#🗣️┃chill-chat
1 messages · Page 35 of 1
Suno admits that Suno, Inc. is a Delaware corporation
It admits it???
A Delaware corporation? Really?
https://www.billboard.com/pro/ai-firms-suno-udio-blast-lawsuit-music-labels-threat-to-market-share/ he'll ya suno
First ever AI Movie is coming and Suno Ai is part of this. Along with Hedra and others. I also using Resolve and capcut. So much in this. If it was not for the ai Companies for everything this could not happen. Thank you Suno Ai. Everyone play a part in this. Stay tune to oct 31st 2024. When the movie comes out. If it has to be push back ill let you know.
When did you start production on it? Waymark Creative Labs released a movie generated through the use of dalle2 last may https://youtu.be/IgPvoPBrlTE
IN THE ICY DEPTHS OF ANTARCTICA, A TEAM SETS OUT TO INVESTIGATE A STRANGE SIGNAL, UNKNOWINGLY EMBARKING ON A JOURNEY THAT WILL CHALLENGE EVERYTHING THEY THOUGHT THEY KNEW ABOUT THEIR PAST AND FUTURE.
PART ONE OF THE FROST INTRODUCES A NEVER-SEEN-BEFORE CINEMATIC WORLD CRAFTED ENTIRELY FROM AI-GENERATED IMAGERY.
Learn more - https://www.thefros...
Even if The Doll isn't the first ever, it's still really exciting that people are even able to make their own movies now! It's absolutely mindblowing what's possible these days 👀
Hi. I see you joined this server 15 minutes ago and have only posted this video (in general chat, where someone suggested posting it here.)
Do you use Suno and is there anything AI-generated in the link you shared?
Oh yeah that's Woodhouse, the resident bot, he doesn't like shouting 😅
We sometimes get people joining just to drop links to their non-Suno music and they never say anything else so I just wanted to check that you hadn't just joined here for that 🙂
if you have any Suno-related questions etc. feel free to post those in #💬┃general-chat - there's also #🐣┃suno-showcase if you generate a song with Suno that you think sounds awesome and needs to be heard by others
drive by link droppers are never popular @potent merlin
@fluid field on your suno profile page these are the songs I would click on first. It has nothing to do with the song itself and everything to do with the imagery.
fair enough, those ones i don't have an image for
oh its just feedback from one person, its not a criticism.
i know that, just saying i may try to figure out an image for them
Just saying they are the ones that draw my eye.
nice, those ones were made by suno
I think it's because they are less busy, than the multicoloured ones that suno normally makes. Any one busy image is fine on its own, but when there are multiple busy images on a page, its the quiter ones that stand out.
true, simpler ones make it easier on the eyes
it me
scent of shoe: made real
I've been known 2 bone
New Midjourney is p. neato
oh hey yolkhead did you see that thing about uhh
hold on what is it called
Flux
what is this?
Oh yea
I read about this already
It looks promising
The only thing is, I noticed the public one has a non-commercial licensing
30~ seconds a gen
the schnell is supposed to be fully open
it is, but I think its similar to cascade
but I could be wrong
I only read about it last night as my body was shutting down
(in that the license doesn't allow commercial use)
I've only just messed with it slightly, comfyui is weird and man I don't like it
me either, and yes, to the comfyui evangelist out there, I know how to use it
Same but it's hardly a "UI"
is it Italian?
(spaghetti nodes)
oh hahaha
well, I I do test other people's prompts to confirm my lora's adherence
and it makes furries.
ahaha the bust
Considering there is literally nothing in the training data
that would do this
big success
Now I gotta work on redoing the negative one with the new parameters
its um... art
kk
man eat an entire butt Woodhouse
you're right woodhouse, but I mean, I'll unplug you
menacingly hands
hands his hands handily
I wouldn't want to have to use my opposable thumbs
to do something neither of us want Woodhouse
also this, like, whole
more like WHOLEhouse
technically sfw, but like, man it shouldn't be haha
Why she got extra lips
hahaha
I don't like this so-called lady
that's how it f---ed up the tongue
Midjourney is pretty good, I normally use it for anything public.
presumably
I just use Italian laborers
sorry I cut across you guys
lmao no you didn't, I WISH this was my private chat estate
But alas, just a security guard 
Forgive me father, last night I did an anime
banger bc bang bang
make a server then you are the king its great I have my own laboratory
he looks like that's what he's singing
Just me and .. just me.
Probably said it before the song started. BD never sings
I like this style personally; also anime is fine imo, I just like variety personally
Yeah old anime is great
They just don't make em like they used to shakes fist Akiraishly
Spirited Away unironically is in my top 3, it's literally one of the most clever metaphors I've ever seen, and the animation is glorious
Miyazaki is dark
when you really get into his work
its almost all about Japan's transition out of Shinto Buddhist naturalism and into the industrial era
and like, that kinda stuff
also, the bomb
ANway, have a thing
hehehe
NY post apocalypse
this one it nailed
I did not prompt the flag in the corner
Imagine the first Fallout game that uses gen AI for textures or even rendering
its going to be nuts
more native 1600x1600
collarbones a lil meh on this one, but its sdxl vanilla, so I'll allow it
essence of bone
welp made an entire song about this dude
and much like star wars, the lore is very tiny so this dude has got to be related to Garbage Tony somehow
Did Garbage Tony have a dad who went out to buy cigarettes but never came back?
I'm getting a bit anoyyed at the Mandela effect
It's ruined alot of things for me
1st my movies changed
And now 2ndly my games change!
I was playing driver SF
And for some reason I can't drift+
I could on the pc edition
But on the ps3 edition I cant+
What the heck is going on here!?!
