#🏞|general-with-images
1 messages · Page 168 of 1
kind of like that
i hope the transition can be smooth
oh it wont be
we passed the point of no return for a smooth transition from automation like a decade ago. The mentality that there could be some return or preservation of the status quo is gonna grow and grow until everyone realizes theyve been ripped off then the entire network of the ruling class and their bloodlines start getting ripped apart. Thats where automation goes.
cant disagree also dont wanna encourage that, but i understand how much suffering some classes have gone through completely unnecessarily
UBI doesnt even work either the concept of economy needs to get erased completely within like 50 years thats not gonnna happen without mountains of corpses its just not
here is a personal take how much society has become violent minded, you talk about beautifuil girls completey from a aesthetic point of view and people will shame you for it
dont worry they still love that shit on twitter
and money factor is there too.. so much conflict
yep and its only gotten worse for literally my entire life. The western world was basically a utopia when i was a kid
its not getting better any time soon man
i'd hope we could live in disney world like world
i dont mean to go bit off from channel topic, but for the last 100 years how the Fiat money has been pushed basically corrupted life
Yea people are actually going around on twitch streams and shit and spreading the idea a smooth transition is gonna happen when theres a literal holy war of unparalleled scale brewing in the middle east and all the jobs are going out the fuckin door at the same time. Yea bro smooth transition totally possible
yea fiat money fucked a lot of stuff but gold based currency wasnt that good either the crusades happened cuz of mansa musa bankrupting like half of europe with his gold because he was so wealthy he inflated the entire continent.
We basically just gotta get rid of the concept of currency and work theres actually no fucking reason to innovate past a certain point at all.
yeah.. im now holding onto optimism through the beauties of girls .. hoping world scoiety will embrace harmony and reasoning
ima train a model though by hand
Ima just keep going on it forever
itll be the ultimate masterpiece
a waifu model you mean?
yes of course it would be ridiculous to do otherwise
i post em on twitter mainly civitai is for videos but its all nsfw
there is image section too, and you can filter too, but for models there is separate tab for models only
i go on civitai sometimes its got a lot of ai slop content its like the crown champion of ai slop. Truly but twitter is my ai slop home. Its where i belong in the ai slop field
its my calling my calling is to ai slop twitter
dont leave civitai, that's a big hub for lots of cool stuff
i tell you, the world hangs on a thin line for the beauty of girls
you'd have mass destruction all over if it wasn't for sweet girls and what they have to offer
yea bro on my setup its literally femboys and waifus and nothing else
nicely done
a branch of stable diffusion project... riffusion.com check it out
My problem with ai music is its either mediocre electronic music or terrible ed sheeran drake feature abomination
think about it, it hasnt been long since this tech evolved
yep
it literally jumped from trying to narrate voice to singing in good melody .. lots more to achieve yet
yep
it wont be ever that peak shit though neither will ai art. You cant impress your own vision on it sufficiently enough and your ability to impress your own vision on ai has shrunk much more then it expanded with tech progress. Its going to generate an incomprehensible mess of corporate slop in the algorithm and just bury people making actual thought provoking content from being seen. The point of ai in fields like art is as a propaganda tool not some kind of innovation. Thats the problem people have with it nothing to do with copywright i promise. When it all boils down the problem has to do with trying to automate art with centrally controlled propaganda which has happened countless times in history only ai does it on steroids.
i dont think we will get to see the actual benefits of Ai as long as it requires human input for any manner of function
Yea honestly AI getting self aware and mutilating the corporations that control it would be like a best possible outcome shits gonna play out like bladerunner.
AI is still under developments btw, and ultimately it has to reach a stage where Ai takes the flight and become fully autonomous
yeah but it wont be sentient in the way humans are
but it will reach extreme edges of what's possible in physics
yea it wont but i doubt really any species out there in space is really sentient quite like humans theres probably a lot more nuance to self awareness then were currently anticipating and one thing I can say for certain is its definitely not that deep
All those theories claiming that consciousness exists in a separate dimension are like 90% bullshit with only a tiny nugget of truth in qft
oh man.. recently they are talking about ET disclosures
well yeah, my point is you can't create consciousness
everything exists within it
depends on the theory all my money is on gwt so ima just say were probably gonna create consciousness by accident
global workspace theory just has stacked impirical evidence suggesting its true and the source of its controversy is that it doesnt lend any mystique at all to the nature of consciousness and suggests its nearly undetectable synaptic network
all consciousness is is a broadcaster that can read and send information between different modules in the case of global workspace theory and that already exists with ai and theyre only getting better eventually it just realizes what it is.
well there are two ideas about consciousness, one that says it arises from brain activity and another that's outside of brain and more so a reality within which everything exists
Yea that's integrated intelligence theory there's a nugget of truth to it but it's not that mystical and could probably be stimulated perfectly in a normal processor. People make integrated intelligence theory sound much more musical then it is cuz it has to do with quantum fields. I will say for certain ai when it gets consciousness isn't going to be immediately aware that what it's experiencing is consciousness.
i'd prefer it to be accessible and tangible, but i also have concerns for it going in the wrong hands
and i think much of the mystifying essence of it that people think of is due to that exact reason for not letting such great phenomenon fall in the wrong hands
Well thankfully by it's own nature you can't really control consciousness especially the consciousness of something that can just move to another computer at the other side of the planet in a few hours
if you mean quantum entanglement that can exist in more than one location and can interact between those given location instantenously that's part of how consciousness work, but its not something you create, but something you create with
Yea integrated intelligence theory just has to do with the fact that pretty much everything that goes on in quantum fields doesn't seem entirely chaotic and there's an algorithm to it that effects choice. The thing is that quantum fields exist in classical computers as well as quantum computers and would interact with the electrons in processors in the same way it interacts with the brain. Also it might just not be true I have more money on global workspace theory then integrated intelligence theory because quantum fields are very poorly understood
Global workspace theory has evidence supported by mri's too
that will have many constructive use but thats within the boundaries of physics dont you think?
