How to Edit Videos With AI (No Editing Skills Required)

Is video editing the bottleneck keeping you from publishing more videos on YouTube and social media? Wondering how to use Claude Code or ChatGPT Codex to edit your videos, add motion graphics, and pick music without learning editing software?

In this article, you’ll discover how to set up an AI video editing workflow that takes raw footage all the way to a finished video with motion graphics, titles, and music.

Why AI Video Editing Works Now

Justin Brown has run his YouTube channel for 14 years, publishing a video every week. Creating just one of those videos used to take 14 to 22 hours, depending on the type of video. With an AI video editing workflow, his team can now hit publish in under an hour, and faster still when the video doesn’t need extra polish. Long-form videos now go live in around 40 minutes.

​The biggest change, according to Justin, is the barrier to entry. Creators no longer need to learn editing software like Adobe Premiere, Final Cut Pro, or DaVinci Resolve, or understand audio levels and music composition. They can learn prompting instead and focus on the human side of video: storytelling, the message, and communication.

​Many people also assume AI isn’t good at video editing, but Justin says that’s down to trying to get an edit done in one shot by typing “edit my video,” and hoping for the best. In the same way that a human editor needs guidelines and instructions to succeed, so does the AI. Clear instructions get to a good result faster, without burning tokens, time, and money on extra rounds.

#1: Set Up Claude Code or ChatGPT Codex to Edit Video Locally

The setup that makes AI video editing work is a desktop app with access to the files on a computer. That means downloading either the Claude app and using Code, or the ChatGPT app and selecting Codex, which now runs inside the same app. These tools can take control of a folder on a hard drive, so editing happens locally instead of in the cloud, which uses fewer tokens.

​Justin gets strong results from both ChatGPT and Claude. His current go-to is the latest ChatGPT model in Codex, and he’s also getting good results with Google’s Antigravity, though he considers it a fair bit behind.

​Pro Tip: There’s no need to know much about AI before setting these tools up. Asking the AI “How can I do this?” and “What’s possible?” are, in Justin’s view, the best questions to ask and get started.

#2: Define the Video’s Goal and Train AI on Standout Videos

Telling an AI to “make a good video” doesn’t work because “good video” means something different in every niche and to every creator.

Justin starts by training the AI on what a good video looks like in his eyes, beginning with the video’s goal: awareness, leads, followers, or sales. Every video is a test. When he isn’t sure exactly what to test, he gives the AI examples of standout videos from leaders in the niche that already achieve that goal.

The AI reverse-engineers what works: the pacing, how many titles appear, what happens on screen, and how sound effects and music are used. That analysis happens before the AI looks at any of his own footage.

A simple prompt to get you started:

I want to make videos like these with your help.
 
The goal of my videos is to bring in leads. 

Here are some standout videos from leaders in my niche that are already doing that well: [link 1], [link 2], [link 3]

Let's break them down. 

What can we learn from them in terms of pacing, how many titles they use, what's happening on screen, and how they use sound effects and music? 

These videos have worked well, so let's reverse engineer them before we look at any of my footage.

#3: Shoot for the Edit So AI Can Make Clean Cuts

AI video editing starts at the filming stage. Justin calls this shooting for the edit: thinking about the finished product before recording, so the footage is easy for an AI (or a human) to cut. Footage doesn’t need to be shot in sequence, as long as the prompt describes how the video should play out.

​Justin shoots on a phone and on webcams, and notes that AI can work with nearly any file type.

How to Record Footage That AI Edits Well

Hit Record Once: Justin records one continuous file rather than starting and stopping. That removes the risk of forgetting to press record again and gives the AI a single file to process. AI can handle large files, and because the editing happens locally on a computer, nothing needs to be uploaded to the cloud.

​Make the Last Take the Default: Justin’s standing rule, for both human editors and AI, is to always use the last take. Moving on to the next section signals that the previous attempt felt right. When a different take is better, he says so out loud while recording, such as “Actually, use the first take.” 

Human editors often missed his clap and raised-hand cues, but the AI follows these spoken instructions reliably. It can feel silly to talk to the camera this way, but it makes AI editing much better.

​Pause Between Statements: A short pause between statements gives the AI room to cut cleanly. Moving straight from one thought to the next makes it harder to get the timing right.

​Restart Full Sentences: When a line goes wrong, Justin restarts the whole sentence instead of picking it up halfway through. That lets him give the AI a clear instruction: don’t cut sentences in half unless absolutely necessary.

How to Plan B-Roll Before Filming

A-roll is the primary camera track: the main shot of the person talking to the camera. B-roll is anything laid over that primary track, including graphics, animations, photos, and other footage. It shows what’s being said, which makes a video easier to follow and more engaging.

​For shorts and reels, filming B-roll that’s planned to go over the top of the A-roll gives the AI material to work with. Still photos also work and can get a slow Ken Burns-style zoom.

