11:36Dan Kieft · Tutorial
I Tried AI Video Editing for 8 Days
An 11-minute field report on three AI tools that survived a week of real production work.
June 12thThe argument in one line.
A chat agent (GPT-6 Astra) can directly control Premiere Pro, After Effects, and DaVinci Resolve through the Higgsfield MCP, letting one editor generate camera angles, b-roll, motion graphics, VFX, and sound design without a production team.
Read if. Skip if.
READ IF YOU ARE…
- A solo creator or small-team editor who wants to cut post-production time without hiring more people.
- Someone already comfortable in Premiere Pro, DaVinci Resolve, or After Effects who wants an AI layer on top of tools they already know.
- A YouTuber who films with one camera and wants multi-angle footage, b-roll, or VFX without more gear or a bigger crew.
SKIP IF…
- You want a one-click AI editor with no setup. This requires paid subscriptions to ChatGPT, Higgsfield, and at least one NLE, plus writing long, detailed prompts.
- You're not willing to review and approve AI-driven edits step by step. Letting it run unsupervised risks wasted credits and ruined takes.
The full version, fast.
Dan Kieft breaks down six ways he uses GPT-6 Astra (ChatGPT) with the Higgsfield MCP to drive Premiere Pro, After Effects, and DaVinci Resolve directly: trimming raw footage from a transcript, generating extra camera angles from a single-camera setup, building b-roll from a character sheet, creating motion graphics through three workflows of increasing complexity, triggering VFX at specific frame numbers, color grading S-Log footage with a manufacturer LUT, and generating synced sound effects. The throughline is that AI compresses the manual labor of editing, not the editor's judgment: every workflow still needs a detailed prompt, reference images, and a human reviewing the output before it renders, and Kieft repeatedly warns against letting the agent run unsupervised because a bad generation burns credits fast.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 02:04
01 · Intro: the premise and the toolchain
A staged cold open (itself AI-edited, per the on-screen caption) sets up the premise that AI can now do real editing work. Kieft clarifies the goal is compressing an editor's labor, not replacing editorial judgment, then lists the baseline toolchain: ChatGPT desktop app (GPT-6 Astra), the Higgsfield MCP, and at least one NLE (DaVinci Resolve, Premiere Pro, or After Effects) with its matching MCP installed.
02:04 – 05:24
02 · Basic editing: trimming with a universal prompt
A single detailed 'universal prompt' instructs ChatGPT to transcribe the raw footage, trim dead space, add editor notes as adjustment layers, flag potential b-cam/c-cam insertion points, and deliver a finished Premiere Pro or DaVinci timeline. The agent visibly operates the NLE itself through MCP, which Kieft notes is slow (around 14 minutes for some trims) and still needs human monitoring.
05:24 – 09:44
03 · Multiple camera angles
A second (B-cam) and third (C-cam) angle are generated from a single real camera by feeding the model a reference photo of the room plus a written description of each camera's position and framing, synced to the existing audio. DaVinci's MCP runs this faster than Premiere's, but only on DaVinci's paid version.
09:44 – 12:57
04 · Generating B-roll
A character sheet built from a few reference photos of Kieft's own face becomes the input for AI-generated b-roll of him doing things he never filmed, like walking in a park. B-roll is targeted at specific sentences in the transcript and inserted at that exact moment.
12:57 – 17:34
05 · Motion graphics: three workflows
Three motion-graphics workflows of increasing complexity: a basic one where the agent drives After Effects directly from voice-dictated (Whisperflow) instructions and hands back an editable project file; a more advanced custom-character workflow where generations are assembled manually to avoid burning credits in the wrong direction; and a reusable pixel-art mascot bumper built from a single screenshot.
17:34 – 21:48
06 · Generating VFX
VFX trigger points are identified by exact video frame number rather than timecode. Each visual asset an effect needs (a creature, a prop, a vehicle in this demo) is generated separately, often through several iterations to get photorealism, then all references plus the frame numbers are fed to the model together to write one frame-accurate generation prompt.
21:48 – 23:32
07 · Color grading
Flat S-Log3 footage is color graded by giving the model the camera manufacturer's own Rec.709 conversion LUT as a technical baseline, then asking for a creative grade on top. Deliverables (flat MP4 and an editable Premiere timeline) have to be specified explicitly. DaVinci's MCP is offered as the faster, more customizable alternative since DaVinci is purpose-built for color work.
