I gave Codex my talking-head video. It cut the pauses and added the visuals.
I have only just started doing this, but so far I have had good results. Here is the simple process I have been using, including the one prompt I give Codex.
I have just started experimenting with Codex and Remotion for my talking-head videos. I record myself, add the video to a Codex project, and ask it to remove the dead space, add captions and bring visuals in as I speak.
So far, I have had good results. If I say “ChatGPT”, the ChatGPT logo can appear. If I say “Claude”, the Claude logo can appear. If I make 3 points, point 1 appears when I say it, then point 2, then point 3.
I am still learning this. This is not the only way to do it, and I will update the guide as I learn more. But this is the setup and prompt I am using now.
Cut the long pauses and obvious mistakes.
Add captions that follow what I am saying.
Bring in logos, images and GIFs at the right moment.
Turn spoken lists into numbered points on screen.
You do not need to write the Remotion code. You tell Codex what you want the finished video to do.
What Codex and Remotion each do
Remotion is the thing that makes the video. It uses code to decide exactly when a caption, logo, image or animation appears.
Codex does the coding for me. It is inside the ChatGPT desktop app and I talk to it just as I would talk to ChatGPT. I create a project, add my files and explain what I want.
You do not need to open a terminal or learn React for this guide. Codex works inside the project and handles the technical part.
Before you start: what you need
The ChatGPT desktop app with Codex.
Your talking-head recording. An MP4 is ideal.
The visuals you want to use. Logos, screenshots, photographs and animated GIFs.
Node.js. Remotion needs it to run. You do not need to know what it does. Codex will check whether it is installed and tell you what to do if it is missing.
Time for a first test. Start with a short video, not a 40-minute masterclass.
Step 1: Create a new project in Codex
Open the ChatGPT desktop app and go to Codex.
Click + New Project.
Create or choose an empty folder for this video.
Make sure the chat you open is inside that project.
A project is simply the folder Codex is allowed to work in. It is where Codex will create the Remotion files and, later, put the finished video.
You’ll know it worked when you can see the project name beside your chat and the empty folder is attached to it.
Step 2: Add the recording and visuals
Add your talking-head video to the project folder. Add any logos, images, screenshots or GIFs you already know you want to use.
Give them obvious names. ChatGPT logo is more useful than image-4738-final-final.png. A clear name helps Codex match the right visual to the right word.
Use your own images or properly licensed assets. I tell Codex not to grab random images from the internet.
You’ll know it worked when Codex can list the recording and visual files you added.
Step 3: Give Codex one prompt
I do not give it a collection of technical instructions one at a time. I give it one full brief, then react to the preview it makes.
Copy this prompt into your Codex projectI have added a talking-head recording to this project. Please turn it into a polished, animated video using Remotion.
I am new to Remotion, so handle the technical setup for me. First, check whether my computer has everything Remotion needs. If something is missing, explain exactly what I need to do in very simple language. Set up the Remotion project and its official skills.
Edit my recording to remove long accidental silences, obvious mistakes, repeated sentences and unnecessary filler words. Keep natural breaths and useful pauses so I still sound human. Do not cut off words or make the speech feel rushed. Keep my original recording unchanged.
Add large, easy-to-read captions. Show them as short phrases rather than long paragraphs. Highlight important words as I say them and keep the captions away from my face.
Use the logos, images, screenshots and animated GIFs I have added to the project. Match them to what I say and make them appear at the correct moment.
For example, when I say “ChatGPT”, bring in the ChatGPT logo. When I say “Claude”, bring in the Claude logo. If I mention something but you cannot find a matching visual in the project, tell me what I need to add. Do not download random images without asking me.
When I make a numbered list, show the points on screen one at a time. When I say “number one”, show and highlight the first point. When I move to number two, dim the first point and highlight the second. Continue in the same way for the remaining points.
Look for other useful visual moments too. Show important numbers as large callouts. Turn comparisons into simple side-by-side graphics. Display direct questions on screen. Use screenshots when I describe something people need to see. Create simple diagrams when I explain a process. Add a short title when I move to a new subject.
Consider every sentence for a possible visual, but do not add an animation simply because you can. Only add something when it makes the video clearer, more interesting or easier to understand.
Use a clean, friendly style with large text, one accent colour, smooth movement and simple transitions. Avoid cheesy effects, wild spinning, constant bouncing and graphics that cover my face.
Aim for a useful visual change every 3 to 6 seconds, but leave the screen alone when my face should be the focus.
Before finishing, check that every caption, logo, image, GIF and animation appears at the correct spoken moment. Make sure nothing overlaps, images are not stretched, GIFs play correctly and the edited speech still sounds natural.
Show me a short preview first and explain how to watch it. Wait for my approval before rendering the complete video. Once I approve it, render the finished video as an MP4 and show me where to find it.
Codex may ask for permission before it installs Remotion or runs something in the project. Read what it is asking to do. If it clearly relates to the Remotion project, approve it.
You’ll know it worked when Codex gives you a short preview to watch.
Step 4: Watch it and talk back
The first version is a draft. Watch it all the way through and tell Codex what feels wrong in ordinary language.
“Make the captions bigger.”
“The animations feel too busy. Make them calmer.”
“Bring the ChatGPT logo in as soon as I say ChatGPT.”
“Keep that screenshot on screen for another 2 seconds.”
“Move the list to the left so it does not cover my face.”
“That pause was intentional. Put it back.”
You do not need to find or change the code yourself. Describe the problem you can see and ask Codex to fix it.
Step 5: Render the finished video
When the preview looks right, say:
Final messageI approve the preview. Please render the complete video as an MP4 and show me exactly where you saved it.
Codex will render the video and give you the location of the finished file.
What I have learned so far
The first thing I learned is that more animation does not automatically make a better video.
It is tempting to ask for something to happen during every sentence. That quickly becomes exhausting. I have had better results asking Codex to consider every sentence, then add something only when it helps the viewer understand, remember or enjoy the point.
A logo can identify a tool. A numbered list can make a process easier to follow. A screenshot can explain something faster than words. Sometimes the best visual is simply your face.
I also check the complete video before using it. Codex can still bring an image in too early, misunderstand a word or remove a pause I wanted to keep. I treat the first result as an edit to review, not a finished video to publish without watching.
My job is to explain what I want and judge the result. Codex handles the code. Remotion makes the video.
The short version
Create a new project in Codex.
Add your recording and visual files.
Paste the prompt.
Watch the preview.
Ask for changes in normal language.
Approve it and ask Codex to render the MP4.
I have only just started doing this, but it is already changing how I think about editing video. I will keep updating this guide as I find better prompts and learn what Remotion can do.