[HN Gopher] The 'Compact' Version of Stable Diffusion 3 Is Gener...
       ___________________________________________________________________
        
       The 'Compact' Version of Stable Diffusion 3 Is Generating Monstrous
       Human Bodies
        
       Author : poopcat
       Score  : 36 points
       Date   : 2024-06-13 21:59 UTC (1 hours ago)
        
 (HTM) web link (www.xatakaon.com)
 (TXT) w3m dump (www.xatakaon.com)
        
       | turtleyacht wrote:
       | I've never seen a drawing book start from outside-in. We learn
       | from the massed blocks, sketched curves, and mirrored
       | proportions. The shapes, the skeleton, the muscles and skin.
        
       | minimaxir wrote:
       | https://news.ycombinator.com/item?id=40661990
        
       | elpocko wrote:
       | SD 1.5, on the the other hand, with the right checkpoints
       | generates flawless human bodies. Older tech, much better results.
       | 
       | Can finetuning fix this?
        
       | haunter wrote:
       | Thanks to the horseshoe [0] puritanism we will live in a safer
       | world. It's all right, everything is all right, the struggle is
       | finished.
       | 
       | 0, https://en.wikipedia.org/wiki/Horseshoe_theory
        
         | spacecadet wrote:
         | Oh man do I mention Horseshoe alot... Its all extremism either
         | way.
        
       | margorczynski wrote:
       | Just what I needed - some nightmare fuel before I go to sleep.
       | But putting that aside - how is Stability still going? The VC
       | money didn't dry out yet?
        
       | tamimio wrote:
       | Plot twist: This isn't monstrous; it's just how AI envisions our
       | future!
        
       | amanzi wrote:
       | Even the good AI image generators struggle with details like
       | fingers, and ear lobes, and other small but important details.
       | 
       | Also, I found this quote interesting: "...Stability AI's
       | insistence on censoring adult content from SD3's training data".
       | Da Vinci figured this out hundreds of years ago, that to draw
       | accurate pictures of humans, you need to understand the human
       | body.
        
         | TeMPOraL wrote:
         | Yeah, it's especially ironic given how big a role nude
         | modelling plays in art schools - it's fundamental to drawing,
         | painting and sculpting.
        
         | lifeisstillgood wrote:
         | I think the point is you need a naked human body - art classes
         | use volunteers, midjourney presumably uses pornhub. I guess
         | there is not a lot of naked human images out there that's not
         | porn.
        
           | TeMPOraL wrote:
           | I imagine if they filtered out just the bits with some action
           | going on, they'd still have approximately infinite feed of
           | naked human bodies in all possible poses, which would be
           | unoffensive out of context (or at least less offensive), and
           | would be equivalent to art classes models for the purpose of
           | training.
        
           | kergonath wrote:
           | It really is not difficult to find breasts, bottoms and well
           | proportioned limbs on the Internet. Even in underwear if you
           | must because you're afraid of something.
        
         | roenxi wrote:
         | Although it'll be _fascinating_ to see how things like the
         | finger situation change as we get models in common use that
         | also handle video well. One thing that becomes obvious quickly
         | with Stable Diffusion is that it doesn 't have a good concept
         | of a 3D world model to work with; or any grasp of physics. It
         | is practically impossible for a human artist with SD levels of
         | competence to accidentally draw multiple limbs and that speaks
         | to how it just doesn't get that humans are a 3D existence.
         | 
         | Once video data is involved though it seems likely that will
         | change. And I reckon a side effect is that a lot of the
         | trickier details will improve. It'll be a great experiment to
         | figure out whether the models also need lessons in anatomy or
         | whether they figure it out through pure observation.
        
         | lovethevoid wrote:
         | Good AI image generators handle those details well already.
         | Even if you don't get it 100%, you can inpaint fix very
         | quickly.
         | 
         | The current problem for them really isn't the details in
         | isolation but rather cohesive details throughout the entire
         | picture in one attempt. It's very lacking and requires a lot of
         | manual input, filtering, reliance on multiple tools, etc.
        
           | kergonath wrote:
           | > Good AI image generators handle those details well already.
           | Even if you don't get it 100%, you can inpaint fix very
           | quickly.
           | 
           | But that should not be the case. A human body is not more
           | complicated than a horse's or a cat's, and those are usually
           | much better. There really is a problem in our relation with
           | our own bodies.
        
             | snapcaster wrote:
             | I do think we have much more sensitivity to deviations in
             | human appearance. We probably wouldn't notice a lot of
             | "weird looking" cats that other cats would
        
         | kergonath wrote:
         | And sometimes that body is nude. On one hand I hope that this
         | will help some people understand that their standards of what
         | is SFW or NSFW need to be adjusted. Also, why the hell are we
         | using workplaces as a yard stick for what is acceptable in our
         | private life? Why on earth is it a problem if kids see tits at
         | home?
         | 
         | Anyway, I hope standards will change in the US faster than the
         | rest of the world finishes shifting to those utterly stupid
         | standards.
        
       | 8f2ab37a-ed6c wrote:
       | Has genAI for images hit an asymptote in the last year or so?
       | Can't quite tell if things have gotten noticeably better since
       | around when DALLE 3 was launched, it all looks about the same
       | quality.
        
         | elpocko wrote:
         | Yeah, the mainstream stuff that you get to use while subjected
         | to heavy surveillance and censorship is about the same. There's
         | a whole universe underneath that, with no restrictions and
         | basically unlimited creativity, if you have the required
         | hardware.
        
       | heavyset_go wrote:
       | You might not like it, but this is what peak human performance
       | looks like.
        
       | philipov wrote:
       | This is what AI creativity looks like.
        
         | A_D_E_P_T wrote:
         | AIs tend to do interesting things only by accident. But
         | interesting accidents happen quite frequently.
         | 
         | What SD3 has done here is "good art," as modernity defines the
         | term: It's unique, it's technically adroit, and, most of all,
         | it's shocking, thought provoking, unsettling. You can only
         | wonder what went through the mind of the "artist" -- what it
         | was trained on, how it attained such a strange result. This is
         | at least gallery-tier if not worthy of MOMA, and I'm 100%
         | serious about this.
        
       | peterclary wrote:
       | "I have no mouth and I must scream" territory.
        
       | peterhi wrote:
       | So they looked at this and said "ship it"?
        
       | bitwize wrote:
       | "Suppose you get a salamander. You know, a salamander can regrow
       | its limbs if you cut them off. So all these nasty biology
       | students do things like, they'll cut a limb of the salamander --
       | a hand, the arm off, just between the shoulder and the elbow, and
       | at the same time, cut the same one between the elbow and the
       | hand, flip around that segment, and re-sew it back on. Nasty
       | thing to do to a salamander. What do you think you get? Does
       | anybody here know? What you get is three elbows. It grows two new
       | elbows to make up for the fact that the wrong thing is connected
       | to the wrong thing." --Gerald Sussman, "We Really Don't Know How
       | To Compute!"
       | (https://www.youtube.com/watch?v=HB5TrK7A4pI&t=2m49s)
       | 
       | These images look like someone spliced salamander regeneration
       | DNA into a human, cut them into many pieces, and connected the
       | wrong thing to the wrong thing all over the place. Maybe that's
       | what the AI wants to do to us, AM style (I Have No Mouth and I
       | Must Scream)?
        
       ___________________________________________________________________
       (page generated 2024-06-13 23:01 UTC)