23:32 – 26:39
08 · Sound design
The model is asked to write out a detailed, frame-by-frame sound-effects generation prompt first, which gets reviewed and edited before anything generates, to avoid wasting credits on a bad direction. Because generating new AI audio visibly degrades the accompanying video, the fix is extracting only the new audio track and re-syncing it onto the original full-quality footage.
Lines worth screenshotting.
- ChatGPT's desktop app removes the 512MB upload cap of the browser version, which matters because most camera footage is bigger than that.
- A single-camera interview can get a second and third camera angle for free by feeding an AI video model a reference photo of the room plus a written camera-position prompt.
- A 3-photo character sheet of your own face is enough reference material to generate AI b-roll of yourself doing things you never actually filmed, like walking in a park.
- Identifying VFX trigger points by exact frame number, not timestamp, is what lets an AI model land an effect in sync with a specific gesture.
- Generating new AI sound effects on a clip visibly degrades its video quality, so the fix is extracting only the new audio track and re-syncing it onto the original footage.
- Approving every AI generation one step at a time instead of granting full automation avoids burning credits on results you don't want.
- Color grading S-Log3 footage with AI starts by handing it the camera manufacturer's own Rec.709 conversion LUT, then asking for a creative grade on top of that technical baseline.
- DaVinci Resolve's MCP integration only works on the paid version, which gates AI-driven color grading in DaVinci behind a license purchase.
- A single recorded catchphrase can be turned into a reusable animated mascot asset that gets dropped into every future video.
- A detailed, reusable 'universal prompt' that defines trim rules, b-cam/c-cam marking, and delivery format can be run against every new raw video a creator uploads.
AI Can Run Your Editing Software For You
AI EDITING WORKFLOW
Six AI-driven editing workflows, from trimming raw footage to generating camera angles, b-roll, motion graphics, VFX, color grades, and sound design, all run through a chat agent that directly operates your existing editing software.
01Intro: the premise and the toolchain
- AI video editing tools covered here require paid subscriptions already: ChatGPT (GPT-6 Astra), the Higgsfield MCP, and at least one NLE such as Premiere Pro, DaVinci Resolve, or After Effects.
- Download the ChatGPT desktop app instead of using it in a browser; the browser version caps file uploads at 512MB, which most camera footage exceeds.
- The goal framed by the creator is compressing an editor's manual labor, not replacing editorial judgment; he still works with a human editor throughout the video.
02Basic editing: trimming with a universal prompt
- A single detailed 'universal prompt' can instruct ChatGPT to transcribe raw footage, trim dead space, insert editor notes as adjustment layers, flag potential b-cam/c-cam insert points, and deliver a finished Premiere Pro or DaVinci timeline.
- The agent actually operates the NLE through MCP, visibly clicking through Premiere Pro on its own, which the creator describes as slow (around 14 minutes for some trims) and worth running only while doing something else.
- You still need to manually review flagged sections, since the model marks ambiguous notes-to-self and thinking-out-loud moments as potential cuts rather than trimming them outright.
03Multiple camera angles
- A second (B-cam) and third (C-cam) camera angle can be generated from one real camera by feeding the model a reference photo of the room and a written description of each camera's position, height, and framing.
- Generated angles are synced to the original audio track rather than recorded separately, so dialogue timing matches automatically.
- DaVinci Resolve's MCP integration runs these camera-angle generations noticeably faster than Premiere Pro's, but only works on DaVinci's paid version.
04Generating B-roll
- Three to four reference photos of your own face are enough material to generate a usable character sheet for AI-driven b-roll of yourself in scenes you never filmed.
- B-roll generation targets specific sentences in the transcript (e.g. the word 'walk' or 'shower') and inserts the matching generated clip at that exact moment in the timeline.
- Cost-efficient video models can be used for b-roll specifically to save generation credits, since the shots are usually short and simple.
05Motion graphics: three workflows
- A basic motion graphic workflow feeds the model reference images plus style-inspiration screenshots, then lets it drive After Effects directly through voice-dictated instructions, handing back an editable AE project file.
