He Never Has to Edit Videos Again (Descript AI Automation demo)

Video editing is becoming a thing of the past, and that's coming from somebody who for the last 3 years has made a significant amount of my income from video editing. And nowadays when someone reaches out to me asking me if I'd be willing to edit their videos for them, I look at their desired outcome, their how they want their videos to look, and I tell them, you know what, you don't need me. AI can do this for you. And so the biggest part of my business the last month or so has been helping people create automations so they don't need to edit at all. And so today I'm going to show you how I think about this, how I do this for actual clients. And today's example isn't even a paying client. This is somebody who volunteered. Um, if you haven't seen my last video where I showed how to set up the Claude automation with Descript, I'll link that up here in the top right corner. But in that video, I put out a call for volunteers and I'm working my way through that list. So I'm going to show you in the next series of videos how I would take a video based on what you want it to look like and based on your existing videos, how I can set up a series of prompts in Underlord or in the Claude connection so that you don't need to edit at all. Today's volunteer was a man named Jeremy, and we're going to look at this together on how I can take one of his raw videos. Create a prompt for it that will completely edit his videos for him. So let's get into it. Here's Jeremy's channel. It's called Real Talk for Men, and he makes talking head videos, um, exploring men's topics, things like anger and dating and mindset, success things, stuff like that. And they're mostly talking head videos with B-roll interspersed, and he throws in some captions, some music, things like that. And that makes him a good candidate for these automations. Talking head stuff, simple videos where you're just doing one shot, those are great candidates for the state of AI right now. This is actually his finished video here, and as you can see, if I scrub through, he's got different scenes where sometimes it's punching in on his talking head. He's got the captions like I mentioned, and then every once in a while he throws in some B-roll. Sometimes they do extreme closeups, quotes, and things like that. So this is the finished product here. We're gonna try to create something in this style again using Underlord. So I'm gonna create a new composition and I'll just call this AI automation video test. And then I'm gonna bring in the source video file. So I'm just gonna drag that in and there we go. We got a 20-minute video. And I can see there's a lot of empty space right at the beginning, a lot of empty space. He did a couple takes there. Okay, so there's gonna be a lot of rough cutting to do, and there's gonna be a lot of adding the B-roll and all that kind of stuff. So the way that I'm gonna approach this is think in versions. So I'm gonna do my best attempt at knowing what I know about his end state videos, I'm going to try to give Underlord a detailed prompt to recreate those edits. And it's not going to go well the first time. I expect that to happen. But what I'm going to do from there is create a version 2. So I'm going to take that prompt that I tried, I'm going to copy it, and I'm going to tweak what Underlord got wrong, or things that I left out, or whatever. And we're going to create a version 2. And we're going to keep going, and usually it takes 4 or 5 versions. Until we get the desired result. And right here I can give it its prompt. I like to use Sonnet 4.6, by the way. This is a good trade-off of quality but not burning too many credits. So the latest model is Opus 4.7 that has a high AI credit cost. Haiku has a lower AI credit cost, but it's not quite as intelligent, shall we say, as 4.6. So 4.6 is a happy medium and it's what I'm currently using and recommending. And by the way, quick tip for you. I don't even type anymore on my computer. I use a transcription service where I can just press a button on my keyboard and then whatever I say gets typed in as text. And so that's what I'm about to do throughout this video. I use a tool called Whisperflow, not sponsored or anything, just a tool that I like to use. Edit this video for me. Remove all bad takes and false starts. Generally preferring the last take. That's usually the best one. Apply studio sound at 80%. Remove filler words, but only remove ums and uhs. Leave everything else. And avoid harsh cuts. So if the filler word is too close to the word before or after it, leave it in. Create a new scene out of every single cut and occasionally punch in, punch in by 10% at times, other times at 20%, other times at 30%. Also, where appropriate, add B-roll. Sometimes use stock media, sometimes generate images for me for the B-roll. At the very beginning, add appropriate music that complements what's being said and lower the volume of that music layer to 35%. Remove long pauses of 3 seconds or more and cut them out completely. Occasionally when there is a particularly strong point, put it on the screen as a quote. Occasionally the quote can be overlaid over the video, but sometimes cut to a text-only page using the Helsinki Blueberry layout pack. And quick side note with the layout packs, if you've got them created, which I recommend you do, and I have lots of videos about that, you should have a layout pack created in your brand. So your colors, your logo, the look and feel that you want for that video. And then you can tell Underlord in this prompt that you're creating, use fill in the blank of the name of the layout pack that I created. But since I don't know what layout pack he has, I'm gonna say the Helsinki Blueberry as a placeholder. At the end of the video, add an outro with a call to action to subscribe to the channel. And real quickly, I'm gonna go look at his original video just to see again Okay, he's got the B-roll, he's got the zoom-ins. I see like some music or sound effect here. Let's see what that is.

