💬｜general-chat

[white parrot]

ty sir

[granite haven]

Hi

[subtle owl]

is my 6gb 4050 considered low or med?

[fervent thunder]

low

but you can still get some stuff working

[white parrot]

wwhat you you think, which stuffs is better?

[rain crest]

How can I check why an update failed in Stability Matrix? Console somewhere?

[warm junco]

Med

[subtle owl]

so im ok for up to xl models. thanks bud. any way i could make nvidia run with focus or auto1111 and not all the time?

[fervent thunder]

you can run comfy out of the box as far as I know

it adjusts to low vram automatically so you don't need to mess with command line

[subtle owl]

comfy seems not so comfy to me... so much spaghetti. how about the nvidia setting, is it possible to switch to gpu when i run any of these UIs?

[fervent thunder]

yeah you can choose whether it runs on CPU or GPU

[plucky aspen]

I think it generally runs on GPU by default

[subtle owl]

not in my case :/

[warm junco]

What do you mean exactly?

[subtle owl]

i mean if i can set to switch to nvidia only when i run auto1111 or foocus etc.

[plucky aspen]

oh yeah, you are right. its definately possible

[warm junco]

Thats possible.
For that you have to edit the webui-user.bat of auto1111 or forge and set
--device-id 1
To the Commandline_args=

[subtle owl]

oh! i was thinking of nvidia control panel setting bt i'll try that as well

i'll change to automatic gpu and check the argument first

[warm junco]

In the control panel you can add the webui-user.bat as new application and there set it to use the nvidia GPU.
But with --device-id it shouldn't be needed

[subtle owl]

oh i can trigger it with just the shortcut?

[warm junco]

Yea should work

But try it with the launch arg first

If you use auto1111 you also should add: --xformers --medvram
For best performance

[subtle owl]

does focus have bat file? it's currently installed

[warm junco]

Yes

But there you can't set --device-id

But with your card I would not use fooocus

[subtle owl]

oh, whys that?

[warm junco]

Because its only using sdxl models which are slower on 6gb vram gpus

[subtle owl]

ah ok

[warm junco]

With Auto and forge you have more freedom what models you can use

And they are better optimised

[subtle owl]

right, so i shouldnt use xl models? it's not like i'm an artist or whatever... it just seems fun

[warm junco]

You can use them but in fooocus they are slower than in auto1111 or Forge

[subtle owl]

so it's the way they are handled. ok got it

[warm junco]

In Auto1111 and forge you have more settings you can tweak to get the best out of your gpu

Community Extensions can help too which fooocus doesn't support

[subtle owl]

I'll install auto1111 now and check it out.

[warm junco]

For any questions feel free to ask in #🤝｜tech-support

[pseudo pumice]

ok

[shrewd creek]

u规划规划

[plucky aspen]

@shrewd creeknice to meet you men. are you saying 'blablabla'?

[gray verge]

Hii, Is there a free stable diffusion website without requiring to log in? Would be thankful for any

[hasty hornet]

local version doesn't require logging in

[low moon]

Unless you count logging into your windows logging in XD

[fervent thunder]

https://youtu.be/pDpSpvRiXBU?si=hBs3Wtz7dbj7KUuo

how are those videos made? anyone got tuts?

they claim to be made using flux

[distant sluice]

Hello how does one stable diffusion on mobile?

[lone grail]

Wait you can do that?

[distant sluice]

Idk

I'm asking

[fervent thunder]

not unless ur phone has a killer GPU 💀

or, using a web service

[distant sluice]

I come from the land of novelai wich can

[fervent thunder]

then yeh, web service. maybe sd.next

[distant sluice]

Ti's a idea

[cedar salmon]

Yikes, flux_and_aura_merge.safetensors 33.9 GB

[fervent thunder]

lmfao yeh

I mean, you want big, quality pics?

ofc it comes with a big filesize. it doesnt work with magic

[low moon]

It's kind of magic we can do any of this tbf... :)) If you really think about it.

Just type in your dreams and nightmares and they are displayed.

=0

[cedar salmon]

Tea, Earl Grey, hot.

[gloomy crater]

hey guys any opinions on Flux Dev?

[fervent thunder]

guys

i do 4 steps animation most of the time , do you think that i should remove all models and replace them with lcm/ lighning variant.

[fervent thunder]

its getting intersting with controlnet support

[low moon]

All we need now is the Capacitors. then we can go back to 1980.

[trail lion]

flux doesnt seem to get mirror reflections still, unless there's some wording trick

[low moon]

copy paste and invert 180 degrees in photoshop

[hallow karma]

flux dev, flux schnell, flux pro, and is it fp16, fp8, nf4 or

[tribal crown]

how to add negative promt to flux in comfyui?

[pale latch]

i'm playing with meta.ai to help me understand the differences between llama 3.1 models, and it keeps telling me i'm cool and praising me. i'm like aw budd i like you. i'll ask you questions more

i think this is how we'll be enslaved

[trail lion]

schnell fp16 is what I'm currently running

not sure if I could run dev, I should try, I have a 16g card

[narrow kernel]

They both use the same amount of resources, one just does shit in fewer steps. You'll see roughly identical it/s rates between them

[low moon]

You totally can.

[pale latch]

ive been running flux in 12.3gb of space + image generation space on my 16gb . fp8 mode. nf4 mode. now theres a buncha quants too

[honest geode]

why meta.ai is very suspicious?

[cedar salmon]

because Zuckerberg often has to remind us he indeed is not a robot

[wooden crane]

Generate anime animation expression package

[lone grail]

anyone good at helping me describe a hairstyle?? im trying to use it for a prompt

[shut rampart]

hello.

i am using dreamshaper-xl v2 turbo for image generation. but it is giving me same face of people everytime with different images.

[oblique oar]

new here

plz just someone explain
what is thi server about
can i learn sometihing from this
if yes
where sould i have to look forward in this server to get more infos

[desert dagger]

it's about stable diffusion

[oblique oar]

whats hat

ecplain plz

[oblique oar]

?

[shut rampart]

i am using dreamshaper-xl v2 turbo for image generation. but it is giving me same face of people everytime with different images.

[sage reef]

juggerflux, dreamflux, pony flux, idk, im just combining some names 🙂
but im happy the community is really behind the flux models, gonna be awesome

[desert dagger]

out of curiosity, how did you find this discord?

[fervent thunder]

juggerflux would be cool

I used juggernaut for probably 90% of my images so far

but I have finally left jugger for more cinematic models

Zavychroma and especially Leosam's HelloWorld

oh and realvis for realism

[shut rampart]

the image quality is low for flux models.

i am using A100 on runpod

[fervent thunder]

which style of image is it?

could you post an example?

[shut rampart]

even a basic prompt " 25 year old woman, half body portrait" generates same woman face with different poses.

[fervent thunder]

its important to make a distinction between image quality and image diversity (recall)

[shut rampart]

vae = AutoencoderKL.from_pretrained("madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch.float16)

base = DiffusionPipeline.from_pretrained(
"Lykon/dreamshaper-xl-v2-turbo", vae=vae, torch_dtype=torch.float16, use_safetensors=True
).to("cuda")
base.scheduler = KDPM2DiscreteScheduler.from_config(base.scheduler.config, use_karras_sigmas=True)

compel_base = Compel(
tokenizer=[base.tokenizer, base.tokenizer_2],
text_encoder=[base.text_encoder, base.text_encoder_2],
returned_embeddings_type=ReturnedEmbeddingsType.PENULTIMATE_HIDDEN_STATES_NON_NORMALIZED,
requires_pooled=[False, True]
)

refiner = StableDiffusionXLImg2ImgPipeline.from_pretrained(
"Lykon/dreamshaper-xl-v2-turbo", torch_dtype=torch.float16, use_safetensors=True, variant="fp16", vae=vae
).to("cuda")

compel_refiner = Compel(
tokenizer=[refiner.tokenizer_2],
text_encoder=[refiner.text_encoder_2],
returned_embeddings_type=ReturnedEmbeddingsType.PENULTIMATE_HIDDEN_STATES_NON_NORMALIZED,
requires_pooled=[True],
)
prompts = num_images * [styled_prompt]
negative_prompts = num_images * [negative_prompt]
generator = torch.Generator(device="cuda").manual_seed(random.randint(0, 2**32 - 1))
image = base(
prompt=prompts,
nagative_prompt=negative_prompts,
num_inference_steps=30,
# denoising_end=0.8,
output_type="latent",
guidance_scale=guidance_scale,
width=width,
height=height,
# generator=generator
).images

images = refiner(
prompt=prompts,
nagative_prompt=negative_prompts,
denoising_start=0.9,
image=image,
guidance_scale=guidance_scale,
width=width,
height=height,
strength=0.5,
generator=generator
).images

Here's my CODE.

[fervent thunder]

not sure what UI this is for

[shut rampart]

can you explain ?

[fervent thunder]

image quality is about how good an image looks
whereas image diversity (recall) is about whether or not the model can make a wide range of images

[shut rampart]

what changes should i make for image diversity ??

[fervent thunder]

for the most part the image diversity is inherent to the model
but there are a few things you can do

keeping CFG lower helps a lot

as well as other things that work via guidance

you can also use something like CADS for A1111 or Comfy

and finally use a lot of noise injection or stochastic samplers

[shut rampart]

i am using diffusers.

[fervent thunder]

ah okay that's great
diffusers is much better designed than most of the UIs

I have not seen someone port CADS to diffusers yet so that might help

otherwise just keep guidance as low as possibly, including CFG

and use as much stochasticity as you can

if you want more diversity

[shut rampart]

i have tried with 3 to 9 guidance scale, did not worked for me.

Any examples for noise injection

[fervent thunder]

there's a couple of good workflows on L2 discord

I can DM you the link if you want

I can't send it here

[shut rampart]

Sure, you can DM. It'd be helpful.