- A more advanced workflow generates a consistent custom-character image first, then manually assembles the AI-written prompts rather than letting the agent auto-generate, specifically to avoid burning credits on an unwanted direction.
- A single animated 'mascot' bumper, like a pixel-art version of yourself saying a catchphrase, can be generated once and reused across every future video.
06Generating VFX
- VFX trigger points are identified by exact video frame number, not timecode, which is what lets the model land an effect precisely on a gesture like a hand movement or a glance.
- Each visual asset an effect needs is generated as its own separate reference image first, often iterating multiple times to get a photorealistic result.
- All reference assets, the raw clip, and the specific frame numbers are fed to the model together so it can write one precise, frame-accurate generation prompt.
07Color grading
- Color grading flat S-Log3 footage starts by handing the model the camera manufacturer's own Rec.709 conversion LUT as the technical baseline, then asking for a creative grade on top of it.
- Specify the deliverable format explicitly (a flat MP4 plus a re-editable Premiere Pro timeline), or the model may only apply changes inside your existing Premiere project.
- DaVinci Resolve's MCP handles color grading faster and with more creative range than doing it through ChatGPT and Premiere, since DaVinci is purpose-built for color work.
08Sound design
- Have the model write out the full sound-effect generation prompt first and review it before generating, instead of letting it generate blind, since a bad direction wastes credits.
- Generating new AI sound effects on a video visibly degrades that video's image quality, so the fix is extracting only the new audio track and re-syncing it onto the original full-quality footage.
- Always verify the generated audio against the source footage's actual duration and pacing before accepting it, since the model can return a soundtrack that doesn't match the clip length.
Terms worth knowing.
- MCP (Model Context Protocol)
- A plugin standard that lets an AI chat model like ChatGPT directly operate another piece of software, such as Premiere Pro or DaVinci Resolve, instead of just producing a file for a human to import.
- B-cam / C-cam
- Secondary and tertiary camera angles of the same shot, generated by AI from a single real camera instead of recording with additional physical cameras.
- S-Log3
- A flat, low-contrast recording profile used by Sony cameras that preserves more dynamic range for color grading, but looks washed out until a conversion LUT is applied.
- LUT (Look-Up Table)
- A preset color transformation file, often supplied by a camera manufacturer, that converts flat log footage into standard, normal-looking color and contrast.
- Character sheet
- A reference image set showing a person's face and body from multiple angles, used to keep an AI-generated character visually consistent across different generated clips.
- Whisperflow
- A voice-dictation tool that lets a creator speak editing instructions out loud instead of typing them into a chat prompt.
Things they pointed at.
03:20toolDaVinci Resolve MCP
07:10toolBlender MCP
14:10toolWhisperflow
Lines you could clip.
00:10
“So I did some research and I found six methods of how you can start using AI to edit your videos. In fact, everything you've just seen was all made using AI.”
15:50
“If you have the approve for me or full access available, it might cost you a lot of credits if you go into the wrong direction. I always want to approve because you will burn true tokens so quickly using this method.”
25:55
“Just don't blindly trust ChatGPT for everything, because that's how you're gonna burn a lot of money. Like, trust me, I've been there.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
metaphorstory
A few days ago, I was doing the final check on one of my videos. Then I noticed that my editor kept putting in these kind of disturbing images. So I texted him.
Like, my hairline isn't that bad, but he just sent me more images mugging my appearance. Which made me wonder, do I even need this guy now that we got GPT -Astra? So I did some research and I found six methods of how you can start using AI to edit your videos.
In fact, everything you've just seen was all made using AI. Okay, before we dive into this, I just want to get one thing clear. The goal is not to replace my editor.
This is just a joke intro. The idea is if editing takes less time, a human can spend more time on creativity and edit their videos even better. And that's why in this video, I'll be breaking down all of the different techniques that go into AI video editing.
I will show you how you can easily trim and cut down your video. Then I will show you how you can make multiple different camera angles just from your A -roll. We will go over cool looking motion graphics.
that are now easier than ever to create and i will even cover a bit of vfx now last but not least ai can do a lot when it comes to sound design so make sure you watch this entire video to the end and let's get started There are five things that you're going to need if you want to follow along with this AI video editing tutorial.
