#๐โai-images
1 messages ยท Page 25 of 1
its not flux tho
character \(series\), eating breakfast with her hands
why not flux?
im actually trying rn
<3 I hope you like my checkpoint, but read the first few lines in the description
you gotta preface with
masterpiece, best quality, newest, absurdres, highres,
flux is amazing but its S L O W
you using full-ass dev or some quants? fp8 or 16?
I mean, there's schnell which is made to run in just 4 steps
but it has lower quality
so u gotta either choose quality or speed 
ofc quality
ye, the Q8_0 atleast saves your ass, since its 99% identical and requires less vram +faster
it's prob the best way to get best quality at the most speed you could
Here is a version of Marty McFly in Handsome format.
4K Image Flux Dev
fantastic job, how did you upscale it?
???
I have a complex Workbench in ComfyUI.
1.- I generate images with Flux Dev, which takes between 30 to 40 seconds per image, I look for the best option, and so that it looks with that quality, I have some Loras on Flux Dev that improves the Image.
2.- I make the DeepFake of the character I want.
3.- I scale to 1080p which takes 2 min
4.- I do the DeepFake again but with the image improved for greater fidelity.
4.- I scale to 4k in Parts, 4 parts, each part takes 2 min, in total it is 8min.
One message removed from a suspended account.
How did you do this
like
Did you pay for Twitter premium for high quality Grok generated images?
You mean ๐??
https://huggingface.co/Borcherding/FLUX.1-dev-LoRA-AutumnSpringTrees
https://huggingface.co/datasets/Borcherding/FLUX.1-dev-LoRA-AutumnSpringTrees-image-captions
Hello all! i have recently trained my own custom Flux dev text to image model, utilizing my own set of autumn leaves and trees photos which i spent the last year or so collecting. I plan to integrate this with a video game by using the flux model to generate training data for an autumn GAN for my video game to skin the charachers with autumn colors ๐ anyways i hope you are able to get something valuable out of it, and if you have any questions about how i made it with FluxGym, feel free to dm me, the instructions for fluxgym are in the model and dataset cards.
as there's no dedicated channel/forum to post loras yet, you should also post some generated samples here
thank you โบ๏ธ
why yours are better ๐ญ
SDXL?
IS THAT TOO MUCH?!
All the same model as yesterday :3
though I wish this one were flux
I wonder what they're talking aboot
@steel haven im gonna try your checkpoint
<3
I love your stuff though; I've been around quite a bit and yours are actually really neat
i used to make actually funny stuff
Yeah me too! Idk what happened
we are old now

