You watch the demo. A 4D scene—dynamic, fluid, captured from multiple angles—rendered in real-time as a cloud of Gaussian splats. It feels like magic. The kind of magic that makes you think: this is the future of VR, of volumetric video, of how we capture reality.
Then you see the number: 640 megabytes per second. Raw encoding rate. For a two-second clip, that’s over 1.2 GB of data. Suddenly the magic pauses.
Everyone is talking about the cool factor. Nobody is talking about the file size. That’s the real story.
Fable’s 4D splat format extends Gaussian splatting into the temporal dimension. It’s elegant: instead of treating video as a sequence of frames, you model the entire 4D volume (xyz + time) as a sparse set of splats that move and deform. The visual results are stunning. The demo on the website runs smoothly, even on a browser.
But here’s the thing. You’ve probably noticed that every new breakthrough in 3D representation comes with a compression problem. Neural radiance fields (NeRFs) were huge until you tried to render them at 60fps. 3D Gaussian splatting solved the rendering speed problem—but now we’re moving to 4D, and the data explosion is back with a vengeance.
If a two-second clip demands 640MB/s, a 10-minute video demands 384GB of raw data. Your phone doesn’t have that. Your cloud budget doesn’t have that. Reality doesn’t have that.
The format is open source. MIT license. The author invites contributions to benchmarking and compression. That’s the right move. But the current numbers suggest this is a research novelty, not a production-ready format. The demo splat file is 7.4MB for two seconds? Actually, the raw file size is 427MB for the two-second clip. That’s still 213 MB/s—not 640—but still enormous for any practical streaming or storage.
I saw a comment on the repo: “Something’s way off with these numbers.” The math doesn’t add up cleanly. But the core truth remains: 4D splats are magnificently hungry for data. And unless someone cracks the compression code, this will remain a beautiful demo on a GitHub page, not the next Apple Vision Pro killer app.
The hardest problem in 4D isn’t capturing it. It’s making it small enough to ship to a billion phones.
So what does this mean for you? If you’re a researcher, this is a fantastic problem to solve. If you’re a developer building the next VR experience, watch the file sizes before you commit. If you’re a hype watcher, keep your skepticism sharp. The technology is real. The barriers are real. The disconnection between the two is where the real conversation lives.
Fable deserves credit for pushing the boundaries. The format is innovative, and the open-source approach is commendable. But we need to stop pretending that every cool demo is ready for prime time. Not every leap forward is a step toward the future. Some are just beautiful dead ends—until someone solves the ugly problem underneath.
The question is: who will solve it? And will 4D splats be the foundation, or just a footnote?
FAQ
Q: Is 4D splatting actually practical for anything beyond short demos?
A: Currently, no. The raw bandwidth is too high for streaming or storage. Research into compression is needed before it can be used in production VR or video.
Q: What's the practical implication for someone working in this space?
A: If you're building a product around volumetric video, you need to watch the compression ratio. This format could be viable if someone develops an effective 4D codec, but right now it's a research toy.
Q: What's the contrarian take on the huge file sizes?
A: The huge file size might be a feature, not a bug. If compression improves dramatically (e.g., 100:1), the quality could surpass traditional video. The high raw data means more information—the challenge is to compress it intelligently.