Hi guys I have a little request
wow we making cool ai images
/chirp
Chirp is retired, and no longer on discord. You can make music now on our website: https://suno.com
hello
you all are awesome!
wait that reminds me of a picture
Hey all
@strange gorge #💬┃general-chat message i see similarities ... (wow the new phone camera is pretty good, you can even see all the trash in my car in the reflection of my sunglasses.. amazing)
Ff lmao
That owns
i just bought these slightly too small 80s sunglasses..
best 4.99 i invested in a long time
how long has the roblox game pressure been out?
new lora is almost done baking
ate 🤩
this is actually exactly how I imagined you
Dominic you rat! The ol' spicy keychain!
I actually had to look this up
wow get some e-culture BUB
hehehe
Not exact of course but reminded me
the prompt believe it or not is: (masterpiece), (best quality), (ultra-detailed), low quality, jpeg artifacts, blurry, poorly drawn, ugly, worst quality, art brut, outsider art, a man walking to the store, waving the corpse of a dog around in the air, screaming, paranoia, illustration, disheveled hair, detailed eyes, perfect composition, moist skin, intricate details, earrings, by wlop
this is what semantic shift excels at. In this case its the negative one
usually they can be combined or used seperately, it depends on what model and what situation
keep in mind, this is CFG 10, so the impact is strong
Just sent in a general application for Suno :3
hey suno team, great stuff you're doing, but there's now major articles about the tool being used by racists to try and stir up conflict in the UK
are there safeguards against things like this?
i hav a secret.... im batman!!!!
OH WOAH
Lyrics are auto-moderated to some extent (users often complain that their lyrics are being rejected due to moderation) but ultimately if you try hard enough it's possible to get around that and it's the user at fault for choosing to make such a song, not Suno.
Suno may be able to determine who created the songs (there is a hidden watermark but I don't know if it would contain that level of detail) if they are available online. If you find examples of this, might be worth DMing links to a moderator so they can get the team to investigate
people try to be normal with ai programs challenge (impossible)
theres a lot of messed up stuff happening with ai rn tbh
have you guys heard that people have been making ai chatbots of real life minors
and they just make more when these minors are begging people to stop
little shadow demon
This is a constant distraction
i dont have a cat 
Find one 😄
That belly!
Fleeing from a kitten on a cat wheel does not seem super-effective 
Makes me think of Third Eye Blind
this is not what I wanted to see as soon as I opened this channel
it's called art mom!
it's called a new phobia
also, wanted to share that cassette tapes work great for cover art!
Funny thing that i just was asking AI language model about the best fit for wedding... And batman got 9/10 (highest from all).
For some reason my another potential candidates for wedding (mosquitos, portable nuclear device, paperclip maximizer machine, the sun, HAL 9000, Rick sanchez, ChatGPT) was less scored...
When AI will learn to be aligned with my individual values to have a fun observing paperclip maximizer machine in actions? Ah?
i just realized how close i am to 50 fans like woah
well done 😄
This rules
@vale lake when you get that lots done be sure to DM me thanks !!!!1
cover art for my "test" playlist where I try to break suno
https://audius.co/Imortalimp/mm-foolish-pride no kidding this was a realy long time ago, im not proud of shit, cause uhm i aint shit.
I'mBlackAsFuck, for making this foolish pride distrack way back when i was a premidschooler, Now just a hint I'm not racist, i just hate all chicks nowadays....
Women are fine!
oh, the LoRa? It's done
this is more a prompt thing, you can get this probably on midjourney too, although not as easy
you can use just 0.5b or -0.5b now. I recommend using high cfg with the negative
like, over 10
in general, the positive is better, but a little negative in there never hurt anything, as it can massively increase creativity for various reasons
the dataset for that one is also availabel
available
on there, so you can actually see the why
yall, I really need to clean up around here
real footage of from the offices
Now that's a blast from the past...
Too Much Coffee Man!

Cute suno art
Thanks guys for the love never expected this much might seem small but I genuinely appreciate anyone who even listens to my songs one time even.
@fast rivet
What is the red dot?
wich browser? android?
Apple
new but not yet listened to generations
let me test. safari?
It’s almost as if everything got zoomed in
Yes, safari
oh yes. i see it. i will report that
try landscape
Hello ★
Guess what - I'm not dead!
Actually I have a lot of stuff I could upload but I wasn't really motivated to do so until now.
This case is a special one though – I'd like to introduce this masterpiece from Zankyou no Terror's OST to you.
Every single song on the soundtrack is simply magnificent, so I highly recommend it to you (Yoko Kan...
The red dot is songs that haven’t been listened too yet, for new renders. So you can keep track
this is a cover version I can't find original rn but lol
sounds nice, is it icelandic or japanese?
icelandic.
lotta international copyright red tape around this OST, Japan itself keeps removing the original from youtube I guess. :/ annoying lol.
I'm glad I have it on CD I had no idea it was this hard to find online these days, damn.
can't even play it on spotify. :/
This is why I hate international copyright law, I don't understand the mindset of wanting to limit your listeners. I get wanting to protect artists work, but this isn't protection it's inprisonment. 😦
most of the OST isn't available but still, von is a highlight of Yoko Kanno's discography and is a beautiful Icelandic song, too bad I can't share it lol.
all I can do I guess is recommend to watch the anime, it's well worth your time.
in my opinion anyway.
not sure if you like anime or not though.