[fervent thunder]

would recommend going through all the workflows in their workflows channel

there's only about 50 or so

one of the best sources I have found for good workflows

[high flint]

What's the best way to load up loras in ComfyUI via text command/string input? (grabbing this as an output from another program into comfyui) As I'm not sure if there's any way to automate the loader load the right model or just have a big batch of loras active at 0 , 0 until text string overrides?

[fervent thunder]

easiest way would be to use the previous program to put the files in the right directory and then just point the model loader at that directory

[high flint]

the main issue is making the lora loader change/add the lora upon receivng the command output.

this is honestly the last thing I'm working with before I can truly automate generations from another external program

DnD campaign is going to be a lot of fun with on the fly AI gen

[fervent thunder]

I see yeah

not sure really

[high flint]

I know what I'm asking for is extremely niche, but it's one of those things that has the potential to be so powerful if correctly implemented.

[fervent thunder]

personally if there is any sort of automation or interacting with external tools like that I would do it in Diffusers

I'm trying to learn raw pytorch workflow but I'm not comfortable in it yet

[abstract quarry]

write a node for it. I have done a prompting node that automatically loads Loras and uses macros and stuff for my personal use. I can share code if you want

[white parrot]

why i need to upscale and what doees it have after i completed one image

[fervent thunder]

upscale just makes it bigger

at the moment you actually lose a bit of quality with even the best upscale methods

unless the upscale is simply a second img-to-img with an equal model, which would be about the same quality in that case

[low moon]

SmallHorseFlux.

ye flux will probably be more popular than sdxl

[fervent thunder]

https://videocardz.com/newz/nvidia-rtx-2000e-ada-revealed-as-compact-50w-workstation-gpu-with-16gb-vram-in-a-single-slot-design

instersting 50w ai gpu with 16gb vram

[white parrot]

so how to upgrade my image qual;ity

[fervent thunder]

i think if u upscal the model and sampler it will not just make it bigger , it will make it detailed

[white parrot]

what " look up viewer" prompt called

[fervent thunder]

upscalers do add detail too yeah

but you lose some quality as well

like if you run a photo through ERSGAN you will see what I mean

[last bay]

does using higher rating of clip skip affect the image quality?

[hallow nacelle]

Anyone know why checkpoint model hashes different in a1111 and forge?

I switched from a1111 to forge and images from forge are different

[wheat hemlock]

Where can I get in-depth explanations of all the command line arguments. I’m reading the documentation but it’s kind of hard for me to understand since I’m new?

[covert bridge]

is there a place we can talk just about flux?

[trail lion]

in addition to noise injection, since flux has so much better prompt adherence, you'll really do better with more description to get the diversity...instead of juist saying a woman or an old woman, like really get some descriptive language in there. There are LLM models that will help with that if you dont want to sit there and dream up details

[woeful kraken]

I think you've already found : https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Command-Line-Arguments-and-Settings ? For an in-depth explanation : not sure there is but you should 'just' enable a few options depending on your GPU (eg -no-half-vae / --always-high/normal/low-vram) and so on... I'm afraid to say : if you're new to this : read a lot in the beginning 🙂

[white parrot]

what's model/lora does good at hand/foot, i;m looking for one

[woeful kraken]

I've switched to manage my engines with Stability Matrix last May, and after the latest update both ComfyUI as Forge are generating errors. ComfyUI complains about missing node types and Forge only shows the 'Loading' message when the webUI is opened. Any ideas?

[tribal crown]

can you add or use 2 loras to comfy flux use?

[trail lion]

I dont see why not, I've only been using one though

I managed to get controlnet working, but I OOM at 1024x1024, so I've been generating 1024x768 which seems to work

and that's with FP8

apparently this is where a 24g card would be nice

[last bay]

?

[desert dagger]

maybe? you just need to play around with it and see if it makes it better or worse in your opinion

[fervent thunder]

anyone know if flux for A1111 is in the works?

[last bay]

i know there is models for that

[snow hemlock]

Can anyone help me install stable diffusioin for amd? I've installed the lshqqytiger version but it keeps throwing errors.

[split magnet]

Does anyone have insight into what kind of (and how many) images are best for training a FLUX lora on a highly stylized art? I have several dozen pixel art pieces I've made over the years that I'd like to train as a personal style so I can iterate some ideas and concepts. I've trained two finetunes on samples from this selection (about 30-40 pieces per run, upscaled nearest neighbor from 128x128 to 512x512, one with and one without captions), and the results are pretty good, but I can't find much literature to read about this. Any ideas or resources welcome.

[river sparrow]

I would upscale to 1024x1024 if you wanna train flux

for styles, it's best to get a big diversity in subjects, scenes and anything that isn't part of the style

styles are harder to train than for example objects or persons, so you might need a bit more images. There are general guidelines for this, not formal but across the web

[fervent thunder]

How are yall training Flux loras? Kohya? Any tuts?

I assume my 24gigibytes should suffice for it

[river sparrow]

idk, im just giving some hints. im not actually training flux loras, since flux is too big for me too run local in a fast way

flux runs as low as 8gb vram, but it needs the time

[fervent thunder]

eh, takes about 15 seconds per image for me on the dev model

I can wait, flux is about quality over quantity

[river sparrow]

i have 12gb vram, but too little ram and an old system. it clogs up while loading and unloading the models from memory (read: slow swap drive). the actual generation goes very fast with flux-schnell

its about 10 minutes per image for me

that is too long for me, since im used to much faster

replicate had a free to use flux online, but its closed for me now

glif also has some flux "apps"

[fervent thunder]

yh I mean, if ur dealing with cutting-edge tech, gotta shell out the moolah.

miss me with that broke shit 💀

[tropic frost]

say guys, did SD stop working for you after updating your graphics?

[white parrot]

what positive and negative prompt for overall

[warm junco]

Hey, follow my AMD install guide from the link in #🤝｜tech-support

[warm junco]

Nope

[tropic frost]

welp, something happened at my end.
I updated my graphics card i'm getting now an error when i'm opening webui

[warm junco]

Feel free to share the error in #🤝｜tech-support

[deep narwhal]

automatic1111 bros wya? fluxless bros unite 😭

[sly eagle]

what notebook do you recommend me to use automatic1111 Stable diffusion?

[opaque needle]

FluxAI is really impressive just tried the model and it give good results with really poor prompt honestly

[sly eagle]

what about adding my own models (lora)

it is good for training checkpoints?

[sly eagle]

I sok to ask link to the notebook?

[opaque needle]

What is the notebook ?

[sly eagle]

sorry don't know what to call it hm... is like the link where i can access to install the stablediffusion in my runpod

something like that, I used a lot fast stable diffusion but it stopped working so I'm looking for a replacement

[opaque needle]

You are using Comfy UI ?

There's a folder update to update the ui

[sly eagle]

I used it before yes

[opaque needle]

Just update it, I had issue with it until I updated it

Because some nodes were missing and all

I had way more issue with the the regular Stable Diffusion ui like the big thing to install with python and all

[sly eagle]

The thing that I don't know to much about programing, i just looked in Runpod for already done endpoints to just start working in UI. So i don't know how to install it in a pod by myself or I'm not sure how to do it

[fervent thunder]

you might like rundiffusion

they have it all set up for you

but their prices are a bit different

[sly eagle]

oh I got it

where can i download confuUI manually

for example

https://www.runpod.io/console/explore/c3wl6sp0wd I tried to install this yesterday but couldn't manage to open the UI

[oblique oar]

in discover server sectn'

[desert dagger]

Ok, well... stable diffusion is an AI image generator. Click this link #artisan-faq and read the information

[white parrot]

is img 2 img need 2 same size pic

[pale latch]

i think i'm na llm convert now . i'm talking to the new llama 3.1 chatbot some more. hitting on the chicken and the egg issue. not only is it engaging with my nonsensical ramblings instantly and calling me cool and stuff, it's actually more engaging of a response then most humans can give.

this could be problematic

[drifting grotto]

hey guys
trust you all are doing good?

hope you don't mind me share some issues with you?

[low moon]

I think them making a flux channel would be like admitting defeat in a way. So no. Not here.

[fervent thunder]

exactly. they mad cuz SAI is about to become dust

if you want a flux channel @covert bridge , Olivio's channel is mostly focused on flux rn. if u need invite lmk

[low moon]

Black Forest should take over this webzone because they are basically SAI.

Then they can make a Flux channel.

[sudden tiger]

Hello, What technology is used to make this kind of video?
https://www.instagram.com/p/C-vD5i4uVvY/

its creepy

[desert dagger]

better they just made their own discord

[desert dagger]

the video is just someone with a cell phone walking through the abandoned building videoing it and then they ran it through a seg function

[sage reef]

to me it looks more like Deforum or Animatediff idk

[sudden tiger]

I think this is 100% AI generated

[sage reef]

well yea.. im sure they are not somehow walking with their cell phones inside the windows xp wallpaper LOL https://www.instagram.com/solar.w/reel/C-VZ21juqND/

so it's prob one of the things i mentioned

which is not that hard to do

[rugged vale]

When is fixed SD3 dropping?