The first point.

Okay, so this prompt that I created isn't going to give the exact result, obviously, but I think it's close enough. So I'm gonna just go for it. This is gonna be my version number 1. So we're gonna hit this little up arrow right here and that drops in the entire prompt. And again, I used, uh, I was, I was just speaking to the, uh, text-to-speech tool that I have and it wrote that all out for me. And this is actually a really strong way to talk to AI because you're just thinking stream of consciousness and you're providing it a lot more detail. Um, you could tighten it up. You could, you could, you know, proofread it and stuff like that, but I'm just going to go with what I got for now. And Underlord is creating its plan. And Underlord tells me what it's going to do up here. So if I look this over, it says, here's my plan. And it's always a good idea to review what it's going to do because sometimes it misinterprets what you're trying to get at, or it makes assumptions where you didn't provide enough context. So it's, it's always a good idea to review this, check for anything that's missing, check for things that it might have misunderstood. So it says I'm going to remove all bad takes and false starts. And it even says the paragraphs where there are retakes. I'm not going to spend the time to actually go look at all of those paragraphs. These correspond, by the way, so if I go to the beginning of my, my thing here, it says N0. Underlord, when it transcribed your video, it automatically makes paragraphs out of your text. And so there is that first bad take that would be N0, which it's correctly identified as something that it's going to remove. Um, just to point that out. Apply Studio Sound at 80% intensity. That's what I told it. And I don't know yet if that's going to be a good level. We're going to apply it, see how it sounds, and then we're going to adjust from there. But we're at least taking an initial stab. Remove filler words. Yep. It did what I told it, or it understood what I told it. Remove long pauses of 3 seconds or more. Yes. Create a new scene at every cut. Punch in at jump cuts. Okay. Its plan is good. And then it asks me to confirm a few things before it proceeds. Helsinki Blueberry comes in variants. Okay, so I didn't tell it if I wanted the dark or the light. It's asking me, which one do I want? And in a second, I'll show you the easiest way to pick that. For the outro, should it reference a specific channel name or subscribe link or social handles, just a generic subscribe for more? For the B-roll, should I lean more towards stock footage or AI-generated images? Okay, so then if you click on this, this style direction down here, this is where it gives you some more capability in this form layout. So you can choose, if you don't want landscape, you could set this to square or portrait, but this is for YouTube, so I'm going to leave it on landscape. Layout pack, there's the layout pack that I told it. And again, if you had your own layout packs created, then for example, if I went into Drive And if this was one of my videos, I could do Descript Mastery. Or for him, hopefully he created his own layout pack and he could select that. And then it ensures that he gets that nice consistent look that he's already made. For B-roll, for the fill type, generative is going to be, it's going to be made with AI. Stock media means it's going to just go off the stock media. Placeholder means it's going to be empty with a placeholder element, which is an object that goes in your video. That you can actually swap out with videos or images that you place in yourself. So I'm gonna do generative. I'm gonna let AI make these for me. And then you can choose how you want those AI images to look. Do you want it to be like cartoonish 3D animation? Or I want to go for a more realistic look with the cinematic. So I'm gonna leave it on what it already has there. And then back in the Underlord prompt, I'm going to finish off answering these questions. So I'm gonna say, let's do Helsinki Blueberry Dark. For the outro, just make it a generic subscribe for more CTA. And for the B-roll, let's do AI-generated images. And then I'm gonna hit send. And it tells you as it's working, like, now let me remove all the bad takes and false starts based on the script. I can identify all the retakes sections clearly. Finalizing plan. And you can watch it work. And it's finished. So you can check the details here. And you can see a whole log of the things it did, how long it took in milliseconds. And some of these, you can see little failed things, and it's smart enough that when it fails things, it'll just retry it. And it can even look at its work, check its work, and if it sees something it did wrong, like the formatting of the text layer is messed up because it's going over your face. It's now smart enough where it'll actually make those adjustments. If you saw my video like a year ago where I tried to show Underlord, it, it could not do that kind of stuff. And the, the choices it made were all over the place, but it's gotten a lot better. So let's just make this bigger so we can see what we got here. Make the timeline a little bigger so I can check down here too. I'm gonna hit the home button to return to the beginning of the video. And let's just play this together and see what we got. And actually, before I do that, I'm gonna click on the sound layer. Okay, it turned it down to 48%. That's interesting. I think it told it 35%, so I don't know how it came up with that number, but let's play this.

