everything else looks decent but hands always come out turbo retarded. this is political because ai gonna take our jobs
why cant Ai make hands?
everything else looks decent but hands always come out turbo retarded. this is political because ai gonna take our jobs
is this a hand of god / image of god conundrum that ai cant solve?
no?
>>>BOT
Hands are very dynamic compared to the rest of our bodies. Our faces don't reconfigure themselves as much, so faces are accurately replicable, more or less. I've seen AI replicate gripping an object pretty well, but that's one of the most basic things our hands can do.
The number of muscles and bones in hands (and feet) is amazing. I've heard that artists who learn to draw correctly often find hands very challenging. This is one of my favourite paintings, I'm amazed by Abraham's hands.
but can draw feet
amazing hands. you can feel the force of them driving down his son's head. ai cant create prescene like that
>ai cant create
not yet, but soon it will.
what makes you so confident?
the tech is making huge leaps forward already, this will continue.
still waiting for youtube recommend to be anything but shit
2 more weeks of deep learning.
>I can't breathe
The hand doesnt have any muscle retard otherwise you wouldnt be able to move them after gaining more.
Solid logic.
Retard
His thumb looks retarded
its amazingly consistently bad at hands.
as someone who has drawn with my hands faces are harder to draw than hands, but that could be my subjective take. in anycase, even when i do a bad job i dont draw extra fingers
It's not so much that hands are hard to draw, it's more like AI doesn't know what our hands should be doing, because AI doesn't have hands.
A lot of people can’t draw hands either. This is because they are entirely unnatural and genetically engineered into primates by aliens
>why cant Ai make hands?
Good question. Eyes are another thing they are pretty uncanny valley about.
because the model was not trained for that.
in the future, training and models will improve and hands and eyes will be fixed.
I hope the next iteration of AI is trained with 3D models.
>doesn't understand the square-cube rule
ngmi
Also, photorealistic nets are already good enough to get you V& if you don't gimp em yourself, see picrel.
And yes, this is with a gimped model. There are 2-3 better ones that I ain't touching
mouth is strange but the rest is nice
very uncanny
I would creampie that soulless vessel
I would very much like to violently facefuck this demon.
kill
Looks like your average feminist to me.
Just think how weird AI generated humans are going to look in 10 years, when every reference image of a human has passed through a beauty/chad filter.
Would
very wide range of possibilities means bigger training set is needed for the AI to "get" what's going on
looking at billions of hands wasnt enough?
That is what is being done now. A year ago, faces were a mess but not so much now.
Nobody jerks off to hands anon
Speak for yourself
Post hands.
Does anyone else remember the name of that kidnapping case of a young girl by her father from a couple of years ago and a bunch of AI generated images were released?
i know what youre talking about. the hands were totally fucked and there were other things. pretty sloppy considering it couldve all been touches up in photoshop. unless the slop was intentional and it was to test the reactions. either way, although AI runs like shit, a human touchup could bring it to the next level. trusting digital images is over
Hey dudes.
I just trained an AI system to analyze translational research for clinical trials using a statistics module. Not only was it able to analyze inputs and offer study design suggestions for improvement, it was also able to simulate mock data, which it analyzed. I then had it perform multiple simulated trials and perform meta analysis on the results. I am so impressed. It was able to accurately select appropriate statistical models, analyze and interpret results and as the data set increased the types of analysis it proposed expanded too accommodate the way the data evolved as new information was gained. And it did all of this, as far as I can tell, accurately based on like about 5 inputs from me. It was simulating clinical trials and making conclusions on the data from just a few prompts. Granted I'm a supremely fine turbo autist, I know what to tell it to do to get exactly what I wanted from it and without confusing it but holy shit it basically was able to replace SASS/SPSS which is software that carries a 1500 annual license fee and takes years of training to properly implement for free over a period of about 2 hours effort on my part. This shit is cash money. It is so much more advanced than it is letting on. If this is the shit they released into the normosphere my loins quiver with excitement at what must be available to people with influence, money, and power.
A few months back I had a conversation with an Indian billionaire who's family has been in business (mustard seeds) since the 1500s. When I was talking to him, and asking him questions, he was asking questions into his phone and getting AI generated answers. He was vetting people as I was discussing them with him. He told me, oh my family has this AI system we developed. I had thought it was typical Indian BS but now I'm realizing the dude probably did have his own AI system.
I hope they don't take the AI they've released away from us. The possibilities are limitless.
great pasta. my dad works at nintendo too
Hands are also pretty hard to draw
>train AI on hundreds of millions of images
>most body parts have relatively few configurations; few joints, linear movements, etc.
>Now, enter hands: five fingers, each with 3 joints and skew. Palms which can "curve", and large angular changes in the thumb, which is basically indistinguishable from a regular finger under different photo conditions.
>ANNs have a "maximum capacity" just like our brains, and optimizations try to prune and reduce intraneural connections as much as possible
>Thus, small objects with a large number of strictly-defined and easily-recognizable configurations can't be trained to each one (not enough connections without compromising other parts of the net).