/chirp
Chirp is retired, and no longer on discord. You can make music now on our website: https://suno.com
rare alien invasion footage
well uh oh
also, I would play the heck out of this game
Im running like hell
Chirp
Making pizza rolls in the barrel of my red hot rifle
Trying to figure out how I’ll use these in my chill style
I should figure out where to use these as well
I've gotten some interesting cuts from using the following styles, which feel fitting for those pics. If you want some real festival vibes, Shamisen is a good addition too: Dreamy, Layered Harmonies, fusion, traditional instruments, mythical, atmospheric, chill, lo-fi
Thanks, I’m adding a few of those in, I’m trying Chinese rock, and also traditional Chinese instruments
boat
/help
Hi there!
If you’re ready to dive into creating music, head over to Suno.com.
For more detailed information, check out docs-guides-resources.
If you have any more questions, just ask in #general-suno where you can always find our community, moderators and guides to help you with most questions. 😊
when ideogram tries to make a ship <.<
this is terrifying xD
what were you wanting to do with em?
Looking for ideas for songs for them
ahh okay.
so @vale lake
Flux is wild
It does text better than midjourney. The biggest issue with it is that I've grown reliant on these generative AI systems 'filling in' the rest of the image but Flux requires you to prompt for it, otherwise it just doesn't do anything besides what you ask.
It's prompt adherence is second to none, full stop
The downside is 30~ seconds per image which is so dang slow
"first person view inside the cockpit of a space fighter jumping into warp. The screens inside the cockpit cast an eerie amber glow, with one screen reading "ENTERING WARP" while the others show various graphs and diagrams. Below the pilot's hands grip the flight stick and controls as the stars outside start to streak with speed. The canopy above shows the stars."
Midjourney text gen isn't really top of the line atm. Hasn't been since release. There is also no real enhance feature for their images, like Krea or Leonardo offer, which would help their images a lot to create those subdetails or sharpness that is often missing from Midjourney. But beside those points, Midjourney is really freaking to create stylized images etc.
I blast through text on midjourney by just spamming generations, with the added benefit that I pretty much image prompt with MJ vs text prompting
because your last point exactly
mashing images together is 
Flux though, man. It's an absolute leap forward for own-hardware generation
Checking it out, wow, it does so well on the small details
and complicated prompts. Like anything though it can mess up
But it messes up so little
Someone has already set up an img2img comfyui workflow so I'm about to mess with that. I am thinking that I can combine midjourney outputs with Flux and get some nice things
I've used MJ for almost two years so I'm kinda biased but I may give flux a try if it's free.
It requires at least 12 gigs of vram
But yeah it's free in the same way that Stable Diffusion is free
The novelty of it is huge right now, we'll see if it lasts
Maybe it has a low ceiling of some kind we'll all find
oof.
yeah I need to upgrade my PC.
I have 3 gigs of vram xD
I have the money I just need to do it.
but I may be moving in the fall so I'd rather focus on the new PC after that stress is gone.
I just returned to MJ after a long break and although i photoshop the images after i'm back at being amazed at how cinematic these images can be.
yeah MJ still has flaws but it's improved so much.
Yeah MJ is the golden started as far as I'm concerned
Plus they just popped to 6.1
also I imagine I would need more vram to use it but wow
that's amazing
holy cow
ugh.... need.... new..... gpu
yea, MJs strength is this
@vale lake #🐣┃suno-showcase message check dis out. I found another new toy lol. and combined it...
it's the best for exploring the latent with ease in that "room by room" type method, which makes it really really good for finding good images
I "gave" kaker the start to the other gen so maybe he'll make something even better lol.
Hmm... Kaker, can you link me the comfyui nodes?
oh, sweet, I can use it then
he gonna get lotta pings in the morning lol.
I'm on an rtx 4070ti
me this morning haah
we both had to sleep eventually
it was inevitable
ooooh
I'm excited
:3
omfg I'm so excited
I haven had a new image gen toy in forever
it does make my sdxl work slightly obsolete, but this is ML
everything is obsolete as soon as it's done
oh yea, I've seen acid jazz on here a lot, it's def a thing
now what is this tennis thing
yeah I found acid jazz today from that "obscure genre" youtuber I discovered last week, figured I'd try it out.
so far suno has known every obscure genre I've seen from the youtuber so it's been a lotta fun xD
anyway bed time for me. o/ laters.
uhh yeah sure there are loads on the reddit though:
https://www.reddit.com/r/StableDiffusion/comments/1ehqr4r/you_can_run_flux_on_12gb_vram/
OK cool, I'll just use this
I'm just on my mobile atm and haven't had my coffee
yeah I don't have a good workflow at all
that seems way easier
But so far: :bison_yes_yes_yes:
"paperback romance novel cover titled "How I Seduced the Robot Cowboy" by the author "robo luvr" featuring an obvious and quite scary skeletal chrome robot wearing a cowboy hat with a pretty, but dumb, woman fawning over them"
this is nuts

Flux seems kinda cool, I just got access to the upcoming Imagen 3 and I’m really impressed by it.
Sometimes the text rendering leaves a bit to be desired, but that’s a common enough issue — excited to see how Flux compares 🙂
Ive never been in.. off topic. It feels... hmmm
what is that one?
I actually think all models should drop text gen and offload it to a small, low parameter post-effect model
bc I guarantee you it has a massive impact on parameter count due to the level of contextual understanding text requires
And it just seems so stupid when it is the easiest thing to add in using AI
Google Trusted Tester Program for the new ImageFX 👀
its truly unfortunate that text got picked up by the communities as seemingly a primary metric, as I think this has gimped models artificially
on local hardware
making these for mty next insta
native 1080x1920
Admittedly, I haven't seen many locally-run models that handle text as well as Dalle3 / firefly / photoshop, etc. Imagen 3 is pushing "great versatility, greater prompt understanding, higher quality images, better text rendering" , so it's included at the end.