So how's being pissed off working out for you, man? Men and anger, you know, like peanut butter and jelly. It's just two things that go together. I've been told my whole life.

I don't love the music selection. It's kind of upbeat. And what I could do, you know, to refine this is maybe he's got a song, maybe he's commissioned an artist to create a song or he's created a song in an AI tool. And he always uses that as his intro. If that's the case, he could save that to his media library and include that in his prompt and say, always use this song. That way he has a consistent thing rather than leaving it up to AI. But that's a good example of how my vague prompt caused a bad result. I didn't give AI any context on the tone I was going for. If I said, I want this song to be epic and I want it to set the tone on a video about anger, then that's a very different prompt than me just saying, add music. So I let AI decide and it chose a kind of upbeat song that I don't think quite fits this video. But let's, let's press on. Let's go look at, I'm just going to skip ahead here to different scenes. Okay. So scene 1 is zoomed out. I can click on it. I can see it's at 100%. If I go to scene 2 and I click on it, I can see it punched in to 115%, which is great. I love that it did that, but it's, uh, his head is now like kind of lower. Whereas if I was doing this, I would put him more like in the top left third like that. So that's something I'm going to tell AI. I'm going to have it adjust that. And then scene number 3, there's an AI-generated B-roll. Let's just see how this fits with what he's talking about.

Like, that's our power. Listen, I'm going to start off by saying Okay, a couple of problems.