The first one is the obvious one, which is ChatGPT. We're using Astra as the brain for all of our projects. The second one is Hicksfield MCP.
This is used for all of our generations. They were also the sponsor of this video, but even if they didn't, this is the best MCP out there. And they also have an API now, which we are using for all of the generations that are about to come that are being done through GPT Astra together.
The third thing that you're going to need if you're editing through DaVinci Resolve, you're going to need the DaVinci MCP. Don't worry if you're editing through Premiere Pro, then you just need Premiere Pro. Same with After Effects.
If you want to do things through After Effects, you just need to have that app on your computer. If you're missing one of the three editing softwares, as long as you have one of them, this can still work. Okay, let's start with basic editing.
We're going to be trimming dead spaces. For this, you want to download the ChatGPT desktop app. or the Codex app.
The reason why is because the cloud version, if you're using this cloud, you have a limit of the file size that you can attach to it. If you're using this on your PC, you can click to on your computer and now you can upload really big files that are already on your computer and it can use it directly from that and work directly from that.
In the cloud, as you can see, you're limited to 512 megabits. And I don't know about you guys, but my camera files are way bigger than that. So make sure you switch to on your computer.
Then you drop in the video that you want to edit. And here you can select GPT Astra. I'm using medium most of the times because it's already expensive enough.
You could go high, but only if you're rich. Then we need to start prompting it. And I have a very detailed prompt that anyone can use.
It's a universal prompt. I will explain to you right now. So you upload your video and then you're going to add in this exact prompt.
And this is a super long prompt, so I'm going to break it down for you. Basically what this prompt does is it first instructs ChatGPT to trim the raw footage. Then it will ask it to transcribe that footage and trim out the dead spaces.
After that, it will add in editor instructions as adjustment layers. Then it will figure out potential spaces in your video where you can have or add in B cam or even C cam. And it uses that with the markers.
And lastly, it will give you the finished version in Premiere Pro or DaVinci Resolve. I'm using Premiere Pro right here because we are using that for most of the editing that we do on this channel. And this is what my team is also using.
But this also works with DaVinci. I will show you an example later. You might get some things that you need to allow.
So I'm just going to do always allow for now. This might cost you more credits. Keep in mind, this process is still quite slow.
You can literally see it and clicking on things for you. Still, some of these things annoyed me as it took like 14 minutes to do some of... these trims so i would only do this while you're working on something else and while you don't want to do the trimming yourself because you still gotta sometimes monitor if it's doing everything correctly so it's literally taking over your pc and it's doing everything for you as you can see we now have a new version where it's marking potential b -roll it is adding some adjustment layers it is doing everything for me like it's now even rendering now as you can see chat chipt has ran all of these commands using my premiere pro and it I actually stopped this because I already had this version to begin with.
And that's the one I want to show you because here it's basically has done the same thing. We're just one step ahead, which I'm going to explain to you in a second, but I always start my videos with a clap to sync the audio. Okay, this is the intro.
A few days ago, I was doing the... And using these adjustment layers, it says intro starts here. Then other notes that I have, like for example here, I say something about I want to add in some VFX here.
And then like it mentions it. I'm trying to do this. I'm not sure if this works.
I'm trying to think like what if we can... Like this is not the part that I want to have in the video. These are just instructions for me and my team to go over later.
So we can manually trim these out or you can even ask JetGPT to do that for you. The other thing that it has done is it has marked the... potential B cam positions and the potential or optional C cam positions.
And here again, optional C cam. Let me explain to you what I mean by B cam and C cam. So this, what you're watching right now is the A cam.
Then this is a B cam. It's a secondary side angle. I don't have a camera there.
Like there's no camera there. I wish there was a camera there. If you have like three different cameras, like that's expensive.
It's a lot of storage. Storage these days is expensive, but we can use AI. So this clip is entirely generated with AI.
And I literally asked Astra to do that for me. Now, the other thing that we can do is we can also create CCAM. That's another angle.
You can literally create as many different angles as you want. you add in multiple camera angles well the first thing you need is you need to have a reference that you can feed chat gpt to what your surrounding looks like so i gave it this image it's with all the studio lights turned off it's a bit of a behind the scenes it's it's not the most clean looking setup like what the hell is happening here i'm not gonna I'm renting this place, okay?
