
I edit videos for a living, or at least I used to, but now clients tell me they’ll just use AI instead. And I get it, AI can generate a single clip now. It’s gotten so good it’s genuinely competing with some of the best VFX artists in the world, but generating one clip is not my job. My job is putting a hundred of them together into something that actually tells a story. So here’s the real question: can AI do that better than me, an editor with 6 years of experience? We are going to find out, and at the end of this video, I’ll show you both the versions side by side. Everyone thinks editing is just cutting the boring part and then merging clips together, but that’s just 2% of the job. The main job is taking clips that were shot at different times of day, from different angles, with completely different lighting, and combining all of them together so it feels like a continuous film. So when you watch a really well-edited video, there are three things doing the job. First, continuity: the face, the lighting, and the world have to hold from one shot to the next, which has always been a problem with AI videos. Second, pacing: how long a shot lives before the next one takes over.
There’s really no rule for it, so my question is how will AI even figure that out? And third, sound design: sound is 50% of the experience—music, silence, sound effects landing on the exact frame, plus the J and L cuts which make the video flow. All three together make a well-edited video, and this is difficult to do even for intermediate editors, so I really want to see how AI will do it. Now for the AI side of things, I’ll be using OpenArt Director. They say it can make videos up to 5 minutes long with visuals, voice, music, and sound all from natural language direction, which means I can just type a simple text and it will give me a complete video up to 5 minutes long. So I was going through some of the examples from Director, and to be honest, I’m very surprised. To show you some of these, it can generate a complete short film, UGC ads, and even product commercials. What you’ll see is it creates a complete timeline, and these are very well-edited videos. The cuts are landing, the whole thing is actually flowing, plus there’s a complete editable timeline so you can change anything as you like. Here is another example: this cut here looks really smooth, and there is complete sound design along with these videos as well that made me a little nervous about this challenge before I even started.
Now, to keep this challenge fair, I’ll use the exact same clips that OpenArt generates, and we’ll see who is able to create the better output. Let’s start with creating the clips first. Since I have full creative control here, I want to make something that I won’t be able to shoot on my own, so I was thinking of a BMW car transforming into a Transformer. That’s what I’m thinking; sometimes my genius is scorching today. So I opened the Director and just typed it out the way I said it out loud, and then I answered some very simple questions like what will be the aspect ratio, resolution, and length. I’m going with 16:9, 1080p, and 30 seconds. Then it wrote the script using Claude Sonnet 5 and showed it to me, and then it built a production plan, which means it broke the scenes into individual shots, defined the camera work, and logged what it called anchors. Anchors here are the car design, the transformation states, and the environment, and these are being generated using Nano Banana Pro. Then it went and generated the actual video with Seedance 2.0, added a music track and sound effects, and dropped everything into a timeline. Now this is the timeline. The first thing is that this is completely editable, so there’s a separate video track and an audio track, and I can also move things around in the video, shorten it, or maybe even replace the video. It’s not a finished video; it’s handing it to me like a project, just like any Premiere Pro project. Second, the cuts: watch how it moves between the driving shots and the transformation shot here, so the entire video is building towards the final reveal, which is pretty good. And third, the total time from me typing that first sentence to having this on screen is about 5 to 10 minutes, and then if we click on export right here, you’ll see that it exports in like 3 to 4 minutes.
Now to be fair, I didn’t put any effort into this; I was just pressing enter after the initial prompt. So this is what I’m going to do: all the clips that it has generated, I’ll download all of them and import them into Premiere Pro, and at the same time, I’ll run these through three or four times and see what I can improve in each version, and then I’ll show the final output in the end. Now before starting the edit myself, because this really impressed me, I wanted to test it on something harder, like an actual client project that I got on Fiverr. I had this brief from a client project about making a pizza oven video, so I wanted to see how this tool, OpenArt Director, will create that video because there’s a real product and it has to look exactly like that. Second, I know it can just compile things together, so I want to make it a little more difficult, and what I’ll do is ask it to mix it with AI-generated clips. So it will be like real footage, then AI clip, then real footage, and then the logo. And third, the cuts have to land on the music beat, which is the part in the intro I said no AI would figure out. So I ran the same process again, and honestly, it’s not even prompting anymore—it feels more like vibe directing because I’m just using simple English. I uploaded four of the videos that the client sent along with a reference image that was also sent by the client and the final logo, and asked it to create a 15-second montage.
