VR Video Test – Looking for Feedback on Quality (demo 40‑sec)

General
1

Hello everyone,
I’m creating this post because I’d like to get feedback on my upcoming attempts at VR creation. Those who follow me know that I previously tried to get into traditional animation, but today I want to explore animation in virtual reality.
After several hours of experimenting with different techniques, I finally managed to achieve what I was aiming for: a 360-degree video with one image per eye to create depth (each image is 1920x2160, resulting in a total of 3840x2160).
In the examples I’m sharing, I think the depth effect works quite well. I’m using an HTC Vive (the first version), and although I’m not a VR headset expert, I assume image quality has improved over time.
That’s why I’d like to share two examples with you that differ only in the number of samples generated per image: one with 1024 samples, the other with 32. This should make a big difference (4 minutes and 30 seconds of rendering for a 1024-sample frame versus 15 seconds for the 32-sample one), but on my HTC Vive headset, I don’t really see much of a difference

For those who will test it, here are my questions:

  1. Is there a noticeable difference between the two examples?
  2. If so, are we talking about a significant drop in quality?
  3. Does my video work properly on your headsets?

Video with 32 samples (MP4 h.264)

Video with 1024 samples (MP4 h.264)

I forgot to mention that the video is 3 seconds long (72 frames). It’s intended to loop, although it’s not perfect yet. For the 1024-sample version, my PC took around 6 hours to render it, compared to just 17 minutes for the 32-sample version.
That’s why I’m curious to know if the quality at 32 samples is good enough — so I can consider making longer and more technical videos.

As for the video itself, there are still adjustments to make: the size of the body representing us, the bed, probably also the camera placement, etc. I used what I had available, since my main goal was simply to produce a VR video and see — with people who have better headsets than mine — whether the quality difference between the two versions is really noticeable.

I’m also sharing a third clip, which doesn’t show the video in front of you — just the room — to better appreciate the depth effect. It’s 5 seconds long, and the idea is to pause and observe your surroundings.

Video with 1024 samples (MP4 h.264)

I’m waiting for feedback from anyone who wants to test it. I know these are just short clips, but I’m working on something more serious (Tifa taking care of you with her hand), which should be around one minute or longer — if I have time to compile everything by this weekend.

1 Like
2

I’m bookmarking this to test it later.
But I do advise that you don’t use x264. This codec is considered legacy these days. It takes up a lot of space in comparison to x265 (which is widely supported these days)

As well AV1 is becoming more and more supported in decode. It’s very slow to encode but the compression efficiency is best in the world so far.
Here’s an example of some codec testing I’ve been doing recently:
Codecs in order:
x264
x265 medium preset CQ20
AV1 preset 4
AV1 preset 6
image

3

Thanks for your feedback :slightly_smiling_face:
And yes, I agree about using H.265. I initially encoded in H.264 based on ChatGPT’s suggestion, as it seemed to offer better compatibility.
Since I already had the files ready, I ended up uploading both examples in H.265:

As for AV1, once I start making longer videos, I’ll include an AV1 version.
(Just to clarify, the videos are currently at 24 fps, but I’ll switch to 30 fps for the next ones.)

1 Like
4

For those interested, the VR project is progressing well. I spent the whole day working on it and I’m currently at 2500 frames at 30 fps, which corresponds to about 83.3 seconds, or 1 minute and 23 seconds. I’m aiming for 3 minutes for the first version (with script), as it will be more enjoyable to test than my initial examples. I’ll also create a non-VR version for those who don’t have a headset. It takes me around 15 hours to generate 1 minute, so I expect the project to be released in the coming week. I’m sharing a few pictures directly taken from Blender to show what it should look like.

Summary

ZTX VR V1_1
ZTX VR V1_11919×986 235 KB

ZTX VR V1_2
ZTX VR V1_21919×987 265 KB

ZTX VR V1_3
ZTX VR V1_31918×980 267 KB

ZTX VR V1_4
ZTX VR V1_41918×988 256 KB

ZTX VR V1_5
ZTX VR V1_51915×983 273 KB

ZTX VR V1_6
ZTX VR V1_61917×985 273 KB

ZTX VR V1_7
ZTX VR V1_71915×983 282 KB

Image with VR distortion

ZTX VR V1_8 vu deformer
ZTX VR V1_8 vu deformer1918×987 244 KB

5

I tried it out on my quest 3 using HereSphere. The difference in quality was minimal to me. If I wasn’t actively trying to look for it, I wouldn’t have noticed any difference. Also, I can’t wait to see the finished results of this project.

1 Like
6

Thank you very much for your test and feedback :slightly_smiling_face:. It reassures me, because with 1024 samples for three minutes (5400 frames), the full generation would take 405 hours, which is about 17 days, compared to around 45 hours with 32 samples. Thanks again!

1 Like
7

I tested it too and can confirm what @Shenanigans mentioned — the difference between the two versions is practically nonexistent. I ran my test on the Pico 4.