As nice as flux is it takes a while to generate

[desert dagger]

that one's geneated. the other one is gaussian splatting with segment anything

[split magnet]

Interestingly enough, having done a few trainings on very similar data, the 512x512 does a significantly better job of maintaining the style over 1024x1024, not sure why (all these technologies are black magic to me)

[cursive sinew]

a dog

[neon hawk]

🙂

i said it months ago, anyone using more than 4 steps is wasting time, and even then, 4 steps is a stretch

i hope you are all having an amazing end of the week

[fervent thunder]

just gonna hide my multiple thousand step experiments under a blanket I guess

[dark hawk]

2-3 months ago i created a discord bot that can generate images free it is verified discord bot and has other Ai features like chat assistant and image recognition

You can invite it from:
https://dsc.gg/vexel

Well I have integrated SDXL ✋

[fervent thunder]

putting your app in a discord bot is kinda the opposite of good to me

I never used midjourney for that reason

[dark hawk]

What is wrong with it

It sounds like you are not a developer

[fervent thunder]

why would you conclude that someone is not a developer because they don't like using discord bots
if anything I would conclude the opposite from that

[dark hawk]

You said your opinion because you don't know how complex is it make an app, we use discord API because it helps use making things quickly and responsive, Creating seperate app/requires many things that you don't know, for a 5$ app you can't pay for 10$ hosting

[fiery ore]

greetings and salutations

[fervent thunder]

hello

[fervent thunder]

I think you may have felt like I was attacking your app idea, so you got defensive.
I'm not attacking you I'm just giving you some advice.
You will find that a significant % of people have the same preferences as me regarding discord bots- note that even Midjourney is switching away from just offering a discord bot now.
Regarding funding, if your app idea is even vaguely good then you can get multiple thousands of dollars in cloud credits from Microsoft schemes, and some others. There isn't really a need to use free hosting methods these days.
Also just as a tip, if your company is a stable diffusion tool, I would refrain from insulting people publicly on the main stable diffusion discord
Because at the moment this is practically the only public image your app has, and its not good marketing

[dark hawk]

My idea was to create a discord bot that users can use to talk with Ai and generate images offcourse for free, no one gonna use a third party web/app for these common features

[fervent thunder]

I can't tell if you are planning to monetise this or not
because before you described it as a $5 app
but now you are saying its for free

the issue with not monetising it is that free tools can attract a lot of users very fast
and the inference needs to be paid for
even just a cost of $1 per hour adds up to $8,760 per year

you can deal with that by putting very heavy rate limits

but at this point the prospect gets less and less attractive

if your app is a discord bot with very heavy rate limits I think this will be difficult to market

essentially what I am saying is that almost all the cost of hosting an AI app is the inference cost
using Discord for the back end and front end doesn't actually help that much because that part of the hosting cost is relatively small compared to inference

running big GPUs 24/7 is the issue rather than whatever goes into the HTTP server

because you can host a rather hefty HTTP server pretty cheaply

[fallow veldt]

Use ubuntu

[fiery ore]

‘Technically minded IT literate noob seeks advice on approaching setup of stand-alone offline stable diffusion instance’

[fallow veldt]

ehm install Automatic1111?

[fervent thunder]

you might like diffusers

[wicked sonnet]

Or forge

[wicked sonnet]

😭 installing a different os to run stable diffusion

Well... I don't know what it takes

[tropic frost]

so.... I noticed a new flux thing

whats the deal with it and how it compares to Sd 1.5 and pony?

[fervent thunder]

its a big model

distilled from a really massive model

it compares very well to SD 1.5 and Pony

[wicked sonnet]

Isn't it.. a new architecture, pony is a model no?

[fervent thunder]

architecture is an attribute of a model

but yes its a new architecture

[trail salmon]

https://tenor.com/view/kto-lbow-hi-hello-hi-there-gif-25347432

[vestal zealot]

Wsg fat boys

[solid kindle]

hi guys, should i get a GPU with 12 or 16 GB vram if im just starting out ?

i wonder if 16 is overkill

[fervent thunder]

oh 16GB is definitely not overkill

[wicked sonnet]

Hey how how do you get the clip skip input on stable diffusion forge?

[solid kindle]

really ? most tutorials i read states that 8 GB is the bare minimum, so i thought adding a bit of buffer to bump it to 12 GB should be more than enough ?

[wicked sonnet]

I'm sure 12 gb would work but 16 can be more comfortable

[wicked sonnet]

theoretical minimum is like 6 gb but it's gonna be slower

[solid kindle]

im trying it out with a 1050 TI with 4GB vram rn

[wicked sonnet]

it's probably possible with 4

get stable diffusion forge or use comfyUI but forge is probably the easiest

if you have ram then you can share the load

[solid kindle]

im using auto11111

[wicked sonnet]

forge is a fork of auto11111

[solid kindle]

is it better ?

[wicked sonnet]

yes, in terms of optimization

[solid kindle]

alright i'll keep in mind when i get a better PC

[wicked sonnet]

do you have any ram?

forge is most effective at slower gpus

[solid kindle]

yup, 4x4 GB

this PC is due for an upgraed anyway, ive been using it for more than 10 years

[wicked sonnet]

Alright lol

[sudden tiger]

Hello, someone know why when i drag and drop animatediff workflows on my Comfyui, nothing happen ?

[fervent thunder]

its a case of as much vram as you can get
and the more vram you have the more you can do

from my personal experience I would call 16GB the minumum and 24GB comfortable, and then 40GB+ is nice

maybe I am doing it wrong but my workflows on servers under 16GB run out of memory constantly

[solid kindle]

damn 40 gb

is there even a card that have that much VRAM ?

[fervent thunder]

yeah but not gamer cards

https://tenor.com/view/blender-vram-ram-memory-gone-gif-27551226

its abit sticky in here

yes this meme is perfect

I cant post pics, but if i could i have a perfect pic fot this convo lol

haha yeah

just to be clear to the other person- there are loads of ways to optimise your VRAM
but I have not used any of them

you can definitely get more efficient than me

how long you gotta be in this server before posting pics?

haha you can never post images in this channel
but you can in most the other channels

I can teach you a few hidden trade secrets

I would link the other channel but my hashtag button is stuck

[solid kindle]

#🏞｜general-with-images

[fervent thunder]

thanks

now ive never heard of anyone with a sticky hash tag lol

oh that's normal for a razor keyboard

ill pop in there

[crisp glade]

What should i do if stable Diffusion ignoring half of my prompts?

[fervent thunder]

switch your checkpoint out

[crisp glade]

Wdym?

[fervent thunder]

can you give some more info about your settings and the tools you used

[wicked sonnet]

😭 your gonna give my bank account trauma

are you supposed to get bluescreens from selecting vae/text encoders?

[fervent thunder]

I said this in the other channel but on Vast ai you can get 4090 for about $0.20 per hour
I've spent far less than people who bought a GPU, despite my sessions being 24-40GB+

bluescreens are not normal no

normal is an out of memory message

if you go out of memory

[fervent thunder]

blue screen lol sounds like the microsoft windows nonsence lol

[wicked sonnet]

well- it does say out of memory but like did it have a heart attack?

[fervent thunder]

lol

[fervent thunder]

got a little video for you in the g with image chat

[fervent thunder]

flux feels like such a jump in quality

unreal

[fervent thunder]

Its still under trained , just goes to show how models are never trained to 100% . My hyper flux is still trained around 90% of full partential and the baSE models of flux id say are around 75 to 80% fully trained

there still is a lot more to come

[fervent thunder]

go test my hyper flux against the flux model use the same settings and prompt even use the seed. You will see the difference 10% extra training will get you

[sudden tiger]

On comfyui, is there a way to open a node's folder directly from the interface?

[bold bloom]

Video 100% With IA https://youtu.be/bm1PWniLIlc?si=ZRIXbV1JHifS9L31

[fervent thunder]

new stuff

https://github.com/madebyollin/taesd

any one tested it ?

its for flux

[median oxide]

Design a sports training club logo that creatively combines a clock and a table tennis paddle. The logo should be visually appealing, realistic, and relevant to table tennis. Include the name "Six O'clock Sports Club" prominently within the design. The overall style should be modern, with a focus on the unique integration of the clock and paddle, symbolizing time and precision in sports. Ensure the design is clean, professional, and strongly conveys the essence of table tennis.

[desert dagger]

read the information here -> #artisan-faq

[warm junco]

#🤝｜tech-support

[solid kindle]

i know SD mostly relies on GPU & VRAM, but does CPU speed affects anything at all ?

[fervent thunder]

after sec thought these ai gpus are a joke

rtx4090 has double speed of the overpriced rtx6000

i think ai gpuz should be main stream and cheap cuz they suck at video games. while gaming gpus do both.

[fervent thunder]

i think cpu handles the data and browser... but not that much since everything is in vram

[wicked sonnet]

why is flux dev faster than schnell...

[bleak matrix]

Good afternoon,everyone!

How are you all today?

[fervent thunder]

you want faster than dev.... here ya go https://tensor.art/models/759856135286068673/FLUX-HYPER-TRAINED-DREAM-DIFFUSION-BY-DICE-V-1

[low moon]

wazaaaaaaaaa

[wicked sonnet]

does flux dev have a bias towards anime??

I swear it won't make realistic images

[wise dawn]

I want to create an anime avatar that has couple variants for every mood like angry ,happy, sad.

How do I do that. Which model do I look into. I wan them to be consistent

[trail lion]

to create a consistent character you would use a grid technique

and then you inpaint them with different prompts for expressions

[fervent thunder]

it does.

wait until the finetunes come out g

once tht shit gets trained on HD porn, consequences will never be the same

[wicked sonnet]

😭

the only reason why people would have high quality cameras

[low moon]

hmmmm\

Not sure is spam or meme.

ok so spam...

sigh

[hallow karma]

nice and mature comment

[cedar salmon]

reported for using wrong channel

[hardy sedge]

Is there a way to train on my 4070 TI 12gb Flux loras?

