
Ever since I started posting AI content, I have been getting a lot of queries about AI video production. One of them was from my school senior and friend who works in the movie industry. He wanted me to make a save-the-date for his wedding, and he wanted something epic and Marvel-inspired because he is a huge Iron Man fan. Now, I have made a few AI videos here and there, but nothing like this. Along with that, Vinayak was also a little unsure about pitching this idea to me. He told me he has never been impressed with AI content—like, he has seen it, but nothing has worked for him. So I decided to challenge myself and take on this project, and let me tell you something: it was not easy
Here is what I want to achieve by the end of this video. First, learn complete AI video production and discover the latest tools. Second, create something with identical faces and consistent scenes, because it’s a save-the-date for a real wedding and has to present the client in the best light. Third, and the most difficult part, impress someone who is already working in the movie industry with my output.
To get started, the first thing I did was ask Vinayak for good photos of both him and his fiancée, Muskan. They sent me intro videos with front profiles and side profiles for both of them, which is actually perfect. I took it into Premiere, exported all the frames, and pulled the best shots for both. With the reference locked in, I went to do some market research to see what others have done and get some inspiration. Shortly after, I realized something: nobody has done this before. There is no reference, no “here’s how someone pulled this off”. So I did what any sane person would do—I went and watched Marvel trailers and Iron Man suit-up sequences on YouTube just to understand the visual language.
Here is where the pressure really started building up. A wedding is arguably the most important milestone of anyone’s life, and this guy works in the movie industry. He is going to send this to his colleagues and his friends—people who know exactly what good filmmaking looks like. So at no point can this project look like AI slop. And that’s the biggest problem with AI content right now: character consistency breaks, faces change between shots, and for action sequences, the physics has to feel real. You can’t have a superhero landing that looks fake. So I had to think like a director, plan every shot, and solve every problem before it becomes one.
Luckily, OpenArt released a new feature called Smart Shot, which is powered by GPT Image 2. I went to OpenArt and created two characters—one for Vinayak and one for Muskan—by uploading their reference images. Then I opened Claude, gave it the client brief, and asked it to write a Smart Shot prompt. I pasted it in, selected high-quality generations, and hit create. As you can see, it generated character plus hero suit references for the front, side, and back of both Vinayak and Muskan, along with environment and set design, and a complete storyboard with what lens is used and how the camera will move—all in just one prompt. Smart Shot doesn’t just generate a clip; it builds the entire production plan first, like an actual director would. This is the kind of document a director, cinematographer, and concept artist create on a film set before proceeding with the shoot.
However, as you can see on the sheets, the faces are completely different for Vinayak and Muskan than their actual pictures. I was very confused about why it was different because it’s supposed to work in one go. Eventually, I realized when I was writing the prompt, I didn’t select the characters. So I went back and did the entire process again, but this time while adding the prompt, I also selected their characters. Now their front profile is actually very similar to how they look—because I have met both of them in real life, I can tell. I immediately sent this to Vinayak for approval. His initial reaction was really good, though he did mention that their side profiles looked a little less similar than the front profile, which honestly made me more nervous because at this point I had no idea what I was going to do differently to fix it.
Before I fixed that problem, I ran into another one. I tried to create a video with this, and it immediately showed me an error, which after all that work was not very motivating. The error said that the request failed because the output video may be related to copyright restrictions. I immediately knew why: the suits I generated were a little too close to Marvel. The male suit was basically a direct copy of Iron Man. So I went back to Claude, rewrote the Smart Shot prompts with suits inspired by Marvel rather than identical, and generated a new character sheet. Now the suits are much more unique, so I figured we should be able to get a good output.
The best part about Smart Shot is that it combines two of the best AI models right now: GPT Image 2 handles the creative direction and character references, while Seedance 2.0 handles the actual video generation—and Seedance is the best video generation model right now. I clicked on Smart Shot, configured the settings, and hit generate. The first two outputs were not a very bad start; it generated multi-sequence fight scenes with sound effects, camera movement, and character consistency, which is honestly impressive. But is this something I can directly send to the client? No. The cuts are not how I wanted, the face is distorting in some shots, and while some scenes look good and realistic, others do not. For a video going to someone in the film industry, that’s not good enough.
So I changed my workflow going ahead. Instead of generating everything together, I decided to generate Vinayak’s scenes first and then Muskan’s separately. I generated a new Smart Shot sheet focused only on Vinayak’s opening sequence fight scenes, making sure to include a close-up of his face. On the updated sheet, the close-up of Vinayak looked really good, and the suit looked pretty epic too. I tried the video creation process again, selecting Vinayak’s character and generating a batch of videos. I really liked how one of them looked—all the face details were there. However, I noticed two problems: first, Vinayak is much taller in real life, and his character looked very short in the video; second, the creatures were completely different across all three generations.