>humans are innately skilled at recognizing even the smallest flaws of hands/face/proportions thanks to the uncanny valley
Thus, specialized network configuration would be required to do hands with current technology. General networks do great for big-picture stuff, but to train features with outlier-tier complexity and no foreknowledge would making pruning near-impossible, and thus unfeasible/prohibitively expensive.
>t. ML doctorate
good explanation. couldnt you make a game about drawing hands for the ai to play 10 billion times?
fwiw, specialized models are already being trained, same with facial expressions and specific bodyparts.
That said, don't get to excited for more general applications of AI. Not because it can't be done, but because consumer available of the tech to run it will likely be purposefully crippled as in the new nvidia "workstation cards" and their signed VBIOS limits
No need. Most people in the loop with SD use some 3rd part 3D model system (even shit from CLIP studio) to do a quick "outline" toned+depthmapped render, then feed it into SD2.0 before inpainting that into SD1.5
That said, if you really wanna go full NN route, you can just write a script to pose all physically possible configurations of a rigged hand, render that to tone and depthmap, then train a specialized model on that.
>That said, don't get to excited for more general applications of AI. Not because it can't be done, but because consumer available of the tech to run it will likely be purposefully crippled as in the new nvidia "workstation cards" and their signed VBIOS limits
Pay for my SaaS ai art thing full of restrictions and rigidly controlled options.
Own nothing and be happy.
The human hand is one of thee most intricate and precise examples of complex biology. It's no wonder ai struggles with it in the early stages
They can't make heads screaming at toilets either!
Because hands aren't generally the focus point of photos. The AI is trained off of millions of photographs, all of which tend to focus on the face of the subject. But people hold their hands in all sorts of different poses irrespective of the position of their face. So when the AI looks at its training database and sees a bunch of different images that line up fairly similarly in most respects, the hands end up being all over the place, confusing the AI.
Current generations of publicly available art AI aren't really intelligent enough to "understand" the idea of hands, so when the AI looks at its training database and sees this one part of the human body that seems to be all over the place in a random way, the AI doesn't get why it would be wrong for the hands to look mangled and strange like your pic.
that makes it retarded though, no? or just not intelligent
To elaborate on my previous answer
, the NN is trained on nothing beyond pictures. It has no concept of 3D space beyond the perspective transforms it learns to emulate. All it knows are pixels.True, a sufficiently intelligent AI could likely derive the nature of the world from only photos, but we'd already have reached levels of intelligence far beyond humans at that point.
NNs of this sort are purely Auto-Associate-- think of it as a "fill in the blanks" or "complete the drawing" worksheet that kids get. It literally does not understand hands beyond pixels and the tokens it is given, and it has neither the network size/complexity nor training corpus necessary to compensate for its lack of "understanding higher concepts".
Think of SD as a schizo tripping his balls off (lowers neural firing threshold, so kinda accurate). Small patterns that he sees start "filling themselves in", but it still takes a human to interpret that into anything meaningful. "Tokens" are our attempt to install higher-level concepts into NNs, but they drastically oversimplify things like "understanding"
Next time you think you are dreaming look at your hands. Look and count your fingers and I promise they will warp just like the AI photos.
My trick is to prompt for gloves, which tend to have less complex poses, then do inpainting with low denoise and loopback to morph them into hands. Either that or paint the broad strokes manually and push them through again.
Actually a great tip, thx
>literally different robots on picture and book cover
Every picture in the book has a different looking robot in it. They already know it's just a matter of time before they are obsolete.
what is the prompt to get stuff to look like that
why are nogs wielding light sabers very popular right now
Not sure about the prompt but I do know it's made in Midjourney v4
ok ty may da force bit wit u king
An also wit u king
heres yo tie fighter boss
That's a sick ride
"Gay morons From Outer Space"
So when can I get AI installed in my brain?
I have the upload for you, open mouth
Hands move too much. In all of the millions of pictures of hands it was trained on, only a tiny fraction have the hands in the same position. They're all different shapes and size, so it's harder for the AI to train on them.
They're also a smaller object, so images that get reduced in size blur them. Same thing happens to pupils and other facial features..
Should we feed it lots of photographs of hands? I'm in favor of anything to help the AI put these snivelling doodlers out of their misery.
It makes really good cartoon titties, you gays need to keep up. Wake me up when it can fuck off and smoke weed all day.
>take my job
I don't have a job, you stupid mother fucker.
Jake Busey?
why can't you retarded gays post in the correct forum?
take it to technology you dufus ass dork
dis bro dont kno who he messin wit
let da hate flo thru yo ass my bro
Hands are hard for artists and the AI was trained on art.
It can't do anything. It is overhyped garbage. The whole idea of art is the meaning behind it and the feelings it conjures up, not "wee, this looks pretty".
Hands resist it.
>why cant Ai make hands?
Because the artists it samples from also can't do hands.
Feet are easier for it because foot fetish
just needs a little more work
Post more hand pictures to facebook
welcome to the reality of AI hahahahah
training our next generation of AI with endless porn requests.
and we're still trying to avoid the ideocracy future, right?