​When Justin isn’t sure what B-roll a video needs, he uses the AI as a sounding board. It can suggest what to shoot, or create a graphic itself, for any moment where showing something would help the viewer understand.

​Pro Tip: An AI editor can flag when a video lacks a strong hook. It can then prompt the creator to record several hook options labeled take one, take two, and take three, and add the best one to the finished video.

#4: Organize Video Assets in One Project Folder

What matters is getting every file into one location so the AI knows where everything is. Keeping everything together prevents scattered files and stops the AI from missing something.

​Justin recommends one folder per video or series of videos, with subfolders for raw camera footage, B-roll (such as product or scenery shots), music, photos, and graphics. A series can also have a single folder for shared assets, such as licensed music and a show bumper.

​Pro Tip: Before the AI starts editing, Justin prompts it to review everything in the folder first and take stock of what’s there. Current models often do this on their own, but he wants to know for certain that it has looked at every file.

#5: Get an Accurate, Time-Stamped Transcript

After the AI reviews the project folder, it transcribes the footage. Justin calls this the most important step in the process. If the transcript isn’t accurate, the edits and timing will be off.

​For editing, the timing in the transcript matters more than the words themselves. The AI needs to know exactly when each word starts and how long it lasts, so it can cut in between words instead of clipping the start or end of one. Justin says many editing tools, including the AI features in DaVinci Resolve, struggle because their transcription timing isn’t accurate. Precise timing also matters for removing filler words like “um” and “ah” from an otherwise perfect take.

​Video length raises the stakes. For short-form content, the transcription built into ChatGPT or Claude is usually close enough. In a longer video, a one- or two-second timing error compounds over time, so cuts later in the video end up way off and the audio drifts out of lip sync. Justin warns that the AI can gaslight its user, claiming it transcribed everything while quietly missing sections.

​Wispr, which the AI may use behind the scenes, is one of the most reliable transcription models. Gemini API credits are another option. For long-form content, Justin has had the most success with Deepgram, a cloud service that charges a small fee. He put $20 on Deepgram’s prepaid tokens months ago and still hasn’t needed to add more.

​Pro Tip: To test this workflow, run a simple edit on existing footage and feed the AI the transcript of this podcast episode as guidance. The AI will walk through setting up tools like Deepgram and connecting to Google, step by step.

#6: Build the A-Roll Story Before Adding Polish

Justin edits long-form videos in two passes: a story pass that builds the A-roll, followed by a polish pass that adds titles, graphics, and music.

For long-form content, Justin’s first editing pass builds the A-roll story, which he calls the most important piece of any edit, whether a person or AI is doing it. The AI removes bad takes and mistakes, then assembles the core content before adding any titles, graphics, or other polish.

A fast way to check the story is to ask the AI for a script with timing, generated from the transcript, that shows what it cut and kept. The script lists each speaker, the timestamp, and what’s being said. It costs almost nothing to generate, and reviewing it is a quick way to remove sections, bring in other people, or catch problems before any polish is added.

The filming rules pay off here. Without instructions like “always use the last take” and “don’t cut my sentences in half unless it’s absolutely needed,” AI tends to stitch together individual words from different takes. The result is a jump cut after nearly every word. When a cut lands in the wrong place, Justin traces it back to one of two causes: an instruction that wasn’t clear enough or footage that wasn’t shot in a way that made the decision easy. Each fix becomes part of a rule set that’s built and refined over time.

Pro Tip: For short-form videos, he collapses the A-roll pass and the B-roll pass (outlined below) into a single step because there’s so little footage that the story comes together right away.

#7: Add B-Roll, Motion Graphics, and Titles With Claude and HyperFrames

Removing bad takes, ums, and ahs produces a clean video. B-roll, motion graphics, titles, sound effects, and music make it engaging, and that’s the polish a production company would normally add.

​The AI can be trained to hit a specific B-roll mix, such as 45% B-roll and 55% on-camera subject, when it has a library of B-roll to pull from. Without a set percentage, Justin tells it he wants the video to be engaging and clear wherever he’s explaining something. The reverse matters too. When a video hits a key point where the viewer should take action, it’s often better to cut back to the person on camera. The AI knows storytelling and story arc best practices and can make that judgment call.

​Claude now has a built-in skill that designs and animates graphics with nothing else needed. Justin says a prompt like “help me make this video engaging, add motion graphics, add titles, add captions” is enough to get started. Pointing it to a brand guide, a PDF, or a website lets it automatically match graphics to existing brand fonts and colors.

​For more advanced work, Justin’s standout tool is HyperFrames, an open-source framework from HeyGen that works across AI models. Codex or Claude Code can be told to use HyperFrames during an edit to add motion graphics and make the video look like a high-end ad. Justin doesn’t use HeyGen’s AI avatars. HyperFrames is strictly a tool for enhancing real footage.

