# I edit most of my videos with Claude in a terminal now Source: https://aaronmakelky.com/blog/ai-video-editing-with-claude Last updated: 2026-10-10 Ryan Doser and I talked AI video editing on AI Rabbit Holes. I still record in Riverside and Descript, but Claude, HyperFrames, and Tesseract do most of my editing now. October 10, 2026 · 10 min read AI Rabbit Holes · AI Video Editing · Claude Code · HyperFrames · Tesseract · Creator Workflows Ryan Doser and me on AI Rabbit Holes, talking about AI video editing. A year ago, AI could generate media and it wasn't very good. You could put in some B-roll or an avatar, and that's about all it was good for. It couldn't go into a timeline and make cuts and edits, or add transitions and sound effects. Claude Opus is so good at anything visual and design oriented, whether it's landing pages, video editing, or motion graphics, that software has now added the tools to let you bring your AI agent into the project and have it make the edits for you. I talked through how I edit now with [Ryan Doser](https://ryandoser.com/) on [AI Rabbit Holes](https://www.youtube.com/watch?v=FcSfa_3QlxI), recorded October 8, 2026. Ryan walked through his podcast workflow in Riverside and Claude Code, and I shared how I clip our episodes and edit my own videos. The full episode is on [Ryan's YouTube channel](https://www.youtube.com/@RyanDoserAI), along with our earlier ones, including [Astra vs Fable](/blog/astra-vs-fable) and [my Grok Bot review](/blog/grok-bot-worth-it-300-dollar-review). ## You don't need editing software to edit with Claude What I'm most excited about, and what I use most to edit my videos now, is a terminal. Nothing against Riverside or Descript, but you're paying a monthly subscription, you only get so many minutes to upload, and you're spending AI credits to use things on their platform. If you take your raw video, say a 30-minute episode saved to your computer, you can point Claude at it and have it start editing. You don't have to connect to any external software that costs money. My most used one is [HyperFrames](https://github.com/heygen-com/hyperframes). It's free and made by HeyGen. You do need an account, but it basically just uses HTML in a terminal. Think of it like the visuals on a landing page: it's all code, and Claude is really good at code. It can set up captions, graphics, and little explainer animations on your screen without a video editing app to connect to. I wrote up one HyperFrames edit I [ran from my phone with Codex](/blog/codex-hyperframes-video-editing-workflow). The other one is [Tesseract](https://github.com/mirage-hq/Tesseract), which had only been out for about two weeks when we recorded. Both are great for clipping little nuggets out of a long video, or adding a personalized style, like certain colors or fonts. You don't need a $50-a-month video tool to connect Claude to. You can take Claude or Codex, whatever you like, and just say, "use HyperFrames to edit this down into the best ten social clips." ## I still record in Riverside and Descript Tools like HyperFrames don't let you record. A lot of people doing content, like Ryan and me, record together, and we don't do it in Zoom or Google Meet because we want separate tracks and high-quality audio. So it doesn't make Descript or Riverside worthless. I still use both every day. Right after this episode, I had a Descript episode to record. Sometimes I record on a camera, and you can do a lot of it on your phone, but for recording multiple people at one time, Descript and Riverside are the best. Once I'm done recording, though, most of my editing isn't done in Riverside or Descript. It's done with a tool like HyperFrames. I also have [DaVinci Resolve](https://www.blackmagicdesign.com/products/davinciresolve), because all my video editor friends told me it was time to learn and step up. You don't have to be an expert with those fancy tools, because Claude or Codex can write scripts to steer them, make cuts, and add graphics. ## Your computer has to keep up Ryan made a point that gets skipped in the "it's over" videos: rendering. If you don't have a machine like a Mac Studio or Mac mini, he said, your computer is going to sound like it's about to explode. Some of these tools don't support Linux, either. If you're running things on a virtual private server, like [Grok Bot](/blog/how-i-set-up-grok-bot) with its own computer in the cloud, it can't use the tools that require a Mac. You need enough room to download your video and enough memory while it's working. Even in the terminal it uses a ton of RAM. It's not something you can do on a MacBook Air or a Chromebook. It needs local access and some pretty powerful hardware. ## How I clip our episodes Ryan focuses on an engaging intro. I use the same recording and do a lot of clipping. I connect Claude Code to the Riverside MCP and say, use this connector and find ten clips based on the transcript. They can be just me, or Ryan and me going back and forth. I just don't want to post any clips of only Ryan on my personal channels. I have a skill to clip those out, and it found ten from our last episode. I always try one new thing. Ryan saves examples of what his skill did well and what it got wrong, and I do the same: "I really like this, keep doing it," and "I really don't like this, never do that again." If you build on that with one thing each episode, within five or ten episodes your skill is going to be more dialed in than any model could be on its own. Even when new models come out, you'll have to give them feedback and context. Some things I like to do are sound effects and tweaking the layout, because two people in a social clip can be