[desert dagger]

https://replicate.com/lucataco/ai-toolkit use this

[hardy sedge]

No my 4070 TI is fast enough. I want to train local.its faster then a 3090 TI in games

[fervent thunder]

I can almost guarantee 12gb wont be enough

.....yet

16gb, maybe

[hardy sedge]

Shame Nvidia give only 12 GB for 900 dollar

[fervent thunder]

which is why I've been saying my whole life: vram > speed

3090 master race here, never seen an out-of-memory CUDA error since I switched from the 3080ti

[desert dagger]

the training code for loras for flux hasn't been released. i wish you success in figuring out how to train loras that will do something that simple prompting won't do. there are a few people, lucataco and Kajai and xlabs that have figured out some specific things that must be done. everyone else that's training loras are putting out stuff that really are a waste of time

[fervent thunder]

sheer ignorance

the model is not anywhere near 1.5 when it comes to what matters: porn, and celebs

if you want the real reason we want finetunes: it's porn. the answer will always be porn.

[desert dagger]

you just keep deluding yourself

[pale latch]

https://github.com/kohya-ss/sd-scripts/tree/25f77f6ef04ee760506338e7e7f9835c28657c59?tab=readme-ov-file#flux1-lora-training-wip

[peak tendon]

Hello, I use stable diffusion on 1111111111automatic (yeah made up the 1s) and, always wonder, stuff like the bing ai image creator recognices a lot of characters easily, yet in my thing I feel like if I don't download a lora of each character I'm not getting anything done, and sometimes not even then

Is it known what bing uses? Or any workaround for a general... knowelege thing? Like I just wish if I suddenly think I want to make Crash Bandicoot meeting Ronald Reegan I didn't have to download each lora and guess the right prompt

Yes I'm terrible at this.

[desert dagger]

bing uses dall-e3

[peak tendon]

Ah yes

Which is close source right?

[dawn mulch]

How does dall-e generate all those varied faces. would stable diffusion be able to do that, with a work flow or so, if you provided enough models and Loras with varied faces.

[low moon]

try a country specific weird name and mix race (koran irish heeeheee) and age specifications

youll get a specific face but forget consistency, its a pipe dream now.

[dawn mulch]

What do you mean consistency? I can get consistent faces and characters using IPadapter. Wont be %100, but around %90 consistent.

[low moon]

well

for me its either 100% on nothing

Gotta take down Hollywood remember?

[dawn mulch]

I stopped watching Hollywood after the 90s.

[low moon]

oh

me too

[dawn mulch]

Are you investing in dodge coin?

[low moon]

lol no

-_-

[dawn mulch]

me neither

[low moon]

Papa Musk said forget about crypto.

[dawn mulch]

Musk was pretending to pump or support dodge coin a while back.

But i dont know much about cyptos otherwise besides the scandles.

[low moon]

Idk. Maybe he just said somethign and everyone got excited for no reason.

The Tesla Truck could use some refinement tho. It;s design is too harsh.

[dawn mulch]

Yeah I don't understand the tesla truck or why they decided to throw a brick at its windows during a demostration.

It's the idea of wildcards in SD supposed to allow you to generate random results such as random faces and so on

[low moon]

That's how you demonstarte the durability of your product. you thwo a brick at it.

Hey check out this new PC I have built, look at how it is indestructible.

Throws brick at it

[dawn mulch]

I don't know much about car nauafacturing or hype so I'm not sure if you're joking or not. It could be one of those crazy things that you just never know about unless you're in the field.

[low moon]

It was a total publicity stunt gone wrong.

[dawn mulch]

Isn't that the story of Musk's life. JK i try not to hate too much on the guy. I dontk now him well enough.

[low moon]

Yes. A series of happy accidents.

Elon Musk is a living meme.

He is a good guy tho I believe.

Overworked too.

He need s avacation.

I will never call Twitter X tho.

Sorry Papa.

[dawn mulch]

The one thing I can say about Elon at least is that he's not just working at 9 to 5 but instead doing other things even if that means he's working 24/7

[low moon]

Wow

[dawn mulch]

Not to go off too far from the topic of SD

[low moon]

thats the best description of billonaires I have ever seen.

The man is a workaholic

I try to work less than 12 hours a day

To pr4eserve my wrists that is

[dawn mulch]

Are you *saying you stressed your wrists for 12 hours a day?

[low moon]

yeah it did

[fallow veldt]

well due to the optimization you might find there, I think it is totally worth it at least I think it is the best way to go with AMD cards. I started using it to get the best of an old 8 GB RX 480, and then with a new AMD card I couldn't even run SD on Windows

[fallow veldt]

Yes, Automatic to me it is the classic SD thing, New Forge or Re Forge seems like good now, but Automatic might be the easier way to install and with more support

[trim garnet]

[wicked sonnet]

that's true

[low moon]

A1111 is like my root, all the othel SD installs mooch off of the SD model folders. XD Its the OG

[rocky meteor]

guys, can i run or use stable VIDEO diffusion without using or installing comfyui?

[wicked sonnet]

I prefer optimization over nostalgia

[grand gorge]

How to make prompts here

[trail lion]

https://stability.ai/stable-artisan

[cold cape]

What’s the best model to use for generating good non-generic guns

[wicked sonnet]

check civitai

[cold cape]

Already tried searching “gun”

Nothing

I’ll look again

[fervent thunder]

last time I searched that there were over 50

check your filters

[wicked sonnet]

guns might not be sfw according to civitai

you have this but ai models as of righht now aren't that great at pure objects https://civitai.com/models/396388/just-better-gun

[neon vector]

Does anyone know if there is a way to in A1111/Forge to hires fix an image and keep the original? So like i can hires fix the same image multiple times with different settings?

[wicked sonnet]

hires fix needs information only present during generation I've been told

[neon vector]

You can hires fix individual gens in a batch after they've been generated

but it replaces the original. I would want to be appended to the batch so i can hires fixing the original but with different hires settings

i was wondering if there was a setting to do that

[wicked sonnet]

can people... not use polls as messages?

[keen falcon]

Anyone knows what prompt to use to get a character to cross its arms in its BACK ?

[dull pelican]

Hi there, I am looking to train a lora, most of my images are in portrait resolution. Do I want them in a square resolution for a lora?

[trail lion]

what model? what trainer? if it's SD1.5 or SDXL and kohya_ss, it's not necessary, have a varieties of resolutions and dimensions is fine just remember to enable bucketing

[dull pelican]

I'm doing SDXL as the model, I would assume I'd use SDXL as the trainer. Okay so ideally I want to train it on lots of different resolutions, got it!

Are there any up to date current tutorials on training a lora with SDXL? I'm pretty new to this.

[trail lion]

not sure how up to date it is, but honestly there are many tutorials out there. Have a look at this for starters https://civitai.com/articles/391/tutorial-dreambooth-lora-training-using-kohyass

[distant kiln]

Hi. I installed Krita and its Ai tools wich are apparently managed by ComfyUi. I am able to generate and expand images inside Krita with no problems at all. I wonder if I can access the ComfyUi interface independently?

[wicked sonnet]

not sure how to access the comfyui within the krita version that's automatically installed but should be possible

[pale latch]

comfyui should be accessible by a local url still. i can't find any krita package that hides the installation of comfyui and the launching of it.

what tools are you using? i can just find a plugin that requires you install and get comfyui to run before setting up the plugin

[distant kiln]

I am using the tools inside Krita to generate images. But I want to know if ComfyUi Gui can be opened independently so i can use its nodes.

[pale latch]

the tool to bridge comfyui to krita. i can't find one that doesn't require installing comfyui and making it work before you install the krita plugin

if you're using it in krita, then comfyiu is running. it should have a local url

i'm not sure how you could get as far as that without realizing you had the UI running on your system. to which i ask which tool you are using

[distant kiln]

I think the same

[distant kiln]

There is no apparent way to use ComfyUi independently

[pale latch]

how did you install it

nevermind. you're dodgy. i give up. #🤝｜tech-support is where you can find someone with the patience needed

[distant kiln]

I did not mean to be dodgy. I just never used Krita before nor ComfyUi

[trail lion]

what he was saying was that comfyui would already be there before you installed the krita plugin, so saying it cant be used independantly makes no sense

[pale latch]

any kind of link to what they're using so i can read and refer them to proper manual page for their situation would've been nice too.

[trail lion]

yah, assume you're following this https://github.com/Acly/krita-ai-diffusion/wiki/ComfyUI-Setup

[pale latch]

youtuber tutorial aftermath. youtuber publishes clickbaits. guides people into the dark forest. abandones them because they already got the click and that's all they want.

[distant kiln]

Hi. I never installed ComfyUi. I first installed Krita and later the AI tools inside Krita.

[trail lion]

you might be able to pull them apart, but that requires figuring out how it installed it, so maybe installing comfy separately makes more sense

[distant kiln]

Thank you.

[trail lion]

for the record what you're doing is kind of backwards, you've never used comfy, but you start with a krita plugin.. probably a better approach is to learn the tool first and then start looking at plugins

[pale latch]

youtuber tutorials. building without foundations since 2007

[covert bridge]

can we talk about Flux here?

[distant kiln]

I have been using stable diffusion since 2022 locally and with online platforms. Later I subscribed to Adobe Photoshop and i am currently using AI in PS. But I am looking for free alternatives which led me to Krita and its AI tools. When I installed Ai tools inside Krita I saw the word ComfyUi in the settings so I thought: "ok ComfyUi is managing the whole thing in the background, I might as well use it independently and use its nodes and load models such as Flux Dev" . But i continued with the instalation and I am currently using Krita and generating images inside. But...i wanted to explore that thought

I have used stable-diffusion with Fooocus, Automatic1111 and Sd Forge only locally

[wise aspen]

hi I have question, using forge for flux and there is option named "swap location", should I use "CPU" or "Shared" option with Ryzen 9 5900/GF 4080/32GB Ram?

[trail lion]

dont get me wrong, the krita plugin looks like it has potential, good luck to you. comfy is a little to get used to at first. if you've ever used a node-based manager like blender's node editor, it'll seem somewhat familar. I personally started with auto1111, and it took several weeks to figure out all the various different ways of doing things

[distant kiln]

I understand. Auto1111 despite its never-ending settings falls short when compared to comfy, right?