It does seem to be an arbitrary metric, especially because those in the know realize that there can be MASSIVE improvements in contextual understanding, but getting clear text will always be some sort of challenge.
However, if that's SUCH a friction or pain point for the general public, it'll be worked on, and while there certainly will be improvements made for "low-hanging fruit", I think we can rest assured that those rising tides will float all boats, including the intermediate-to-advanced products that would benefit
is this with imagen 3?
1980s anime, classic anime style, drawn by Studio Ghibli, water droplet shape, colored outline, bokeh, colors, swirling colors, a rose, transparent water, silhouette, breathtaking composition, at the height of his craft
above is the prompt
trying it
it may be that it prefers nat language
P good for plain old sdxl vanilla
the lora massively improves water scenes mostly just bc of the great wave painting being targeted
hmm, it may be related to Studio Ghibli then
for whatever reason when combined
idk though
that one is quite nice
I took the ghibli thing out first, and the classic anime style by itself was still giving ssues
I like the vibrance
SDXL still vbeats it on this prticular prompt, but its p close, and it has better adherence
plus, this is running at 120 steps
Yeah, I thought the second was closer to what you'd asked for, but the first one just stunned me with the detailing on the petals
also, I'm generating at a high res
yea it's got good detail, but its a lil midjouirney trick from preference data
you'll notice most things that have preference will get those little droplets or beads
on everything
when there is stylistic room to
we naturally draw positive conclusions bc it makes it seem more complex; same thing it does for moist eyes
I think the interesting test would be to see if it can do, for example, Mona Lisa clones
This and the Kiss 2: electric boogaloo
are like my go tos
100% -- that's a good point. I still find the level of detail subjectively nice, but you're right that it's simply having a simple pattern for "visual texture"
DALLE does them fine, but that's because it has a preprocessor
Oh, I agree, it is good detail
But I will say the images Kaker posted above for first attempts are absurdly good
I think Flux has something special
they cooked that model really really well for it to be just 12g
gig of VRAM
like, the architecture or something must be novel, bc that was extraordinary
🙋 Me also. I'm constantly dismayed at how far behind I am suddenly technologically.
Stop sharing interests and being well-reasoned in your conversation and making me re-realize that we're living in the future already -- you're making it difficult to want to do anything "traditionally productive"
That was me a couple days ago watching the video for "End of the Road", reading the video description and realizing this is something that people can just do now.
I used to pride myself on being "up on tech", and instantly I felt like... 😛
IO haven't even seen this, what is it?
End of the road was Noisee though. It's an online AI video generator. With a paying account you can get image references and use text to prompt what you're looking for. After that it's just a matter of editing.
Web services exist and I do make use of them when I can. But I'd really prefer to do stuff locally as that would hopefully mean better control and some time saved by not waiting in a queue.
Hepcat posted a video, and for whatever reason the description just was the catalyst for things to "click" that "...this is the future. You don't need to invest a crazy amount in overhead to get started making this kind of stuff."
🌌 "End of the Road" - A Contemplative Lo-Fi Journey 🌌
Embark on a thought-provoking journey with Hepcat Harmonies' "End of the Road," a slow and captivating lo-fi instrumental track featured as the last track on The Hepcat's Lofi Mixtape Volume One. This AI-generated video, created using the innovative Noisee technology, invites viewers to conte...
Same. Like I said, Flux might be limited in some way but it is absurdly good. With img2img it's even better since that "fill in" fixes that issue for me. Generate an image elsewhere, img2img using it for the bones, ez pz
Knowing how seriously the International Olympic Committee takes themselves, I'm utterly horrified to think what that woman is capable of to be placed in the same weight class.
tried to make the archer karna. the closest. another wasted day of spending time on ai, to create something thatll never be perfect
https://www.reddit.com/gallery/1ej07fq @strange gorge @vale lake okay, y'all are convincing me to check this out locally -- am I able to drop it into my existing Auto1111 or ComfyUI?
As long as you get the model and vae and the text encoder thing ComfyUI supports it natively. Not sure if A1111 does
Usually auto1111 needs to specifically add new frameworks or whatever, like how we needed to wait for sdxl or sd3 to be supported.
Okay -- I'm gonna dive in. Are the files on civit, or are they hosted on their own site?
Not sure, I snagged them from elsewhere. When I get to my computer I can link you them
There are 2 models, 3 variations afaik
With a 4090 there is little difference between them so I've been using the full dev one
Yeah, I gotta get them from somewhere else then.
I'm rocking a 12gb 4070ti that I got for dirt cheap
Does this help
Also I should try non-photo styles with flux. Lots of people stoked about it's anime capability
Perhaps it's good at other illustration styles
Nice, I appreciate it dude -- wasn't trying to rush ❤️
Nah I just linked Yolkhead that yesterday since I was in it at the moment, lot easier to c/p a single link than looking for a few that I don't remember where they are 
ComfyUI is such a departure from what I know that I've gotta get comfortable (heh) with it.
But img2img works really well, and the little bit of inpainting does as well even though I haven't dialed it in
For less than 24 hours after it's release though 
yee howdy
lmao
never change civit
yay!
I see what you (unintentionally) did there
I'ma try it out now
it's a comfortable position I am assured
ffs, now I need to explain to @daring echo why I need to buy another new graphics card.
Hey honey, maybe you can use my 4070ti to play Dreamlight Valley in SUPER high res.
