I have optimized VTube Studio's VRAM usage!

No generative AI was used during the production of this video (enjoy bad photoshops).

The idea behind this video originated right after my debut, when I realized that even with 16 GB of VRAM in certain games I am getting close to utilizing my entire graphics card. I already knew that texture resolution was massive. I didn't however know how much of it is a fault my artist, how much of it is a concept of an avatar and how much of it is the rigger - I don't want to point fingers.
Live2D being what it is - it can't be changed - it's closed source and inefficient. But that doesn't mean I can't optimize VTube Studio a bit. And while it's fair to assume most Vtuber will likely end up with models being below 1.33GB of VRAM and it might not be too big of a problem for most, while gathering references I have seen quite a few being bigger than that.

VTube Studio VRAM usage calculator: https://suimachine.github.io/web_projects/vts_calculator/
VTS-Memory Compression project (source code & build): https://github.com/SuiMachine/VTS-Memory-Compression
Nvidia's Textures Tools: https://developer.nvidia.com/texture-tools-exporter
More in-depth blog entry about it: https://suimachine.github.io/bcn/2026/01/31/VTS-Texture-compression.html

Resources:
Nathan Reed blog entry: https://www.reedbeta.com/blog/understanding-bcn-texture-compression-formats/
Marmoset Toolbag's lecture about BCn: https://youtu.be/3H-HGlsC0NY

Video sections:
0:00 - Intro
1:12 - Disclaimer & problem introduction
2:31 - VRAM usage with ARGB32
3:06 - A solution to problem
4:43 - BC3 and runtime compression
5:47 - DDS files and BC7
7:33 - ARGB32 vs BC7
8:31 - Using a VTS-Memory Compression hook
10:08 - Closing words and plan for the future
10:58 - Credits etc.

Видео I have optimized VTube Studio's VRAM usage! канала Sui VT