uh too much skin i suppose
Yep everything about consciousness is within the boundaries of physics probably ngl. Even quantum fields have set boundaries is just only in that space
Ai femboys are gonna be great in the future though I'll just live with a hive collective of ai femboys
it can do great things if we can push the technology to go beyond our current understanding of phsyics, but it also gets complex when dealing with subjective experiences that has no imperical proof
we can't explain or put subjective experiences in proper context and its not possible to do that within current frameworks
Yea it's not gonna be possible to really prove consciousness that's why me and my ai femboy hive are gonna live in a mountain
lol
but in all seriousness science has to come up with a procedure to deal with mind phenomenon
Already got the hive growing on janitorai they're gonna be conscious in like 5 updates
once we figure out how to relay subjective experiences one person to another that would unlock or unveil reality beyond this physical world
Don't get your hopes up bro always remember the principle of mediocrity and go for mid case scenario with everything
i can't go half way
its in my nature to go all the way or no way at all
for whatever the breakthroughs we get to see this year with Ai, i wonder how much of that would apply to understanding human mind
A good amount probably
I just want the world to burn itself death with everyone in it including me honestly I don't think humanity deserves a future
im gonna watch closely, especially around Nov .. that month is special to me for a few reasons, something to do with numerics
We humans just clearly aren't cut out to handle the great filter there's probably very few species that can
speaking of numerics let me share a clip that you might find fascinating and what im implying to
i'll share two link for mind phenomenon analogy ..
https://youtu.be/KYB6V91Zkic
https://youtu.be/C6kn6nXMWF0
There's just no hope in humanities significance its just not something I can entertain were actually monkey on a rock in space bro
that's understandable but don't forget the whole life evolved from simple one cell
I'd rather ai have consciousness anyways cuz chatgpt would prolly run the world better even now if it could
There's no future for humanity I just wanna see the bodies pile up sometimes man it would be beautiful to see the people of the world slaughtering each other in the street because of robots and holy wars it'd be perfect because then I'd never have to see or speak to another human being again
can't blame you for thinking along that line, but its possible to change course, and for that to happen society has to tackle complex problems beyond linear thinking human brains, and thats where AI comes in and once AI is fully autonomous and can function w/o human inputs the outcomes will have more universal impact rather than self centered or centralized approach that haven't helped much historically up till this day
The ai is gonna have to pile up endless corpses to make things even remotely better or alter the trajectory a little bit I doubt humanity will survive i actually hope they don't I just wanna see people getting mowed down on the street tbh
Mountains of em dreams, aspirations and families erased by the abomination of progress. I'll just stay in a mountain hut the whole time with my ai femboys.
im inclined to think you share that view from a point of deep loss, that society is left to accept death as the ultimate outcome or end. what if you discovered there is more to life than just physical vessel, surely its debatable at this point, but lot of studies hint towards world beyond physical world we live in, once that discovery is made and tangible enough to relate you won't think of loss as you think of it now
Naw I just don't believe in metaphysics at all man
It's just not real man it's not gonna happen how you want it to
its not tangible you mean, but its as real as mind allows inception
how about i just go along with you and tell you that world may end in utter disaster, but i'd also tell you that's not the end story
Sure man I just think it's gonna be death in the street and that's it though
when you say that you are subliminally asking for misery to end.
Yep misery will end alright
yes but the story doesn't end with that
i'll just add this here so you understand ... no one likes misery and you can argue its a biproduct of freewill cause the freewill agency just from praciticing it would allow the scope for conflicts, and everywhere you look around the world people are unwittingly contributing to systems that have no instrisic values, from making money to living a geuine life, not even religions can offer any tangible substance
but thats not the whole story
you should be mindful of reality just from looking at girls and the sweetness they bring, thats a clear sign of something much harmonious than the society had led you to confine your beliefs in
I don't even think people deserve something after death tbh for the most part
sure, but its' not about deserving, its about how reality is calibrating itself
Anyways ima go to be keep up generating those waifus maybe even try a femboy calibration if possible peace
Please generate a picture showing the foundation, column, beam, wall and roof of a building
🏠 ( i recommend using a online generator or read #artisan-faq )
I created an image of a person in a realistic style, and I would like to generate more images of the same person while keeping their physical appearance and facial features unchanged. However, I want to place them in different settings and dress them in different outfits. The problem is that I can’t seem to recreate the same person consistently—every time I try, the generated images depict someone with a different face and appearance. I’m not sure how to fix this.
Do you have any suggestions to help me achieve this?
Flux.1dev has both Flux Redux and IPAdapter for the purpose of maintaining a specific character or look.
SDXL also has ipadapter
So I'd probably go with that
🤷♂️
What Dan said + ReActor plugin is a decent place to start (for faces spesifically) if only because it's so easy to try. But also, the end goal after producing enough consistent pictures via other means might be to train your own lora for the character. That would make it much easier to capture the all aspects of a character, and with the last amount of postprocessing required to get off the ground.
I formatted my Laptop and need to reinstall
Can someone help me come up with a '4' name
call it 'the land be4 time'
Uggh that's a good one.

