A passthrough for any video?

deni12

fenderwq would like to know aswell! Capcut is very handy and easy, but also lacks quality and options. and i cba to used after effects or premiere with nice quality but bad workflow and redicolously long rendering times.

fenderwq

deni12 Gonna play around with U-2-Net when I have a bit of spare time. Not a program but seems easy enough to implement (in action: https://github.com/xuebinqin/U-2-Net#usage-for-salient-object-detection). There are also quite a few wrappers already available, like this one for python: https://github.com/nadermx/backgroundremover.

kruk23

Try this model:
https://github.com/PeterL1n/RobustVideoMatting

It handles VR scenes surprisingly well (especially if lighting is "normal"):
https://www.redgifs.com/watch/coordinateddarkredbuzzard

I wrote a python program that uses gpu to decode video frames, feeds it to model and then encodes it back so it quite fast.

fenderwq

kruk23 Very cool thanks! Tried another model the other day with good results but that one was very slow. This seems far more practical.

dudu32

i just do it with affter effects and rotoscoping here... xD
don't know before that there exist such nice tools and better ways to do this...

would be awesome to see in future a AI app for this which does it on the fly and for any video and good quality.

keep doing and research on this guys...very cool stuff !

Vrsumo2017

dudu32 would be awesome to see in future a AI app for this which does it on the fly and for any video and good quality

Version 13.3 did that on the fly...
Every video you watch you just had to start off like this

In this order
Pass-through on
Saturation on
Color range about -10 and pull it back till you got a decent enough Pass-through
And opacity to about -99 (if needed)

Worked really good 👍

Work well with many videos, the videos that didnt work so well THEN you needed to find the background color, which I rarely did.

Was very simple. And the Pass-through you got worked through out the whole scene 95% of the time, this new version it looses Pass-through when it cuts to a new position/camera angle, not good

And 13.3 version with the guardian off for quest 2, it was very easy to move around in your room, position the model in different locations in the room or house, standing doggy or if they did a doggy on a chair I like positioning the model on my desk chair, then I might mover to the edge of my bed...

13.3 version was very simple, fast and very fun to mess with it. It was crazy good compared to this new version. 🤷‍♂️

fenderwq

Vrsumo2017 I agree that this is harder now (only the XXLayna scene really works for me) but still even when it worked at its best it was nowhere near what is possible with a good AI. For reference, this is an image from the upcomming scenes that went through one of these models. This kind of thing simply isn't possible with chroma keying (making a color transparent).
Really curious to see what the SLR devs will come up with too. I hop it's based on a precise pre-rendered "mask". In the meantime I'll use some of the AI's to give me some content to bridge the gap 🙂

kruk23

fenderwq
Is it U2-Net model? Very nice results (although I would prefer to mostly cut off the guy).
What scene and minute it's? I would like to compare it with RoboustVideoMatting model.

"This seems far more practical."
I wouldn't be so sure 😃 I can transcode 5k video at 100fps so pretty fast but it's heavily optimized code running on rtx 3090. Would be probably hard to run it on other machine (people would have to compile two c++ projects [or trust my binaries] and download bunch of tools from nvidia [cuda + cudnn + tensorRT])

Vrsumo2017

fenderwq but still even when it worked at its best it was nowhere near what is possible with a good AI

13.3 version worked well, decent enough, but I could find quiet a few scenes it worked really well, it varied and depended on the background. And I found many good Pass-throughs with very less effort.

They got this one scene coming out soon, guy wearing all black, man that's gonna look really good, I've seen scenes where I could get rid of the guy and just his stunt cock was there, sheeeesh! 🥵

Literally look like she on top of you, and when she rub his chest, slides her body "tits" on his chest or rub around his stunt cock her hands look like they rubbing on ur skin, and rcg with her legs under his... her legs look like they under yours, I got these results with 13.3 version, I'm sure this new update u could get them results but it just takes to much effort. 🤷‍♂️🤦‍♂️

Male body in or out? I like having the option for both 👍

But most definitely when ever they get it to work perfect, OMG! 🥵🤯👍🥂

sheldoncooper

Vrsumo2017 the man body is the main reason that breaks immersion

fenderwq

Vrsumo2017 yeah this passthrough thing is definitely gonna be awesome.
Made a few vids with the male body out and it merged perfectly with my own body and bedroom. Very immersive. This worked really well for a vredging movie where the guy is practically still and the angle doesn't change. Still as soon as the guy starts moving "no male body" just won't work. So I agree, having the option for both would be great.
Also starting to think it's better having no bed in the scene. It's nice for matching the perspective but it breaks the immersion more than I expected.

Vrsumo2017

Vrsumo2017 Male body in or out? I like having the option for both

Wow! and looks like we'll have the options for both 👍

Video demonstration toggling with body or no body.
https://www.sexlikereal.com/blog/456-advances-in-ai-background-removal-for-passthrough

👏👏👏🥂 SLR 👍

Vrsumo2017

fenderwq Still as soon as the guy starts moving "no male body" just won't work.

13.3 version G, I'm saying...it didn't have that problem, you set pass-through faster and it would stay on that same setting when positions changed. And combined with guardian off in quest2 even better.