Your raw prompt
Not toooo bad
Hey it's his prompt, not mine. Stole it 'cause images are workflows
It's just flux takes it literally. Here, let me fix it
wasn't trying to clutch at pearls, I just legit wasn't sure what was considered "too violent", even though it didn't actually show any explicit gore
Nah you're right
In your defense, my go-to test for image gen is asking it to make a roadside accident cleanup as depicted by felted woolen puppets, so I'm 100% into the content ftr 😛
"1976 ektachrome photo of an anthropomorphic cute mushroom dressed as a police officer writing a ticket for a speeding grasshopper"
lol the hat just stuck on the guy's head
I am in lost of words. Of what going on with this youtuber. It crazy after. I make this song all this goes down. But this song was made on suno ai. https://www.youtube.com/watch?v=buB-GSiERnE Please no hate on it. Thank you Suno AI for making our dreams come true. This song was written by me.
Provided to YouTube by IIP-DDS
Dance To The Beat Of Mr Beast · Skippy The Evil Clown
Dance To The Beat Of Mr Beast
℗ Skippy The Evil Clown
Released on: 2024-07-02
Lyricist, Composer: Skippy The Evil Clown
Auto-generated by YouTube.
god damnit that's so cute -- and wow. I know yolkhead and I were talking about how textual accuracy is sometimes a poor indicator of a model's overall capabilities, but heck.
Did you need to roll many generations to get that?
No, that's the first shot. Honest
I think that text accuracy might be a side benefit of how accurate these things adhere to prompts
These are just images after all
What does it matter if its text or a mushroom police officer
Like the guy has 'police' on his coat
the finger gun to end all finger gun
he so smol
Flux doesn't do prompts like these either which is the other struggle
the classic 'smatter words'
It wants a kinda clear indication of what you want
interesting
what does is make if you just do something vague?
It's a real struggle
or rather, try this
I am adapting your prompt you've been messing with
cartoon, stick figures, low quality, poorly drawn, ugly, worst quality, art brut, outsider art, deranged and manic MS paint girl shooting a rifle
at running people, with the Album titled: "shooting men in the f---ing head"
not anything I'm using
Yeah I'm going to do my style of it then I'll try yours
I'm actually getting it set yp
I mean it's your image prompt, sans negative since Flux doesn't do negatives
yeah but what if I want to generate women bending over? ever think of that, smart guy???

its hugging face
you can gen anything, they don't care
lol
they're all furries
never change huggingface and also civit
oh hell, they really dont care
its 4 step distilled model
hmm, I wonder if that's the base model, since its 12 b params
that's fast inference if so
The dev needs 20 steps
actually, I realized that might not be appropriate
Probably not
haha, I forget america is actually insane
We do this sort of thing irl, never for pretend 
Sorry all, will switch to crazy cat ladies
sorry, I wasn't trying to lol
I just had partially typed already
Woodhouse u prick
this is the actualy space
the other one was the distilled 4 step
this one is a LOT better
28 step
I also noticed that yeah
"LP album titled: "Gun Laws and Cat Paws" with a child's poorly done drawing in crayon depicting crazy old woman hurling fries using her hands at a police officer, the police officer is scared and raising their hands to shield themselves"
Yea, I'm going through my crazy prompts
get hyper hyper specific it seems to really like that
I'm having a hard time adjusting but I'm getting there
... I think Flux wants me to stop using it.
its not exactly right, but it'll do
you can just copy the stack of restatements
and it will likely be a sufficient expansion
I'll make something specific for flux
later on
sell a guitar
kill me
lmao didn't even see that
nothankyou
I'm growing rather fond of your company.
Clearly it is slightly behind her
it's true

pretty accurate
Gongriel Ferdinando's masterwork, titled, "laptop playing gonzo" the composition calling into question
that's the whole prompt
Just in case
also, I made Gongriel up
Gongriel really lost their way when they went through there "Never a Lamp of Lava" phase
I just imagine Kaker visiting the Leaning Tower of Pisa and just passively commenting to people "you know you're not doing anything to actually support the building's structural integrity, right? I just wanted to make sure you were aware of this."
kill me
its a lot less edgy
Why does everyone keep saying that to me this week
or rather, I'm gonna have to adjust my prompts
I have been seeing people just...ad stuff at the end of flux prompts
by comparison, thsi is sdxl to the same prompt
Things like "this was the best photo ever and I won a prize"
a whole new era of cargo culting
It was weird enough when people would walk up and say that to me in the grocery store, now it's just becoming a concerning trend.
why such a big suit???
is it because of ur large automobile
"very early 3D computer generated image from Bryce 3D that shows a car tire with raised white letters around the edge that read "REAL MEAT". The tire is mounted to a pickup truck in the desert"
The suit's the same size as it ever was -- the same as it ever was
'meat' is wrong way around but
and the lugs aren't exactly correct
If you are a car person that's your AI tell
the wheels
I'm not even a car person, and for that Civil War movie, I noticed they used AI images in their ads when it showed a broken down car that appeared to have 2 front ends
I remember that yeah, lots of lazy AI work
and a weird non-door in the middle -- just right off center frame, like... someone got paid to advertise that movie.
It is so easy to correct
hahaha
I think flux will def overtake SDXL if the community is able to run LoRas on top
uh.... wat
"ur about to make such a big mistake girl"
bruh, is flex grooming me
wooooodhouse'd
ffs nevermind
well, my Op Art ones are amazing
it's so much cleaner
Like, other models all struggle on this
bc they get sort of muddy in parts

yea, it needs to go away lol
hmm... I wonder what happens when I combine prompts
hmmm
all model struggle on ballet
I need to figure out where the overtraining is
It's so direct with how it interprets prompts.