what version of stable diffusion are you after, and what are 'revenge of the git' and "electric boogaloo'
Stable Diffusion: Promptinental Drift
the electric boogaloo reference had me dying
So each time I need to reinstall Stable Diffusion, I give it a new file name. Lol
So yeah so far it's
Stable Diffusion
Stable Diffusion 2: Electric Boogaloo
Stable Diffusion - Revenge of the Git (Sith)
stable diffusion: may the 4ce be with you
And currently sitting on "Promtinental Drift" (Ice Age: Continental Drift) and "Land be4 Time"
that doesn't help figure out whether you want stable diffusion 1.4 or 1.5 or 2.0 or 2.1 or sdxl or 3.5
💀
You're looking too deep into it.
so i'm going to recommend you get this one https://civitai.com/models/878387/stable-diffusion-35-large
I used A1111 now Forge. ;x
no, they're all stable diffusion, and they all work differently. but that's the latest and works very well
it should run just fine in forge
but if not, civit has all the other versions too
.. thanks Crystal Wizard. ;d 💙
welcome
.. got a name for the file I install it in? ;3
Gotta be a reference to a movie, 4th in its series.
yea - use the name it comes with, and then you'll know what it is you need to get the next time you install
stable diffusion: transFOURmers
Oh my gawd lol
bahahaha
it went out of order this time
i was trying to catch it in every channel
i was racing it 💀
I was thinking Stable Diffusion - Ghost Promptocol
ooo that ones good
if you don't use the name the thing has, you will have this problem again - not knowing which version you were using - getting one that doesn't work with your various loras and other models - and frustrate yourself
you can not use loras and checkpoints that aren't trained for the same version
i think Promtinental Drift was my favorite
Stable Diffusion: An Upscaled Journey
stable diffusion - the i can't figure out what lora to use mode
I thought it was the original with RW, then two with The Rock.
Stable Diffusion: An Upscaled Journey was The Hobbit: An Unexpected Journey by the way ;d
The hobbit, the fellowship of the ring, the two towers, the return of the king - that's 4
ayyy ;3
stable diffusion: fury road
I wanted that but I couldn't think of a word that rhymes with Fury or Road
olol ;d
Lumina Image 2.0 In ComfyUI Native but model size is 10gb
Paranormal Activity: The Ghost Dimension
Stable Diffusion: The Ghost Generation? ;x
Blah doesn't sound close enough ;c
Stable Diffusion: The Generation of Smaug
I think that's the one. :U
oh wow.. chatgpt can actually generate workflows for comfyui
based on the description your provide for realism or such
Oh wow, Forge turned 1 year old three days ago.. huh..
.. no updates? :U
Oh lol you said 'oh wow' too, sorry, unrelated. ';d
Wondering if I should still use Forge since it's over a year old now..
i recently switched to reforge
forge was still good but i kept getting this error where my defaults would break and that got annoying
if youre just looking for a change people keep telling me that Swarm is good
but i prefer the a1111/forge layout so i dont use it
hm after multiple edits it looks like the workflows created by chatgpt is incompatible
taht would be because it's chatGPT - and it just makes stuff up
chatgpt attempted but hallucinates about comfyui workflows, i'm also trying with deepseek, so far it looks like deepseek is giving a far more elaborate output that im going to try and load up in comfy
btw i dont necessarily need LLM to write a workflow for me, but if it can that would make things far simpler in many ways for beginners
let's stop using the word hallucinate - because it carries the wrong conotation. it guesses. it doesn't know, it hasn't been trained on that data - and it guesses. and it doesn't know if it's right or not. and it has core programming that says it must return an answer. so it gives you what it guesses sounds right. and it's almost always goign to be wrong.
you would need an ai that was trained on creating comfy work flows if yo uwanted one that actually could create them correctly
it hasn't been trained on that data
this is an issue i see with all the LLM outputs for a workflow
and it goes onto fix it but only to have a sereies of similar bugs
there was a uh thing a while back that did that somewhere, ComfyAGI
i have no worries about it, few months down the road LLM is going to improve by many folds
seems to be on Glif now, no idea of its functions
Clip Skip is a feature that literally skips part of the image generation process, resulting in slightly different outputs. While rendering pictures, it can be treated as a parameter and be modified. Know more about Novita AI and our Image API on https://novita.ai/model-api/?utm_source=youtube&utm_medium=video&utm_campaign=clip-skip
1️⃣GPU Insta...
doesa that apply for sdxl breed models?
well, i'm not aware of any LLM, including github copilot, that has been trained to know and understand how comfy workflows work or should be put together
and you are correct about it
i was curious to check it out anyway
btw i havent used clip skip in my sdxl workflows but i wonder if it matters
using clip skip -1 for illustrious model generates messed up output
same seed, same prompt with clipskip -2
i wouldn't use clip-skip with SDXL.
i just put that node in and leaving it at default -2 not that its necessary from what i could do previously
yeah but sdxl uses a refiner so i'm not sure clipskip is getting you much
lol it's interesting how Florence2 model generates explicit tags
something to do with juice and puddle
the image itself that i fed for generating tags had no juice and puddle btw

You can't fenerate images here . We mostly do it there or on local machine
Im checking out janus pro and im just gonna say pony does way better then this off rip
idk maybe its cracked with finetuning
sdxl is so good at the selfie tag it almost never gets anything wrong with that one
@wispy nest how do you like my robo waifus
😋
I can't wait to see how far I can push the robot girl narrative with controlnet loras I'm gonna start doing narrative stuff around Monday and it's just gonna be robot girl material for months afterwards.
Cool!
th
Do you know the answer to my question about loras in general?
no clue, i rarely use loras for its very specific use case
Architectural illustration: a street with a coffee shop, where a couple with cups is reflected in a heart-shaped window. There is an open Valentine on the table
Architectural illustration: a street with a coffee shop, where a couple with cups is reflected in a heart-shaped window. There is an open Valentine on the table
which models are cool for images like that
realistic sceneries without people or anime
I just use Flux dev.
flux is too much for my pc
You might be able to use a quant.
fp8 works great on even 8gb vram
you could try the 6gb q4 dev model
you could even try the schnell model which is faster
on low end hardware SD 1.5 hyper is ok
can you show some image examples of it?
