August 13, 2026
Macs, magic, and comment-section gasps
Better Gaussian Splatting in Julia
Julia’s 3D glow-up has fans losing it over smoother scenes and Mac support
TLDR: GaussianSplatting.jl 2.0 makes photo-based 3D scene building easier to use, prettier, and available on more devices, including Apple machines. Commenters were overwhelmingly wowed, especially by the interactive demo and the fact that this kind of visual tech no longer feels limited to expensive hardware.
A niche graphics tool just had a very public main character moment. GaussianSplatting.jl 2.0 lets people build walkable 3D scenes from photos with smoother loading, a less frozen-looking app, and support for more kinds of graphics chips — including Apple laptops, which instantly became the crowd-pleaser. The loudest reaction wasn’t nitpicking the code; it was pure awe. One commenter simply declared, “this is amazing!” while another called Apple support “huge for accessibility,” basically summing up the biggest mood in the room: this stuff used to feel locked behind pricey gaming hardware, and now it feels a lot more reachable.
The biggest scene-stealer, though, was the project’s demo image of an amulet from Kyiv. People weren’t just impressed — they were delightfully ambushed by it. One user said the homepage image was cool, then admitted that after mousing over it, it became “absolutely incredible.” That’s the kind of accidental drama the internet loves: the old bait-and-switch, except the twist is “surprise, it’s even better than you thought.” Another commenter got unexpectedly emotional spotting a familiar Ukrainian fountain on the front page, turning a graphics demo into a tiny wholesome plot twist.
There wasn’t much outright fighting in the thread, which is almost suspicious by internet standards. Instead, the hot take was basically a victory lap: better visuals, fewer weird floating artifacts, and a program that doesn’t look dead while it’s busy. In comment-section terms, this was less a flame war and more a group gasp.
Key Points
- •GaussianSplatting.jl 2.0 supports running the same Julia code across AMDGPU.jl, CUDA.jl, and Metal.jl backends via KernelAbstractions.jl.
- •The application now uses separate frontend and backend threads so the UI remains responsive during tasks such as dataset loading, JIT compilation, rendering, and training.
- •The release adds an MCMC-based densification strategy that the article says offers finer control over scene Gaussian counts and relies less on strong initialization.
- •Depth supervision can be enabled using dataset depth maps, which are refit to a common scale against a prior point cloud and discarded if refitting fails.
- •Geometry regularization and Sky Dome are added to improve reconstructed geometry, surface smoothness, and separation of sky or distant background to reduce floaters.