ChatGPT Can Edit Your Descript Videos for You
Ross Zeiger shows how to use ChatGPT to automate Descript video editing with plain-English commands. You can remove silence, fix audio, strip green screens, and crop footage by typing what you want. AI handles about 80 percent of the work, but you still need to review and fix mistakes.
Chapters
If you connect ChatGPT with Descript, you can do almost all of the editing you need just by talking to it in plain English. So in this tutorial, I'm going to show you everything— how to download ChatGPT, how to connect it to Descript, and then how to edit it. So I'm going to take the video that I'm recording right now, this intro, and I'm going to use this as the sample. I'm recording into Descript right now. I'm gonna take this video and I'm gonna talk to ChatGPT to edit this up into the video that you're watching right now. Okay, let's see how it goes. First of all, you need to download the appropriate app. So I'm gonna download for Mac. If you're on Windows, obviously download for Windows. I'm gonna click on download for Mac. I'm gonna hit save chatgpt.dmg. I'm gonna click on that little arrow, install ChatGPT. I'm gonna drag this over to my applications. I'm gonna open up Applications and I'm gonna open up ChatGPT by double-clicking on it. And the first time I open it, I get this warning. I'm gonna say open and it says, welcome to the desktop app. I'm gonna ask you a few quick questions. Doesn't matter too much what you say here. I'm gonna say marketing and I'll hit continue and I'm gonna hit skip. I'll say go to ChatGPT. If you've got Claude on your computer, you can import your stuff from there or from other AI apps that you might be using. You can import that data. I'm gonna turn that off for now. And that could be really beneficial to do because all your work over there could transfer over. But I'm gonna hit continue. And here we are in a brand new ChatGPT instance. You can talk to ChatGPT with your voice if you want, but What I'm going to do for the sake of our setup is click right here, this little thing that says Plugins, and then say Connect Plugins all the way at the bottom, Browse All Plugins. And then, oh, right there actually is Descript. But you can also, if it's not there, you can type it in, start typing in Descript on the search bar, and then you'll see it right here. And I'm going to say Install. And it opens up my browser and it has me sign in and I can choose the drive that I want to sign into. If you have multiple drives, make sure to choose the right one and hit allow. And it's going to open up ChatGPT again. And there we go. It now says Descript plugin installed and then click try now. It tags Descript. So this is letting it call on the tool, call on the connection that it's made. And it dropped in this line that says, edit this recording for clarity and clean up my audio. I haven't even provided it a recording yet. That is just a placeholder or template that it's dropped in. But what I could do is click this plus button, add files and folders. And then I could find a video on my computer and provide that to Descript to go ahead and start editing. So what I'm going to do is pause this recording. I'm going to take my recording of the intro of this video you're watching right now and then provide that to Descript. And that's what we're going to work with. What we can do is take the video that's already been recorded into Descript. So here it is. This is the video that I already recorded at the first, what is it, 7 minutes of the video that you're watching right now. And just to point a couple things out to you on what we're going to do, you can see right here there's 33 seconds of silence before I actually start talking. That was me starting the camera and kind of arranging things. And then there's 46 seconds of silence. I can see that because there's no words transcribed right there on the timeline. And so there's some issues with this. There's some retakes, there's some things I misspoke, and that's the kind of stuff that ChatGPT is going to edit for us. So in order for ChatGPT to actually see this project, what we can do is come up to the top where the name of my composition is. It's— it— AI titled it Descript Mastery Sound Quality Check. And I can hover over that composition, click on the 3 dots and say copy link. And that's going to be enough for ChatGPT to find this video in my project. So I'm going to say, I'm going to leave in actually this default that it said, edit this recording for clarity and clean up my audio from this project. And you just tell it in plain English. So the exact wording I use Isn't important, but I'm gonna say from this project colon, and then I'm gonna paste in that link that I copied. And I'm gonna, I'm gonna bump up the thinking on the model here. So it's currently using 5.6 SOL. You can do anything from light up to extra high. The higher it is, the more credits it's gonna use. And by the way, I'm on the $20 per month plan. With ChatGPT. So if you're on the free plan, your limits are going to be really low. You're not going to be able to do a whole lot, but, uh, just FYI. So I'm going to hit send. It tells me what it's thinking. It says, I'll open the Descript project, review the recording, then tighten the transcript for clarity and apply the available audio cleanup tools. I found the 6:55 recording. So that's good. That tells me that it is connected. It's seeing my Descript drive and my project. And now it gives me a thing to give it permission. It says allow Descript to run descript.prompt. You don't need to worry about the coding that it's going to use. It's going to throw in some composition IDs, project IDs, things