What HyperFrames Can Do in an Edit

Place Text Behind the Speaker: HyperFrames handles background removal and places titles behind the person speaking. Justin says text appearing slightly behind the speaker is a current trend that signals extra polish beyond a static title.

​Generate Graphics From the Transcript: In one video, Justin mentioned a Venn diagram. With no guidance beyond “using HyperFrames, make this engaging,” the AI created an animated Venn diagram that appeared behind him. It works because the AI has the transcript and knows everything said in the video.

​Improve Screen-Recorded Tutorials: For tutorials, HyperFrames can screen-record websites and add scrolling footage as B-roll. When Justin appears as a picture-in-picture in the corner, it tracks the mouse and moves him if he’s blocking the action. It also zooms in automatically when text is too small or something on screen isn’t clear for the viewer.

​Emphasize Key Moments: The AI can slowly zoom in on the speaker when they make a key statement, the way a film editor brings the face closer during an important line. It can also create thumbnails.

​The key prompt, according to Justin, is asking “What can you do? What should I do for my video?” The answer varies for each creator and each video. He also asks for six or seven different versions, picks the direction he likes, and tells the AI to do more like that in the future. That trains it over time.

#8: Find and Place Music With an MCP Connection

Music and sound effects are key parts of an engaging video, but picking the right track on stock music sites can take weeks. Top music libraries like Artlist and Epidemic Sound now offer an MCP connection that links directly to the AI. Justin uses Epidemic Sound for his videos.

​Timing matters here, too. Once the AI has the transcript, it understands what the video is about and its vibe. It then finds music and sound effects that fit and places them in the timeline without any manual work. If a track doesn’t work, a request like “something a bit more upbeat” results in a replacement being placed in the video. The music is heard playing back in the finished video, which is where it matters.

#9: Review and Refine Edits in ChatCut

ChatCut is a free tool for Mac and Windows that has a visual interface for AI video editing and review. It suits creators who have done some editing and want to see a timeline of clips or edit the video’s text like a Word document, while AI does the work. It integrates with Claude Code and Codex and connects to AI video generation models like Seedance to create new clips.

​Justin compares ChatCut to Adobe Premiere or Final Cut Pro, except that the timeline is visible and doesn’t need to be touched.

​For short-form videos and reels, he sends small fixes back to the AI, such as a missed section or an extra “um” to tighten. For long-form videos, having the AI reprocess for every minor tweak drags out the process. In ChatCut, he makes the small adjustment himself during preview, and the AI tracks the change and learns his editing style from it. HyperFrames has its own interface as well.

​ChatCut also includes a player that previews titles and motion graphics without waiting for the video to render. Watching playback and giving feedback without rendering shortens the time to a finished long-form video.

#10: Bring In an Orchestrator to Manage the Whole Edit

Justin’s biggest level-up, beyond accurate transcripts and HyperFrames, has been moving the edit out of a single chat thread. In a long thread, the AI loses context. By the time it reaches titles, it has forgotten details from earlier in the process.

​The fix is an orchestrator: a single AI chat explicitly named the manager overseeing the edit. Justin’s prompt tells it that it’s the orchestrator, that it will oversee the addition of captions, titles, and motion graphics, and that it must check its own work before he sees it. He gives it the overall goal, such as a good, engaging video for YouTube, and asks it to confirm the edit is accurate, the timing is good, and the video is engaging. Naming the orchestrator gives it permission to spin up agents.

​The orchestrator then creates threads below it. It spins up agents to create captions and checks whether they’re accurate and on-brand, looping back and forth until the work passes. It reports back only when the video is ready for review.

​Each sub-manager can follow its own guidelines. A caption agent, for example, can have rules for caption style, on-screen text, and pacing, and it can’t send its work up the chain until it meets them. Standing rules build up over time as well. Justin told the AI once that whenever he says links are in the description, it should add a title saying so, and he never had to repeat it. When that title is missing, the orchestrator catches it and fixes it before he sees the video.

Other Notes From This Episode

• Connect with Michael Stelzner @Stelzner on Facebook and @Mike_Stelzner on X.
• Watch this interview and other exclusive content from Social Media Examiner on YouTube.

Where to subscribe: Apple Podcasts | Spotify | YouTube Music | YouTube | Amazon Music | RSS

✋🏽 If you enjoyed this episode of the AI Explored podcast, please head over to Apple Podcasts, leave a rating, write a review, and subscribe.

---

Stay Up-to-Date: Get New Marketing Articles Delivered to You!

Don't miss out on upcoming social media marketing insights and strategies! Sign up to receive notifications when we publish new articles on Social Media Examiner. Our expertly crafted content will help you stay ahead of the curve and drive results for your business. Click the link below to sign up now and receive our annual report!