I don't like how it does a smart transition when it's cutting into the B-roll. So I'm gonna clarify that with AI. I'm gonna tell it to do hard cuts. And there is a captions overlay, which is great. The text it used, the font and everything, that comes from the Helsinki Blueberry layout pack that I told it. So again, if you had your own layout pack, you could have that be your own fonts. But let's see here. Let's just keep moving. And I see that the music keeps playing. I didn't want it to play the whole time. Again, this is probably a problem with my prompt. I should have told it to cut the music out once the intro has wound down. And what this actually did was it looped the song. So the song's playing throughout the whole video. So we're going to fix that. There's a text-only screen. That's okay. Again, if he had his own layouts here, this would be more personalized to his branding. Okay. And then let's just scroll to the very end. And there's the outro page. So it says, if this hit home, subscribe for more, download the free guide and leave a comment plus subscribe. Okay. That's fine. And this is that placeholder element I told you about earlier, where if you had elected to. It can drop in these elements and you can say add media, and then you can replace that with an image or another video. Um, if you want, but this is just a, a CTA outro page, so we can actually just delete that. I'm just gonna delete that. Now that we've reviewed the video and seen what AI has done, you have broadly two options. Well, let's say three. Option number one is you could leave everything as it is. You could say, I like this prompt, this did a good first pass for me. And then now you take over as a human editor. You apply your human judgment, go through, fix the little things like I was doing, like, you know, changing the positioning of his video, lowering the audio volume, changing where the audio starts and ends, that kind of stuff. Option 2 is we refine the prompt based on the changes that we see need to be made. And we make it a little bit better. We run it again. We see how it does, and we keep repeating until we nail down this prompt to get the result that we want. And option 3 is you call this good. Maybe you were lucky enough to get a one-shot prompt. And in that case, what you can do is go to Underlord and find our prompt here. Copy it. That's the original prompt that I had. Right-click, say copy, and then click on this, this plus button right here to create a new chat. Click on start with a prompt template. Click on new template. And then this is kind of a weird bug in the interface at the time that I create this in May 2026, but we got to click on start with the prompt template again. And that opens up the, the, uh, thing to save this prompt. And then I can paste that in and I can hit continue. It automatically creates a name for this prompt. And I'm going to call this, I'm just going to, at the end of it, call it V1. And actually I'm going to call this Jeremy template. So we got Jeremy template V1, and then there's a description. And again, that was generated by AI and you do have to have a description in order to save this. It won't let you if that's empty. So if you're trying to do this at home, be aware of that. You can upload a custom thumbnail. So one note on all this, by the way, is eventually where Descript is headed with this is there's going to be like a publicly accessible template library, which is down here. You can see this share publicly. They don't have that capability set up yet, but Eventually you might want a custom thumbnail that you can show off. And so then people can— makes it more attractive for people to find your templates and use them themselves. I'm gonna leave this on private and I'm gonna hit publish. And now it returns me back to the template browser here. And it says all templates, which all these ones that you're seeing here are Descript templates. They say by Descript, but I can click on this dropdown and I can say created by me. And you can see I've created a bunch of templates over time, but if I scroll down to the bottom, there is my Jeremy template V1. And if I click on it, I can see details about it, the tags and the prompt itself. Now, one annoying thing about this whole interface is there's no edit button. So I can't just click, I can either use it and it's gonna drop it into this project and run this prompt again. Or I can click these three dots and delete this template, but there's no edit button. So what I have to do, it's— we're now gonna get into the phase of our version 2. And so what I'm gonna do is click on this little button right here that copies this whole prompt. I'm gonna close out of here. I'm gonna modify the name of this composition. So I'm gonna rename and I'm gonna say V1, and then I'm gonna create a new composition. And I'm going to completely start over. I'm going to call this one V2. I'm going to bring in the source video file again, the unedited. So this one is 20 minutes and 24 seconds, and our V1 after the cuts was 14 minutes and 20 seconds. And Jeremy's cuts, the human-made cuts, he had it down to 9 minutes and 25 seconds. So he cut quite a lot out, but I'm going to go back to V2. I'm going to come back to Underlord and I'm going to paste in our prompt. And by the way, instead of saving these all as templates in Descript as like your V1, like I showed you, and then V2 and et cetera, you could do this in Google Docs. So if I come over here to a blank Google Doc, or you could do Microsoft Doc or Apple Notes or whatever your, your tool of choice is, and I could start a thing here that says V1, and I'm going to paste in. So there's my V1, and then I'm going to do a V2, and I'm going to paste it in again. But this time in V2, I'm going to make all the changes that I identified it needs. So Studio Sound was a little bit intense. I don't know if you could hear that through the recording, but I'm going to lower that to like 65%. It was a little bit intense and I couldn't hear all his words well. The music was a bit loud, so I'm gonna change that to 20%. I'm gonna have it cut a little bit more aggressively. So it says long pauses of 3 seconds or more. I'm gonna say 2 seconds or more. And now I'm gonna start adding my net new instructions. Do not use transitions, hard cuts only between scenes. After applying punch-ins, make sure that the face of the subject is in the top left third or top right third of the image. And if they're centered, make sure they stay centered with their eyes on the top third. Cut any tangents or parts where the speaker