You do not need to know how to code. You need footage, a point of view, and a willingness to tell the editor what to change. You describe what you want in plain english. Codex handles the technical work, you make the creative decisions and watch the drafts.

One honest note before we start. When I say ChatGPT here, I use Codex for this workflow. In my experience it has been better at getting the visuals to match what I am picturing than Claude. That is my preference, not a universal ranking, both can work with the right setup. This guide follows my Codex workflow so you have one clear place to start.

The setup has three parts: Codex does the work. HyperFrames supplies the video-making skills. Your editing guide tells it what good looks like to you. Expect a few feedback rounds, because the first draft is how you discover what you actually need to specify.

Before and after: the raw clip, then the edit that came back. Same footage, my fonts, my captions, my pacing.

The editing starter kit

Free zip. The project templates, a completed example guide, and every copy-paste prompt in this article.

Download free

The kit contains plain-text templates, a completed example, and the prompts. Bring your own footage, reference edits, font files and sound assets. It does not contain installed software, my paid fonts, CapCut sounds or private media. The guide and templates are free; your app plan and any assets you buy are separate.

Your first goal: one 15-second draft you can play, then improve. You do not need to set up your entire content operation in one sitting.

1. Give Codex a home for your editing

Here is exactly how to get the kit into Codex, click by click.

Step 1: download and unzip. Hit the download button above. The kit lands in your Downloads folder as editing-starter.zip. Double-click it and it becomes a normal folder. Move that folder somewhere permanent you will not delete, like Documents, then a folder called My Video Editing. Do not leave it in Downloads and do not work out of the zip.

Step 2: open it as a project in Codex. Open the Codex desktop app and open that folder as a local project. The exact button labels vary by app version, but you are looking for the option to open or add a project from a folder on your computer. If you already have a project you use, point its folder settings at this folder instead and make it your primary editing folder. The official project instructions are here.

Step 3: there is no uploading after that. This is the part that confuses people. You do not attach the files to a chat. Once the folder is the project, Codex reads the files right off your computer, every task, automatically. That is the whole point of giving it a home.

Step 4: start a task and make it prove it. Start your first task inside that project and paste the prompt below. If it names the files back to you, the kit is working. Keep using this same project for every future edit.

The one gotcha: the files need to actually be on your computer. Not still inside the zip, and not a cloud placeholder that has not downloaded yet. If Codex says it cannot see a file, that is almost always the problem.

What the files do:

  • AGENTS.md: the entry instructions telling Codex which editing files to read.
  • editing-brain.md: the process: inspect inputs, make a short draft, revise, export.
  • editing-guide.md: your creative preferences: fonts, captions, pacing, graphics, audio.
  • editing-rules.md: lessons you want reused after review rounds.
  • sample-editing-guide.md: a filled-in example to help you answer the questions.
  • videos/first-video/brief.md: what this particular video needs.
  • videos/first-video/feedback.md: notes tied to this video and its drafts.

A .md file is just a text document. You do not have to write it yourself. Tell Codex your preferences and ask it to save them in the right file.

I don't know how to code. This is my video-editing project. Read AGENTS.md, editing-brain.md, editing-guide.md and editing-rules.md. Tell me which files you found and what needs to be filled in. Keep this folder as the home for my editing setup. Explain technical decisions in plain English.

2. Set up HeyGen HyperFrames

HyperFrames is HeyGen's framework for composing and rendering videos, with skills that coding agents can use. It works with Codex and with Claude Code. It is the video-production setup used here, not an avatar upload. The official HyperFrames setup lives here.

Paste that link and the prompt below into your editing task. Let Codex inspect what is installed, follow the current official instructions, and prove the tools work before you move on.

Read my editing-brain.md and the official setup at https://github.com/heygen-com/hyperframes. Inspect this project before changing anything. Install the core HyperFrames skills for this environment using the current official instructions. Set up the media tools needed to inspect, transcribe, edit and export my footage, including FFmpeg if needed. Use a small source clip I supply to create a short playable MP4 test. Save the working setup and versions in setup-notes.md. Tell me what is ready and what still needs my input. Don't call setup complete until the test plays.

If you want the exact install reference: the current official agent command is npx hyperframes skills update, and the interactive alternative is npx skills add heygen-com/hyperframes with Core Skills selected. You can ask Codex to run the right setup, you never have to type commands into a terminal yourself.

Check before moving on: Codex can see your clip, load your chosen font, and hand you a playable test export. If setup fails, paste the error back and ask it to fix that step. You do not need to know what the error means.