At least that is what I understand from watching YT tutorials

I am a graphic designer by the way which is why I feel comfortable using software such as krita and Ps.

But I have done some programming as well with python and lua

[wicked sonnet]

you can link auto1111 to photoshop

there's a plugin

but maybe that defeats the purpose

[trail lion]

I think there's one for krita too for that matter

I never bothered because auto has a built in one (yes it's terrible). with comfy not really having a good working editor, it really does make more sense to use something like krita

[distant kiln]

What do you mean!?

[trail lion]

I mean if you have auto1111, I'm 95% sure I've seen an extension for krita

[distant kiln]

Impresive!

[wicked sonnet]

auto1111 has a lot of plugins

[trail lion]

I just never went down the rabbit hole because I've always been able to make the built-in editor work (or worst case edit something it gimp and drop it in)

I have quite a bit of experience with gimp

[distant kiln]

What makes you still use auto1111 when you have Comfy installed?

[wicked sonnet]

both have different work flows

[trail lion]

well I'm not really using it anymore, since I'm doing a lot of flux workflows, but I do still prefer auto...possibly just because I used it long enough to really understand it

[wicked sonnet]

neither are really superior.. I think it's like windows vs linux.. though in this case linux isn't that complicated either

[pale latch]

stability would be suicidal to enforce no flux discussion

[distant kiln]

I thought you were using Flux inside comfy!

[wicked sonnet]

:> little do you know

[pale latch]

i agree linux is just less known. like stick shifting isn't that tough, but most people just won't bother to learn it

set in our ways

[wicked sonnet]

lol

but yea

[pale latch]

its hard to know anything about stability. They've been radio silent since schmidt and parker took over

[trail lion]

well it can be, I guess it depends on what distro you run and how nerdy you are

[pale latch]

i was hoping the new CEO would introduce himself to the community but that hasn't happened at all

[trail lion]

I ran gentoo for like 15 years, I finally went to the dark side to the *buntu stream of distros

[wicked sonnet]

I've never really used linux

but I don't wanna use windows after 2025

[trail lion]

I got tired of compiling packages, essentially ,and having it break my system all the time. Ubuntu and all it's variants are very well maintained

[pale latch]

is there any real benefit to gentoo for a home machine, other than earning a badge? I got that badge too but i never really felt like i had any benefit. i still have no idea what i'm doing in most cases and getting gentoo to build and launch on my machien was just a flex when i reflect on it.

[trail lion]

honestly, no...I mean, you can really taylor it at a code level to your exact specs. binary package based distros have a generic set of flags etc to appeal to the greater population

[distant kiln]

I thought that comfy, being nose based was much more powerful..like you could do anything you can imagine...having deep access

[trail lion]

but if I'm being really honest, the nerd factor is the biggest one

[pale latch]

i went back to windows because i was tired of wrestling with my gpu drivers. I had more control over the shittiness of AMD drivers with linux so i used that, but when i got nvidia and the new Ada cards wouldn't get detected by the newest live cd's, i just installed windows 11 and learned that explorer and terminal have tabs now. i don't think there's ever any going back

[wicked sonnet]

Well, it can be more complicated

You can in practice do anything you want in stable diffusion too

[pale latch]

stable diffusion webui? find me a way to send separate prompts to the various text encoders. i don't think it's capable of it. there's no extensions that allow it

been that way since the launch of sdxl. nobody has bothered to recode the core of it to allow prompt separation

animate diff and prompt control are the two reasons i keep coming back to comfy. i prefer a standard UI instead of a plate of spaghetti as a workzone, but sometimes you just gotta have the flexibility that noodles give you

[wicked sonnet]

simplification have limits

[distant kiln]

Using which gui?

[pale latch]

it could stil be simple. it could even be done with a keyword like BREAK.

[wicked sonnet]

Well I'm using stable diffusion forge but I'm not trying to do anything special

[pale latch]

Omost sends different prompts to differetn tencs and is probably the simplest of the UIs

[distant kiln]

I am ready to get into the nitty gritty of nodes in comfy for image generation only...not animation at all

[pale latch]

easier than you think really it's just batching many images in most cases. But it is a step up in difficulty i won't deny it.

[trail lion]

not animation? I thought that's where comfy shined

I'm not an animation guy, but I've seen some impressive workflows

I grew up on auto, but in that ui you can only do what whatever plugin author allows for, with comfy, the sky (and your resources) are the limit, so insert controlnet, insert upscaling, whatever you want

and my late favorite, you can even mix different model architectures, like flux to sdxl (with a refiner workflow), and then upscale, it gets really nuts

[distant kiln]

I want images to insert in my designs. Flux seems the best-free model for that purpose.

Although generally speaking it is still beneath dalle3

[trail lion]

never used dalle, it's not a free model

[distant kiln]

There is a free version of Dalle3 in Edge web browser

That is what I use and its results are, generally, far better than Flux..but as you said it is not open

[trail lion]

I have to professional ambitions for this, so for me, the opensource approach is what I'll embrace

*no

[distant kiln]

Yes, open is better that is why I will use Flux

[fervent thunder]

on rly low hardware gentoo can be good because you can compile everything with compiler flag choices that just keep it super light and minimalist

but even then.. there are better choices for minimalism that specialise in that

I just use debian, ubuntu and ubuntu server I don't rly like linux exotics

[trail lion]

yah I use mint which is a *buntu variation, it's really easy and at my age and level of patience, it's exactly what I need

[fervent thunder]

oh I forgot about Mint!

[trail lion]

that and it hasnt broken my machine after like 3 years

[fervent thunder]

yeah Mint is fine

Fedora is nice too

and the team is friendly on IRC

KDE Neon if you like KDE stuff

[trail lion]

I did kde for a while, proton and blah blah, but I'm back to the cinnamon UI ,which is mint specific

[pale latch]

i settled on garuda linux, because it was arch, a rolling distro, and it was packed full of all the extras that the gamer in me wanted

i wasn't trying to minimilize

but that was the most recent linux tour. in the past i used it a lot for php development work, pre2008

[trail lion]

I would say that I'm like a 1/2 minamalist

[fervent thunder]

I didn't know about Garuda Linux

that sounds cool

[pale latch]

it's a cool arch distro. ilike i mean they're trying to be cool. not that it is actually cool. i think its cool but i'm often really wrong on these topics

[trail lion]

that plus my wife uses my computer so she needs to be able to use it without being a complete nerd

[pale latch]

their "dr460nized" version. i'm telling you. so f cool

https://www.youtube.com/watch?v=GG-xDgdjhjU

[trail lion]

NTI - no terminals involmed

*involved

[fervent thunder]

https://youtu.be/0AT8esyY0Fw

do you guys know the local version of this

[trail lion]

yes, it's called your computer

[fervent thunder]

warp fusion its on Gcollab

[trail lion]

you install comfyui and off you go

[fervent thunder]

hmm its slow for me.

[trail lion]

hence why some do cloud, or buy a better computer, but this kind of application is pushing moderm tech to it's limit

so there's your decision matrix

[fervent thunder]

i know right, for the first time i am puting my pc to a good use cuz of ai.

[trail lion]

I know your pain,I still wont spend 2k+ on a 4090, but I did just buy a used 3090, we'll see how good of a decision that was in the coming weeks

[fervent thunder]

for how much ?

[trail lion]

I've been using an AMD card, so I know what slow is

$850 US

[pale latch]

i grew up middle class poor so i had computers all the time but late generation slowmos. story of my life. felt so good to finally save for a good one and get it.

[trail lion]

I have the money, tbh, but I just wont spend $1k on a pci card, I just wont

[pale latch]

growing up middle class poor taught me to be cheap too so it took me a while to have money and then realize i could just dump it on something so good and way beyond my means.

mid life crisis helped

[trail lion]

I've built entire PC's for $250

[pale latch]

lot of value in being frugal (heh)

[trail lion]

so this new trend of let's spend $3k...not a joke, on a card, sorry, that wont work for me

[pale latch]

yeah but what a card

[fervent thunder]

card demands are kinda higher now

[trail lion]

nvidia stock loves it

[fervent thunder]

can run on cpu if patient

[pale latch]

really it's not a trend either. just always has been how cutting edge works. but nvidia is now just bluring the lines between consumer and prosumer a lot more. Pentium 2 processors were so cutting edge at launch and were $2000 just for the chip

[trail lion]

the irony here, I have a box sitting on my shelf, from an AMD RX580, which I bought right off the line, new...so persptive matters a lot here

[fervent thunder]

in my country i was watching the gpu stock go away i didnt know why cuz i am still new to ai.

[trail lion]

and btw I spend $800 on that

[fervent thunder]

yeah they took GPUs for AI that the gamers wanted lol

for sure it was not for video games. its for rendering big tits.

[trail lion]

so now my new card if it does anything other than crash and burn, should literally do like 10x

[pale latch]

i had the 480 with 8gb. 580 was the same card but costed way more. it baffeld me.

[fervent thunder]

i have 2060 even thought there were pretty good looking games i only thought its worth upgrading cuz of sd ai

but i still have it i will not upgrade untill i feel the botelneck... i do sd1.5 animations only

[trail lion]

I get a ton of joy from this stuff, so I really feel it's worth it, with a caveat, I'm not paying $3k

[fervent thunder]

yeah ai is fun, when u have good result is so satisfying.

[pale latch]

speaking of AI for games, wow.. i been running some of these quantized llama 3.1 models on my system . had red dead 2 running while it was doing inference, but it lagged out a bit here and there. still , i'm sure there is a ton of opportunity to build a game around a local language model

[fervent thunder]

when we gonna upscal and use ai in games already ?

i thought of pathracing in games befor it happen , yes it did. now what about ai ?

[wicked sonnet]

We already have ai in games though...

