Stable Diffusion 3 is... something
Вставка
- Опубліковано 8 сер 2024
- Stable Diffusion 3 released and it's... not great. It's plagued with issues and the AI community doesn't feel like it lived up to the hype. Let's take a look at some SD3 image generations, some cursed, some not bad.
In this video I use ComfyUI and Stable Diffusion 3 to create AI generated images.
00:00 Stable Diffusion 3
00:49 Stable Diffusion Subreddit
01:34 SD3 Comparisons
2:20 My Good Generations
Follow me!
Gaming channel - @Greenskull
Discord - / discord
Instagram - / greenskull
TikTok - / greenskulltok
#ai #artificialintelligence #stablediffusion #cool #funny - Розваги
I love a good cursed AI image though, I must admit.
Same it’s the only reason I use ai it’s so funny
Greenskull you interact with your community a lot for how big you are as a channel
It's cool
Btw when more MOVTIR
Admit what
@@Jelly_shy_guy_man that he loves a good cursed ai image
Those humans are:
Cursed.
Scary.
Weird.
Strange.
Creepy.
Extraterrestrial.
Goofy ahh.
CSWSCE for short
Ai is telling us to touch grass, by making itself unable to make anything but grass
I'm guessing this is what happens when you train a figure artist but forbid them from starting at the obvious base point that every figure artist has started at for hundreds of years... nudes.
One of the funniest videos ever. I must've belted out, "hlahahia" a thousand times.
hlahahia
Emphasis on the hla
Hlahahia
Do you remember me I’m the Bobby guy I love how much you’ve grown I remember *old man voice* back in my day green skull ai was a new channel
Gotta get this up higher
1:03 The rant below is Unfathomably based
“Ai will take over!”
AI:
Clearly you didn't see the video before this one that he posted...
@@Pookerbears that one was just a game tho
Lol not all a.i same
This is the worst, Stable diffusion is mistake, glory to midjourney 6
They are purposefully gimping it
THE HECK IS THAT!!
Thanks for doing the tests n sharing w us
Stable is just craiyon but high quality
Craiyon is painful
More like "craiyon but HD"
@@_CuteHamster_ yep that sounds better
@@UA-camsViewerReviewer YES like 99% of the things I ask it to do it turns it into a mess.
@@Racecar-roaster Yeah
As a regular Stable Diffusion enjoyer, locally, I had already switched back from SDXL to SD 1.5 because I actually had the VRAM to run it without it crashing after every few usages. And SD3.0 is supposed to require even more VRAM? No thanks, I'll wait until I can afford my next GPU upgrade.
Depends which model you use. There is at least one model that works well with 8Gb VRAM and others have claimed it works with 4Gb VRAM. If you can't run SDXL without crashes you may struggle but I don't know what your card specs are.
@@Elwaves2925 my GPU is an RTX 2070 Super with 8 GB of VRAM. I was using the RealVisXL model.
Sounds like they must have trained the AI on some photos that have anti AI noise
Will the 8 billion parameter version ever be accessible for local install?
It isn't just grass it's people 'laying' on anything, whether that be a sofa or the ground. Even then the issues with anatomical meltdown goes beyond just that word and appear to be part of the model in general. It's fun to play with for a while, but I'd be willing to bet that with SAI's more corporate future ahead, SD 1.5 and SDXL will be around for a long time to come.
You forgot me😢
I said remember me when you reach 100k
Dang that sucks
Ohhhh ... my bad, Bob right? Hi ... how are ya?
I don't see a comment of you saying that 😹
how does this only have 511 views rn?
For some reason this is getting no views lmao
maybe he was right when he wondered if that picture was YT safe
@@edu.33 lmao 💀
because its rambling about nothing...
1:17 You've heard of hot pocket, now we got girl pocket.
That's a sick mad max jacket
The licensing agreement has changed.
It's a bad omen.
Iv had wierd images like this in Midjourney lots of times and the same with DallE
I hope it improves soon but the results are funny
It will not. They did it on purpose ("safety")
Even if they release the 8B model or a 4B model, people likely won't latch on like SD 1.5 or SDXL because SD3's license is absolutely terrible.
SD 1.5 AND A1111 FOREVER!