3. Build the editing guide around your taste

"Make it look good" leaves every decision open. Give it a reference you love and say why: the title appears immediately, the captions are easy to read, the little clips explain the story, the sound feels intentional.

And the references do not have to be yours. Tag in editing you like from other creators. Save the reels that stop your scroll, drop them in as references while you build the editing files, and tell Codex what you like about each one. "I like how fast the captions land here" teaches it more than a paragraph of adjectives.

Use the completed sample for ideas. It is an example, not a requirement to copy my look. The blank editing-guide.md in the kit is the one you customize.

Help me fill in editing-guide.md. Interview me in small batches about my audience, reference edits, title and caption fonts, font files, capitalization, text size and placement, colors, pacing, transitions, supporting clips, music and sound effects. Ask what I dislike as well as what I like. If I can't name a style, help me describe it from my reference. Save my answers in editing-guide.md, show me a short summary, and list any missing files. Keep this separate from the brief for one particular video.

Things worth adding to your guide

Every one of these comes from something I actually had to work through in my own revision rounds. Decide which apply to your style.

  • Actual font files. A font name, or a screenshot of the lettering, is not the font. Add the real .otf or .ttf file and the exact weight. I needed the licensed Ugly Dave Regular file for my handwritten accents. "Something similar" was not the same look.
  • Regular versus alternate letters. Even the correct font can substitute joined or alternate letter shapes. If the letters look wrong, give a visual reference and have it check the font file and ligature settings before swapping fonts.
  • Different jobs for different fonts. Name the title font, the spoken-caption font, and the accent font separately. Video typography does not automatically apply to a carousel.
  • Caption behavior. One word or a phrase? Lowercase or sentence case? Where does it sit, at what size, with what alignment? Check the words against the actual audio.
  • Readable text. Plain, shadow, outline, or card. My feedback rounds repeatedly asked to remove unwanted caption boxes. Test over moving footage, including a light shirt.
  • Your face and the app controls. Keep eyes and mouth clear, leave room for Instagram's buttons, and check the whole clip, not the first frame.
  • The first second. Say what should be visible immediately. I wanted the title on frame one, with motion coming from supporting elements, not a wobbling headline.
  • Your sound vocabulary. Codex does not have CapCut's sound library. Supply the pops, clicks, whooshes and typing sounds you are licensed to use, name the files, and say which action each belongs to.
  • Sound timing and volume. A pop lands with the image. Typing stops when the typing stops. Test on a phone, effects export too quiet or too loud all the time. Leave quiet gaps.
  • Supporting-clip audio. Decide if the little clips are silent. I had to specify muting their embedded audio so it did not fight my narration.
  • Real motion versus stills. Ask for moving clips when that is what you want. Say how long a photo holds and whether it fills the frame.
  • Edges and backgrounds. Straight edges, rounded corners, shadows, borders. For a full-background image: "fill the frame, no padding." These details caused several of my revisions.
  • Relevant visuals. Clips should explain the current sentence. No decorative filler, no reusing the same asset under a different crop, and use the finished moment of a screen recording, not the clicking around before it.
  • Lists and graphics. Do earlier list items stay on screen? Do labels appear when you say them? If a graphic travels behind you, inspect the cutout for halos.
  • Your real footage. State whether retouching or generated imagery is allowed. My rule: preserve my real face and footage, add text and graphics on top.
  • Private information. Check screenshots for names, messages and phone numbers before they ship. Supply a cleaned example when that proves the point.
  • Export format. A vertical Reel starts at 1080 x 1920 MP4. A carousel video slide is 1080 x 1350. Ask for a playable file and keep the original footage.
  • Versions. Number the drafts, keep the last approved one, and say exactly what stays unchanged. "Fix the captions" should not quietly redesign the edit.
Look at the font, sound and supporting-media files I added. Create an asset-index.md with the exact filenames, what each is for and anything missing. Ask me which sound belongs to each kind of visual event. Do not guess a paid font or replace a missing asset without telling me. Reference this index from editing-guide.md.

4. Keep the guide available for every video

The guide and rules live in the main editing folder, and AGENTS.md points Codex at them before every edit. Keep the project pointed at that folder. Local folder access follows the project you open, it does not follow you into unrelated chats.

Start a fresh task for each video inside the same project, and open every one the same way: ask it to name the files it read and summarize your style back to you. That checks access and understanding in one move.