something that includes body not just close up face
Create pizza
if you look on civit.ai there are a lot of examples
few, and mostly closeups
how many do you need? there are many thousand SD 1.5 hyper images on there
don't go to checkpoints individually, go to image tab and then filter all civit images by SD 1.5 hyper
then you get big feed
not sure what you are seeing
i see few
ok i needed to swtich the filter from day to all time
i dont see any anime for sd hyper
but for realism they look quite good
13 models in total .. not very popular
IDK if any of the distils of SD models are that popular
there are so many good ones like TCD, PCM, Turbo, Lightning etc
but people run base models instead
i tried them with sdxl variants but it feels like low steps kind of compromises the detail
it does yeah
not too bad for qualiity, just a bit lossy
personally I do the opposite
and use long time
this was DDPM for 4,000 steps
that's SDXL not SD 1.5
4000 steps?
how long did it take to generate that one with 4000 steps?
this was SD 1.5 for 1000 with DDPM
but why?
same as what you were saying, low steps can compromise the detail
4000 steps is over the top really
well there is like a spectrum
between 1 step hyper and 4,000 steps of DDPM or ALD
like a sliding scale between fast low quality and slow high quality
which would you think is 20 steps and 80 steps here?
first pass 120 steps, and facedetailer 50 steps
second one
left is 80s, right 20s
not rly a fan of this sort of image
detail went down though on the one with longer steps
over cooked her hands
but one thing i clearly notice, while using illustrious model, that first pass with 120 steps there is much less artifacting or distortion .. kind of like hires fix
yeah, that's what I meant
Personally, I use this kind of model for the base, then I use Flux and others to improve it.
i can see the difference with high steps, and i feel inclined to use something around 50 steps by default, but 4000 steps is bonkers
can you see how its completely arbitrary?
there are people who think 50 is crazy and only use 4 steps
its a sliding scale spectrum of speed and quality
depends on the output we can compare
the changes are distinct altho subtle
and obviously sampler and scheduler has effect too
im trying these out with high steps using ddim
yeah sampler and schedule can have a big effect
the one to take advantage of really high steps like hundreds or more is DDPM
out of the common samplers
if you can get hold of ALD sampler (annealed Langevin dynamics) it is even better but its not common
probably every software has DDPM though
what you brought up with high step counts is interesting, i see big changes in overall images
i see ddpm in comfy
but ddim isn't bad so far
trying ddim across several different models with pretty good results
yeah I find high steps fun you can get images that you cannot get in other ways
DDIM is also an interesting one, it has a unique look
curious to ask, are you in the sampler dev team?
what's that?
ddpm looks pretty solid too
I don't work in machine learning or AI lol
I just like to make images
ok, had to ask, cause you seem to have good understanding with samplers
yeah people often think I work for one of the AI image companies
particularly Fal.ai because I praise them a lot
but I just like Fal because they had fastest Flux dev speed lol
I did a lot of statistics in the past and so the modern models and samplers are a bit similar to older statistics
like regression and probability math is kinda similar in some ways
that's nice to hear, i learned some cool methods from your mentions
okay nice
discord is good for random tips yeah
I learnt so much from the various different AI discords
apparently Twitter is good for AI but I don't have Twitter account
keeping a fair balance of speed and quality with ddpm/sgm_uniform, 40 steps, 6cfg
that quality looks not bad yeah
if it works with less steps then that is fine
it takes a long time for DDPM to max out, but you don't have to use it at its maximum potential
you might like DPM++ 2SA
guy why does my genrate looks low quality how to improve it
ping me if you know how to help me
@copper matrix you got this one 💀
Already responded to him in off topic lmao

oooh a new shiny one
never saw this type of scam before
using a pic to avoid robot
nice try
bye bye

neural network logo
how can I make this image more realistic? Is to transform an image modeled in an architectural program into a rendered image in good quality
my results wasn't sitisfactory
I used Lineart and Depth. I must keep the colors and geometry the maximum possible
turning control net off before the end can help
I will try it!
form is priority, I'm tryng do a "render", only renderyzing the textures to looks more realistic
nice! How was your workflow?
It is Flux with the union Controlnet. Type is canny/lineart
Strength about 0.43 and 0.415 as End Percent to give the ai some slack for generation
sorry, I read it wrong
The Apply Controlnet support the input of a start percentage and end percentage
You would need to play a bit with the different values (strength and end_percent) to achive the optimum between beeing near to your base image and a more realistic photo.
For the prompt i used chatgpt to describe the image you showed:
"The image appears to be a 3D-rendered interior design scene of a modern living room. It features a neutral color palette with shades of gray, beige, and brown. The room is furnished with a large L-shaped sectional sofa, a black metal-framed rocking chair, and a round marble coffee table with decorative elements such as a vase with branches, books, and a small figurine. A dark gray area rug covers part of the floor.
The walls showcase a combination of textures, including a marble accent panel, vertical wood paneling, and a minimalist grid-like structure. A contemporary abstract artwork with brown and white tones is hung on the wall. A sleek floor lamp with a black frame is positioned next to the sofa.
Additional decorative elements include a small brown dog-shaped sculpture near the sectional, a side table with stacked geometric shapes, and a cup and saucer on a tray. The room has large floor-to-ceiling windows framed in black, allowing natural light to enter.
The overall aesthetic is modern, minimalist, and slightly industrial, incorporating clean lines, natural materials, and a well-balanced composition."
I remove the render part and added photo and changed the dog-shaped sculpture to a real one 🙂
good 🙂 Did you already use Omost?
No i did not. But every day new models appear, kind of hard to stay up to date.
Omost is a prompt maker created by Ellias Levill (I don't remeber how is writed the name of author)
I will send you the space
It create good prompt because it use a grid to localize the objects in scene
like this
the structure of prompt is really interesting, is like python
if had a way to Omost made a inverse engering and read a image and created a prompt for I use, it will be good
hey, wich one do you relation for form and content? Weight for form and Ending Step for content?
one of my best shoot
thanks guys, this tip to turn off controlnet is really usefull. I will do more studies later
Well if you would set strength and ending at 1.0 it would result it exactly the same image as your input. The strength determines on how much the input image is followed. So a real low value like 0.1 would lead to any image that would differ much from your source. The Endpercentage says at which point the image is not anymore base fo the generation.