13.4 version? Not working so well like previous version. 🤷‍♂️

alina

After I have experienced Passthrough I felt the same leap from 2D porn to 3D with VR glasses. Passthrough is obviously the future. SLR with this form and all your resources I think it’s obvious your team should invest more in making Passthrough more friendly. I would even go as far as producing and directing videos with Passthrough as primary focus (meaning, contrasting backgrounds with flesh, clothing or lingerie not the same color as backgrounds etc.).
I mean after watch a good Passthrough then switching back to non is okay but we all are searching for the next best video that can have a good Passthrough. Same with 2D porn, I mean is still nice but it’s nothing in comparison to vr porn.

dudu32

alina
ye passthrough is clearly the future but the deovr needs work...don't know about these downgrades in version...that sad thing =/
i tried my first passthrough on quest 2 with unity and some porn scenes and already knew back then that this gonna big in the future.
now i have pico 4 with deovr and its now more amazing with the color rbg camera and passthrough porn.
for 400 euros its rly ok but could be better and hope that passthrough color and quality improves and gets much better with pico 5...

most amazing stuff will be to do this all live with live cam girls or just with friends or partners...where you can interact with each other and this gonna be whole another experience...aka metaverse porn ^^
probably soon possible too...
there already exist live cam girls sites which sit infront of 180cam and you can join there room in vr and interact with them...
amazing stuff and combined with passthrough like here on slr...this all gonna be so way better on many levels...

edit: but real future of porn will be full 3d scanned body videos with passthrough where you can move around and view from any point like in virtamate or realgirlsnow...
i think this is the the real thing for passthrough and future porn industry.

Vrsumo2017

alina Passthrough is obviously the future.

With version 13.3 it's now G, worked great 👍

Now I got a reason to transfer files from my computer to my q2, until slr adds samba support, I think it's called samba 🤔🤷‍♂️ where it works like pigas and sky box easy to stream from computer.

But yeah that pass-through is awesome. 👍

Vrsumo2017

dudu32 but real future of porn will be full 3d scanned body videos with passthrough where you can move around and view from any point like in virtamate or realgirlsnow

BRO! with hand tracking right? The other day I put the hand tracking on and I haven't used it since b4 the update, now hands have color "quest2" and I was moving hands around watching a video with pass-through, it crazy cus you can put the hands over body parts with no see through on the hands like b4, but Anyway I paused the video just to move my hands around and I thought "it be really cool if you could move body parts with hand tracking, lol I thought damn what if could edit this like a photo and make tits big with my hands, lol I just be thinking of stuff...

But I know what you mean about virtamate, I'm thinking pass-through with handtracking and being able to move the models around, I seen how they move the model around with the mouse, kinda looked like they moving a sexdoll, but he'll in vr with pass-through and hand tracking. And the 3d scanned porn models...Sheeeeesh 🔥🥵🤯🤯🤯🤯🤯

dudu32

Vrsumo2017
with hand tracking that will be another level and next step probably...
some guys probably gonna figure out soon how to rig some body parts like "tits" in these videos and add physics to them so you can play around with them.

realgirlsnow has good 3d scanned models already but its still not the real thing and these body scans lacks of quality...

but you probably already can rig these models and add physics to these models and then able grab her tits, legs, head, hands like in virtamate while the animation is played and in action...
i can imagine that this all already possible to do right now with the scanned models and probably looks way more realistic than the 3d models in virtamate even if still in bad quality...

edit:
virtamate would be so much better if users could add faces from photo scans and add skin texture from photo...
there so many way to improve...
but its impossible probably because virtamate is based on daz studio...sadly =/

we need app like virtamate but with photorealstic features where users can create and add faces from photo and skin texture from photo...this would be awesome great app and next step too.

this guy already posted this idea like 4 years ago and how it would look like:
https://www.reddit.com/r/VirtAMate/comments/8sqta0/this_is_why_we_need_user_texture_import/

just look at the skin difference and details:
https://imgur.com/a/TBwuEIC

or here skin from photo meshed with metahuman:
https://www.youtube.com/watch?v=BK-ayKR8NlQ&ab_channel=Avaturn
https://www.youtube.com/watch?v=U2yderQOMUo&ab_channel=KeenTools

so much better and probably there ways to make it even more better...

fenderwq

dudu32 Checked out the highest fidelity of unreals meta humuns in vr some time back. That quality is sooo much better than any human I've ever seen in vr. If a game ever gets to that meta human level quality, with those facial expressions and ease of customization... 3D games might actually become really cool 🙂
For now even upcoming games like VAM2 are just waaay to fake and lifeless for me. It's gonna be some weird ass future though 🙂

Vrsumo2017

dudu32 with hand tracking that will be another level and next step probably...
some guys probably gonna figure out soon how to rig some body parts like "tits" in these videos and add physics to them so you can play around with them.

Well I found this
https://vrporn.com/primavr-pc-vr-game/

Guess it's possible now? 🤷‍♂️ ...crazy 😮

dudu32

we need this:
https://www.youtube.com/watch?v=zSmN7aXSd5Y&ab_channel=RalphVirtual
https://www.youtube.com/watch?v=KdfegonWz5g&ab_channel=celsovfx
deepfake + metahuman = realism 100%...and then all in passthrough...

this level of detail would be just crazy for vr porn games like virtamate...

future is great.

dudu32

Vrsumo2017
WOW that was fast...
crazy...
my brain need some time to figure out how that all works but it looks amazing and on point...

don't know that this already possible...^^
first time see something like this...
this clip just pure magic...haha lol.

rascellion

That looks like it was filmed with a Kinect, which is 12 years old technology. The interaction to me looks like it's just a tiny disturbance applied on the point cloud. Will be cool to see what they can do if depth cameras at some point go up to 4k range, though. The best depth cameras currently are the Realsense devices, and they too still go only to about 1280x720 resolution, which leaves something as densely packed as the surface of the human body quite inaccurate.

« Previous Page Next Page »