I cannot do too much to the place. But I then use the same Chachapiti text. But this time we're going to use this entire prompt.
What it does is we're going to ask it to add in the b -roll. on specific segments. So I'm just going to copy this prompt.
We're going to go back to ChatGPT and now we're going to enter it in there. And now you can just ask it and it will start up Premiere Pro again and it will start generating all of these things. If you have the Higgs field MCP connected, that's the main thing you need to do.
So you're basically going to start up this new chat by pasting in that prompt. Then you're going to add in that extra reference of your room, of the space, of the side angle. And then you're going to say you the Higgs field MCP.
Make sure you have the Higgs field MCP connected. Basically what you can ask ChatShipTip to do, I will show you right now. Install Higgs field MCP.
Install DaVinci MCP. While you're at it, you might as well install Blender MCP. And literally by just sending this, it will install all of these things for you if you have a subscription.
If you don't have a subscription, get a subscription first. If you want to support my channel, use my link. Basically this is the task that you need to do.
Okay, once you have done that, you can hit send. And now it will get to work using that prompt. Again, this process might take some time as it's still quite slow and we don't have an actual Premiere Pro plugin yet.
With DaVinci, this might be a lot quicker, but now it's also like generating these videos using Seedance 2 .5. And again, like that's a lengthy process regardless. Now you can see it working here right now.
So you can see it use these Higgs field MCPs. And if you then go over to Higgs field itself and you go to the video tab, you can literally see it. prompt this and it is using cdance 2 .5 it is asking it to replace video one using this as a reference pretty sick right i'm just gonna tell you like it might stop occasionally because right here it's for example waiting on the video to confirm but this is the b -roll that it made for that video then i noticed that my editor kept putting in these kind of disturbing images keep in mind this is entirely like ai generated i didn't record this myself it's just using it based on this reference and it made up itself what i look like from the side and looking at everything it's done quite an amazing job like we can even see like you don't see it but i see there a shadow on my wall like this shadow then if we look at the ccam even this one it's time a human can spend more time on creativity like it's really really good like this thing there that light
actually it's positioned like that the light behind me is positioned like that just this mouse mat thing is not positioned like i don't have that but overall this is quite good the only thing is this is if you want to have ccam i don't think this this angle is flattering of me so i just want to use the bcam and the acam you can also do this in davinci because it is much much quicker as someone has created a mcp for davinci so basically by just prompting it use the davinci mcp within a heartbeat you would have a trimmed version of your timeline so if you are davinci editor or if you are trying to get into editing yourself you might actually want to try out davinci keep in mind you need to use the paid version for this as the free version doesn't have the mcp to my knowledge but this is to show you how you can generate multiple different angles of your main a roll all right so you now understand how you can trim up your video and how you can create multiple different angles of the person that's talking on the screen but the next thing that i'm going to show you is b -rolls b -rolls are a great
way to spice up your videos the problem however is that generic stock footage b -roll looks boring and might not fit your vibe the solution is generating those b -rolls yourself using ai you can completely customize them based on the prompt you put in anything is possible really the other thing is is that you can get exactly what you want in like 10 minutes you can save credits by using cost efficient models like cling and they still do the job quite well Cool.
So how do you generate the B -rolls like I've just shown you? The first thing you're gonna need is you need to have multiple different references of the character that you want to have B -roll from. In this case, it's me.
So what I did is I used my iPhone and then I made these three images to get a full picture of my face. But if you have something a bit more complex, then you might want to use a character sheet like this one or this one. this one is like 100 accurate like i literally scan my body and this is actually how muscled i am i'm just trying to make a joke here because i'm not that muscled but as you can see this is super simple i use my face as a reference then i just add in a prompt explaining what i want to have on the screen if you want to see all of the prompts that i'm using in this video then make sure you're joining the free school community here for every video that i make i'm uploading all of the assets all of the prompts sharing them with you so you can go through at your own pace.
To create your B -roll, I will just give you a quick rundown. What I did is I uploaded the three images of me. Then I used this skill MD file that explains how to make a character sheet.
And then I just gave it a simple prompt like this. Using the skill and my three reference images, can you help me generate a character sheet in Higgs field? You also might add in Higgs field like this and then explain what you want it to look like.