like that. It's just going to make sure that you're giving it permission to do what it wants to do along the way. And if you're comfortable with what it wants to do, so here's its plan. This is the prompt. That it's going to basically do with the steps. So it's going to preserve the speaker's meaning tone, whatever, remove, clean false starts, abandoned takes, keep the screen recording. So you can read over those. And if you like what it says, then you can say allow once, or in the bottom left, always allow. That's going to make it so that you get fewer of these checkpoints where it's asking for your permission. Or if there's something in here that you don't like, That, so for example, it says, do not add music. If I do in fact want it to add music, I could say deny. That's deny this plan. And then it's going to give you an option to put in a new prompt where you can say, do everything we talked about before, plus add music. But for now, I'm going to hit always allow. Let's see what it does. And you'll get this loading bar that says running with the progress going across. And for something like this on a 7-minute video with Not a whole lot of edits, it's probably going to take a couple minutes. But the longer the video that you're working with, the longer this is obviously going to take. So if it's looking at an hour-long video and you give it a giant list of things to do, that's going to take a lot longer than a short video. And it actually says connection failed. So I'm— I leave this kind of stuff in because I'm going to show you how I troubleshoot it if I need to. Um, but ChatGPT might on its own go ahead and retry it. So let's see what it does. Okay, so it did go ahead and figure out, it didn't require any extra intervention from me. It just updated with edited successfully in Descript. And this in blue is a link to what it did. And what it did, instead of editing the composition in place, it created a new composition. That it named Clarity Plus Clean Audio and left the original untouched. Now that might be because I'm recording in the same project while I'm trying to do this. So that might be causing issues. And so it found out a workaround and it gives me bullet points of exactly what it did. It removed a false start and repeated conjunction, tightened 14 excessive pauses, reduced runtime from about 8:48 to 3:45. Without removing substantive content, applied studio sound at 70%, and preserved screen and narration synchronization. Okay, so again, I can click on this link right here and that opens up the project inside of the ChatGPT app, which isn't good for our purposes right now. So I'm going to close that and I'm going to just check in the project itself. I'm going to click on the composition. And there it is in that dropdown. So there's the one I'm on with the checkmark. That's the project I'm currently in. Untitled is where I'm currently recording the video that you're watching right now. And then Clarity Clean Audio is the copy that it made. So right off the bat, I can see that, you remember I had 30 seconds of silence before I started speaking. It trimmed that down to 1 second. And I can see some things that it ignored right there, that, that strikeout. So it says, I said you can do almost all, and it took out the can do. So you almost all of the editing. That's not a good change. That's probably not going to sound good. But anyways, you can see how I'm going to show you in a second how you can take what ChatGPT started doing for you. And edit it, take your human eye and fix the mistakes that it made. And by the way, this is a really important point. With a lot of my workflows, I've gotten to the point where I can have ChatGPT or Claude do 80% of the edit for me, but it still requires me as the human to go over it and fix the mistakes it made, fix the things that it forgot to do that it messed up, or put in the wrong place. So we're not at 100% of this being able to edit for you in most cases. There are some workflows I do where it can do 100%, but for what I'm showing you right now, this is exactly how I would handle it. So let me go ahead and play this. I see another thing crossed out there. You need it down. So I had a take there where I was like, I said the wrong thing because I said, okay, so first download for Mac. And I didn't mean to say that because some of you are on Windows. So I actually changed how I said it. And that was a good cut. I like that. However, it cut it a little bit abruptly on the next word. So right there, it kind of clips the word you. So what I can do in the editing process is smooth that out, restore a little bit of that cut so it's not so abrupt. And by the way, one thing that I want to show you is in the top right here, I have my credit usage. If you're not seeing that, it's something that you turn on in the settings of Descript. So if you click on this dropdown, you go to settings, and then on this first tab here, this general tab. All the way at the bottom, the very last one before the labs box is this thing that says always show usage info. If you have that toggled on, you'll see how many Descript credits you've used and how many you have left for your current billing month. So if you see up here in the top right, it's currently on. If I toggle that off, now it's gone. That is the default is for it to not be on. But I'm going to toggle it back on and there it is. So you can see I have 422 credits, AI credits remaining on this cycle. And if I click on that, there is the feature I just used. So that edit that it just made, it says Descript Mastery, my Drive name, via MCP. That's the name of the connection that Descript has with ChatGPT. And then on the right side is the