Eject!
just put masterpiece, best quality, newest, absurdres, highres, all up in the front
and good luck, this model is wild
hell yea
1024x1024 & 40 steps
I think tpm++2m is fine
but euler is the one I've been using
o dayum!
D:
Yeah the backbone of the model is based on illustrious
Illustrious is like pony, where boob size is scaled with age ;|
why tho...........
I have no idea
fucking porn brainddead people
Ah I gotcha; that's always a tricky endeavor
even tho too skinny
Now that's just AI being biased lol
lol
(plump:0.8) works pretty well
i want to say too much but mf is actually right :(
I hate it when AI is right
AYO??
It don't know youuu
i go to burger king every week, it should be a fair parameter
LMFAO
@wheat radish sorry for double pinging, but, where is the prompt enhancer thing?
The prompt enhancer thing?
for SD
You mean that other guide I made some time ago ?
no the HF space
Yeah that
link?
blessed ty
hires with noob ai v1
hi
hewwo
me after work
the fuck?!
huh huh?
how did you make this lora
by queer?
I think to send to civitai
aaaaaaaah
?
yes he is the creator
i love her sm
yeah qwizi made some drawings of her
So that's why I thought it was the right choice, right?
i think so
my beloved
AiA
ai hub chan coding mangio fork
yeah!
pamer ๐ผ
you could make tonsnof emotes
me when no ai hub chan model
now make it do 9/11
oh right it was already done
by me
wdym
On January 4th of this year 2025, Michael J Fox was just awarded the Presidential Medal for his work in defending research on Parkinson's disease.
And I came up with the idea of โโtrying to get a photo of that moment with FLUX DEV in 4K
the fun police
amazing
thank you
cursed
super cute
wow prompt is very large
prompt: "super cute"
IL
welp.. literally no i just added it to the base tags
Give me a prompt for my cats >.<
?
illustrious
what is it
Are you using flux correctly?
no
no?!
XD
huh
if i set CFG to 1.5 instead of 1 it looks better for some reason
what model are you using?
flux... well my version
cute
im not sure wdym?
u trained checkpoint or lora? ๐ฎ
trained checkpoint XD
only took 3.5 minutes to generate >_<
i mean, the style doesn't belong to me, i just trained it XD
aaaaaaaaaaaah
then you don't have to
i don't think it's easy to train, the style is very complex. i've tried in SDXL or other models and it didn't work
it's okay, I did it before xdd
Well, styles
I don't know this style
and I don't want to do it anymore
you fixed it!
and you draw well xdd
fax
I did it in 4 seconds
I'm speed
yes she has only 1 leg
thats style
xdd
this is not...
where's the ai hub chan lora btw?
I have on my laptop xdd
I think I will send it to civitai tomorrow
like is it public? iirc u talked about having trained it on weights.gg once
Did I say that?!
maybe I remember it wrong
Are you sure I was talking about flux?
Did you train it or got it from hobqueer?
ai chan lora?
I created a new one
hobqueer? He didn't send me the data set 
ohh
I created my dataset from what is here
A set of audio files compressed into a .zip file, used by RVC for voice training. The quality & length of the dataset are the biggest determining factors of the final quality of the model.
and on Twitter
You mean ๐??
@wheat radish I remembered!
I sent AI-Chan's Dataset and Models to HF (I recommend using IL because the pony turned out very poorly I think)
using IL
IL?
iirc pony is mostly used for nsfw?
aaaaaaaaaaaaaaaaaaaaaaaaaaaaah
Illustrious
is that a model or smt?
yup, 100%.
Oh I'm sorry, it's just the best at it so people use it for that. Sorry for the correction.
u so horny!
sassy baka
i dislike pony >.<
huh? why?!
it doesn't look very good
sure it's good at NSFW but that's all it's good at
wait!
It's not a smart checkpoint
Illustrious is much better at that
huh I don't understand what's the base model
sdxl with alternative training on booru tags
could u send where to download it?
you can find it on civitai
hoooold up
base pony is disgusting looking
@shadow bane PONY!
I'd look for a beautiful merge like https://civitai.com/models/783814
All I know is that pony is a meme nsfw model
it's cute, but still >.<
Illustrious better
Dam
i trained my own flux NSFW model that i use for inpainting, works just fine for me ๐
@wheat radish IL (illustrious) is a Model created for SDXL Models
For example, like a pony, but not a pony. This model can write text better without LoRA (Ponit ak can't)
idrc about nsfw
if pony is mostly just nsfw sdxl, why do people use it for other things too?
Bro wants a skibibi toilet LoRa
nooo!
Ngl I would train it
pony not for nsfw only
pony backgrounds suck
it has been already trained a flux.1-dev one on weights.gg
flux for life ๐ฅ
Flux is boss.
What the Sigma
I love flux, only bad thing is that it has way less LoRAs
Living that fp16 life
leaving the chat
Q8_0 is 99% the fp16, with less vram and faster
Grrrr
yeah but not much faster in my experience
:3
i only use full precision for my flux stuff
I mean who wouldn't want that little percentage more of speed
also, doesn't fp16 use like 21 gb of vram?
22.8
i don't wanna give up quality for speed

I managed to upscale a 430p Image to 4k Resolution in 13 Seconds
q8_0 requires only 13gb
it still fills up
I'm proud of myself
something something quants
there is still some quality loss and i notice extra limbs / fingers
tend to happen more often
what? I use it completely fine
I never had an extra limb/finger yet
I mean I just recently used it but still it's good quality
it's good quality but not the best possible quality... XD
I think I should just stop pretending I can see a diff
and switch to q8
But I like using all that VRAM
sometimes i can get it to 0
๐ญ
honest talk, I don't see a difference
what gpu do u got
4090
i also have a 4090 ^^
I just recently got a pc
I can't say that it's bad tho, I'm finding myself nice in games & using AI
uh oh is discord cdn not worky?
?
even more expensive
4060ti great for games and good for AI
i love playing games at 4k 144 fps
it's a bit of an obsession
It's why we use flux fp16
we videophiles
which site are you using?
Hmm.. I think he is looking for metadata in the photo or something like that
still can do crazy things with that, i have been making everything i send here with a 3060 12gb
yub that's true
I find myself pretty good in both AI & Gaming
ill?
as a checkpoint?
yep
also, link? :3
@graceful lark MH
Maybe you just clicked on that file? And automatically copied the name
how
oki
I recommend using IL because for some inexplicable reason it understood the dataset better than the ponies
or use pony 1-5 eph
something went wrong
As of now, I still haven't contract the author of AI Hub by Weights chan. I still draw them my own. 
(2024, very aesthetic, newest, best quality, masterpiece)
hm? ๐ฎ
it's something similar on score_9, bla
Humanized AI Hub by Weights chan real. 