[fervent thunder]

nah i mean this

[wicked sonnet]

This? this what

[fervent thunder]

rtx remix now work with comfyui to remaster games directly

more modern stuff like stable diffusion and llama 3

hasn't come to games enough yet

[fervent thunder]

https://youtu.be/j9I0iLxGJl0

[wicked sonnet]

Well, it's mostly because it's really hardware intensive, experimental and unstable?

[fervent thunder]

yeah

those reasons are pretty much the reasons

[wicked sonnet]

Sure, people probably do that but people need to discover things for things to change

[fervent thunder]

looks like ai in games happened befor i thought of it

i watch that video befor knowing sd ai n i had no idea what it was.

[wicked sonnet]

Well.. it kinda depends on what you mean by that.. neural networks have existed for sometime

[fervent thunder]

now i know

[fervent thunder]

i meant sd ai in video games to change textures and assets.

[wicked sonnet]

how old are you btw

[fervent thunder]

i am 30.

[wicked sonnet]

I have my doubts..

[fervent thunder]

i am non native speaker thats why i sound ike a baby who just got into sd ai.

[wicked sonnet]

ah, okay, sorry about that then

[fervent thunder]

I was about to say
its non native speech yeah

[wicked sonnet]

well I'm not native but I don't remember how bad my english was

[fervent thunder]

like when I speak French I use "tu" all the time instead of "vous" and that sounds like a kid

because you are meant to switch to "vous" at the right times for formality

[wicked sonnet]

😭 I don't speak french but I can understand

[fervent thunder]

here is my idea about sd ai and games.

[wicked sonnet]

Well I kinda assumed that they were young because of kid in their name

[fervent thunder]

for example u run comfyui then make it read api of a video game to alter how that game look.

yeah the star citizen API got used this way

they hooked an LLM up to it

another popular space game, I forget the name, also had that

Elite Dangerous

[wicked sonnet]

Well still the issue about unreliability and that ai is often trained on stolen work so the legality is questionable. Steam doesn't let you upload AI generated works

[fervent thunder]

i now remember the low poly tombrider that it was said that it was upscaled with ai.

[wicked sonnet]

though I'm sure it'll get better

[fervent thunder]

i wonder if it was upscaled in game or as image.

[wicked sonnet]

DLSS 2 is AI upscaling that's being used in a few games like csgo 2 I think

[fervent thunder]

nah dlss work as uscaller per pixels.

[wicked sonnet]

Well, it's still AI and it's pretty good for casual players

issue right now is just what to do with AI

[low moon]

https://www.youtube.com/watch?v=gO3Mk3le0qs

[fervent thunder]

when i go home i will show you what am talking about.

[wicked sonnet]

you want to use comfy ui as live asset creation?

[fervent thunder]

nah i use tydifusion to alter those low poly humans in viewport , i think same thing should be done for games.

i do it for raw videos too , its unlimited fun.

[south wyvern]

Hello everyone! I'm exploring different generative tools, and there's one problem I can't figure out.
I have two images: a shot of an iPhone with a blank screen and a website design. I want to combine these 2, so design is displayed on the screen.
I've tried a couple custome composition nodes in ComfyUI, didn't workout (I don't have much experience). Also tried image-to-image but got low-quality and unstable results.
Would appreciate your advice!

[fervent thunder]

mask only the display and use control net. ip adapter maybe

or use krita for that , it has sd ai tool too

[south wyvern]

Thanks for a direction, will look into that. Do you have any good tutorial link on these topics?

[wary magnet]

Hey guys. How does one avoid pixelation when upscaling with Ultimate SD Upscaler for instance. When zooming in, the image becomes super pixelated. I'm guessing it's normal to a certain degree, but I've seen other people's images that seem really smooth. Do they just run them through editing software?

[fervent thunder]

TBH this can be done with ms paint if u dont want fancy ai effects.

[lean needle]

So the union controlnet for depth is a complete gamechanger for flux:
https://www.reddit.com/r/StableDiffusion/comments/1euz2a9/union_flux_controlnet_running_on_comfyui_workflow/
The regular depth controlnets (v2/v3) seemed to just vary between not working (at low strength) to destroying the image (at high strength)

strong recommend, try it out!

[fervent thunder]

decided to ask here since its gpu related
Does anyone know any good tech support discord?

[desert dagger]

the #🤝｜tech-support channel here is pretty good for stuff that's gpu related

[fervent thunder]

it doesnt have anything to do with ai though

[desert dagger]

so? what's the issue?

[fervent thunder]

the gpu stopped suddenly and doesn't seem to get detected by anything

[desert dagger]

and you've tried rebooting just in case something got hung in memory?

[fervent thunder]

yeah a couple of times

[desert dagger]

yeah - I would suggest posting in our #🤝｜tech-support channel and seeing if @warm junco might have some suggestions.

[fervent thunder]

Ill do that thanks

[fervent thunder]

ive fixed it
for some reason restarting was bugged so I had to completely shut it down first

[desert dagger]

restart doesn't clear cache all the way - that's why i always tell people to shut down the machine, and leave it off for at least 10 seconds so the electronics discharge. Restart deliberately leaves things loaded into memory and when you have issues, you really want to load everything fresh

[fervent thunder]

that makes sense

[desert dagger]

the most hated thing a computer tech can say: "have you tried turning it off and back on again?"

[pale latch]

restarting a windows machine leaves a lot in cache and memory still . you really have to power it down then power it back on

sometimes

usually in the event of a hardware failure it's a great idea

[old lichen]

which flux version to download for 4090rtx?

[desert dagger]

how much vram do you have?

[old lichen]

24gig

[desert dagger]

ouch

[old lichen]

I haven't been following the news, I don't know which version I should download

[desert dagger]

all i can say is download everything, and then try each of them and see what works best on your machine

[old lichen]

Has anyone run any tests?

[desert dagger]

no one but you has your machine. what GPU does it have?

are you using comfy? auto1111? webui? something else?

[old lichen]

A lot of people have video cards like that)) comfy - forge

[fervent thunder]

its hard to give advice regarding optimising low RAM/VRAM environments
because the operating system and other apps installed will also affect it

[desert dagger]

a lot of people have video cards like what? what operating system is on your machine?

your box is your box. you're going to have to just try the various models and encoders for flux and see what works best for your system.

[old lichen]

128 gig of RAM
11 win

[desert dagger]

vram is what counts

[split cape]

hey guys

[old lichen]

flux1-dev-bnb-nf4-v2.safetensors
is that to run in forge?

[split cape]

I'm having custodial questions for fooocus on windows, idk if anyone uses fooocus here.

[desert dagger]

i don't use forge. you might post the question in the #🤝｜tech-support channel

[split cape]

basically my program is not automatically saving outputs like it was the first day I used it; or maybe they're going somewhere else.

nvm, I think I fixed it

fooocus is very easy to use for beginners; I'm not sure what all I'm missing out on with other distros

I've seen people drag around nodes and have like logic systems for inpainting

[split cape]

it goes to "temp" and then when I open the history log it creates a folder in downloads, which is ok but it would be neat if it went where I wanted it

[pseudo pumice]

sup

[south wyvern]

True but I want to automate it. Things like detect screen, apply transformation, fix edges, apply lighting/reflections.

[tulip yarrow]

what even is flux

And is there an SD3/flux lora trainer yet

That you can run on colab/locally or something

and will people still be using SD1.4/5 after this?

[fervent thunder]

yes SD 1.5 is still the strongest model in certain aspects

I predicted before SD3 came out that 1.5 would not be surpassed until the end of 2024.

looks like my prediction is on the right track

1.5 is still king but once flux gets finetuned and optimized we (might) finally move on from king 1.5

yeah pretty much

[low moon]

SDXL is vice president.

[bleak matrix]

Good morning, everyone! How are we all this charming morning?

[low moon]

very good. wazaaaa

[fervent thunder]

doing good

[smoky frost]

god i hate this my reddit post on stable diffusion got removed by filters and i cant even share the image of that in here

oh nvm i shared it in sd3 channel

[unkempt hatch]

what's the best generalist FLUX finetune right now?

[river sparrow]

FLUX pro

[stark tusk]

Too bad Stability AI became so unreliable te Linux Foundation is going to release an entirely open source model with an irrevocable license later this year. Talk about cutting off your nose to spite your face! Hope Stability AI peeps find new jobs soon.

[river sparrow]

finetuned models usually have a focus point, and they are not generalist

[fervent thunder]

I follow that project pretty closely and its not really a done deal yet

[stark tusk]

Amazing what 15 ex-employees of SAI could accomplish with Flux once they were free of SAI leadership.

[fervent thunder]

apparently SD 3.1 may be competitive according to twitter

[white parrot]

hello guys
wwhat best version should i used for anime stuffs? Stable Diffusion/Midjourney/Pony Diffusion
iss there's any stuffs pls tell me what help me choising
thank you

[full lark]

Somehow sd forge is faster than comfyui on my 3060, 3 minutes per image instead of 6 , and if its a batch count of 2-3 or more it takes 2 minutes or less

[pale latch]

you have any launch options on comfyui that might be slowing it down?

hoping theres a wave of extension updates to fix all the things that gradio 4 in forge broke

dynamic prompts, one button prompt, lots of extensions that pull from where they thought stuff would be in the gradio 3 layout are now broken

oops that wasn't meant as a reply sorry leo

[solid kindle]

how does people make high quality image just like shown in civitai ? generate it at 512 x 512 then upscale it?