amount of credits that it used. So that action where it did, what was it, 6 edits, that used 30.8 credits to do that. So just be aware when you have this connection made, there are things that use your ChatGPT credits, your ChatGPT allocations. And for example, if I ask it questions about the project, and I type in a question about the script, like any feedback on this script, anything that could be clearer or more concise, and I send that, it can offer advice or feedback on my prompt. Now, people ask all the time, why, what is the advantage of this? Well, a few things. First of all, if you work with ChatGPT a lot, it knows a lot about you. It has context on your voice, your style, your audience, if you've talked to it about your customer avatar or your audience avatar, like who's typically watching and consuming your videos. And so with that background information about you, it can provide tailored feedback. Whereas if you're talking to Underlord, Underlord is this thing in the bottom right corner, which you also have access to GPT or Claude, et cetera. But every single time you start a session with Underlord, it's starting from zero. So that's one big reason why you would consider using ChatGPT, the connection to work with Underlord. Now, another thing is if I click on the models here, the auto button, you can see I have the Claude models and I have GPT-5.5, and I have a lot of the older models as well. However, I don't have 5.6. So that's another reason to use ChatGPT in the app is you can have access to the latest models and you have more granular control. So again, that slider I showed you, I can go all the way to extra high. I can go to extra light to be more conservative with my tokens. Whereas it's not obvious, it's not transparent in Descript. which version of ChatGPT 5.5 you're using. So you get more granular control and it has more context on you as the user. So that's 2 big reasons for this connection. Now do note, it burns my allocations that I'm paying for or not if I'm on the free plan in ChatGPT. It burns down my 5-hour allocation and it also uses Descript credits. So that last thing that I just did, that last question I asked it, it used 3.6 credits of my Descript AI credits. So it is drawing down both. Okay, so now you've seen a couple different use cases of this. I'm going to show you how to actually edit the video. So I'm going to tell it apply green screen to the webcam speaker and crop in on the screen share so only the browser window is showing. So what I want it to do is 2 actions. I want it to remove my background on the— let me make this a little bigger— on the video of me. And then I want it to crop out all the background around my desktop so that we're only seeing the browser window. One thing you can do is have it always ask for the approval. This is where it asks for the approve or deny. As it takes action, I'm going to say approve for me and just let it go ahead and run and let's see what it does. And it's done. And it says I applied green screen background removal to the webcam speaker throughout, a 1.65x crop on the screen share. So we'll see how that did in a second. Existing bottom left webcam placement retained, all prior cuts synchronization and audio cleanup preserved. Okay. And over here it says finalizing video for playback. So I'm just going to do a refresh on that. So it looks decent. It did the webcam green screen. However, it didn't completely center my browser window. So this is another example of where me as a human editor, it took a good first stab at zooming in, but I'm going to need to come in here, double-click and reposition this. It actually wasn't far off at all. I don't know if you saw how small that movement was, but I would do something more like that to center it. And again, the quality of the output is often the quality of the prompt. I didn't tell it I wanted it centered, or I didn't give it any specifications on that. So that's an example of something that I could fix, I learned from, and I'd be a little bit more precise next time. Now, the next thing I'm going to tell it to do is add layouts. So if I go to my properties, let me see what layout packs I'm working with. I'm going to go to change layout pack. I'm going to look through my drive and I'm going to have it do this one, the YouTube Descript Mastery Brand Kit. And the reason I'm looking is I want to precisely use the name of this pack. So when I'm reference— referencing it to ChatGPT, I'm going to say, now go through this video and apply the layouts from, and then in quotations, I'm going to say the exact name. So I have YouTube in brackets like that, Descript Mastery Brand Kit, end quotes. And I'm going to say, I'll let you decide where and when to use them. Now, again, this isn't a good example of a prompt. I'm more just demoing the capabilities of this. But a good prompt would be very specific, like use an intro slide from 1 to 10 seconds, use a quote layout from 30 seconds to 45 seconds. Like the more specific you are, the better your results are going to be. But sometimes too, AI will surprise you and do things that you wouldn't have done and you're happy with it. Um, so that's kind of what I'm demoing here. So let's go ahead and send that and let's see what it does. And by the way, I have a layouts masterclass as part of my bigger Descript Made Easy course, or you can buy the layouts masterclass as a standalone course through Udemy. I'll leave links to those in the description. And my Descript Made Easy program, in case you're wondering, it's got lessons on every feature in Descript. It goes deep, deep, deep into everything you need to know to learn how to use this tool, but the primary value