got ai chan
Alr
Ai trained on ai lol
original stuff wasnt ai
I was not sent a data set
I created what was at hand
I just know better to draw by hands. It's not like I got that talent since I was born, but I practice to get good. 
smexy,.........,
@wheat radish I will soon create for weights
Alr
That "AI" logo on AI Hub by Weights chan's hat is outdated like tutorial videos on YouTube. 

stop i got confused..
ai chan already a model on the weights site?
nono
"AI HUB by Weights" refers to the server name where you're talking right now.
otaaaaaay
@wheat radish
Alr let me know
otay!
hyped fr
Upscaled from 704x704 to 2816x2816
dam
aah I just disabled it automatically
I always forget to disable reply pings 
I don't want to distract you
How?
* vencord*
D:
I upscaled an Image to 10GB in 30 seconds I think
4 million x 4 million Pixels ??
idfk
but i like pings
@odd lance ping
Nobody ever pings me x.x
@shadow bane
@shadow bane ping
@shadow bane
Holy moly I'm going to make Moana 3 AI Version
@odd lance / @wheat radish https://www.weights.gg/loras/cm5l85fc30jejl1185e1pmmj8
damn it kinda suck with flux
cool
You try to find 10 - 30 Good pictures of Good quality without data set created by AI
๐ฅน
Trying to find non-AI furry art these days is sooo difficult >_<
first world problems
Then I'm stupid and I don't know where to find good pictures
:3
ill check later if i have something
no no it is not necessary
Yes
If anyone needs a better LoRA AI-chan for ponies then I think I'll write to you
if you don't mind
ofc you can add me if you want
I will do so xdd
:3
mh?
and i paid 200 euros for one
wowie
this is my favorite gif
I created it a long time ago to be honest
oh !
you can try upscaler
so that the text and other things are very high-quality
- add the word negative
๐
ai-chan was reborn in a fantasy world where 300 years ago she taught pigeon magic
new anime name xdd
dam
But it's better than the anime "jidou-hanbaiki" xdd
true
how
how what? ๐ฎ
how did you do the same prompt
I don't know her trigger word
I didn't do that
are you using SD?
yes
you upscaled it?
ComfyUI
i just turned off my pc ๐ญ
yep
I want to know too
^
bro I did Ai Chan NSFW without wanting to, help me, I didn't put anything too much in the prompt
masterpiece,ai-chan, white fur, purple eyes,1girl, solo, blue hat, headphone,1 girl, simple background, black background, looking at the viewer, glowing eyes, blue glow, light, light hitting face, expression, one eye closed, tongue out
NO NSFW!!!!!
what checkpoint are you using?
it wasn't supposed to be that tongue sticking out, nor that expression
just use "ahegao"
it's not really nsfw, but the expression was strange
thanks
๐ญ
and the upscale, how did you do it?
BRO IT GOT WORSE
but I don't want nsfw bro 
no problem Lol
add negative prompts
something like this
Better
wifi
ethernet
Use 1024x1024
I still don't understand, does it have to be in another workflow or in mine?
e-girl
Honestly, I don't know, I only answer colab version xdd
Can I have it?
just to see how it looks
Here you can use upscaler for SD & SDXL
but unfortunately this version has a bit of a limit
it doesn't matter
It works very well
crazy we still have colab
Created specifically for dumb people like me (Honestly, I don't even know this author, but it's cool that he created and updates it)
- there is a thing that does not work well
Created especially for @past moth
oh nice

this one I did on my PC, but I didn't understand one thing
he made two images of 25, only one came out, and the upscale was normal, which I didn't understand because he made two images and one came out and the upscale was normal.
So the code creates 2 pictures
1 low-quality photo is just rubbish
2 High quality
yes, but in this case there were three
but showed two
but it's not that I want 3 images, it's that you've done 1 extra process that's wasted time, isn't it?
Which do you think is better for the first image? 720x720 or 520x520
yes I put 1024x1024 in the upscale
hmm
i need to remake ilaria upscaler
I use 1024 to make the image 1536
For colab, it takes less than 30 seconds to create + upscaler
WTF
Your Baby
even more my baby
me in a bit
me like rn
me too
how do you do that
yes but you draw them?
aah nope
how do you do them
I honestly don't know who drew these GIFs but I like this character so I use these GIFs
ah sowwy i lie
i so much love sona
ph so its not yours
ita from someone else
and you adopted it
yep :C
Ahh, but I'm not saying it's me
my fursona is different
They are two different characters
ohhh i thought it was the same
Does anyone have a good workflow to give me Txt2Img
I just lost mine 
If possible, give me a simple and well optimized one, please co upscale