"poorly done crayon drawing of a nuclear reactor with the word "jumbo!" above the reactor"
"poorly done crayon drawing of a nuclear reactor with the word "jumbo!" scrawled on top"
It is
But the second one
"scrawled on top" and it went "ok"
But the first one is "above the reactor"
and it went "ok"
And yeah they are fantastically done, not poor at all
ok, it does substantially better on this prompt that was all but impossible on any except DALLE
The 3D tires I was doing earlier, I asked for the words "on the surface" and it gave me this
with "nano-hair"
uuuuuuh
It's on the surface alright
I had to change it to "around the edge"
and it got what I wanted it to do. You have to be extremely specific
which is good!
But takes changing since I would assume lots of things. Flux doesn't assume and tries to follow instructions
Ha! look at that left hand
ok, so it is overtrained
ie if they allow LoRas I can improve it
massively
a mona lisa kudrow
but its replicating existing artwork ie
this is why the above image had the weird hands
(its got mona lisa in it)
ask for a named work by your guy that made the skeleton from eariler
trying it out
"A Mona Lisa Kudrow by Gongriel Ferdinando"
yuuuuup
she burned in

I mean, I'm the only one, to my knowledge, that thinks this way abt them. I think there were others, but they borked SD 2.0 when massively decreasing the weight of this kind of data and the model was awful. However, they also censored it horrifically sooooo
that is a huge component
"A Marna Losa painting by the artist Gongriel Ferdinando"
I think proper nouns and text ability both inflate parameters by many orders of magnitude
hhehehe
oh gongriel
so accomplished
try anything with flaming june
I will get there next actually
but presumably it will wreeeeck it
"pointillism work depicting a lava lamp on a table by artist Gongriel Ferdinando"
My guy, Gonriel has it going on
I wonder if, with data being cyclical like this, fake artists already exist in the synthetic data
Probably
Ok, now I'm going in for the moana lisa
Aw, damn
she's really burned
I mean, its good and bad
this came from
"half-length portrait painting by Italian Leonardo da Vinci
the lady sits, a clever smile held on her lip, the deceptively simple artwork a true masterpiece
the artist captures her visage perfectly on the canvas: she grins with her twinkling eye, 30 in × 21 in
renaissance, a maiden's knowing face captured by the lines of paint, the only of his portraits whose authenticity has never been seriously questioned
del Giocondo smirking, her right hand resting on her left. The period piece is rendered on a white Lombardy poplar panel
the most famous oil-on-canvas in the world, a destination masterwork, there has been much speculation regarding the model and landscape depicted
Dimension 77 cm × 53 cm
Year c. 1503–1506, perhaps continuing until c. 1517
Location Louvre, Paris"
This kind of style allows me to get clarity when using something like SDXL
but here, I'm still getting those burned in fade
the good news is, if LoRas are relatively cheap to train for it
I already have the dataset to fix this
and the model will magically uplift, in my experience, in its generalization, which it already has due to the great text processor
"a copy of a famous work by one Leonardo da Vinci, you know the one, it has a famous smirking woman in it, by artist Gongriel Ferdinando "

hehehehe
you did it!
a flaming june mona lisa
Hmm, flaming June is actually ok
Gongriel!
"a copy of a famous work by one Leonardo da Vinci, you know the one, it has a famous smirking woman in it dressed up in an 80s aerobic outfit with the word "SWEAT" on the headband, by artist Gongriel Ferdinando"
Compare the background to the one you just posted

they like
hell yeah
Does this technique have a name, where you round-about prompt for something without directly invoking
I do it a lot with Suno
its my own technique so no, basically
I called it "diversifying"
dang mine too!!!
Right yeah, something I just 'came to' since how else are you supposed to do it
I love this though, cracking apart a new thing
oh same
Flux is weird and different than what I am used to but not that different
long convo about Flux is long lol.
It's because it's new, cool and on
fair lol.
is this flux?
nope, MJ 6.1 lol.
My PC is too weak to run Flux, I have 3GB of vram bud xD
As I told Kaker yesterday I need to upgrade my PC.
It's free
But I'm also likely moving "soon" so I'd rather deal with that and focus on a new PC at my new place.
Yeah you can hug ur face
I'm using it here atm
I'm running local because I've never regretted going BIG GPU in my life and it keeps paying off
hmm.
I can fit it as well, but I'ma wait a bit
since it will prob run really slow
and atm my SDXL workflows still win out
once I can train loras, then its in
I'm not used to SDXL's UI but I'll give it a try xD
I'm at about 28~ seconds per generation at full strength, 28 steps. Not bad, really

I mean, from one image, apart from the custom aspect ratio, DALLE-3 follows my prompt better lol.
lemme see maybe it needs a tweak.
ok, I prompted it in huggingface, but without any text input, and I got this
so guyz, meet flux
Dalle always looks like dalle to me and I don't exactly like it
You can obscure midjourney and SD gens using prompts but Dalle always seems to be dalle in a way
But that's also probably what people do normally
kinda like SUno all sounds like suno because people don't try to make it sound like not Suno
I like Dalle I just wish it was more user friendly. I couldn't figure out how to get custom aspect ratios on it via chatgpt and i just gave up lol.
closer to what I asked for but Flux still wants to make her human, can't do furry art I guess rip. Unless I'm doing it wrong xD
is a nice quality Kitsune version of her I guess.
lemme try my villain now.
Flux has a steep unlearn process for me
And like I said earlier, doesn't 'fill in' as much. You have to ask
I'd assume the prompt style would be closest to dalle but I guess not?