Ai is so goofy tho 🤦♀️😂
This video is so hilarious XD
I tried using it and it generated me a cursed girl with 2 fingers on her elbow. 😂
I have a strong suspicion that this is mostly to do with economics rather than technology. The upshot is that A LOT of people have recently gravitated to making people so having the system f*ck up their anatomy plays into the makers of these AI image generators. Having you spend a lot of credits just to get a good image just makes absolute economic sense to me.
The thumbnail- wth
Stable Diffusion 1.5 sucked too. But the community trained models that made it a decent generative AI. But Stable Diffusion changed its licensing agreement to prevent that from happening with SD3. So what you see is as good as it's going to get.
I'm sticking with 1.5😂
SD3 Medium (Local) SD3 Medium (API) SD3 Large (API)
"Look at me "Look at me "Look at me
I'm
I'm Yynchhhla". hahaha "
i ^ m ty'ping
hathha"
So... No humans and a five paragraph essay should do the trick!
even the local medium "free" model looks better than their larger paid ones in the latest incarnation. It's like the company is going backwards, they're getting worse and worse with each new model version
Me personally I am more of an sdxl kind of person
Yooo wassup dude
yoooo
... think I'll stick with DE3 and MJ6.
Meth is one hell of a drug
Licensing issues will make this useless regardless of how bad it is. You only really have limited rights to your work if you subscribe to their paid service and as soon as you stop paying that you will lose any kind of ownership. This will also cause big problems for those wishing to finetune models also from what I understand, so might not even get much finetuned models that will fix these issues.
Are you ai
This is so sad.
MIDJOURNEY 4 EVEN MORE GOOD THAN THIS
Erm what the sigma
It's 2:30 am bro
Timezones exist.
@@Cara.314 no way Sherlock
On the one hand, none of the base models have been great. Okay to mess around with IMO but not for actual use.
SD3 is failing with humans in many case outside of portraits, it clearly needed more training in that department. For the rest I've found it's okay for a base model but whether it's usable in the long term will depend on what the community comes up with and how easy it is to get around the in-built censorship.
I prefer the problems
Am i the only one struggling to just look at ia generated cursed humans and body parts (hands for example) ? Idk it's so weird that it's makes me very unconfortable
Cool
Nice
I’m an ai
are you?
@@GreenskullAIare you?
@@GreenskullAI idk am I I thought you were an expert at this greenskull
5
0:05
Ḅ̶̡̭̫͇̲͖̞̥̠̠͉̒̈́͊͑͜E̷͎̞̬̬͍͔͕͖͇̟̻͙̱̦̹̎͒̾͂̽͑͐̿̍̃͛́̌̕͝ ̸̛͉̳͕̣̩͙̤̫̱̣̼͋͑̅́̈́͛̒̕͜͝ ̴̢̛̛͔͕̥̥͓̘̜̟̫̝̹̺́̔̃̈́͋̽͘͜͝͠N̴̢̨̛̟̜̺̤̜̼̜̻͔̞̍͆́̈́̉͊̃̒̿͜͠Ợ̵̡̩̲͎̰͈̲̜͖́̈̊̎̿̄̔͗̎T̷̡̧̰͕̖͎̜̹͔̳͚̝̳̈̽̀́̈́̋̾̽͑̈́͘ͅ ̶̯̯̠̬̅̓̂̂́͌͐̕͘̕̕͝ ̷̺͕̝̞͎̖͂A̸͕̬͍̳̽̓F̸̨̨̭̥̰̪̗̱͈̻̎̉͒̆̈̐͐͑͑̅͆́̈́̕͝R̴̰̤̄̒́͋͒̑͗́͆͆͛͜͝Ȃ̵̧̛̳̺̯͍̐͘I̶̗͔̳̜͍̹͚̱̣̟̰̮̞̣͗͜Ḑ̴̱̻͔͈͎͓̺͙̂̽̇̉̎̄̀̉͒̕
this early?
W
To me it looks like i’m first but i’m actually not
🤣🤣🤣😂😂
👽🥄 Spoon gang checking in! 🌀 If you also believe spoons are superior to forks, leave a like. Remember, the spooniverse is watching. 🌌👀 #SpoonGangOrbit 🛸✨
What
@@TheMiningCrafter it’s a copypasta made by ai in one of his vids
666 likes
lol 💀💀💀
only 23 comments, 33 likes, and 178 views? come on!
first :D
49 th
first
Congrats on being first my guy