# Animated SOG: Check Scrubbing and Audio Before Publishing a 4D Splat Clip

Convert one owned TRiGS clip to .asog, then compare first playback, motion and audio on your target browser.

By Paras Madan | 2026-11-22


A smaller volumetric clip is useful only if it plays well enough for the viewer to inspect it. When you're publishing a moving Gaussian splat on the web, check what happens after someone drags the timeline, as well as how quickly the first frame appears.

[Arrival.Space's Animated SOG project](https://github.com/arrival-space/animated-sog) packages moving Gaussian splats and optional audio in an `.asog` file. The initial tooling appeared on November 20. On November 21, a [three.js player change](https://github.com/arrival-space/animated-sog/commit/bc976cf106363a4bad25a21405f9281622ce1a65) addressed blurred artifacts while scrubbing by changing which splats are included in the sort. This is a reason to pin the player version and test playback behavior, not just compare file sizes.

## Start with one clip you can inspect closely

Use a synthetic clip or an owned capture with no rights questions. Pick something easy to grade: a small object moving across a plain scene, with a clear audio cue when it changes direction. Retain the original scene and timing so you can tell when the converted clip drifts.

If you already have a `.trigsbin` export, use it for the first pass. The project documents a direct route from that file to `.asog`. If you only have a sequence of PLY frames, the [flip-book pipeline](https://github.com/arrival-space/animated-sog/blob/main/docs/pipeline.md) is a separate route, not the same command with a different extension. The project identifies the Open4DGS trainer and exporter used there as separate AGPL-3.0 tools; the Animated SOG repository itself is MIT.

With the project's encoder installed, its documented commands would look like this for an owned clip:

```sh
asog encode desk-demo.trigsbin desk-demo.asog --audio cue.mp3
asog info desk-demo.asog
```

Keep the output together with the input and the encoder revision. If you later change compression settings or upgrade the decoder, that gives you something concrete to compare.

## Measure the first frame and the finished download

The [format specification](https://github.com/arrival-space/animated-sog/blob/main/docs/format.md) explains how an Animated SOG file is structured. The project describes streaming decode during download and provides PlayCanvas and three.js players. Use the player you plan to ship, not just the one that happens to be running on a developer's computer.

Load the same clip with an empty browser cache, then again under a connection speed representative of your users. Record three times separately: first visible frame, first usable playback and full download. A quick preview followed by stalls is not the same result as smooth playback.

Measure bytes for your own clip. The compression result the project reports for its samples is not a guarantee for a different scene, encoder setting or input format.

## Scrub to a point where an error is obvious

Play through the direction change. Then jump directly to just before and just after it. Inspect whether the object appears in the expected place and whether an image from an earlier moment lingers after the jump. Repeat that test with a moving camera if the page lets users orbit the scene.

Check the sound separately: listen for the cue on first playback, then pause and seek before replaying it. If the browser requires a user gesture before playing audio, test the actual play control users will click. Don't infer sync from seeing that both the object and its audio exist in the file.

If you're choosing between the included PlayCanvas and three.js players, record which renderer and browser you used for each result. A working desktop preview does not establish that the same page plays well on a lower-powered device.

## Keep a fallback for the first publication

Start with one embedded clip, not a whole gallery. Keep a still image or ordinary video available if the viewer cannot start, and review the page on the devices you actually expect people to use. Capture the file size, playback result and exact player revision in the project notes.

For this first test I'd ship the clip only if the viewer can start, find the cue and seek back to it without losing the scene. That's a stronger result than a smaller file on disk.

I'm Paras. I work at the intersection of AI, engineering and distribution.

I help AI teams explain technical products through clear documentation, practical guides and developer-focused content.

[Book a 30-minute call](https://cal.com/paras-madan/30min).