I don't use dalle so that's ignorance on my end
I mean I know how to prompt a custom version of SDXL so I'm trying to shift to that.
I mean, Dalle-3 can fill in the blanks for you or it can be super specific. One negative is just how super censored it is, more so than midjourney.
and the lack of custom aspect ratios, or ease of access of them at least.
flux still has the too many fingers problem lol. oh well.
the consistency of flux is good though, at the cost of flexibility.
I always feel like there needs to be a balance of those two things.
I don't really see the balance here personally.
you need a bit of randomness to AI Gen's, otherwise you'll only see what you know how to make and not anything new to spark new ideas.
it's still very good quality though.
Just no reason for me to personally use it other than a quick curiousity. 😛
this is actually amazing adherence wow
hmm, try again, basically. Is it on civitai?
I honestly know very little about training on concepts, everything I do has to do with generalization
there are a ton of guides though
dalle-3 is spectacular
it's def up there, I'd say flux def seemed better though
but I love dalle
@strange gorge I was so close to being productive today, you're the devil aren't you
the devil has more skin
hmm, well, what's the goal?
also if you're just doing it at home, using Claude scripts, something might just be wrong
I'd feed your code through successive to several different LLM brains
to see if there's a basic error
in parameters
but I just basically always have the best success on the same setting, bc what I'm doing is super super weird lol
my dataset is literally just burned in images I make using base sdxl
but ya, if it's a character or something like that, civitai allows you to use buzz
to train
hmm, like Mario type stuff?
like, n64m
I'd check your earliest epoch
like, the first one
that's a big dataset
for a relatively simple concept
so it may be overtrained
very fast
ah, ok so a couple things
-
batching can cause overfitting
-
your dataset is so big you're prob getting "there" before finishing your first epoch
-
I'm not sure you need to, hang on
yea, I'm seeing that
Ok, so I think you'll likely want to honestly reduce your batch size
And see if that helps
I do 1 exclusively
but it takes a lot longer
you can prob do 4
2, I'd want to know how they are captioned
bc that will have ab ig impact
it could be there's an issue: with that many images, its hard
you may just want to get like 30 images
and caption them carefully
Can you record songs with iPhone voice memos?
and try that first
bc the concept, it should be able to learn it on a small dataset
its already "in there"
so its less teaching, so much as "signal boosting"
I'd be amazed if there isn't a ton of that in SDXL base's knowledge
ah, ok, so that's prob why here
another component
You can get auto captions
from a number of places
you can also just caption them arbitrarily with works of art
and see what happens
but realistically, minimally you'll prob want to caption them with a "trigger" word
and then basically whatever concepts you want shifted. I think using terms like "beauty" or "pleasing composition" and focusing on that
will work
bc it will glob onto "good aesthetics" as being this
and then normal prompts
will translate
so yes and no
if its this big of a dataset, I think it is going to run a good chance of straight lobotomizing your model lol
But idk, i haven't done training without captions before
oh wow
so maybe something was just set up wrong
that's not right
yea, if it has 0 impact
then its something not set up right
I'm def not the person who could do it, sorry about that, aside from jsut not having nearly enough time
You'll want to ask on the stable diff discord
I'm into my own niche in SD, but SD itself is a massive community
And I know only what I need to to function there
This was funny to watch in real time, @fast rivet , thank you for that
#💬┃general-chat message
oh hey @vale lake
Guess Flux does quite well at 512x512 also
If that is of any use to you, not me lmao get the heck outta here
probably
I actually did have trouble getting it to make bad stuff
but I prob just need to mess around
Right so did some cool tweaks and I got image gens down to 14~ seconds
Using the full model heck yes, 24 gb of vram stay winning
It is, yes
The 'standard, so 1024x1024 and the aspects that fall into that
yeah and this is with the dev full fat model
I meant it when I said earlier I have never regretted going big for GPUs
I had SLI titans back in the day when I had even less sense than I do now
That was extremely stupid and dumb and SLI is the same
well, was, not a thing anymore
It was for me because the games I played didn't support it 
"an early 3D model of a tall, thin mouse wearing flood pants and a stupid looking vest with a unicorn design on it with a speech bubble that says "nice!", as well as brown rubber boots, with their arms open wide almost like they expect a hug"
Not exactly what i asked for but might as well have been
The design is not on the vest, and the Unicorn doesn't have a speech bubble
Not really that tall either
cool wheel
One wheel at a time 
I bet upscalers would work pretty well with this stuff, yeah.
Snap a mouse's fingers? I mean... probably?
Flex or Flux?
Plex and Flux is going to mess me up.
Plex is something my good friend in Tokyo runs, what with is 2gb internet connection 
his plex, not "PLEX" plex
I figured -- I have a Plex setup for my less tech-savvy mom -- I call it MattFlix for her
Just noticed the mouse has a unicorn horn. No! Not what i asked for this is a trash model for trash lords
I am the mom in this case
Am I allowed to borrow your car this weekend?
I have my ancient forums I'm a member of, longer than, I'd guess, most users in this discord have been alive
Of course honey
:)))
The biggest one is NVIDIA control panel turning "CUDA - Sysmem Fallback Policy" to "No"
Otherwise this:
You need all the vram for these comfy UI tweaks
"photo of a hand written note on notebook paper sitting on a table, written in blue pen that reads:
"Flex or Flux? What does it matter? If you are here by the time I get back I am calling the po-leezo."
There is a little doodle of a dinosaur chasing stick figures"
To be clear the tweaks I made are to use the entire fat model and 'best' most memory hungry settings
Not the low-vram options
Agreed lol
ah yeah you got one of them unified memory macs?