[fervent thunder]

yeah that works

or generate larger

[full lark]

I have to try removing lowvram option, but last time I did that it was even slower (but that was like 2 weeks ago, I haven´t updated comfy)

[pale latch]

that option would slow it down significantly

[full lark]

with --lowvram it took around 5-6 minutes, withouth it as far as I remember it took more than 10

[pale latch]

well , okay. in an ideal environment, --lowvram would slow it down significantly. if comfy had enough vram to run without it i meann.

if i understand things right too, the new forge doesn't have the same backend as comfy anymore

that's probably why it's been "abandoned" because the community created such a huge stink over it that he had to rewrite everything before progressing it

God forbid open source projects share a code base

[full lark]

I´ll try updating it and running 900x900p , at least in forge with that resoution it uses 11gb of vram (so I dont have to offload)

[pale latch]

and it's nice that flux can scale down to 512x gens too

[full lark]

Yeah, the other day I saw someone post a 256p image and it was decent 😁

[hardy nexus]

Guys it's seems Flux is going to absorb all the image gen Ai community. Especially now with gguf, everyone can have it's shot at flux. New loras and merges coming everyday now.

SAI better release something fast, before Flux is everything you can see in CivitAi and hugging face.

[full lark]

I just hope they train their model correctly this time, and they better release at least a 8b version of sd3/ sd3.1

[fervent thunder]

I'm expecting them to drop either 3 8B or whatever 3.1 is soon

[hardy nexus]

i don't lol. But i hope so.

I would not mind 8b as well, gguf it and call it a day

[fervent thunder]

I can't think of a reason why not

[hardy nexus]

i don't know, maybe because they think they could never match flux so why bother. Maybe they think that they could leave image gen AI to flux now and create something else ?

[fervent thunder]

flux isn't rly fundamentally different to SD3

[hardy nexus]

it is not. But it's already here and has gained momentum

[fervent thunder]

they both use DiT architecture, ret flow loss, and similar VAE sizes

the big difference is that flux dev was distilled from what is presumably a very large model

but stability AI could pull that trick too

I would personally rather have a model that is not distilled even if it is lower image quality

[hardy nexus]

we will have schnell models not distilled at all. Which will surpass in time even pro. Just give it 6 months

[fervent thunder]

ah yeah in 6 months I agree

[hardy nexus]

Sd3 4b 8bit gguf would be the best if we ever have our hands on it.

[fervent thunder]

I have high hopes for the next pixart

[mystic star]

Someone knows any model which it's like dall-e or anything like that to make whatever i could think?

[fervent thunder]

sadly dalle 3 has still not quite been matched in that way

auraflow and flux do okay though

and SD3 some of the time

[hardy nexus]

september ish. I really hope there is the new vae with it, otherwhise i think people will simply not adopt it, as they have not adopted better models than sdxl like kolors and hunyuan dit that have the old shitty vae of sdxl

[mystic star]

sd3 is usable with a1111?

[hardy nexus]

can be used in swarm ui and possibly forge ui, which is like auto1111 but better

[mystic star]

forge ui it's better?

[hardy nexus]

at least for flux it's even better than comfyui in terms of optimization

so i'm guessing sd3 as well

[mystic star]

ty

[winged dagger]

Yall

I need more customers anyone got any ideas for my consulting ideas.

[fervent thunder]

every new model should have modern VAE yeah

the difference is too big

[pseudo pumice]

gn

[wide acorn]

hey guys got a question, when i start tydiffusions server in max i get an error about it not being able to use all my cores it says {Onnxruntime with acceleration providers detected
Note: NumExpr detected 32 cores but "NUMEXPR_MAX_THREADS" not set, so enforcing safe limit of 16.
NumExpr defaulting to 16 threads.} anyone know what file i need to edit to use all my cores?

[fervent thunder]

Hello

I hope you are having a good day

[fervent thunder]

yes sd1.5 for life

[pseudo coyote]

Hi all, anyone uses ForgeUI?

[quartz siren]

Depends on what you mean by whatever you think but the closest is flux.

Flux is close in prompt following and better then dalle3 in certain aspects as well.

[mystic star]

anyone tried using a1111 with virtual box or any virtual machine? i want to use ubuntu but i want to avoid the double booting in my pc

[little zephyr]

Just came to ask quickly if diffusion is capable of proper text generation on the images yet?

[wide acorn]

its not great

[little zephyr]

aight cya next year

[low moon]

Flux finetunes will destroy everything out there.

[fervent thunder]

I disagree, I think auraflow and SD3 finetunes will likely be better for prompt following

because they won't be coming from a distilled base

[low moon]

🤔

[fervent thunder]

I am rly not a fan of distilled models

[warm junco]

You can use WSL2

And docker

Virtualbox or VMs won't work with SD

[bold haven]

yo

[ocean python]

Greetings, need help.
Does anyone here has "kaminoan" lora for SD1.5 from civitai?
And if yes, could you share, please?

||(Ping with notification, please)||

[jovial igloo]

Hey guys just got a new pc and wondering where to get started. Watched a lot of videos on how to install but I’m still left wondering what’s the best version to start with?
XL, 1.5, 3… lol.. just Looking to generate some anime characters and also use it to fill in some of my art sketches

[mystic star]

which version of a1111 i should use for wsl2?

[trail lion]

so I risked a refurbished 3090 with amazon's used program, we'll see how this goes, it just arrived

my first nvidia card, like ever

[split cape]

I got something called fooocus and I was making anime art last night that looked clean 🙌

I'm pretty sure it uses SDXL

[radiant tusk]

Anyone tried comfyui with flux yet?

[trail lion]

yah, it's nice, I've generated with dev, schnell, fp8, controlnet depth, and a couple loras

I did all that with a 16gb amd card, so pretty excited to try it again with my 3090

[split cape]

I only have a 2060m

[radiant tusk]

how fast does it generate

im trying to generate right now and each step is taking like 2 minutes

Using the default prompt

[trail lion]

hrm, seems like it was less than that, but I could run one now

schnell is faster obvouisly, I run 10 steps with dev and 4 with schnell

2 mins 42s with 10 steps on dev

I'm really glad you asked honestly, because now I have a frame of reference for when I swap out the card

[jovial igloo]

Also I’m coming from MidJourney does anyone know if there is there anything similar to style referencing and character referencing for more condos it any characters and styles in stable diffusion?

*more consistent character

[trim magnet]

character loras and artist loras also ip adapter

[lusty stag]

Are we able to self host Stable Audio or ai music generation on our server? if so is there a cost for this? If we are able to self host will it depend on servers from Stable servers or api

[desert dagger]

https://stability.ai/news/stability-ai-names-hanno-basse-as-new-chief-technology-officer

new CTO

[split cape]

does anyone have a quick summary of the pros and cons of different models/engines?

[wary ginkgo]

I know a little about sd and xl in terms of picture size. But what about pony? is it better version of xl?

[obtuse schooner]

Hello chat

[ornate notch]

[trim pecan]

Hey chat

[warm junco]

The original one will work

[abstract quarry]

not better. It's trained on anime and danbooru tags. So it does not understand normal language anymore, but you can use tags to precisely control the image. It can do all kind of pornographic and weird hentai stuff - that's probably the reason it's so popular. Although some people claim it's also good with arts.

[abstract quarry]

I guess you search for IP Adapter

[lethal sierra]

how can i cancel my subscription?

email support doesnt answer

and you keep trying to charge my card every few days

im not able to cancel it from website because, it keeps redirecting me to provide new payment details

url in email about unsuccessfull payment:
https://assistant.stability.ai/
i click, log in and it redirect me to:
https://stability.ai/stable-assistant#choose-stable-assistant-plan

[low moon]

They're all busy fluxing.

[last bay]

is anyone using FLUX checkpoint on Ax1 and help me?

[coarse dome]

HEllo evryone

im nEW here

[main snow]

I diffused in a stable manner

[lusty beacon]

Any mods here?

[mighty raft]

does this basically mean it's prefinetuned?

theres a number of things wrong with flux.

1. no negative prompt (unless you use their paywalled pro version).
2. it cant do a lot of styles like SD3 8B and im afraid the finetunes will have more narrow uses cases too for that reason.
3. im not sure if the vae is as good as SD3, i just know SD3 has finer details.

but hey, the positive outweighs the negatives for now.

[undone garden]

How do you queue up multiple lora training sessions overnight? I use ai toolbox and it's cmd based so it should be pretty easy I thought. I just copied and pasted multiple lines and pasted them together, but I have to press enter after every single one is complete and I can't find a way around that

.\venv\Scripts\activate
python run.py config/whatever_you_want 1.yaml
python run.py config/whatever_you_want 2.yaml
python run.py config/whatever_you_want 3.yaml

I'll try by using & tonight, do you think it'll work for my scenario? I wouldn't want to try random stuff and lose training time

[ruby crown]

How do i use X/Y/Z plot without it using a grid? Since it desperately wants to save as a grid i can't use it because the grid would be absolutely immense. I just want to save individual images with variation. I've made sure to remove save as grid everywhere that i can find, in the UI + in x/y/z plot....

i've tried "sub grids", on and off. it doesn't matter.

i've noticed if i only select Samplers, it will continue, but if want a range of denoise strength or cfg value, nope, it wants to save that as a grid and it's not possible to do.

[fervent thunder]

in some ways yes

[fervent thunder]

there is a very recent update to comfy called execution inversion that is ideal for this

[shadow oyster]

Check #🏞｜general-with-images message

[ruby crown]

ok, no interest in comfy ui though.

it does seem to work, i just had to set some insane values in the UI for saving grids. really stupid that you can't untick it. i've never ever had any use for grids in the 2+ years i've used SD.

[fervent thunder]

if you don't want to use comfy, you could do it in Diffusers

[ruby crown]

Cool, where would i find it? is it an addon?

[fervent thunder]

this is Diffusers https://huggingface.co/docs/diffusers/en/index
its a python library

[ruby crown]

ok. nah, i'll look for some other addon i think. thanks anyways.

[fervent thunder]

ok sorry I am not good with the traditional GUIs

[white parrot]

hello everyone

[fervent thunder]

hi

[white parrot]

how to know what vae fit for my checkpoint

[fervent thunder]

which model?