Here’s the final video—pretty decent. I would like to change some things, so let’s see how much control I can have on the timeline. This third shot I would like to use a different section of it, so what I’ll do is just extend it, like when the ingredients are falling on it. I really like the physics of it when they interact with the base, so let’s see if I can trim it. Yeah, I am able to trim it, and then it ripple deletes; if you know Premiere, it moves everything on the timeline, which is cool. Then I think there’s a small flow problem that I would fix: I would move this pizza clip to the other side, which is also surprisingly easy to do. I’m not very sure about this shot, so I think I would replace it with something simpler, and then it left the original audio from this video, so I’ll just mute it and sync it again to the beat. I think I’ll add one more clip in the middle here. I mean, for 15 minutes of work, I think I’m very happy with it. Personally, the thing that I actually like is the amount of control I have here. Even if, let’s say, AI gets something wrong, it’s very simple to edit; anyone with basic knowledge of any editing software can make changes, and of course, a lot of this can be avoided in the planning section when it provides you all the shot lists and things like that. Also, I really like how realistic the product looks; it’s very close to the initial video that we sent. But yeah, as far as I can remember, when we created something like this for the client, we took the entire day and we just merged clips to the beat, and it was a very simple project, and the same thing I was able to do here in 15 minutes. I think the final challenge will be very interesting. Okay, so now it’s my turn. I’m downloading the same clips from Director and I’ll bring everything inside Premiere Pro. The first thing I do on any project is watch everything once and pull selects; that’s called assembly. While in Director, I ran the whole process again. The car and the Transformer scene were already logged, so I created a rival for it and added simple text to create an explosion scene.
Then I dropped in a shot list of how I wanted it to cut, whereas in Premiere, I started building the rough cut based on how the shots actually flow. Now the biggest problem with these clips is the continuity, so the shots keep jumping back to the original scene, and I could regenerate them of course, but I want to see how the AI will handle it. So I’m following a very simple editing rule in my timeline, which is never cut between two similar angles, and just in case, when I have to, I add a zoom-in so it doesn’t feel very abrupt. I’m cutting between medium close-up, slow-motion shots, wide, and using magic cut to make it flow, whereas in Director, I built the entire timeline in one go. Once it was done, I’m still adjusting minor things, replacing some shots just so both the videos flow well. Of course, I’m not adding any effects in both the videos; we’ll just judge both the videos based on continuity, pacing, and sound. Now for sound design, these generations already came in with really good sound effects. All I’m doing in Premiere is creating a J and L cut so the audio feels immersive instead of hard cutting with the picture. Then I added two music tracks, timed it to the buildup in the scene, and the edit was done. In Director, sound design was already completely done since starting, so it was good to go. Now Director’s version took 20 minutes start to finish; my entire editing process took 2 hours. So here are both the final versions. I also sent them to my team to see if they could tell which one is human-edited.
Here is the first version, and here is the second version. Comment one or two: which one do you think is edited by AI and which one do you think is edited by me, and more importantly, which one do you think is better? I want to see how many people get it right because I think it will be closer than you think. If you want to try this yourself, the link is in the description. Here’s what my team said: the first editor thinks that the 40-second version is AI; he preferred the 30-second cut because it had a more dramat
ic feel and better continuity. His reasoning was that the 40-second version felt like it was trying to use every clip, while the 30-second version felt intentional. The second one could not tell at all—these are his exact words—but just based on continuity, he selected version one to be AI. And the third one went the other way completely; he liked the first one because the music made more of an impact, so he assumed the second one was AI-edited. And these are all full-time editors, so if it was close for them, it’s going to be a lot harder for everyone else. Before I tell you the final result, make sure you subscribe. The first one was AI and the second one was edited by me. Let me know what you think about this video and we’ll do more challenges like this. Bye-bye.