A Stable Diffusion WebUI named Forge, which works just like Automatic1111 but better. 
Does it Support zluda?
(But what about ReForge?)
Do you think I know about API of things? I'm just basic about using AI image generator. 
take anyone of your generated images and drag it into the blank workflow to see if any contain the workflow metadata. (Assuming you're using comfyui)
i use SwarmUI for 99% of things
how many people here are furries O_o
Not me
not me
what'cha need?
but you're a cat... that almost counts
prompt{"8": {"inputs": {"ckpt_name": "noobaiXLNAIXL_vPred10Version.safetensors"}, "class_type": "CheckpointLoaderSimple", "_meta": {"title": "Load Checkpoint"}}, "9": {"inputs": {"sampling": "v_prediction", "zsnr": true, "model": ["8", 0]}, "class_type": "ModelSamplingDiscrete", "_meta": {"title": "Legacy?"}}, "10": {"inputs": {"width": 520, "height": 760, "batch_size": 1}, "class_type": "EmptyLatentImage", "_meta": {"title": "Empty Latent Image"}}, "12": {"inputs": {"upscale_method": "nearest-exact", "scale_by": 1.6, "samples": ["10", 0]}, "class_type": "LatentUpscaleBy", "_meta": {"title": "Upscale Latent By"}}, "13": {"inputs": {"text": "bad quality, worst aesthetic, blurry, open mouth, oldest, old", "clip": ["45", 1]}, "class_type": "CLIPTextEncode", "_meta": {"title": "Negative"}}, "14": {"inputs": {"text": "no humans, puppy, animal focus, general,\n\nall fours, sitting, looking at viewer, fluffy, fur, animal ears, animal ear fluff, brown fur, blue eyes, hand up, dog paws, waving, blue aura, smile, chibi,\n\nbest quality, very awa, highres, newest", "clip": ["45", 1]}, "class_type": "CLIPTextEncode", "_meta": {"title": "Positive"}}, "15": {"inputs": {"seed": 1337, "steps": 28, "cfg": 1.5, "sampler_name": "euler_cfg_pp", "scheduler": "beta", "denoise": 1, "model": ["8", 0], "positive": ["14", 0], "negative": ["13", 0], "latent_image": ["10", 0]}, "class_type": "KSampler", "_meta": {"title": "Hack to Include Metadata"}}, "26": {"inputs": {"upscale_method": "nearest-exact", "scale_by": 1.25, "samples": ["115:4", 0]}, "class_type": "LatentUpscaleBy", "_meta": {"title": "Upscale Latent By"}}, "36": {"inputs": {"samples": ["117:4", 0], "vae": ["8", 2]}, "class_type": "VAEDecode", "_meta":
it's just a cat, no furry stuff
ah cool, NAI I have the perfect settings
hold on
honestly, this is even more text, but there is a limit here
"Image Width1664
Image Height2432
Bit Depth8
Color Type2
Compression0
Filter0
Interlace0
parametersmasterpiece, best quality, newest, absurdres, highres, 1girl, business attire, sleeping in office chair, drooling, sprawled, Negative prompt: masterpiece, best quality, newest, absurdres, highres, 1girl, business attire, sleeping in office chair, drooling, sprawled, Steps: 40, Sampler: Euler, CFG scale: 5.8, Seed: 14406105880194, Model: ILLUSTRIOUS\IllustriousBlackMagicClean20.safetensors, Model hash: 023308ec79, VAE: sdxl_vae.safetensors, VAE hash: 235745af8d, Hashes: {"model": "023308ec79", "vae": "235745af8d"}, Batch index: 1, Batch size: 4"
if I haven't forgotten, there was a bot looking for metadata
of a txt2img Workflow, but with upscale
@keen bramble right?
and well optimized
Main KSampler: 28 steps, 3.5 CFG, euler sampler, normal sched, 1 denoise. Add masterpiece, best quality, newest, absurdres, highres to the beginning of positive prompt, Use: nsfw, worst quality, old, early, low quality, lowres, signature, username, logo, bad hands, mutated hands, for negative prompt
that's why i copy and paste my images into Paint before saving them, so people can't see my (totally innocent) prompts >.<
find your generated images folder, and use one of those
if you've made images before, it'll have the workflow there