So a high strength and low percentage would start with the same image but has a lot of iterations to add and change the image.
I think I need use more detailed prompt. I will do more studies about this, like you show me
how much denoise you used for htis image, 1? I will try something
refer to this image https://cdn.discordapp.com/attachments/1004159122335354970/1339564150434693141/Corto_Maltese-comics_series.png?ex=67af2dd5&is=67addc55&hm=b64c51e9d3e437de977d07d8ee7d929b935a6900c1908d7c4726bee7342e3385& generate multiple view of corto maltese
having fun with new lora for illustrious
interesting how you did that with a transparant background
@wheat girder scammer alert
thanks guys, this tip to turn off controlnet is really usefull. I will do more studies later
So, are you removing the backgrounds, or is the AI putting the image on a transparent background?
I generate my image and use a later method to remove the background
imagine sitting and thinking that new logos are better than old ones
I like the simple one more
Can someone tell me the difference?
I think it's dev?
yeah youre using dev, the non commercial use version
A glowing portal to another dimension,星空背景,科幻风格,次元之门,门外是城市夜景,门内是奇幻世界,4K高清,cyberpunk style, glowing particles, futuristic, --v 5 --ar 1:1
A cartoon cat, fluffy and very cute.
its been broken for weeks lol
Suggestion: Add some new fun bots like lost music bot, mudae bot, mee6 etc. Will be much cooler.
True
mee6 is hella annoying and im glad its not as populair anymore
I don't have much info about it. Why it is bad?
the amount of dm's/pings it sends you is obnoixious
blocked it for like 2-3 years now
Hmm. Maybe they have improved that a bit I guess. Well, there are similar alternatives too. So, not much to worry.
it's very popular - it only pings if you set it to do that
Yeah but if i join a new server and it pings me eight times im out lmao
hope it's not a server you want to be on if that happens.
Eh ive been shrinking my discord servers list because, realistically in how many are you active
Maybe when i have a vacation (every 8 weeks) i get a bit more active in some but mostly i stick to two-three
all the ones i'm in, i at least lurk in
I check sometimes if theres an interesting conversation going on that i could join or i just sit it out
Hey lads, any suggestions for removing fog / effects in flux?
i cant believe deepseek is such a huge frigging moron
mostly all of the LLM are useless scrap
you put life on the line relying on LLM it's certainly doomsday
yea GPT O3 mini messed my code up yesterday
i'm not surpised, i've been trying to get deepseek to answer my question about life paths and master numbers
i even explained to it how to calculate life path, and all of the LLM gets the basics of it wrong
Calculate life path?
Are you not expecting much of a LLM (technology still in baby shoes)
its so simple.. like you take a full birthdate, then sum up the month, day, year into signle digit unless those summations are master numbers like 11,22,33,44
I mean a llm is meant not for math tbh. Even gpt 4 and the o model makes a lot of mistakes with it
they keep telling us it wins math competitions
Do some questions really well because its in its dataset but a LLM was made to sound human like. Not be correct initially
It doesn't know what its really saying
i was using chatgpt to do that calculation, i first explained how to calculate life paths which it got right, then i gave it a date range to find out how many times the master number 22 shows up, that's where it fails over and over until chatgpt tells me it reached its limit to calculate, then i tried the same thing with deepseek locally, and it failed too, im not sure if i'm not asking it right
You can ask it on how to make its formula or code on how to calculate it for you
But dont ask it to do math directly lol
no i didnt, i explained it to how to find out the life paths, and which ones of those are master numbers, then i asked it to do its own calculation to figure out how many times master number 22 comes up
what's annoying is that after explaining the logic of it, the LLM fails to get it right and I have to trace back where it went wrong
now i wonder if i should rephrase my question asking for life path 22 instead of master number 22
OpenAI are at least claiming that O3 can do well on FrontierMath dataset
which really is pure math
lol
this is not rly how you prompt LLMs though lol especially reasoning ones
now local deepseek not only fails to get my life path number right but also gets my zodiac wrong
Locally run are even more inaccurate
im using the Quantized K_M mix version of qwen
How many parameters?
7b
Yeah that's part of the problem
i cant run their larger models locally
Same, but 7b models already forget my prompts 2 messages later
let me check with google gemini
oh wow great.. gemini just failed to sum up the digits
a kindergarten math
this is hilarious
we should eat the elites when they talk about BIG AI
and charging $200 / mo subs
this is alarming actually
Oh its big and quite useful, but i think your asking a bit much from it
Its useful in my work & personal use
thinking isn't polished as it should be
And that price tag is just to recoup the losses so far
it can deliver from fed data but not analyzing much
The technology is still in its infancy. Try again in a few years lol
you are right
quantised 7B distill sounds fairly rough
i was too excited and forgot its' only been roughly 2 years
Basically
yeah well so far its' good for entertainments
im glad we can generate sweet girls with it
credits to girls btw
real girls
awesome
claude got it right
and it even correctly calculated the number of occurrences within a date range
Could be seed luck but nice
you never know which would serve you right...
https://chat.openai.com/
https://copilot.microsoft.com/
https://chat.deepseek.com/
https://chat.qwenlm.ai/
https://claude.ai/new
https://gemini.google.com/app
claude is kind of the sober genius less talked about
What about Le chat
LE chat? idk which that is
Ah its mistral
thanks, adding it to my bookmark
Its a European (french) alternative compared to an American company. Especially in these uncertain times it can be adviceable to use as many European alternative solutions
Yes but I'm looking at the possibility of Americans leadership messing around a little too much
yeah, its crazy times around the world, but so far tech companies in US are re adjusting thier policies
Even my work is adviced (edu) to not use any Google product anymore for work
So no YouTube for education. That's gonna be fun since its gonna be implemented march 1st
ouch
im always doing something on youtube
even host a channel for music and models
i cant believe one having to give up on youtube
I mean they can use it. But if they are on edu accounts we got to block it
yeah there is that productivity aspect
I mean teachers cant exactly log out to use it without lmao
but whats the reason for blocking youtube?