gets too far off subject. And by the way, on that last one, that is something that not everybody might be comfortable doing. That's giving editorial decision control to the AI. And you can test this out, and if you like the results, you know, stick with it, but for most people, that's giving the AI a lot of decision room and it's gonna, it might cut out things that you want to keep in there or jokes or things that add your personality to it. And so I'd be careful with that, with giving Descript editorial decision control beyond cutting out retakes. Retakes and bad takes, those are generally safe. It, it can do a pretty good job of that. But something like this might be a little dangerous, but we'll see how it does. Aim for scenes to be no less than 4 seconds, but no more than 10 seconds. So with this last line, there was times in our version 1 where it was just making too many scenes. It was just making tons and tons of cuts. And then there was also scenes where it dragged on for 20 or 30 seconds. And so what I'm trying to do with this is set some parameters to make sure that things aren't too fast, but also not too slow. And so this can be, if you develop a, a rule of thumb for your pacing, a clause like this can be really helpful. For static images, apply Ken Burns. For the music, make sure it only plays for no more than the first 30 seconds and then fades out. Okay, and I'm gonna leave this as is for our version number 2. So let's copy that. I'm going to select it all. And then we return to Descript. I'm going to paste this in and I'm going to hit send. And while this is working, you might be wondering how many credits is this all taking? Well, I'm going to give you a final accounting at the end, but let's check so far just to get to our version 2. So if we go to our settings and we go to usage and we scroll down, you can see an itemized list of all the credits you've used. And my allocation's just reset. So I've used roughly 200 credits so far in developing this, and some of these were generating the AI images. So if I had used stock media instead, that would've been a little bit more efficient. Um, talking to the Underlord, that was like 4.5 on my last Underlord send. Applying studio sound was 10 credits. One of my previous Underlord sessions, I think the big one where I give it version 1, that was close to 100 credits. But anyways, roughly 200 credits so far in developing these AI prompts. Okay, so now Underlord is ready with our results from version 2, and it's got apply studio sound to 65%, remove filler words, long pauses. I won't go through all these again. This time it wants me to confirm the Helsinki Blueberry variant. Oh, and that's something, if it keeps asking that, That's something that I should clarify in the prompt. So let's go back to our document and let's start a V3. I'm just gonna copy all this again. I'm gonna hit return. We're gonna call this V3, and then I'm gonna find in here where I said Helsinki Blueberry, and I'm gonna say Helsinki Blueberry dark. And that's gonna just make it so that it doesn't always ask me that. It's gonna know. Exactly the one I want. Okay, let's go back to Descript. Outro, should this be talking head scene? Okay, and so since it asked me that before, I'm going to clarify that as well in my version 3. And it's this clause right here at the end of the video, add an outro call. I'll say this should be— I said this should be overlaid over the talking. So the video is going to cut out, but the audio is going to keep going in the background with our outro CTA. Okay, so then speaker 2 appears to be a second person. Since we're only focusing on Jeremy, I'm just gonna tell it to ignore that. It's asking if it wants me to detect the face to adjust the position. I'm gonna, yes, that is what I want it to do. And quote selection, no, I don't have any specific quotes. So once again, I can click on the style direction, choose the aspect ratio. I'm gonna leave it on landscape for YouTube. I'm gonna leave it on Helsinki. and we're gonna have AI-generated images. Okay, so to clarify on Underlord's points, this'll be Helsinki Blueberry variant number 2. The outro should be overlaid over the video, so we shouldn't see the face. We should still hear the voice, but just overlay the outro CTA information. Ignore speaker 2. Speaker 2 should be cut out entirely. For the face positioning, yes, detect the face position automatically and adjust according to where the subject is. And yes, do some alternating left and right thirds to keep it interesting. For the quote selection, you decide. Okay? And then I'm going to say send. It's telling me an overview of the updated plan, and I'll come back when it's done. So this time I want to show you something. Two times Underlord— so I gave it that prompt and it said, oh no, something isn't right. Here are the details, network error. So it stopped the work. It did maybe a quarter of it. And then it gave me this error message. And I leave that in because this will happen to you. Descript has errors all the time with this kind of stuff. This is all brand new. And officially Underlord is still in beta. So expect errors. This is part of the game. However, this was an easy one to work around. I just said, keep going. It started it again. It said, let me requery. And it was able to look and identify where it left off. But it had another error and all I said was, go. You can just tell it in plain English, like start over, keep going, whatever, and it'll figure it out. And this time it worked and it finished everything that it needed to. So let's take a look once again at its work. Um, okay. It did a punch in. I like this punch in better and you can see it actually punched down and to the left so that it kept him centered. So that is awesome. Love to see that. And then it went, it punched back out. It went to a quote. I don't like this, the speaker. It just did a, they just said speaker instead of his name. So, and that's kind of, again, everything is ultimately my fault for the, the quality of the prompt. Um, because Underlord is just doing what it was told or what it, it thinks. And so if I go back to my Google Doc, then I'm going to say, anytime you are quoting the speaker, the speaker's name is Jeremy spelled J-E-R-E-M-I-E. So we got that set. And then let's press on. Beautiful. It did 120% punch in and it put his face in the top left. That is awesome right there. That, that is so cool that Underlord is smart enough to figure that out. Um, let's just glance quickly at some of its B-roll choices and I'm just going to actually play that.