So now you hit generate. After it's done generating, it will give you the carrot sheet that would look something like this. You now are ready to generate your video.
But first you need to have a video that needs B -roll. So I recorded this segment for that. So the sentence is, if you're stuck for ideas on what to generate next, go for a walk or take a shower.
Sometimes the best ideas come when you stop trying so hard. If you're stuck for ideas and you don't know what to generate next, the best thing that you could do is you can either go for a walk or take a shower. Because sometimes the best ideas come when you stop trying so hard.
so at the part of go for walk i want to have b -roll and at the part of taking a shower i want to have b -roll just for visual context so at the part where i say you go for a walk i want to generate a b -roll for that so i'm basically asking it to generate me with this character sheet, a Seedance 2 .5 video of me walking in the park.
And then I describe what my shot looks like. I do the same for the shower part. Then literally what it does is it uses Higgs field again to generate the video and it will give you the video right here.
And then if you combine that, you get something like this. If you're stuck for ideas and you don't know what to generate next, the best thing that you could do is you can either go for a walk or take a shower because sometimes the best ideas come when you stop trying so hard. Okay, let's go a little bit more advanced.
Let's talk about motion graphics. Here's an example of the motion graphics that AI made for me. Video editors need to start using GPT -Astron ASAP.
Not only does it create small motion graphic elements like this. It can also create full range animations that used to take hours in XFX. All of that on autopilot.
If you want to learn how to use GPT -Astra, then check out the full tutorial that is in my free school community linked down below. Here's the entire timeline that AI helped me made for all of these. And I'm naming this a bit more advanced because the misconception is AI is this magical thing that will do everything for you and it will look amazing.
You won't get this result. by just prompting it you need to combine a bit of your own skill together with ai and that's why i'm saying like editors are not replaced like me and my editor actually collab on this so we're trying to figure out what can be done and what cannot be done with ai and we have used a lot of ai it was also a bit of a painful process sometimes to figure out how to do these different style of motion graphics i'm going to show you three different workflows the first one is this basic animation workflow what i did is i uploaded references of what my community looks like, because I would love to have some cool looking motion graphics of that community.
Here's the banner of that community. Still need to update this. Then I also added in some style inspiration.
So I found these different type of motion graphic inspiration. And this is kind of like what I screenshot it. And then I asked it to help me.
Then I basically used Whisperflow to give me these motion graphics. And it just went to work. Like it just used After Effects and it made all of these things for me.
And after going a bit of back and forth, it finished the After Effects motion graphic for me. It even gave me the complete project file so I can go in there and I can edit it myself. And I have the finished video.
So again, I am not an After Effects user myself, but my editor is. now give this to my editor to tweak little things the next one is more of an advanced workflow and this is quite hard to explain because there are so many different prompts that went into this but i'll try my best to give you a summary first we're talking about this kind of motion graphics and this is cool because we're trying to create a new style that would add to my videos where we have custom animated images of me that we can use in different situations the way i made this is i added in a bunch of different references of styles that i liked and then again i just started talking using Whisperflow about the kind of animations and things that I want to see.
So then it started to create these references of me first, and then eventually it made these images of GPT Astro 6 and me again. And then we got this generation. So it made like all of these different images and it kept me updated.
And then it gave me the prompts of what it wanted to use. And then I manually use these prompts inside of Higgs field because I didn't trust Astro there and didn't want to waste too many credits. Keep in mind, if you're doing this and you have.
the approve for me or full access available it might cost you a lot of credits if you go into the wrong direction if you have money you're good if you don't have money please don't do it like i'm not doing this either like this is risky like i always want to approve because you will using this method so then i got like all of these different prompts i manually created them inside of hicksfield and then i tweaked like all of the different settings so this is me going back and forth before i did the final generation and then eventually like this is the whole chat like you can see how long it is eventually like it made like all of these different things for me and then i imported them into my timeline and i added like a small little things like this white transition right here but all of this is done hand in hand with astra okay so there's one motion graphic which is this pixel version of me and i'm gonna share with you how i made this because i think this is awesome for adding in this into your video so i just uploaded an image of me
And I just told them like, this is a screenshot of me and I asked it to create a pixel version of me. So it made like these three different versions using the Higgs field MCP. Then I basically asked it if it can now give me a Seedlens 2 .5 prompt.