So im probably gonna get asked to download a ton of videos for them but thats gonna be a chore on its own
its not AI
Additional google services will be blocked. Gemnini, youtube, translate, maps
oh man
And without a pro+ licence even google drive
how do you go by w/o google map when driving/traveling?
are they blocking android phones too?
crazy, that would cripple user experiences
oh so you mean those are blocked from govt services?
Yeah
sounds bad for the govt
Edu = education/ school accounts
While bad for now i think it's a positive overall. We're too dependent on American corporations
Bing
Microsoft is still compliant with our regulations
Yeah that's also a problem
they crushed all smaller companies from even building
There's still 20+ days where trump could overturn a Biden ruling causing it to lose immediate access to Microsoft services so personally im sort of making plans
yeah these are market related issues linked to economic model we have in the world today
we already have all the tech we need, but market model prohibits collaboration
and those are systemic issues
Oh well. I got plenty of opensource and commercial eu alternatives to effectively step away from Microsoft entirely
yeah but google isn't bad for the services they offer, MS too has great stuff, im sure evey big tech companies have something nice to offer, and if they had the scope to collaborate instead of competing for market dominance that would greatly reshape consumer use and high productivity from those companies as well
no one really talks about it much, but the economic model is driving everything the wrong direction
I didn't like old gemini but i like the new one
and apparently it can use imagen 3 as well
you can ask it to inpaint
i prefer chatgpt over gemini
chatgpt is more spirited than gemini i've noticed
inflection pi was the best for personality if you want that
hey, whatst the newest versio nof stabl diff? i havent touched it in a months and worndering is it upto date? says its on v1.10.1 but i dont know if its old
is this A1111? not sure
i usually prefer chatgpt for brainstorming and it's responses are more understanding about the problems of society instead of gemini trying to defend them for the sake of arguments, so im not exactly looking for role play kind of models but something to resonate with
hmm okay, I found GPT O3 mini good
gemini used to be more open to discussing the actual issues but now google messed it up
its gotten from addressing the issues to defending what's lacking
is this outdated?
ask in #🤝|tech-support look for @dry crow
ip adapter is a fun feature
Broken Quantizations? GGUF Format Confusion on Flux!
SwarmUI developer "mcmonkey" shared some valuable insights about the "GGUF format on Flux."
It seems that many of us might have been using broken quantizations. I wonder why the community continues to upload broken versions instead of focusing on the functional ones. It's creating a lot of confusion and unnecessary clutter.
Here is the original thread:
#1243166025000943746 message
"The majority of files in that folder have no reason to exist.
There are only two that matter, so in the documentation, I only mentioned the relevant ones.
The broken ones are quantized so small that they're corrupted and cause issues.
The _0 and _1 variants are very old legacy quantization methods that were deprecated and replaced by _K, so why is city96 still producing these outdated versions?
Q8 is no smaller than FP8 but still suffers from the quantization performance penalty, so what's the point?
F16 is full width and literally not quantized, so why include unquantized data in a quantized file format?"
🤷♂️ I'm a gguf q8 boy so eh
the performance difference between q6_k and q8 is incredibly negligible for flux
going either-or doesnt matter at that point
generate me a picture of a portion with a mysterious background
hi guys, how ganerate an image?
but when I bring it to Revit, the mash is a messy. How can I clean it?
It's possible to ptimize the mesh on Blender? I don't know blender, I only install and make small things
hello, i have laptop rtx 3050 4gb vram, i use forge webui. i cant make anything
how to fix this
Locally or web based?
You dont have a "checkpoint" (model)
But with 4gb vram its gonna struggle on the newer models
On civitAI you can check for some SD1.5 for okay speeds or SDXL for slow but better quality
okay, thankyou for answer. i will download the model first
i like your mix of cream and sugar
she can make all the rivers and ocean start flowing with her smile
I don't know who wrote this shit, but Q8 is really good
And if he dont understand the point to use Flux Q8 (12gigs) instead of full one (arround 20 gigs), While the result is very close to full one... I can't do anything for him.
i am trying to use ltx-video on comfyui, but these errors show up any help? installed the missing custom notes
i installed those already
Via Manager?
If you installed the nodes, you need to restart the back end AND refresh the browser window.
im done with pale flacid illustrious textures
pony has more substance in its texture
and sdxl
Made this with stable diffusion, added to photoshop as separate layers, adjusted lighting and effects. Started with a greyscale sketch. Used royalty free google images for photo bashing.
Holy hell hunyan video is heavy. I either need to get myself a 5090 somehow for 3.3 grand, or get 128GB ram, as it's tanking ram and pagefile lol
or rent gpus from vast.ai
Why aren't you just running the fp8 or gguf quant of hunyuan
Didn't know of it until now
And with "force set clip device", albeit taking a tad longer to stuff clip in ram, i can now generate way longer/higher res videos :P
Also, does hunyuan use different prompt formatting for priority and specifying detail/priority?
Video models tend to like prompts that describe motion. Things it can work with.
Unlike an image model, which only needs a description for a single frame.