I went back to Claude, generated one more prompt for Smart Shot, and locked in both Vinayak and the creatures in the same sheet. With the new sheet selected, I started creating the videos. That is the thing with AI creation—every generation teaches you something; you don’t figure it all out upfront, you iterate your way to quality. Immediately, the quality was much better. His face looked really good, all the creatures were consistent, the physics looked great, and his face stayed consistent even with motion blur on. I sent these videos to Vinayak to preview, and he seemed to love them.
From there, I switched to generating individual scenes—flying, fighting, getting hit—each moment as a separate generation. I was doing a lot of back and forth with Claude to get the prompts right, and then I noticed something: I had been generating longer shots until now. So I asked Claude to write a prompt for a shorter, single-action scene, switched my video generation settings, and adjusted the prompt accordingly. The output was much better—the quality jumped immediately. Shorter clips meant more detailed and more controlled shots. Detailed prompts to Seedance with the Smart Shot sheet and character reference were finally giving me back the kind of shots I was actually looking for.
After generating enough scenes for Vinayak, it was time to add Muskan in. The challenge here was keeping everything—all the creatures, outfits, and environment—the same even with Muskan’s entry. I created a new Smart Shot for Muskan and gave the last Smart Shot as a reference, so the creatures and environment stayed consistent. The sheet looked really good, except the faces changed again. Then I realized a big thing: faces don’t have to be accurate on the sheet. While generating the video, you can select both the sheet and the character as a reference. The face on the Smart Shot does not actually matter as long as I select the right character—Smart Shot just does its job for the suit and the background, while the character reference handles the face separately.
With that figured out, I started generating Muskan’s scenes. Her entry had to be epic, and since I wasn’t familiar with the vocabulary of cinematic shots, I asked Claude to give me examples of epic entry sequences. I generated the same shots with different camera angles so I would have options to cut between during the edit. I had multiple shots for both of them, though there was still no “wow” factor. The best thing about AI is that you can create things that aren’t possible in a standard production. So I asked Claude to write prompts for truly epic shots, such as freeze frames, impact pulses, and Deadpool-style fight sequences. The output looked crazy—even the face detail stayed completely in sync, and it already added speed ramps in the shot. At this point, I downloaded everything and imported it into Premiere Pro, went through all the clips, laid the ones I liked on the timeline, and wrapped up the generation phase. Usually, my work as an editor starts here, but I had already spent a full phase figuring out how to prompt and get good shots with consistent characters.
For the editing phase, the first thing I did was select music from Epidemic Sound because that decides the pacing, the structure, and how the story goes. I picked four or five songs, finalized one, and started arranging the clips to match the story suggested by Vinayak. I also tried to cut on movements because all AI clips have different framing, and the more match cuts and cuts on movement I use, the harder it is to identify that it’s AI.
Once I finished the rough edit, I ran into a new problem. I had generated all the clips with sound on, which meant they had audio synced to the movement, but they also had different types of background music playing in every clip. I couldn’t directly use that in the project, meaning I had to do the entire sound design from scratch. My biggest worry was that there was too much happening in these clips and it would take a lot of effort to match the sound design quality that the AI produced. I went onto Epidemic Sound and downloaded every possible sound I might need—ambient sounds like thunder and distant fire, transition and movement sounds like whooshes, impacts, and laser beams, and creature sound effects. Then I started building the sound design in Premiere and color-labeled every layer so I wouldn’t lose my mind: magenta for ambient, green for transitions and movement, blue for original clip audio, purple for the music track, and yellow for creature sounds. This was easily the most labor-intensive part of the entire project. I have so much respect for sound designers now, especially people who work on similar action sequences.
Once sound design was done, I grabbed a title card from Envato for the save-the-date reveal. The hashtag the couple had chosen for their wedding was perfect for a closer: “#OurLifeIsAMovie”. I replaced the placeholder, created a Dynamic Link from Premiere to After Effects, and adjusted the pacing. With that, only color remained. The footage was already looking pretty good, so I added an adjustment layer with slight contrast and brightness, exported the video, and sent the final draft to Vinayak.
While waiting to see what he would say, I want to mention I am hosting my first masterclass, where I will walk you through similar client projects from brief to final delivery along with answering your specific questions—the link is in the description. Vinayak replied and really liked the quality, and the best part is that he said this is the first time he has been satisfied with an AI video. That right there made the entire effort worth it.
Here is what I learned from this project specifically: AI creation is all about iteration. Every generation teaches you something; you go in thinking you will figure it out quickly, and you come out later with a completely different understanding of how these tools work. I have said this before and I will say it again: the people who are going to win are the people who combine their core skills with AI. For example, for me in this project, it was video editing plus sound design plus AI. I knew how to cut for emotion, and I knew how to layer sound. AI gave me the visuals, and my skills made it a film.
Now, before I show you the final output, nobody in the industry provides this kind of information for free, and 80% of people who watch this channel regularly do not subscribe. I have set a pretty ambitious goal of hitting 500K, and the only way it happens is if people like you who actually get some value from this content subscribe. So make sure you subscribe. That being said, here is the final output. If you want to try OpenArt and Smart Shot for yourself, the link is in the description—use code SMARTSHOT for 15% off on the plan.