Excited to see the final result; it’s looking promising :slightly_smiling_face:

1 Like
8

Thank you, Slibowitz :slight_smile: This second feedback on the quality helps me plan for longer projects. I’m also reassured that the video works in VR on two different headsets. Regarding the final result, I think it’s important to remind those who see this message that I only started learning Blender in September, and with my other videos I haven’t had much time to dedicate to it. I’m doing my best to make the outcome as clean as possible in terms of movement, and also detailed enough so it doesn’t feel like a mechanical motion. The first three‑minute version will surely have things to improve, I’m still an amateur, but very motivated :slight_smile:

2 Likes
9

You’re welcome! Good luck with Blender. Respect for learning it. It’s way more complicated than it looks ^^ I tried getting into it once, but let’s just say it didn’t go so well ^^ Fortunately, there are so many resources available now, and with AI assistants, it’s even easier. For a first attempt, it already looks great! I can’t wait to see what you create as you get better.

1 Like
10
  • Thank you :slight_smile:
  • Yes, it’s true that it’s more complex than it seems and it requires a lot of time. Don’t worry, my first days with Blender didn’t go very well either :joy:
  • I agree, ChatGPT helped me save quite a lot of time while learning Blender. As for the available resources, it’s a bit more challenging since I’m only looking for royalty-free or similar content. I had to search extensively to find what I needed
  • Thank you :slight_smile:
  • I’m also excited to see the final result. Version 1 will simply feature Tifa in an empty room interacting with the viewer (duration: about 3 minutes). I’m a bit concerned because I don’t think I’ll have the resources to add proper sound to the scene, so I’ll probably go with very calm background music. Then, for Version 2, I’ll decorate the room with paintings and other visual elements. I’d like, if possible, to add a second model next to the viewer, performing simple actions like placing a hand on the shoulder or whispering softly, to create a sense of closeness with the audience. (A bit more than that, but ChatGPT prevents me from translating what I wanted to write at the beginning. :rofl: )
1 Like
11

For the start, this already seems very good. Plus, you can’t go wrong with having Tifa in a scene ^^

Because it’s too NSFW? In December, OpenAI will allow more adult themed content, but only with age verification.

A good tip I can give you, if you are looking for assistance in writing for NSFW stuff: Use Nomi.AI. It’s normally a companion app, but it can make a great creative writing partner as well. And it allows full NSFW writing. Create yourself a Nomi with a backstory, that says that your Nomi is your assistant for content creation. Define the Nomi helps you writing NSFW / erotic / porn stories and the Nomi helps with writing. Give the Nomi some creative traits and you will get really good results. I use Nomi.AI for NSFW creative writing as well, works fantastic. If you want some examples or need help with hit, feel free to dm me :slightly_smiling_face:

1 Like
12

Yes, I noticed that the AI was bothering me a bit too much with explicit terms, but I guess I must have been too explicit ^^

As for Nomi.AI, thanks for the advice. But the creative process is something I truly love: I enjoy searching and exploring ideas and concepts. AI mainly serves as a tireless teacher to whom I turn to shed light on my blind spots. I’ll still take a look at Nomi.AI—if it allows me to translate without bothering me, that will already be a good thing :slight_smile:

13

Quick update on the project’s progress: the 3 minutes are almost finished (2 minutes 30 already done). The most time-consuming part is compiling everything, but I managed to compile the first 40 seconds (1250 frames ×2, one per eye). I’m sharing them with you. Personally, I’m quite happy with the overall result, though I noticed two issues:

  • Tifa’s hand is not closed enough (this will be fixed in V2, since compilation has already started; I’ll finish this version, which is ultimately a big test for V2).
  • The torso movement of Tifa, when she leans in close, is probably a bit too fast.

This is just a first test, but I’ll refine it as much as possible. I think the final result could be pretty cool.
The demo version has no sound. I tried an AV1 version, but the output wasn’t good—I haven’t yet found the right settings.

Format: MP4 H.265 30 fps (ZTX Animation demo V1_injected.mp4)

I didn’t take the time to write a script for this demo. My usual script tool doesn’t work very well with VR, but I’m planning to calculate between frames and milliseconds to make the script as accurate as possible.

1 Like
14

For almost everything I do:

ffmpeg -i file.mkv -c:v libsvtav1 -preset 6 -crf 30

9 times out of 10 this only smoothes out film grain (which imo makes the quality better)
You can decrease the preset, which will make the file size smaller but P6 is imo the best sweet spot for space/time.
Tools like Tdarr/handbrake/fileflows etc. all use ffmpeg behind the scenes so you can use those settings in them, they’re sometimes behind “advanced settings”.

I’ve never tried transcoding 3D animation so I don’t know if these settings work well for it, but it’s worth a try.

15

I just tested something that’s the closest to what you suggested. Personally, I don’t have a player that can read AV1 files on my headset (I use Virtual Desktop and run the tests on VLC), so I’ll let others judge the results. I can’t encode it in MKV: I tried, but the program I use to inject VR metadata doesn’t seem to recognize it.

For anyone who might have advice, here are the video format details:

  1. Blender generates two images per frame with an offset for VR (1920×2160 for each image).
  2. I combine these two images in DaVinci Resolve into a 3840×2160 format.
  3. I then encode in MP4 H.265.
    After that, I use HandBrake.

AV1 settings: (AV1 10-bit (SVT PSY), CRF: 30, speed: 1).