Those look sick

The cup and the sign
"vintage photo of a 1986 Nissan 300ZX as a station wagon parked in front of a convenience store in Venice Beach, California. The store's sign reads "Lil Sippy" and has an anthropomorphic cup flexing its muscles"
That isn't a station wagon
The 300ZX isn't at all, which is the point
But that is a cup flexing
THAT'S better
double y's
You can inpaint with this, so stuff can be 'easily' fixed
❤️ oh hey, I know you 😄
Flux is blowing me away since I've been playing with it, so good
yeah the text seems to be better than any AI Art I've seen so far.
Vänern Biggest lake in sweden
Will people take me seriously if i wear a facemask?
Or will they if I don’t wear a facemask
@limpid violet
"gorgeous anime of a nuclear reactor in the background with a cute chubby extremely fine-haired creature playing a Fender stratocaster while a child sits on the ground, surprised and in awe. A man stands next to the child with his arms behind his back, grimacing as they are unimpressed. His shirt says "Wish I Were Dead" in pink letters with a picture of a muffin"
Flux is lol, lmao even
you won't get me this time woody
My DMs r open :)))))))))))))
big woody boy
I'll pit it in showcase
and probably even above space
so no gang violence
space laws
@west hull looking at those two images you posted closer, those are fantastic
The 'Feline University' is what really sends it home
It's back there and blurry, out of focus

@rich moat I thought you posted this
Cricket may have posted it. I suspect that cat of doing many dark things when I'm not looking.
They've got a degree, how dark could those things be?
Same. In addition to the other generative image AI we got this is a fantastic addition
Especially since locally run
https://youtu.be/sENc8-WSPE8?si=KWR-T9hCBMbgNXGu
I've been using AI Suno music snippets in our videos and it's SOO GOOD!
Answering all the hard questions about sports, Did you know that playing golf can be dangerous?? Two Sport novice parents go up against their children and the competition is sports trivia. what is the difference between soccer and football?
🎁 PATREON
Join Our Patreon Dance Party
https://patreon.com/8elbows
▶ SUBSCRIBE to see more!
http...
Used your prompt as a baseline and tweaked it a bit. Works quite well:
An 80s movie poster showing an older sophisticated and stylish Asian Japanese woman wearing shades leaning up against an awesome 80s sports car holding a pistol in a dark, wet alley somewhere in Tokyo. The Japanese license plate says "69NICE". At the top center of the poster, above the title reads "A FILM BY DERBY JONES", the title "STREET VIPER" is made out of a single snake that ends in the head of a snake hissing. Below the title, a captivating tagline reads "She WILL Catch You".
At the bottom center of the poster a block of text contains typical movie credit information:
"Rated R"
Not quite
It's kind of a mix of a few, looks like
I didn't specify
hi :3
Hey, welcome to the server!
Caught me just as I was heading to bed, but it's a pleasure to make your acquaintance. 😄
Have you made any music with Suno yet? It'll get blocked if you try to directly share the song link here, but if you share the song in #🐣┃suno-showcase, @ me and I'll listen in the morning 🙂
You're just living on Flux now, aren't you?
I'm so excited to really dig into it
It's good stuff
https://www.instagram.com/reel/C-NwmU1ooMD/?igsh=emxlM201NTJncHk1
Flowers from yesterday
a book you used to read to me from my album Late Bloomers
.
.
.
.
.
.
.
#rain #dewdrops #rose #rosepetals #artificialintelligence #aiart #aimusic #reels #explore #fyp #suno #sunoai #sunomusic #digitalart
(my wife runs my insta btw)
but incidentally, I think the song she chose is by far my personal fav I've made on suno
but it's v sentimental
but I made it back in the v3 days
I agree, I've had a bit of time to dig into it
I think it's going to be interesting, because it seems that finetunes won't be happening, at least according to the devs, which implies loras won't really be a thing either
however, it's really good on it's own
and this doesn't preclude them from releasing their own "preference lora" improvements
so basically you could end up with a local model that is highly managed by a company
which is kind of a cool concept
@strange gorge is going to be thrilled when he wakes
I'm still awake buckwheat
hahah
and yeah that looks sick, and since they delivered with FLux I have no reason to doubt the video thing
also did you see this
https://www.reddit.com/r/StableDiffusion/comments/1ejlvuw/simpletuner_now_supports_flux1_training_lora_full/

yup, same, it looks great
I guess the devs weren't the actual dev devs, but where some other devs
oh interesting, checking it out
~devs~
oh yea, simple tuner, I've seen this; it was the only one that supported cascade
My understanding was it was more related to the cost to do it
makes it prohibitive
but if we get a few people together
Considering the waves Flux has caused
I'm down to try out erasing baked in stuff
and see if it improves the model
which is already amazing
it also is likely to unlock um... things
in my experience
SDXL is a lot less censored
after removing overtraining, likely bc it generalizes
and then removed whatever weird aftermarket dataset they pushed over the top to bork nudity
6.6
In using Flux there is this depth to it
I bet you SDXL would be a genius if I could get people to stop training it on text lol, its too much for the thing and a waste of brain
but people don't listen even when I get results, somehow
anyone tried kling ai premier yet? its 50% off so i am lowkey tempted to try it
true
I mean, its def better trained
at base
on top of having more params
and it's super precise
this imo will also make it have a little more wiggle room when fine tuning
That Street Viper poster, you can just kinda....tell it where to put things
anywhere in the prompt
in my experience ie it is less finnicky
and it just...gets it
now this is worthwhile
more like
NICE
a joy
oh, its the new ayers
the pop one, its on the big playlist
I put it third honestly
the sound is good
listening now
second I mean
