mrsarac/ff-tracking tracks the characters an AI agent types into the terminal at the glyph level and produces a 6-second video that looks as though a virtual camera filmed the screen. The latest release published on GitHub is v1.0.0. It is a Rust project and uses the MIT license.
The video consists of two-pass rendering with fframes. The first pass draws SVG text for each character to obtain glyph coordinates, and the second pass handles perspective projection, depth-of-field, chromatic aberration, bloom, and an LED grid with a single SkSL shader. Depth-of-field is the effect that blurs the parts that are out of focus. Both passes and the sound read the same hud crate, so the tracking coordinates and the focus are designed not to drift apart. The sound effects are synthesized by tools/sfx.py, which uses Metal on macOS and Vulkan on Windows and encodes the video with FFmpeg.
That said, this project is closer to a narrowly scoped style test for video presentation than to a reusable tool. Of the 67 files in total, there is just one test, a parity test for the score crate, and there is no continuous integration setup. No benchmark numbers for judging performance or quality are presented either, so the details of the rendering design can be checked, but the actual performance evidence is limited.