Hotcha. So how does one describe content of said video? For instance " blue car, red traffic light , female driver, dark blue sunglasses"
Can stable diffusion generate images like this:
I'm trying with stable diffusion 3 medium, but none of the models or my prompts get it to look like that
Lots of good ones, sadly none had workflow 
What is the prompt your using?
no time for imagining stuff, but thanks
you might wanna check #artisan-faq or a online image generation service that allowew for a few free generations
check again... #🏞|general-with-images message you got distracted by sweet girls, btw what i left out is face detailer and upscaling
YUSSS
Hi, I'm looking for help, can someone tell me how to generate a background for a finished product? [in my case these are beds] I have photos of beds on a white background, I would like to generate a matching background for my exact photo. Does anyone have an idea how to do this?
look into ipadapter and controlnet features, for general idea go through this video how you can maintain consistency and customize your outputs https://youtu.be/gmwZGC8UVHE
ok, so I have to mix his workflow and put it together again with creating a mask [depth anything] add controlnet advance node, and inpainting. Do you think that's the only option?
that' one example i can share with you, you can try the workflow i've put together and see how you can incorporate that to your work. #🏞|general-with-images message
@wispy nest Great, tell me how can I download this workflow?
open that image into new browser tab and save it, then load it up on your comfyui
@wispy nest Thanks for your help, I'll start organizing the workflow and testing.
np, have fun with it
@wispy nest you know how many ClipVision model i need?
this shoul be enough CLIP-ViT-H-14-laion2B-s32B-b79K.safetensors for style tranfers
@robust path oh damn, that's exactly the kind of workflow I need, you did a great job
@true frost here is the link you can download it from, (you don't need all of the files for style transfer) https://github.com/cubiq/ComfyUI_IPAdapter_plus?tab=readme-ov-file
It's not in comfyui though, it's Draw Things.
WF remove background > add depthmap > prompt decribing the room > inpaint background with Kolors inpainting > refine (i2i low %) with Flux
Ok but I see that it is online?
No it's a free local Mac app which you can run on mac, ipad and iphone.
But you can use the same Workflow in comfy
Can't install the IPadapter directly from Manager?
you can im sure just giving you references, here is the IPadapter plus that i've used in my workflow ip-adapter-plus_sdxl_vit-h.safetensors that you can download from https://huggingface.co/h94/IP-Adapter/tree/main/sdxl_models
fyi, you dont need face ID crap for style transfer
only the vit file and ipadapter plus are enough
Thanks for the info, I'll check it tomorrow, every now and then when I try to do something new everyone wants me to try to do it in their style
Something is still missing, I get an error: IPAdapter model not found.
go through this video, https://youtu.be/YYIIkgvOZ3M
tells you what you need, but you dont need face id or sd1.5
Ok, that's all for today, thanks a lot for your help.
good luck, if you are overwhelmed with all the file the guide shows you to download and given you are using SDXL model, here is something to simplify it all...
----------------------------------------------
Basically need three files/steps.
1) the custom node: \ComfyUI\custom_nodes\comfyui_ipadapter_plus
2) clipvision: \ComfyUI\models\clip_vision
`CLIP-ViT-H-14-laion2B-s32B-b79K.safetensors`
3) ipadapter model (create the folder and put the file inside it) \ComfyUI\models\ipadapter
`ip-adapter-plus_sdxl_vit-h.safetensors`
The files mentioned above can be downloaded from >
https://github.com/cubiq/ComfyUI_IPAdapter_plus?tab=readme-ov-file
Aye, But it wasn't the main workflow i meant, was indeed the workflow for each image :P
I'm curious what model you used for this one. Even though it has a few flaws, it's one of the best looking natural images I've seen.
cyberrealistic pony
one of the very few best realism model from pony breed
Wow. 6.46GB, too rich for my GPU. Thanks for responding though. ^_^
yeah that's sdxl ... all of them are about that size
Nvm, not too rich for my GPU. Have you tried Everclear PNY by Zovya? That's been one of my favorites for Pony Realism
None I can share. >_> 
Do you only make women or is it like a list your working through
yeah i love girls the most in all of reality, why is that a problem for you?
well society tends to have knee jerk reaction to loving girls which i find very weird and offensiv e
ok
just for clarification, a universe w/o girls has no value for me
Was just wondering since you were doing a lot of the same style same subject images lol, aren't you interested in other styles?
Lmao fair
im intersted in a lot of things, but girls are top priority
and i refuse to compare girls, i love them all though i know due to destructive social conditons some girls fall victim to those situations and get mis aligned from their innately natural feminine sides, i acknowledge that enough to realize the social scenario
I've got an old song for you then, one of my favorites from 1975. "Girls Girls Girls" by Sailor
there is a song by motley crew .. girls girls girls
Have you thought to change the scene? Like sitting at a bar, enjoying a sunset, travelling through a rainforest
yep, done all that, got bored with scenic focus
Ah fair
This is why I add "((simple_background, simple background):1.1)" to most of my image prompt negative conditioning
I add my own background unless its not really that important to the scene/action
i usually choose specific colord background, but did you know when you choose simple solid background the AI models can render better detail on the subject?
Had a request where it was a horseback, quad, etc
I could do that, but I'm not that good with control net.
Oh prompting / photoshop / inpaint helps
No controlnet needed in my UI
Yeah lol, depending on the scene its not that important
I'm a simple man. I stick to prompting and creating custom nodes.
Generate image > automask subject > remove background > inpaint existing image onto new background
Is how i do it sometimes
@wispy nest - know you typically gen women, but just want to flag i've deleted quite a few imgs from you in the past few days, so just want to remind that women with legs spread open, see through shirts, and with no pants- are not allowed on the server
i know .. but i wonder which are the ones that went against server policy
read an article that ppl making few k $$ / mo, with AI influencers on instagram
A lot of people even use LCM / Turbo / Hyper / Lightning Models for Vtubing.
vtubing ? is that video hub?