The first point, anger gets the job when you're a little boy because it showed up for the interview all full of piss and vinegar in a really loud voice. And it said, I've got this. Perfect. So the first point. When you're a little boy.

Okay, so there was like a bad cut right there where he says perfect. I don't, that's something that I would probably as a human editor have to remove. Again, I could put things in the prompt, like making this flexible to say, if he looks down and says just one word, then delete that. But that's probably something that's easiest removed as a human editor. And one thing I didn't check is the intro music. Let's take a look at that real quick.

So how's being pissed off working out for you, man? Something about men and anger, isn't it? Like we're told it's like peanut butter and jelly. Like somehow it's like—

Okay, that's not bad. It's really slow and you probably can't hear it through the screen recording. It's, it's really quiet, which is great. We want it to be soft in the background. Um, I might bring it up a little bit actually, but it's much better than it was. And let's just check the outro.

Check it out and let us know what you think of it. Finally, just leave a comment.

I— okay, so that's kind of cool. This time the placeholder, it replaced it with his video. He zoomed out and he's this like little square above the subscribe to the channel, leave a comment below, et cetera, CTAs. And so that worked out perfectly how I hoped it would. So now I'd make all my edits to the V3 here based on what I found problematic from version 2. But what I'm going to show you is a way to have Underlord help you refine this prompt. So I'm going to copy all of V3. I'm going to come back to Descript. I'm going to start a new chat with the new chat button in the top right corner. I'm going to paste in the prompt. At the very beginning of the prompt, I'm going to add in a couple blank lines. And I'm gonna say, help me refine the following prompt. Edit this however you see necessary so that it is more clear for you when I provide it to you as my editor. I'm just gonna put a colon there after it just to make it clear that everything that follows is the prompt I want it to look at. And I'm gonna hit send and it says, here's a refined version of your prompt. Edit this video for me. Follow these instructions in order. And so put it into a nice list. That's probably the main thing. It probably made everything a little bit more clear. And so I'm just going to read over this and I'm going to copy this. There's a few small tweaks I'm going to make, but I'm going to do that over in my doc. So let's start a V4. I'm going to paste that in. And so there's my prompt. And I could call it, I could say with AI help so that I know this was the one. That Underlord helped me with, because it may have new problems now, but this is part of the process. For the Ken Burns part of the prompt there, I'm going to say, make sure that the subject stays in the center of the image. If it does not, do not apply Ken Burns. So I'm just going to give a little bit more clarifying there to make sure that Ken Burns will actually work. And just so that it's not confused here with this, the speaker attribution, I'm going to say, if the speaker quotes somebody else, Make sure the attribution is to the appropriate person. And then it's got instructions on the outro, hard cuts only, no transitions between scenes. Okay, so then I could copy all this one more time, come back to Descript, start a new chat. Remember, you have to start a new chat in order to save this as a template. Click start with a template prompt. New template. And remember that little bug I showed you? We got to click start with the prompt template one more time. We could paste in our version 4, hit continue, call this Jeremy V4, saving it to private, hit publish, close out of this. We're going to run this one more time. New composition. Project, bring in our Anger file, go to Underlord, start with prompt template created by me. I'm gonna scroll down. There's Jeremy V4, use template, and I'll be back in a couple minutes. Okay, so it finished with its latest set in our version 4, the, the prompt that was created by Underlord or with Underlord's help. I'm not going to waste your time going through it all again. I think you understand the process. It's just a matter of iterating and as a mistake or something you realize you forgot to provide enough context on emerges, then you keep refining your prompt and you just keep on going with this process until you have a result you're satisfied with. And as for me, if you were my client and I was editing this for you or I was creating a prompt for you, a system for you, then the next step would be I would pass this off to you. You would test this out on your actual videos and see how it performs in the real world because there's all sorts of things you gotta think about. He is recording this in a park with probably a lavalier mic. I didn't actually look at what kind of mic he has. Um, but it's gonna be a different situation when he's recording at home in his home office, and that's gonna require a different set of prompts. So you might have a version that works when you're out and about, a version that works when you're at your home office, a version that works when you're in your car or whatever. And so there's different prompts that you'll need. And the thing I want to point out to you is this took me, you know, I didn't spend a whole lot of time on this for the sake of this demo, but you might spend a couple hours making this prompt and you might burn through a lot of credits doing this. I'm going to show you my credit allocation in a second, but it's a couple of hours of work up front that will save you hundreds or thousands of hours later. And if you get your, your prompt really tight, then yeah, you, you burnt a lot of credits making this, but each time you run it, it might only cost you, well, let's take a look. Let's go to settings, usage. Um, so the last Underlord prompt, it looked like cost about 60 credits. That was my version 4. The one that was refined by AI, and then it generated 3 images throughout, it looks like. So 12 credits. And again, there's ways that you could make that even more efficient by doing stock media instead and not burning credits in that sense by not generating it. And so it used about 500 credits for this, this entire process. And that's not entirely what I've shown you because the tutorial I'm recording right now that you're watching, I'm recording into Descript, so that is burning some credits as I'm using studio sound. And, but let's call it roughly 500 to 1,000 credits is what you can expect to budget while you're developing your workflow. And then every time you run it thereafter, you're looking at burning 50 to 100 credits going forward. And so if you want to see the rest of this process, if you want to, I'm going to, I'm going to send this recording to Jeremy. Right now, and I'm going to get his input. I'm going to ask him to test this on his videos and see how it goes. And then if you want me to, comment in the description below, say part 2, comment part 2, and then I'm going to take Jeremy's actual feedback and I'm going to show you the next steps on how we can make this a perfect prompt so that he never has to edit again.