Before editing, read AGENTS.md, editing-brain.md, editing-guide.md, editing-rules.md, asset-index.md if present, and this video's brief. List the files you read and summarize the current font, caption, sound and framing rules. If a file is missing or inaccessible, tell me before rendering. Reuse the working setup in setup-notes.md.

5. Edit one video, in rounds

You also do not have to hand it the whole edit on day one. Start small:

  • Just the closed captions. One job, nothing else.
  • Just cut the dead air. Any blank space longer than 0.2 seconds, gone.

Start with a couple of jobs like these, watch them come back done your way, and keep building from there. The full treatment can come once you trust it with the small stuff.

Gather the inputs. Raw clips, a reference, supporting media and the brief go in a new folder under videos. Say the audience, the hook, the key points, the target length, and what the viewer should do at the end.

Ask for the opening first. Fifteen seconds is enough to judge the font, captions, pacing and sound. Correcting a small preview is much cheaper than re-treating a whole video.

Read my saved editing files and videos/first-video/brief.md. Inspect the footage and identify the take and transcript you'll use. Create a 15-second opening draft using my reference, font files and approved supporting media. Keep my real voice, meaning and face unchanged. Give me a playable MP4, not just an edit plan, and tell me which choices need my review. Number the draft and preserve the source.

Give specific feedback. Watch it on your phone with sound. "At 00:04 to 00:07, use the moving clip named office-broll.mp4 and remove its audio" gets applied. "Make it more dynamic" gets interpreted.

Update draft [filename]. At [timestamp], change [specific thing] to [desired result]. Keep [approved elements] unchanged. Record these notes in this video's feedback.md, save a new numbered draft and show me the changed section. If the change affects other timings, explain that before applying it across the full video.

Extend and check. Once the opening works, have it finish the video in the agreed style. Then watch the entire final export: transcript mistakes, covered text or faces, repeated assets, quiet or competing audio, awkward cuts, the closing CTA, the last frame. Inspect the exact file you will upload.

6. Turn the feedback into lessons it can reuse

This is the step I do not want you to skip. After the rounds, make it update the guide, or every decision stays buried in one conversation.

Lasting style choices go in editing-guide.md. Reusable lessons go in editing-rules.md. One-off timestamps stay in that video's feedback.md. The goal is one consistent guide, not a growing pile of contradictions.

Review the feedback rounds and the version I approved. Update editing-guide.md with the lasting preferences I explicitly chose, and update editing-rules.md with the reusable lessons, date and a brief example. Remove or revise outdated guidance that these decisions replace; don't add duplicates. Keep one-off timestamps and clip choices in this video's feedback.md. If a preference is unclear or conflicts with another rule, ask me about that point. Show me what changed and the files you saved. Read the updated guide before the next edit.

The difference in practice: "move this image at 00:06" is a note for one video. "Keep all overlays away from my eyes and mouth" is a rule forever. And "no white cards on spoken captions" can be a video rule without banning every white panel in a carousel.

Finish with a check: ask Codex to read the saved guide back and name three decisions it will apply next time. You are building an editing process that reflects your taste, not relying on it to remember every chat.

If you get stuck

  • It can't see a file: unzip the kit, open the correct local project, verify the file is downloaded on this machine.
  • It gives you a plan instead of a video: ask for a short playable render and the exact blocker if it cannot produce one.
  • The font still looks wrong: check the file, the weight, and the alternate-letter settings. Compare a sample phrase before rendering again.
  • It forgets your style: ask which guide it read. Correct the project path, then reconcile any conflicting rules.
  • It changes too much: name the last approved draft and list the only changes you want in the next version.

Everything to download

The editing starter kit

Free zip. The project templates, a completed example guide, and every copy-paste prompt in this article.

Download free

Or grab the pieces individually:

References: the official HyperFrames setup and local projects and folder access. Setup references checked September 10, 2026. The creative examples come from my own editing feedback, and the sample is a starting point for your taste, not my complete private production system.

What to notice across the whole system:

  • You are the creative director, not the technician. Plain english in, playable drafts out.
  • The first draft is a discovery tool. It exists to show you what you forgot to specify.
  • Specific feedback beats vibes. Filename, timestamp, desired result.
  • Lessons go in the guide, not the chat. That is the difference between a one-off edit and an editing system.
The edit did not get easier because I learned the software. It got easier because the editor finally knew my taste.

This is the same build-it-with-an-agent pattern as my email command center: describe the outcome, review the drafts, and save what you learned so the next one starts smarter.