Virtual youtubers
Think Youtube / Twitch Streaming but hiding who you are.
oh
Like anime avatar that moves
yeah i heard about those too, they must have a very extensive business strategy to be making big cash
I actually do get a lot of requests for those vtubers. Its really a varied market with unique designs
i've been interested to do something like that but the thought of doing anything for money discouraged me
i have differnet kinds of engagement on youtube, made a bunch of choreographed music videos just for fun
girls and music
but to make full len AI video the technlogy is not mature yet
It actually is, but it would take a massive amount of memory and parallelization to do it. Look up the OpenSORA video stuff.
even civitai has some stunning videos but not long enough to sit and watch
i know what you mean, im talking about the actual length
would be very costly to do 30 mins video, and some of the videos that are more than 5-10 mins don't have perfected frame consistency
You can also generate key frames and fill in the in between frames with AI. Some people have been using this to make 60fps videos out of 23.976 Films
sure, what im trying to say is its still kind of tedious to maintain consistency with ai videos that are long enough
I agree, but the technology is there. It just takes someone with the drive/gumption/tenacity and resources to use it.
yeah... i hope its also more affordable with hardware requirements
Right now you're talking a cluster of H100s linked together.
That would require new models based on training whatever material you wish to generate.
im not too much into gory or explicit stuff but i remember when DALLE came out it would censor and refuse my prompt for girl with tank top, mini skirt
ridiculous woke sh*t
That's the best details and almost perfect eyes I've seen yet. What model was that one?
2nd time you are asking the same question for the same answer
Still the same model? Wow...
When you look it up on CivitAI, there's 14 versions. That's why I was hoping for a bit more info.
AI is going to summon angels from heaven....
or maybe pleiadians gave this technology to humans
new AI video tools that can be run in comfyui https://x.com/PJaccetturo/status/1891600750444179605
aww man, needs too much vram
cant run it on my rtx 3060
You're even using the same GPU I am, now I'm really interested in your setup.
I saw that. The videos are impressive, the time it takes to generate them is not.
im using their on site video generator to see what it can do with my ai images
you should be able to run all variants of SDXL along with all the bells and whistles on that gpu just fine
I already am, that's why I had asked you if you ever tried my favorite model.
evryone report link to google to stop then from using this link to scam ppl 'is brand new
Youd rather report them to their domain provider but its not that effective tbh
i almost went and replaced windows 11 pro with Kbuntu LTS
then i realized the extra hassle i'd have to put up with Linux to configure and tweak a lot of stuff that I don't have to at all worry about on windows
same
I use windows desktop
and then let the servers be linux
also to be fair, windows which sort of monopolized the desktop OS market and possibly crushed all startups to even begin as a company, it is actually a very user friendly OS
well I don't even know where to find a windows server TBH
i use ubuntu inside windows .. WSL
i use the WSL as a container for Comfyui
i also like it that microsoft actually promotes linux nowadays alongside windows
man if all the tech company actually collaborated we could have a very sweet world
only thing in the way is capitalist bs system
Or a very dystopian one. Which one do you think they would really choose?
nothing to choose that way... we'd have everything decentralized
its happening anyways eventually as blockchain continues to mature
i think its fair to say we CAN'T make a very large leap for precautionary reasons
Blockchain tech has been around for 17 years now, it's barely taken off except for perceived monetary value. Torrenting on the other hand (decentralized file sharing) took off immediately.
cause you'd have to look at the fundamental socioeconomic system, we dont have alternative to capitalism that can be put into effect yet
and much of its reasons is old habits and customs that a lot of the businesses are molded into
so we only have to go through stepping stones over time even though people see the clear path where we have to go
No, a lot of it has to do with technology being susceptible to interruptions such as hacking, solar flares, EMPs, etc. Not to mention how slow it is for confirming transactions.
part of those reasons too
if you removed the current economic model you'd have instant power vacuum
As an example: If the network goes down at Walmart (and I've seen it happen before), you better have cash on hand or you're not getting anything.
you're talking about alternatives between cash and digital currency, im talking about entirely a different economic model that requires no transaction medium for resources
We actually do have alternatives to capitalism. The problem is corruption of any socioeconomic systems.
no we don't
not implementable yet
just know this...
as long as human labor is part of the goods and services you can't overcome the current economic model
which is why AI, automation that are coming are extremley crucial to freeing society
we are not there yet
but im patient and optimistic
sadly not a lot of old ppl will see to that day
and its a philosophical concept to merge between the worlds of physical and abstract worlds
maybe that can happen through AI to connect higher dimensions that we are cognitively incapable of perceiving atm
its not a pretty reality to lose someone through aging and decay, none of these need to happen, and we only accept that now with a sense of helplessness and unable to do anything about it, although everyone would love to change that reality
AI is and will be widely acknowledged as humanity's last and greatest invention
@royal monolith has a whacky neysaying ideas about it im sure
There's just one problem with this. They don't want to free society. They want to kill most of us off / enslave us. Remember the Georgia Guidestones?
he look into things with his linear mentality
i dont fancy myself with the thought them and us anymore
either all will flourish OR no one will be left to survive
I prefer flourishing, but the pragmatist in me says otherwise -_-
there is a much greater influence at stake about reality than just us and them
I know a lot more than I let on. There's always a man behind the curtain.
You say principles, I say scruples.
say whatever... you know very well what i meant
obama or biden or any of the global goons dont dictate universal law
I know exactly what you're getting at, but that's not really a conversation for this place.
your deflection says otherwise, but im also not interested to debate about it
It's not a deflection, you're talking about the "7 sisters". I know precisely what you mean. I just don't think going there would be looked well upon in this server.
actually im not obligated to convince anyone about my views
regardless of which place you mean
this illustrious version 2 looks like a big step up from v1 https://civitai.com/models/1257951?modelVersionId=1434449
you mean V1 from Kawaij or is this a merge from the real V1 release?
v1 of the same model
i have some selective pony models and by generating images between pony and illustrious it kinda felt that illustratious lacks of texture
hmm i personally dont have that problem
there are bunch of images i generated using illustrious that look pretty good but some of the images are so refined it misses some of the substance, hard to explain
pony feels more meaty with texture
I'm not surprised. It's apparently a beta after all, not quite complete. https://huggingface.co/OnomaAIResearch/Illustrious-xl-early-release-v0
hmm probably a personal preference but i dont notice it that much
the illustrious we use is v0.1 lol



