[HN Gopher] AI-Time Travel-Street View
       ___________________________________________________________________
        
       AI-Time Travel-Street View
        
       Author : _Microft
       Score  : 31 points
       Date   : 2024-11-07 11:52 UTC (3 days ago)
        
 (HTM) web link (timetravel-streetview.glitch.me)
 (TXT) w3m dump (timetravel-streetview.glitch.me)
        
       | michaeljx wrote:
       | This is cool but I do wanted it to be more realistic. This is too
       | idealistic
        
         | tantalor wrote:
         | The style looks like a comic book to me
        
         | andybak wrote:
         | > This is too idealistic
         | 
         | I think you meant something more like "stylized" - I've never
         | heard idealistic used in this context (although I've been wrong
         | before with things like this - feel free to correct me)
        
           | svara wrote:
           | "idealize" can work in this context, meaning (in art
           | specifically) abstracting from the concrete form to show the
           | essence of a thing.
           | 
           | Though looking at the images "stylized" seems to fit better.
           | 
           | https://www.merriam-webster.com/dictionary/idealize
        
       | jtxt wrote:
       | Neat, I played with 360 equirectangular projection gen for a
       | gamejam last year. It has some interesting possibilities and
       | challenges.
       | 
       | I started to image segment objects then generative infill to
       | replace the background and make objects in the forground to work
       | have z depth for vr, used switchlight ai to make objects into PBR
       | textures so I could have dynamic lights...
       | 
       | I later found this tool for generating 360 panoramas:
       | https://skybox.blockadelabs.com/025e37f6fb27d9f2495cbbbf874f...
       | 
       | What would be cool is to have ai generated scene that is
       | spatially coherent and explorable. I think it's possible with a
       | bit of tooling now...
       | 
       | Curious what the author's plans for this is.
        
       | mhitza wrote:
       | Looks less like a street view, and more like 360 images that link
       | to one another. For example with future LA, I could only jump
       | once forward and then back and the images didn't have any
       | continuity between them.
        
         | nacs wrote:
         | Agreed. There is almost 0 continuity between the images as you
         | move through the world.
         | 
         | I think a true AI streetview would require rendering these
         | images on to 3d primitives then re-rendering areas with some
         | overlap so there is a smoother transition while navigating
         | through it.
        
       | lnauta wrote:
       | I'd never seen generated dinos before, they are way messed up!
        
       | milliams wrote:
       | I think the title should be "AI time-travel street-view" if it's
       | going to have hyphens anywhere,
        
       | GiorgioG wrote:
       | Great, random themed location images.
        
       | pimlottc wrote:
       | This is cool but the interface could use some tweaks to be
       | useable, or at least more consistent with actual Google Street
       | view.                 - transitions do not animate between camera
       | positions       - positions are spaced so far apart that it's not
       | clear how they are connected.  Google Street View normally has
       | many positions close together, so you always have a sense of
       | where you are relative to your last position.       -
       | acceleration for spinning the viewpoint is too high, I kept
       | overshooting my target and having to go back and forth, dropping
       | down to a slower speed for accuracy.       - the icons disappear
       | when you click, which no explanation to get them back.  I didn't
       | even realize they were off at first and just thought there
       | weren't many icons       - the "info" icon enlarges on hover,
       | suggesting it's interactive if you click on it, but nothing
       | happens       - arrows on the ground would be nice
       | 
       | Some of these are a bit nit-picky, but if you're going to call it
       | "Street View", then you set certain expectations. If you're going
       | for a different experience, it should be called something else.
        
         | drdaeman wrote:
         | It also lacks scene consistency. E.g. in the France level I
         | walked into the cathedral (arrow on the building visible at the
         | starting location) and got really confused as the exit was
         | nowhere to be seen (I turned 180deg and it wasn't there), arrow
         | icon was on a stained glass window.
         | 
         | And if we're going for a wishlist, I'd say it's a shame there's
         | no WebXR support. It has stereoscopic mode already, so WebXR
         | feels like a natural choice.^K Never mind, just realized it's
         | not a software project - just an art project on some platform.
        
       | pwdisswordfishz wrote:
       | What's a "Travel-Street" and "AI-Time"?
        
       | mvdtnz wrote:
       | I don't get it. Clicking the arrow buttons just seems to
       | alternate between a couple of images. Clicking the 'era' buttons
       | just catapults me to a different scene with the same problem. I
       | assumed I would get the same scene but in a different era but
       | nope.
        
         | throwaway314155 wrote:
         | There's nothing to get. It doesn't do what it claims to do.
         | What's more concerning is that it's still on the front page for
         | so long somehow.
         | 
         | Almost like people upvote titles without actually clicking on
         | them...
        
       | toasteros wrote:
       | AI image generation is complete crap, a total waste of energy and
       | resources.
       | 
       | The Houses of Parliament most definitely has only one clocktower.
       | 
       | Cool idea, but maybe stitch John Atkinson Grimshaw paintings
       | together instead of crummy A"I" images.
        
         | GaggiX wrote:
         | >AI image generation is complete crap
         | 
         | It depends on the model, nowadays you can generate images that
         | cannot be distinguished from stock photos, illustrations, anime
         | images etc...
         | 
         | I guess some people think it's bad because the only time they
         | notice it is when it's done poorly.
        
           | stephenr wrote:
           | The linked images are very noticeably bad. Bad enough that I
           | honestly don't understand why someone would post it.
        
       | dvh wrote:
       | I think better way would be to have very crude 3d procedurally
       | generated world and then use style transfer. This way the world
       | wouldn't change with every step. The important parts would still
       | be there, for example if you going towards building it would
       | still be there in every frame just getting closer. At worst it's
       | style would change but not the actual building presence or basic
       | shape.
        
       | quantadev wrote:
       | Tried to use this, but the mouse drag to change view vector is
       | jank. Doesn't work right. So frustrating I left in about 15
       | seconds. I'm on Linux Brave browse, but I don't think the issue
       | was on my end.
        
         | mhuffman wrote:
         | Yes, I think it is hard-snapping to certain rotation angles
         | when it gets close.
        
           | quantadev wrote:
           | ok, maybe that was it. I didn't stay around long enough to
           | notice the pattern.
        
       ___________________________________________________________________
       (page generated 2024-11-10 23:01 UTC)