AI-Time Travel-Street View
(timetravel-streetview.glitch.me)39 points by _Microft 7 days ago | 23 comments
39 points by _Microft 7 days ago | 23 comments
nacs 4 days ago | root | parent |
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.
milliams 4 days ago | prev | next |
I think the title should be "AI time-travel street-view" if it's going to have hyphens anywhere,
pimlottc 4 days ago | prev | next |
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 3 days ago | root | parent |
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 180° 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.
jtxt 4 days ago | prev | next |
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.
quantadev 3 days ago | prev | next |
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 3 days ago | root | parent |
Yes, I think it is hard-snapping to certain rotation angles when it gets close.
quantadev 3 days ago | root | parent |
ok, maybe that was it. I didn't stay around long enough to notice the pattern.
michaeljx 4 days ago | prev | next |
This is cool but I do wanted it to be more realistic. This is too idealistic
andybak 4 days ago | root | parent | next |
> 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 3 days ago | root | parent |
"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.
tantalor 4 days ago | root | parent | prev |
The style looks like a comic book to me
dvh 3 days ago | prev | next |
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.
pwdisswordfishz 3 days ago | prev | next |
What’s a “Travel-Street” and “AI-Time”?
GiorgioG 4 days ago | prev | next |
Great, random themed location images.
lnauta 4 days ago | prev | next |
I'd never seen generated dinos before, they are way messed up!
toasteros 3 days ago | prev | next |
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 3 days ago | root | parent |
>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 3 days ago | root | parent |
The linked images are very noticeably bad. Bad enough that I honestly don't understand why someone would post it.
3 days ago | prev | next |
mvdtnz 3 days ago | prev |
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 3 days ago | root | parent |
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...
mhitza 4 days ago | next |
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.