that one. I'll go ahead and check out on there as well! I appreciate it! I'm just wanting people to get use out of it more than anything. I originally made it for myself but realized I could get it working so others could get some life out of it, too.

[atomic mortar]

no need to be sorry you went the right way about it! good luck with your project!

[glacial igloo]

Thanks! I'll do my best 🫡

[warm junco]

Congrats Eurotypo!

[atomic mortar]

thanks! i got so tired of the spambots/scammers i decided to give it a hand

[unique ravine]

hell yeah

[dry forge]

Hi

[atomic mortar]

Hi hi

[dry forge]

I was just wondering if you guys had a bot I could use I want to make a movie but it would be like a lot of messages in a row over long periods of time, hopefully not too long 😔

[atomic mortar]

nothing like that here, we do have #artisan-faq but its a paid service. theres also sources like CivitAI that allow video generation but also renting a cloud gpu and generating it that way is also a option

[dry forge]

Cool beans, I'm somewhat new to this whole ai video stuff just have an idea I always wanted to bring to life is all

Thanks

[atomic mortar]

video AI is still pretty pricy and tricky to learn, even with the right hardware it takes multiple tries to get a solid result

LTX is making some solid progress on that front for video + audio, could be worth looking into

[uneven plume]

hi everyone

[marble junco]

hyeyy hii

[atomic mortar]

Crazy how its a copy paste from a reddit post 1:1

@marble junco care to elaborate?

[atomic mortar]

Deleting just in case then

[foggy ice]

Does that unstable diffusion server no longer exist?

[atomic mortar]

when i last saw it there was some shady stuff going on. reddit has a LOT of complaints on that group and eh the website seems dead seeing the last post was from 2024

[foggy ice]

Ah, can't say that I'm surprised I guess, although I thought they moderated it pretty well