wait 5.8 @~@
why
How long have you been doing this again?
Yeah, i've seen your model pages. 
hm? wdym
what about 5.8
no nsfw? ๐
" CFG scale: 5.8"
But if the CFG is less then the photo will have a lot of noise no?
wait
Does my photo have a lot of noise?
i use 3.5 CFG, plus depends on how the ckpt is trained
Neither do the thousands I've made before this so 5.8 is def safe
just don't go below 4
Epsilon Pred models break under 4, Vpred works better lower.
I'm using EPS models
Yep
oh dur
eps?
Epsilon Prediction, it's the way the seed and prompt talk to each other, if the seed is the same, but the prompt changed from last generation, most of the image will change. On V-Prediction it'll change significantly less.
Most models use Eps, but some newer SDXL trains use VPred
Ahh now I understand
Apparently
Oh, and SD 2.1 - 768v (yes that exact version) was vpred only, though no one used sd v2.
might wanna chill on those images, it's bordering nsfw on a few
Breh
I don't see NSFW??
๐คฎ
im sowwy
Whqt
i didn't say straight up NSFW, but close to which with this server being 13+ is iffy.
Furry nsfw shit
some people have extremely sensitive eyes i guess
You can name this NSFW.
Bruh
no nsfw? again
She can't help that she have big booba
๐
are all big booba nsfw?
At least with this one the ass is covered
or the model just didn't learn gen furry
Goofy It needs Update ff
Try this:
Anthropomorphic wolf with gray fur, the wolf has red eyes and is wearing a black sweater and gray shorts holding a bubble tea
is it for flux yes?
Second try
1 min
it's more like furries
why res so low
No upscaling
no I mean the native res
And its from my Discord bot
why isn't it 1mp at least?
I'm on mobile rn so I can't really use my GUI directly

@rustic ingot

The bot uses multiple 4090s ๐คท๐ผโโ๏ธ
That are on a local GPU rack
Do you mind sending all your results to Civitai?
because Ai-chan looks better on you, so I wanted others to see it too
well, if you don't mind ๐
I feel really special here HAHAHA
I'm not 
Everyone is special to me if they tried my bad LoRA and it turned out really well
I don't know the link to it
So i can't test it
ยฏ_(ใ)_/ยฏ
Thx
Hi, I was just wondering why you feel it is a good idea to have your Twitter profile linked to discord? A profile on which you follow several very nsfw furry accounts?
You mean ๐??
Bro because its his Account?
Hmm.. let's start with the first fact
because I used to send LoRA newbies there
the second fact is because all nsfw are "Friends" (which I have never seen) who are engaged in AI
Only because you don't like it?
You'd think they at least would keep it hidden but damn most of them just put it out everywhere
You didn't have to do the third faet if you weren't interested
Ah yes, it's for research purposes
I haven't been on Twitter for a long time, so I don't even know what's there
You mean ๐??
I know that they updated tos that they will use all media for ai
๐
Perhaps you ought to check what you are following lol
It's a hell of a lot of very very degenerate shit
No real animals thankfully just drawn
hm?
That means people have generated with his shit 184k times.
Don't believe the numbers
Ohhhhh
I never used civit so I habe no clue about all that
*have
Yeah dude, welcome to 2025
fucking clown world
no worries, I'm a fan
I've made nsfw videos for free
:3
I created them for free and he sold it on Onlyfans...
Please don't feed their unhealthy addictions to furry nsfw
No nsfw in this chat
No NSFW in this chat ๐
yep
Just kidding lol
i don't use civitai for gen img only post models
I don't often create art for myself
or others
I first saw them when Nick88 asked me if FlareAI (Old CivilityAI) has something to do with them
After that I had to argue with Discord Support
That they change rhe bots namr
๐
hm?
Hello there Sky!
Thanks for reaching out and for the screenshot!
For verified apps, we only allow name changes in situations that are absolutely necessary, such as legal issues, business concerns, or other such matters. So I can review whether or not your bot would qualify, can you please share the reason for your request with us?
Update me when you can so I can proceed, and if you have any questions or concerns about this process, please don't hesitate let me know.
Best,
Julie
Discord:
@~@
