Gracia moving volumetric scenes hands-on: I’m impressed
One of the top three experiences I had on the AWE Europe show floor was not even an official experience of the AWE Europe show floor. It was an on-the-fly demo of the moving volumetric scenes offered by Gracia that I had in the corridors of the event. I’ve been impressed, and so I absolutely need to tell you about it.
Gracia
Gracia is a startup that aims to become the YouTube for volumetric content rendered via Gaussian Splats. Its long-term vision is to have creators upload 3D scenes (probably with their phones) to Gracia’s servers and then have many users navigate through the available scenes and enter them. Thanks to the power of Gaussian Splats, the viewers should be able to enter the scene (alone or with friends) and live it as the creator could see it.
Currently, Gracia is at the beginning of this journey and offers an application for Quest and Steam VR where it is possible to navigate a few scenes scanned by the Gracia team itself. The Quest version is pretty limited, in the sense that in it you can only enjoy the default scenes, while in the Steam VR version, it is also possible to see some scenes that have been scanned by the community. Gaussian Splats can be quite heavy so the Quest version must use a simplified version of them, resulting in a lower quality of the visuals, too. It’s still impressive that this experience can run on Quest, though, and the Gracia team is proud of the optimizations that it has been able to perform.
A few weeks ago, I reviewed Gracia on Quest and said that the experience it offers is surely interesting, and it shows the potential of this platform, but I also highlighted its issues, like the fact that some of the scans had clear artifacts, or that the UI was horrible. I couldn’t review the PC version for comparison because I had not a PC that was fast enough, though.
Luckily, during AWE, the investor Tipatat Chennavasin came to the rescue. He offered me in the venue of the event an on-the-fly demo with Gracia on his powerful laptop. This way he could show me how it performs better on PC, and especially how it is the newly released feature: moving volumetric scenes.
Moving volumetric scenes
The team at Gracia has digitalized 3-4 small animated scenes, usually involving people, of 10-30 seconds each. It is so now possible to enter the Gracia app in SteamVR and enjoy something similar to a short volumetric video happening in front of you, always rendered using Gaussian Splats. This is innovative because similar programs (like Meta Horizon Hyperscape) until now usually tried only to show static scenes. This comes at a price, though: as Upload VR reports, these scenes are very difficult to scan and heavy to render, so they are still an experimental feature:
Gracia says these clips were captured using expensive 3D scanning studio rigs, and took six minutes to train a single frame. That equates to weeks of total AWS compute time per minute, and though it can be parallelized to take less time in reality, it’s still a significant cost. Further, the download size of these clips equates to over 300 megabytes per second. That means streaming this kind of content would theoretically take a 2.4 gigabit internet connection, which barely anyone has. But Gracia expects its compression to improve by at least an order of magnitude within a year, making it possible on modern fast internet connections. It’s also exploring using cloud VR streaming.
Hands-on moving volumetric scenes on Gracia PC
Tipatat launched Gracia and then he gave me a fancy Quest headset with an orange headband and an orange Link cable that connected it to the PC. As soon as I put the headset on, I saw around me first of all the mediocre UI of Gracia, which I immediately moved away. Then, as soon as I focused on the scene, I saw a blond girl dancing in front of me. And I went in awe. (Pun intended. Because I was at AWE. Ok, if you want to close this article after reading this bad joke, I understand you)
First of all, the quality of the visuals was MUCH better than on Quest. On Quest, the girl in the Cyberpunk scene has good quality, but because of the higher compression of the splats to make them run on mobile hardware, there are some small artifacts on her body. It is like her skin has a bit of fur.
But on PC, this is no more: the girl had a very convincing appearance. I can not say she was fully realistic because our brain is very picky in finding even small imperfections when we see the depiction of other humans. I could see she was not real, for instance, because some parts of her skin looked like were painted with a brush. But we were very close. The quality of the realism was amazing, more than with a Metahuman.
And even more impressive than the quality of her visual aspect was the fact that she was moving. She was performing a dance in front of me, and the reproduction was very smooth. Since it was a volumetric scene, I could move around her and watch the scene from every direction. This increased the realism because I could get close to her as I could do if she were real. At a certain point, I crouched down to see the scene from another point of view, and while I was appreciating this, Tipatat jokingly exclaimed “You perv, you’re trying to look under her skirt!”. I tried to explain “No, I was just looking at things from another point of view”, but I guess no one in the room believed me.
Still, this fun episode made me think of the value this kind of content can have in the adult industry ecosystem. A moving person, reproduced with such a high quality, that you can see from all the points of view, maybe in AR in your room… I mean, this is clearly the future of adult entertainment. This could also be a good thing: adult entertainment has always had the power to push technological advancements forward, so maybe it could also help a lot in the development of Gaussian Splat technology.
It was also cool that I could use the controllers to move and resize the dancing girl and make her very small: I put her in my hand and it was fantastic. When she was as small as a Barbie, the “brush stroke” artifacts on her body were less visible and her appearance was even more convincing than when she was at her real size. It was impressive.
The video could also be paused, and so I could see her in all her full details, including the semi-transparency of her black dress.
After the dancing woman, I tried also two other moving scenes. One depicted two parents dancing with their small baby. This scene was very convincing, too, and I could feel their happiness while watching them. The third scene was about a chef teaching how to cook a certain dish. It was still very good, but it was the worst one of the group, first of all because it showed more scanning artifacts, and also because it was a bit colder and more impersonal.
Final Considerations
I didn’t expect the moving volumetric scenes in Gracia to be THIS good: I was genuinely impressed by them. I think long-term they could become the future of content sharing: instead of making just a video, maybe people could remember their best moments by capturing a scene and then having it rendered as 3D moving Gaussian Splats so that to be able to relive it any time they want. This would be amazing.
Of course, it is not something that can happen tomorrow. On the quality side, I think that what I’ve tried is actually already good enough to be used by general consumers. However, the capturing tools are still bulky and expensive, and the resulting content is very big to download and very heavy to render. When we will be able to just use something like a phone with a mobile headset to do both the scanning and the rendering, then we will truly live in the future of 3D memories.
For what concerns the present, instead, I truly invite you to try these moving scenes on Gracia if you have a PCVR setup that is powerful enough because they are really great.
(Header image by Gracia)
Disclaimer: this blog contains advertisement and affiliate links to sustain itself. If you click on an affiliate link, I'll be very happy because I'll earn a small commission on your purchase. You can find my boring full disclosure here.