And I wanted to do this part manually again. So now it gave me the prompt where I basically asked it what I wanted to have happen. So in this case, I wanted it to walk into the frame and say, don't forget to subscribe.
So now if you upload the part where you want to have that in your video together with your asset. and then you ask it to replace it. And we get this cool looking result where I walk into the frame and say, don't forget to subscribe.
And I jump out of it. Not only does it create small motion graphic elements like this. So now I can use this asset over and over and over again in all of my videos.
One of my favorite things to do with AI is making VFX. So I'm now going to show you how you can do an effect like this. So you want to learn how to do VFX.
Well, let me actually... let me show you wait what the hell is that thing behind me okay and for my next trick let me actually move back a little bit so now we're gonna do this oopsie So there are a couple steps that go into making a video like that.
First of all, it starts with your recording. Now, this was my raw recording and I already envisioned what I wanted to do. So I wanted to have something like there.
Then I wanted to use some kind of like magic wand. In my case, it's actually a pen that I hold in my hand. And then something needed to happen at the end.
Let me actually move back a little bit. So now we're going to do this.
Okay, I hope that works Yeah, so I hope that works. So I've identified three points.
First point being is when I look over my shoulder. So I'm gonna leave a mark right here. An important thing I want you to do right now is if you're using Premiere or if you're using DaVinci and your timeline looks like this, where you can just see like normal time of how many seconds it's into the video, change this to the view.
I do that with command and then clicking on this. Probably would be control of your own windows. And then you can see the exact frames.
I'm going to leave a mark and remember this frame for when you're prompting. Then the next thing is here when the pen arrives in my hand. So I'm just going to leave a mark right here at 279.
So I'm just going to add in a mark and then I'm going to do the same when I'm clapping into my hands. So right here, I want something to happen with like a car going through the wall. That is the first things that you need to do.
So now you've identified these three points of where VFX need to happen. Then we're going to walk through them step by step. So the first thing we need.
is a creature my first version of the creature was this i made the mistake here where this didn't really look photorealistic until after a few generations i finally landed on something like this this looks a lot more creepy bit more realistic and then this ended up being the creature that i wanted to use the second thing that we need to generate is the magic wand so here i made a magic wand like this again the prompt is simple i'm just describing like what i want to make what the background looks like the lighting the style and i'm using cdream 5 .0 pro for this so now we already have two of our things then the last thing is the car and here i have this guy i call him bruiser and i wanted him to be in a tesla bruiser is gonna run into me and this is the reference that we're gonna use for it so we now have created the three of our references then what you can do is you can take all three of these references put them inside of chatgpt or cloud Then you add in your video and then you're going to precisely describe what you want to have happen per frame.
So as we remembered, I have here frame 107, then I have frame 279 and then I have frame 567. So that is what we're going to prompt. And if we do that correctly, then it will give you a very detailed prompt of everything that's happening precisely on the timestamp.
Like it would come up with that timestamp and it will tell you exactly where. that thing needs to go and will reference these images. Like this deserves like a separate video on its own, but I just want to show you like what's possible.
After doing that, you can look at a side by side and then you have results like this. So you want to learn how to do VFX. Well, let me actually, let me show you.
Wait, what the hell is that thing behind me? Okay, and for my next trick, let me actually move back a little bit. So now we're gonna do this Oopsie So that is exactly how you can generate these VFX if I went too fast Please let me know I can make a separate video about this if you want to Moving on, I will show you how you can use Astra 6 to color grade your footage.
You can do this both through ChatGPT itself and also through DaVinci. Let's first start by doing this in Astra 6. Basically, what you can do is I have this footage.
This was shot while I was in Malaysia. This is S -Log footage. I'm using the official Sony conversion log that I found online for my camera.
And then basically I'm giving it a simple prompt where I'm saying that the uploaded MP4 file is my S -Log3 footage. And the other file is the official Sony recording 709 conversion LUT.
I want to use this to color grade my MP4 by using the conversion LUT. And please, for the deliverables, can you put this in a Premiere Pro timeline so I can manually edit it further if I need. That's all I do.