tough. My experiment this week was loops. If you've watched a reel or a TikTok where the last frame and the first frame match, it replays seamlessly instead of looking like it's sliding to the next video. I had Claude loop three of the ten clips I'm posting this week, last frame to first. ![The last frame and the first frame of Aaron's Liquid Death fridge video side by side, both showing the empty room with the case stack and the caption Deep beneath a suburban home.](/images/blog/liquid-death-nature-documentary-edit/loop-seam.jpg) The same loop on my [Liquid Death fridge video](/blog/liquid-death-nature-documentary-edit): the last frame and the first frame, side by side. ## Captions need a brand guide too I think the biggest one is a brand guide, a design system, even for video. For a landing page you have typography, spacing, and animations. For social clips, the version of that is captions: where you put them, what fonts and effects you like, whether the spoken word glows or animates in and out a certain way, and where they sit in the frame. When it's just me as a talking head, I like the captions somewhere below my chin, and my rules say don't cover my face. Anytime I talk about a tool, it puts in the Claude logo or the GPT logo, with little graphics that come in. When I do a two-person clip with Ryan, whoever does most of the talking goes on top, and the captions go down by the divider. That's all Tesseract with Claude Code, with no Riverside timeline or Descript subscription, just Claude running in a terminal. ![Punch-in on the car display with Aaron's brand caption band placed above the screen instead of over it.](/images/blog/one-shot-video-edit-claude-tesseract/carplay-punch-in.jpg) From my CarPlay edit: the caption band sits above the car screen so it never covers what I'm pointing at. ## Make the skill a little better every time Like a lot of the writing skills, landing page skills, and email skills people have built, your first version isn't going to be very good. I turn it into a skill, use it every time, and try to make it one to five percent better every time I use it. You're always going to get a better output that way than by asking, "What if I just try this with the new model OpenAI dropped next week?" ## Turn the finished edit into a how-to When you're small and just starting out, you don't have a bunch of sponsorships. People say they don't have the money to hire an editor or a copywriter, but you know you need a well-edited video, and you need to turn it into a blog. It's a fun time to be a solo person or a small team, because you can just use Claude for this and build a skill. On the episode I screen-shared my [one-shot CarPlay video edit](/blog/one-shot-video-edit-claude-tesseract). I knew some people would ask how I one-shot a video, and it was a real one-shot. I gave it no iteration, tweaks, or feedback after the first output. So at the end of the project, I asked Claude to use all my real prompts, go through our conversation, and turn it into a blog post I could share with people. The post starts with the finished video. All I gave Claude was one raw video I recorded on my camera and said, edit this. It put in the graphics, framed me, and zoomed in on the CarPlay screen. People want to know what camera I used, what editor I used, and the exact prompt, and those are all my real words. The post has the skill I used in Tesseract, the exact fonts and colors in my visual design system, and a screenshot of Claude Code on my phone, because I finished the edit from my phone instead of my laptop. Now I do that with most of my projects. At the end of the conversation, when I have the final edit, I say, turn this into a how-to guide and put it on my website. The editing project becomes a how-to, and if you're a freelancer, it shows people how you work. ![Filmstrip of frames from Aaron's finished CarPlay edit showing captions on the torso, punch-ins on the car display, and the closing card.](/images/blog/one-shot-video-edit-claude-tesseract/filmstrip.jpg) The filmstrip preview from that edit, one of the outputs the how-to walks through. ## Don't get locked into one video tool What I tell people is that if you already pay for a subscription like Claude Max or ChatGPT Pro, you should learn to do this on your own, because standalone software is going to wax and wane even more than the model labs do. You don't want to be stuck in Riverside or stuck in Opus Clip. They love to change their pricing on you, so never pay for a year, because you don't know what their product is even going to do three months from now. I'd rather do it with HyperFrames or Tesseract. If you're ready for it, DaVinci Resolve is amazing, but it's intimidating for most people. I'm not a real video editor, and I get a little anxiety when I open that app, because there are so many knobs and menus. I'd start with Claude and say, this is what I have for media, and this is what I want to get out of it, like five Instagram reels. How can I do this without paying for any external tools, and what do you need from me? It'll walk you through it. ## Generated media is the seasoning Some people think AI video editing means generating avatars and generating the A-roll, the main content. YouTube, Meta, and X are starting to knock down the ways to monetize those videos and changing how they label AI video. I've always liked this analogy. The generated media in your clips is like B-roll: a graphic that demonstrates what you're talking about, while it's still you talking on camera, not an avatar. Generative media is the seasoning on your meal. You don't want it to be the meat and potatoes. You want it to be the salt and pepper on what's