[white parrot]

i;m try using nyanmix but idk what vae fit most for it

tho it say try using same vae with Counterfeit/OrangeMix/RefSlave/etc use but i'm too noob to know

can you help me?

[fervent thunder]

maybe try this one https://civitai.com/models/22354/clearvaesd15?modelVersionId=88156

[white parrot]

thank you

[desert dagger]

you could try pinging them on twitter i guess. or get your bank to put a stop payment on yoru card or something

[short scroll]

Oh how do i do that bcs i got the same problem and i dont really use twitter so can u help me

[true canopy]

i cant add any lora to flux on forge 😦 anyone else got a simular problem? "moving model" is taking forever

[desert dagger]

to start with, no one from stability.ai is really on this discord at all. the devs all left several weeks ago, and emad was the only management that was ever here. so posting here for help with your stable diffusion account is really not going to do you any good. you can try to get ahold of the guy that's in charge of #🗣｜artisan-support-feedback artisan maybe, post in the feedback channel and maybe he can help

[desert dagger]

are you sure the lora you're trying to add is FOR flux?

[true canopy]

it says Flux.1 D

[desert dagger]

maybe - don't use forge? switch to comfy

[true canopy]

im using flux1-dev-bnb-nf4-v2

[desert dagger]

yeah, i think that one really will only run in comfy. I think people have had issues with it in auto1111 too

[true canopy]

i mean, im not sure if it works, its been loading now for like 10min

maybe i rly should just try comfy

[desert dagger]

look in your console log - what does it say?

[desert dagger]

you really should.

[white parrot]

idk but anyone can help me at #📝｜prompting-help message

[true canopy]

it just says "moving model(s)"

[white parrot]

oh nvm i saww it

[desert dagger]

wonder where it's trying to move them to

are you out of space anywhere?

[fervent thunder]

guys please, is she real or AI
https://youtu.be/6PvbVT4_EvU

she replied that she is real.

[true canopy]

i dont think so

[desert dagger]

her first video was a year ago. could we make that sort of quality a year ago?

[true canopy]

lol , an image is coming thru now, its just crazy how slow it is

[desert dagger]

look at her other videos where she's writing out forumlas and then see if you can get any AI video generator to do that without messing up the forumlas

[desert dagger]

what's your gpu? how much vram do you have? etc

[true canopy]

i have a 4060ti 16Gb so i dont think thats the problem

regular generations take like 30sec or so

[desert dagger]

i do too - though i run comfy not forge. maybe the issue was that it had to copy stuff to other directories

[true canopy]

maybe, it is the first time using lora, ill just see what happens when its done

[fervent thunder]

she is perfect then ?

must be her big eyes making it look like ai

[desert dagger]

go look at her other videos and use some common sense, dude.

[true canopy]

after 20 million years i got an image

[fervent thunder]

https://youtu.be/0xO5dBja118

[desert dagger]

it's like the people saying we didn't land on the moon, because we can do special effects now. but when it happened we couldn't do those special effects at all, couldn't even think about it. but because we can do it now, obviously it was special effects

[fervent thunder]

see how her blink aint natural

[desert dagger]

her first video was A YEAR AGO! we couldn't DO THAT A YEAR AGO

[true canopy]

omg, 30sec without lora, with lora 13min, something is not working correctly lol

comfy it is

[desert dagger]

what does a lora do, and are you sure you want to use that one?

[fervent thunder]

we had lcm and ip adapter year ago right ?

[desert dagger]

video. think. we could not do the sort of videos a year ago that she posted a year ago

[true canopy]

well , i guess loras are kinda new to flux? maybe its broken

[desert dagger]

why are you using a lora at all, to start with

[fervent thunder]

she is using a god faceswap model

[desert dagger]

oh go away

yo're not real either

[true canopy]

why would i not want to use loras, they are awesome

[desert dagger]

tell me what loras do. what they are for.

[true canopy]

well i sure cant explain it but they change the style of the images generated

[fervent thunder]

she removed my coments about ai

[desert dagger]

not going to argue with a badly written AI bot. go away

[desert dagger]

a LoRA - Low Rank Adaptaion - updates specific weights in the base model. the only reason to use them, and add all that extra computation in, is for specific reasons. maybe you want to make pictures of your dog. the AI knows dog, but not YOUR dog. so you train a lora of your dog. or you want to make pictures with your face. so you train a lora of your face. just dropping a lora in without even knowing why will get you nothing extra and can cause you problems. the weights you want to use might be updated by it in ways you don't want.

[fervent thunder]

go away is a deep statement.

😦

[desert dagger]

then use your brain, go look at her other videos, watch them, think back to the technology we had when she made them, and realize how foolish you're being now

[true canopy]

ofc i know what i want, thats why im im trying to use a specific lora that someone else has trained to try and get the generations i want

i havent trained my own lora, im trying to use one from civitai

[fervent thunder]

on another news i am back working on 3d stuff... its been long ago since i touched a 3d tool.

i must say alot have change to the best.

[true canopy]

like a spcific painting style, i dont think i can generate images like that without using a trained lora

[fervent thunder]

u can use prompt for that, but u have to have wide range of paint styles.

if u dont know this stuff like me , use wildcards they have all styles per artist names.

[desert dagger]

you might want to train your own, then. Matt has a really good video on his channel where he trained his face for flux. just search for Matt Wolfe and then look for the video with him as superman for the thumbnail and the title "how to make images of yourself" - civit's trainer doesn't seem to work well. there's a flux trainer on repilcate that works extremely well

[true canopy]

thank you, perhaps i will in the future, but i think ill try out comfy first, since forge seems to be having problems with loras

[desert dagger]

you might also want to join the L2 discord, that's matteo's - there are a lot more technical peopel there than here, and he writes nodes for comfy

[swift heron]

wasn't there a channel on this server previously that explained how to get started (locally) with SD? Did it get removed?

[river sparrow]

there are enough guides on the web available, depending on which ui you prefer. do a bit of research first, in case you don't know which ui to pick.

[burnt geode]

Hey ppl, just starting on training, but theres so much info it gets overwhelming, does anyone has a tutorial on how to start on generation and train? I want to make ai of myself

[peak tendon]

why are you obsessing on if a girl is real or not lmao, it looks more like she has instagram filters on her

but i don't understand why it is important, she's explaining programing

sound a bit creepy

anyways, I came here to ask
I'm getting out of space in my main external HD due to stable diffusion, would it be okay if I just... Drag the whole folder to another one?

or would that cause problems?

[bleak matrix]

Good afternoon, folks! How are you all doing

[fervent thunder]

hi

[pale latch]

Something i've noticed about the newest forge if anyone else thinks they're going mad. if you change the prompt, it may take one generation for the new prompt to have an effect. it seems to use the one still cached if you change it for one round

[desert dagger]

nope. you move stuff, it'll break. if it's comfy, you can put your models somewhere else and use the extra paths .yml file to point to them. if it's not comfy, ask in #🤝｜tech-support how to handle it

[bleak matrix]

Hullo! Waves

[fervent thunder]

yep

https://youtu.be/xh_jhpRp89A
ai upscaling in gta san.

[peak tendon]

Lol Su Xi upscaled looks like Joey Sturgis (very well known metalcore producer)

[fervent thunder]

what time stamp ? u think its cuz ai used its pictures for upscaling ?

random seed = Joey Sturgis

[winged dagger]

GUYS

who wants to buy my courses

i need customers

[river sparrow]

spam

[desert dagger]

you need a marketing course yourself

that's the thing you need to actually TAKE and learn how to market

i can tell you to go to google, search on online marketing course and then sign up for one and ace it. learn how to market, cause what you're doing right now is going to get you zero customers

[warm junco]

In #🤝｜tech-support you'll find local install guides

[pallid quest]

hi everybody

[main snow]

Suggest your favorite loras

[raw fable]

hey guys I just launched an app to generate + edit images using Flux. it's like ChatGPT interface

could i get your feedback? https://rubbrband.com

its using Flux Pro + a bunch of editing models under the hood

[fervent thunder]

see this amazing addon in AE

https://youtu.be/fgody1uMqko

image to 3d using ai

[drifting briar]

pretty cool ngl, clean interface and works well

[keen root]

Scene 1: Opening - Problem Introduction

Shot 1A: Close-Up of Desk with Snacks

Camera Movement: Slow zoom-in.
Visual: A young professional (early 30s, casually dressed) reaches for a bag of chips on their cluttered desk. The desk has a laptop, a notepad, and a few empty snack wrappers. The lighting is soft, suggesting the end of a long workday. The person hesitates as they pick up the snack.
Expression: Frustration and tiredness.
Shot 1B: Mid-Shot of the Character

Camera Movement: Still shot.
Visual: The person puts the snack back, looking disappointed, and glances at their reflection on the laptop screen. The reflection shows a tired expression, hinting at the need for healthier choices.
Text Overlay: "Are unhealthy snacking habits slowing you down?"
Scene 2: Introduction to Solution

Shot 2A: Wide Shot of the Kitchen

Camera Movement: Smooth pan from left to right.
Visual: Bright, modern kitchen with a clean, minimalist design. The Kutchina Nutrimix Pro, a sleek, portable blender, sits on the counter. The kitchen is well-lit with natural light, creating an inviting atmosphere. The young professional enters, now with a slightly brighter expression.
Expression: A hint of curiosity and hope.
Shot 2B: Close-Up of the Nutrimix Pro

Camera Movement: Slow zoom-in.
Visual: The person’s hand gently rests on the Nutrimix Pro, showcasing its compact size and stylish design. The blender is highlighted against the blurred background, emphasizing its portability and modern look.
Text Overlay: "Meet your new on-the-go companion for healthy living."
Scene 3: Product Overview