So now you hit send and then it will do everything for you. So it has now given me this folder. And if we look at these side by side, you can see that we now have a color grader.
video this is the super simple way again sometimes this might take a while in this case you don't want it i specifically told it don't use my premiere pro directly because that's the slow version like for simple tasks like these you might not have to use ai if you have more advanced ones then i would definitely use ai but keep in mind like it can still be quite slow the other thing you can do is you can use the davinci mcp if you have the premium version of davinci then you can ask it to color grade directly DaVinci, it is much quicker and it will also give you much more customization options because DaVinci is best for color grading.
You can even ask it to be creative when it comes to color grading your work. The last segment, which is also a very important segment, is sound design. Sound design is what sells your videos.
If you're doing sound design manually, you can now use AI to do sound design for you. There are various different ways. And again, this also might deserve a video on its own because you can literally we have Astra 6, control your sound design software, which is crazy on itself, but I'm gonna show you a simple version.
Remember when we had this video.
I personally don't really like the sound that Seedance made for this one. So what we can do is we can upload this into Astra and then we can just give it a super simple prompt. I want you to craft a Seedance 2 .5 prompt to generate an audible synchronized sound effects track for this eight second UI animation.
The reference is the video, so you need to analyze it frame by frame and understand what sound effects. Keep in mind, you can do this for any type of video. In this case, I used this animation, but you can also literally do this for a intro video.
That's how we do it. sometimes then you're just gonna ask it to make a prompt in this case i'm gonna make the prompt because i want to save some credits before it does this in hicks field and i get a version that i'm not happy with because i always want to check the prompts that it's going to use so as you can see it wrote a super detailed prompt of the audio and now you want to go through this and see if you agree with this if you disagree then go back reiterate to change a few things always do that with your prompts just don't blindly trust chat gpt for everything because that's how you're gonna burn a lot of money.
Like, trust me, I've been there. If you have any changes, change them. Otherwise, copy this.
Go into Higgs field. Upload your original track in CNS 2 .5. Then paste in the prompt right here.
Make sure you add in video one to refer to that video. Then match the length of the video. Choose your settings.
You could do this in low settings. You don't need to do this in high settings because I'm gonna show you why. Here I have the finished result of that prompt.
like it sounds a bit better but the video quality has noticeably gotten worse so you just want to download that video then you want to separate those audio tracks and then you want to replace that with the original video literally you could even drop in that video in there and you can just use like whisper flow and you can say something like i want you to replace this audio of the file that i just uploaded and replace it with the original file that i gave you And make it match.
Then Astra will do the work for you. And you will have your original high quality clip. With the new audio that you just generated.
Alright. That is the complete guide of how you can edit your videos using AI. And particularly we're using Astra 6 as the brains of this.
If you want to try this out yourself. I would recommend using Higgs for that. I will leave the link in the description down below.
And if you have any questions on this video. You want to see more things. in depth.
Lastly, make sure to click the video on the screen right now if you want to learn what else you can do with Astra 6 and see that.
The Hook
The bait, then the rug-pull.
Dan Kieft opens with a joke about his editor sabotaging his face, then reveals the real premise: everything in the cold open you just watched, the footage, the camera angles, the graphics, was generated with AI, and the rest of the video is the six-method breakdown of how.
Named ideas worth stealing.
02:30list
Five things you need for AI video editing
- ChatGPT desktop app (GPT-6 Astra)
- Higgsfield MCP + API
- An installed NLE with its matching MCP: DaVinci Resolve + DaVinci MCP, or Premiere Pro, or After Effects
The baseline toolchain: a chat agent for reasoning and prompt-writing, the Higgsfield MCP for the actual generations, and at least one NLE the agent can operate directly.
Steal forany creator deciding on a minimum toolchain before starting an AI-assisted edit
03:50concept
The universal trim prompt
One long, reusable prompt that instructs the agent to trim raw footage based on a transcript, add editor notes as adjustment layers, flag b-cam/c-cam insertion points, and deliver a finished NLE timeline.
Steal fora repeatable first-pass edit on every new batch of raw footage
How they asked for the click.
VERBAL ASK
26:30link
“Join my FREE community where you can find my prompts & chat about AI”
A direct verbal ask near the very end, reinforced by a title card and a description link, pointing to a free Skool community rather than a paid product.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
OTHER LINKSAlso linked in the description.
Visual structure at a glance.