already there. If you're trying to generate the whole video, everybody else is doing that and it's garbage, and the platforms probably aren't going to let you monetize it or push it out anyway. You're just burning money and credits. Find ways to dress up what you have, not replace who you are. ## Volume isn't an advantage anymore The floor got raised. You can clip faster and put out more content. Ryan and I both think volume is no longer a viable strategy, because everybody can crank out blog posts and videos with AI, and it can run 24/7. I once posted one TikTok a day for a year. That was three years ago, and it was hard to do. Almost nobody I've met can say they did that. Now with Seedance and about 50 cents in API costs, you can easily make a video a day, run it in the cloud, and automate it. The volume advantage is going away quickly. It's more about the quality and value of what you make than how many posts you can pump out. The people who already know video are the ones who get supercharged. I've watched professional video editors use these tools, and they blow me out of the water. I'm a self-taught weekend warrior. They went to college, and they know the vocabulary, the techniques, and the shot angles to describe. Claude can do whatever you tell it, but if you give it a simple prompt like "edit my video, make it good," you get the new version of AI slop. I see it on YouTube all the time now with Claude motion graphics. The really good people are going to fine-tune it into amazing visuals, and everybody else should just go raw instead of competing with AI slop videos that all look the same. ## How Ryan runs the podcast side Ryan's workflow starts with the same Riverside recording. After we record, his podcast editor skill goes through the project over the Riverside MCP and makes the basic cuts: pauses, ums, and the other typical cuts. He exports that to Google Drive, and a podcast intro skill builds a hook-style intro. He trained that skill on the intros his human editor made and on the AI versions he kept or threw out, using Opus 5.5 on high, and it calls Seedance 2.5 through OpenRouter for B-roll. He called the latest intro about 95 percent there. After an episode is published, his packaging skill writes title options, the description, tags, and thumbnails. His YouTube repurpose command turns the video into an SEO blog post with screenshots and a diagram, plus a promo email for his list. He shares a lot of those AI video skills in his [AI Marketing Insiders community](https://www.skool.com/ai-marketing-insiders), and the rest of his systems are on [ryandoser.com](https://ryandoser.com/). ## Bring yourself on camera The moat is your original ideas and your authenticity. You don't have to be intimidated by how to make a decent 20-minute YouTube video with an engaging intro anymore. You bring yourself on camera, let Claude, DaVinci Resolve, Tesseract, the Riverside MCP, or whatever tool stack you want handle some of the intimidating technical work, and spend more time on what you're going to say. You don't need a team of editors. If you want help bringing tools like these into your business or team, [book a 30-minute call with me](https://cal.com/aaron-makelky/intro). I'm also [@theaaron on X](https://x.com/theaaron), where I tweet about this stuff stream of consciousness. ## Related pages - [All field notes](/blog) - [I one-shot a CarPlay video edit with Claude and Tesseract](/blog/one-shot-video-edit-claude-tesseract) - [I made a nature documentary about my gym fridge](/blog/liquid-death-nature-documentary-edit) - [Editing a video from my phone with Codex and HyperFrames](/blog/codex-hyperframes-video-editing-workflow) - [Astra vs Fable, my earlier AI Rabbit Holes episode](/blog/astra-vs-fable) - [About Aaron Makelky](/about) ## Sources and references - [Watch the AI video editing episode with Ryan Doser](https://www.youtube.com/watch?v=FcSfa_3QlxI) - [Ryan Doser](https://ryandoser.com/) - [Ryan Doser on YouTube](https://www.youtube.com/@RyanDoserAI) - [HyperFrames by HeyGen](https://github.com/heygen-com/hyperframes) - [Tesseract on GitHub](https://github.com/mirage-hq/Tesseract) ## Common questions ### Can Claude edit a video without video editing software? Yes. Aaron points Claude Code or Codex at a raw video saved on his computer and has it edit with HyperFrames, which is free from HeyGen, or Tesseract. Claude writes the code for captions, graphics, explainer animations, and cuts in a terminal, so there's no editing app or subscription for it to connect to. ### Do I still need Riverside or Descript if Claude does the editing? For recording, yes. HyperFrames and Tesseract don't record. Aaron records with Riverside or Descript to get separate tracks and high-quality audio, especially with more than one person, then does most of the editing with Claude. ### What computer do I need for AI video editing? Something with local access to the file and powerful hardware. Some of these tools require a Mac and don't run on Linux, so a cloud server may not work, and editing uses a lot of RAM. Aaron says a MacBook Air or a Chromebook isn't enough. ### How do I get better results from AI video edits? Aaron saves his editing instructions as a skill, including a brand guide for captions, logos, and two-person layouts. He uses it on every episode, tries one new thing each time, and records what to keep doing and what never to do again. ### Should I generate the whole video with AI? Aaron doesn't. He uses generated media like B-roll that demonstrates what he's talking about, while the main footage is still him on camera. He expects platforms to keep limiting how fully generated videos are monetized and labeled.