How to Edit YouTube Videos with Codex and Hyperframes
A practical guide to using Codex and Hyperframes to auto-transcribe, cut, and animate YouTube videos, reels, and ads with AI.
What are Codex and Hyperframes used for?
Codex is OpenAI’s coding agent, available as a desktop app that can work inside a local project folder, run commands, and pull in outside repos. Hyperframes is an open-source repo built to sit on top of Codex specifically for video work: transcribing footage, cutting out dead space and stutters, generating motion graphics, and animating scenes based on plain-language instructions. Together they turn a text prompt into an edited video, complete with cropped face cam, background templates, animated title cards, and B-roll pulled from your own files.
TL;DR
- Codex plus Hyperframes lets you describe an edit in natural language and get back a video with cuts, cropped face cam footage, animated cards, and synced subtitles.
- The workflow starts by transcribing raw audio, which lets the system identify and remove stutters, mistakes, and dead air automatically before any visual editing happens.
- Hyperframes can switch between a full-screen talking-head view and a template view depending on whether you’re showing something on screen, a distinction you teach it through prompting.
- For short-form content, the same setup can produce reels and ads with fast scene switching, paper-style 3D animations, music, and sound effects synced for engagement.
- The tool can go out and collect its own B-roll, including grabbing screenshots, past YouTube clips, and logo assets from a connected Google Drive, based on the context of what you’re talking about.
- Setup starts with a new local project in the Codex desktop app, pointed at a folder where you paste in the Hyperframes GitHub repo URL and ask Codex to install dependencies and pull in the skills it needs.
- This is a skill-based system, meaning the editing behaviors (when to zoom, when to go full screen, how fast to cut) are defined once and reused across videos rather than re-prompted every time.
How does the Codex and Hyperframes editing process actually work?
The process breaks into two stages: transcription and structure, then visual assembly.
First, you feed Codex your raw recordings. In one demonstrated example, that meant two separate files, a face cam recording and a screen recording, for a 14-minute video. Hyperframes transcribes the audio, then uses that transcript to find and cut mistakes, stutters, and dead space. In the example given, the 14-minute source was trimmed down to roughly 9 minutes purely through this cleanup pass.
Second, once the audio and cuts are locked in, Hyperframes handles the visual layer. It decides when to show a full-screen talking head versus a cropped or “template” scene, based on cues like whether there’s something worth showing on screen. If you’re mid-explanation with nothing to display, it can zoom into a full-screen shot of you talking. When you move into a demo or tutorial section, it can shift back to a template view with your face cam cropped into a corner and the rest of the frame free for screen content. Animated title cards can be inserted between sections or concepts, acting as transition points through the video.
Can it actually edit short-form videos and reels, not just long-form?
Yes, and this is where the tool’s strengths show up most clearly. In demonstrated examples, Hyperframes was used to generate full reels: one built around a breakdown of an AI workflow method, and another built around a free API access announcement. Both followed a similar structural pattern: an open-ended hook line that creates curiosity, followed by numbered points revealed one at a time, with scenes switching roughly every one to two seconds.
The visual style leaned on constant movement: paper-style textures, 3D animation, and a mix of scene types (B-roll with a face cam inset, full-screen animated graphics, full-screen talking head) so no single element sits static for long. Subtitles were generated automatically. Music and sound effects were synced to match pacing. B-roll wasn’t limited to stock footage either. It was described as either AI-generated or collected by the tool itself, then pulled into the edit based on the topic being discussed.
A separate ad-style example (built in both 16:9 and 9:16 formats) went further on the content side: it pulled real screenshots and past video clips relevant to the pitch, including footage from previous live events and specific segments with named guests, then animated a brand logo retrieved directly from a connected Google Drive account. That level of contextual retrieval, grabbing your own past assets and matching them to what’s being said, is what separates this from a template-based reel generator.
How do you set up Codex and Hyperframes for the first time?
The setup starts simply, using the Codex desktop app:
- Create a project folder. Make a new folder anywhere accessible (desktop is fine) and give it a clear name.
- Open a new project in Codex. In the Codex desktop app, choose to start a local project, name it, and point it at the folder you just created. This becomes the workspace Codex operates inside.
- Pull in the Hyperframes repo. Copy the URL for the Hyperframes GitHub repo, paste it into a new chat inside your Codex project, and ask Codex to pull in the repo, install any dependencies, and set up the skills needed to use it for video editing.
- Populate the project with your files. Once the repo is installed, the project folder is where you’ll add video recordings, project files, and rendered outputs as you start working.
One coffee. One working app.
You bring the idea. Remy manages the project.
From there, editing becomes a matter of prompting Codex with instructions about what kind of video you want (long-form tutorial cleanup, a fast-paced reel, a calmer ad) and letting Hyperframes apply the visual logic you’ve defined.
Is this worth setting up if you’ve never edited video before?
The appeal of this setup is that it doesn’t require traditional editing skills. You’re not manually cutting timelines or keyframing animations. Instead, you’re describing behavior in natural language (when to zoom in, when to go full screen, how fast to cut) and letting the system apply that logic consistently across videos. Because the behaviors can be saved as reusable skills, the setup cost is front-loaded: once you’ve taught it your style once, later videos should require less explanation.
That said, this is still an agent-driven coding tool at its core. Comfort with a desktop app that runs local projects and installs repos will help, even though no actual coding knowledge is required to use the finished workflow. Anyone expecting a polished one-click editor should adjust expectations: this is closer to configuring a system than opening software.
Frequently Asked Questions
What is Hyperframes?
Hyperframes is a GitHub repo designed to work inside Codex projects, adding video-editing capabilities like transcription-based cutting, scene switching between full-screen and template views, animated title cards, and B-roll collection.
Do I need to know how to code to use Codex and Hyperframes?
No coding knowledge is required to use the finished workflow. You do need to install the Codex desktop app, create a local project, and paste in a GitHub repo URL, but the actual editing instructions are given in plain language.
Can it edit both long-form YouTube videos and short-form reels?
Yes. Demonstrated examples included a long-form tutorial video with cleanup cuts and section cards, as well as short-form reels and a 16:9/9:16 ad, each using different pacing and animation styles suited to the format.
Where does the B-roll come from?
B-roll can come from footage the tool generates itself, footage it records, or assets it retrieves from your own files, including past YouTube clips, screenshots, and logo files stored on a connected Google Drive.
How long does a full setup take?
The core setup (creating a Codex project and installing the Hyperframes repo) is a short, one-time process. The larger time investment is defining and refining the editing “skills” you want the system to apply consistently across future videos.