This repository reproduces one video, not a video product
Astra ChatGPT Hyperframes contains the source and assets for a single ChatGPT-themed motion response. The original motion belongs to Rajmoni, whose public reference is linked in the README. The adaptation keeps the reference's backgrounds and control movement in 180 cleaned source plates, then draws replacement typography with JavaScript. The project is candid about that boundary: it did not invent the motion from scratch.
The final piece is 15 seconds long and repeats the same 7.5-second visual loop twice while the original audio continues slightly longer. Hyperframes renders 360 lossless PNG frames at 1,920 by 1,080 and 24 frames per second. FFmpeg converts them to H.264 with an explicit BT.709 path, copies the existing AAC packets, and writes validation details. That pipeline is the repository's subject, not ChatGPT itself.
What happened when we ran it
Our unprivileged Debian sandbox installed commit 8b9a6a3 in 22 seconds. npm pulled 272 packages, and the installed tree occupied 694 MB. Before dependencies, the repository used 25.6 MB across 205 files and about 173 lines of source. One CI workflow was present, but there was no Dockerfile and no tests directory.
Our runner found no standard build script or target and skipped the build. It also found no standard tests script or target, so tests were skipped. npm audit reported 1 known vulnerability rated high, with no critical, moderate, or low findings. The repository has its own check, render, and compare commands, but those are not results from our lab block. We cannot claim that this container reproduced or validated the MP4.
Rendering needs Chrome, FFmpeg, and a separately licensed font
The documented path starts with Node.js 22 or newer, Chrome, FFmpeg and FFprobe with libx264, plus at least 1 GB of free space. After npm ci --ignore-scripts, setup pauses for explicit acceptance of the Switzer font license and downloads a pinned font from its official CDN. Rendering stays local and needs no AI model, API key, private prompt, or agent service.
Chrome can be installed through Hyperframes or selected through HYPERFRAMES_BROWSER_PATH. FFmpeg and FFprobe must already be on the system path. The project also supplies a GitHub Actions render workflow, but the user must accept the font license before running it. These dependencies are reasonable for browser video work, yet they make the 173 lines of source a misleading guide to total setup. Our installed JavaScript tree alone reached 694 MB.
Hashes make drift visible without promising universal identity
The render scripts verify hashes for the 180 plates, audio, raw reference, comparison video, and downloaded font. A verification report records the decoded output, frame count, and audio hash. The project reports that a fresh install on its recorded macOS environment produced a byte-for-byte identical MP4. It also says the included delivered video is only a comparison artifact and never becomes render input.
That is a credible reproducibility method with a bounded claim. Browser versions, operating-system font rasterization, the system monospace font used in the opening prompt, and FFmpeg builds may change individual pixels or encoded bytes. Linux can render the same composition without guaranteeing the same file hash. The comparison command is therefore more useful than a blanket assertion: it gives you a way to locate drift on the machine you actually use.
The editable layer is small because the motion lives in plates
film.js owns the title, prompt, result typography, and frame seeking. index.html defines the composition, dimensions, and audio. Measured positions live in assets/tracks.js, while the backgrounds and controls are committed PNG plates. A Python script documents the cleanup and tracking work that produced those plates, though rebuilding them with different image-library versions may change their hashes.
This structure makes text edits approachable, but it is not a flexible scene system. Much of the visual motion is baked into source-derived frames. Changing the timing, camera movement, interface choreography, or art direction means more than replacing a string. If your goal is an original series with reusable scenes, a framework such as Remotion or Motion Canvas gives you a cleaner starting abstraction than adapting this one composition.
Attribution does not grant reuse rights
The README credits Rajmoni for the motion reference, identifies the font designer and distributor, and labels ChatGPT and OpenAI names as their owners' trademarks. It also states that media and source-derived plates retain their owners' rights and that attribution does not create a new license. GitHub's API reported no asserted repository license when checked. Those facts should stop a casual copy-and-publish workflow.
GitHub showed 141 stars, 11 forks, and 0 combined open issues and pull requests on September 30, 2026. The repository was created and last pushed on September 6, with no published release. This looks like a preserved production artifact, not an evolving library. That is fine for study. For distribution, replace the protected identity and source-derived material, establish a license for every asset, and build motion you can clearly claim as your own.