of it is that we meet 3 times per week on Zoom calls. And I'm there, I make myself available for an hour to answer your questions about Descript, about YouTube, about making videos for social media in any form that you're doing it. And even AI tools, if you want to ask about ChatGPT or Claude or anything like that, more than happy to go into those. And we meet Tuesday, Wednesday, and Thursday morning. In US time zones. And we have a lot of fun and we help each other grow faster than we would on our own. So if you really want to learn how to master this tool, check that out. I'll leave a link in the description below, or you can go to descriptmadeeasy.com. Okay. So I can see it's starting to do stuff in Descript. It dropped in a layout and I'm assuming this text is going to get replaced soon. AI is going to override the placeholder text with text that's relevant to this video. And let me just scroll through here. Let's see what else it did. So here on the last scene, it applied a speaker view where I'm full screen and it took away my screen recording. And there we go. If I go back to the first scene, it's now populated the placeholder text. It says connect ChatGPT to Descript. Okay, and it's done. It only did 3 layouts. So it did the intro that I showed you. It applied the screen recording layout to the middle meaty part of it, where I'm actually showing the screen. And then the speaker name one at the very end, which is the one that's on the screen right now with my face. And then it just says Descript Mastery. You could have that be your position. Like it could say founder, for example. And that's a case too, where again, I didn't tell ChatGPT what to put in there. So it just took a stab. At what it thinks I want, but I know that I prefer the term founder. So that's what I'm going to put in there. Or you could say YouTube tutorial or something like that, but it's up to you as a human to go and fix the things that you were either too vague on or AI messed up on. Okay. The last thing I'm going to demonstrate to you is sort of a catch-all. You can tell ChatGPT, I said, go through this project and make suggestions. Did I miss anything? What would make this project better? Keep in mind, my audience is Descript beginners, and I'll hit send. And here's what it came back with. So it says, this currently feels like part 1 of a tutorial that promises part 2, which it was. That's solid advice. And it says fix before publishing, and it gave me some numbers here. So like number 1, deliver the editing demonstration or narrow the premise. Yep. Gonna do that. Add beginner prerequisites. That's a good idea. That before downloading anything, briefly explain that a Descript account is required. The viewer should already be signed in, et cetera. That is good things to mention. I will incorporate that. Explain MCP. So the terms, that is useful. Add zooms and callouts. Record a real ending. Okay. So you can see how you can work with it as your partner and sounding board, et cetera, to get better ideas. Now, the thing I'm going to show you that will tie everything I've shown you in this tutorial together. Is how to systemize this so that everything you did today, you can take it, put it into one prompt and use it from now on. And I said, based on all the edits we did today, put it into one prompt that I can use again for next time when we edit together. And it says it's currently writing it. And there we go. You can see how long this is. I won't read all of this, but what it does is it gives it a lot of context on what you're looking for. And then the exact steps that it needs to take. And the way that I would use this is I would copy this prompt and put it into your note-taking software of choice. So I use one called Standard Notes and I just called it ChatGPT prompt. And I would paste that all in there. And then the next time I would take this whole thing, copy it, give it to ChatGPT, see how it does. If it does well, then, and there's no reason to make any changes, then use it exactly as is. If it, for example, applies studio sound and 70% is a little bit weak, I can still hear the air conditioner hum in the background, then I'm gonna take this prompt and I'm gonna edit it to say 80%. And you can just keep a dynamic prompt like this, a living, breathing prompt that you update and edit as you go. And the key unlock with all prompting is using AI to have it make prompts itself. And you tell it in plain English what was wrong with it. So like I say, this is great, but the studio sound is too weak. And I send that. It's going to update that prompt to include the adjusted studio sound volume. So there it is. It says now 90% intensity by default. And so I'd take this and plug that into my running prompt. So I'm just gonna copy all and then paste it in. And now I have my new parameters and I can clean up the, uh, you know, the redundant audio heading there, et cetera. But I would take exactly this and paste it into ChatGPT next time. Or you could also run that same prompt in Descript's Underlord and you could save it as a prompt template there if you wanted. So it's not perfect, but I recommend Work with it, play with it. If you get a prompt that works for your workflows, it can save you a ton of time. I have some workflows now where I drop in my raw video, I give ChatGPT my prompt or Underlord or Claude, it doesn't matter, they all work together. And I step away, make a coffee or walk my dog, and by the time I come back, my video pretty much is ready. It just needs me to go over it, fix some things, and go from there. Let me know in the comments, how is this integration working for you? How much editing has it replaced in your workflow?
