I Made an AI Samurai Short Film. Here's What I Learned.
Two weeks ago I paid for a subscription to Grok Imagine and decided I was going to make a samurai short film. In my head, this was going to be relatively straightforward. Type a few prompts, generate some scenes, stitch them together in Shotcut and upload the finished masterpiece to YouTube. Judging by some of the advertising, I assumed I was only a few hours away from becoming the Akira Kurosawa of Yorkshire.
The reality was somewhat different.
![]() |
| Promotional poster for The Keepsake (2026), my first AI-generated samurai short film. |
From Prompt to Production: Planning an AI Samurai Film
What I quickly discovered is that AI is a bit like employing an incredibly talented artist who has consumed several pints of strong cider before arriving for work. Sometimes it produces something astonishing. Other times it appears to have forgotten how swords, people and basic reality function.
My original plan was simply to generate video clips and build the film from those. That lasted about five minutes. Very quickly I realised the best approach was to create images first and build a storyboard shot by shot. Once I had the images looking right, I could then use them to generate video. It was slower than I expected, but it gave me far more control over the story and probably saved me from disappearing down several very expensive dead ends.
Even then, things frequently went wrong. A sword would mysteriously change hands between scenes. A kimono would alter colour. A character's face would subtly change from one clip to the next. At one point a samurai mask appeared to fuse itself directly into a man's mouth, creating something that would probably require immediate medical attention.
The strange thing is that AI can create breathtaking visuals while simultaneously struggling with details that any human would immediately notice. It can generate a beautiful autumn forest with sunlight filtering through the trees, drifting leaves and the sort of atmosphere filmmakers spend a fortune trying to create. Then, moments later, it forgets what the main character looked like twenty seconds earlier.
The Biggest Challenge: Keeping the Same Samurai Alive
Keeping the ronin consistent became one of the biggest challenges of the entire project. I would finally get him looking exactly right. Same face. Same beard. Same clothing. Same expression. The next scene would arrive and he'd look like a distant cousin who'd wandered in from another province. Correcting one problem often created another. The character might finally be right, but now the background had changed. The background might be perfect, but the armour looked different. The armour might be fixed, but the sword had developed a mind of its own.
Creating an AI film often feels less like directing and more like negotiating.
Editing the Chaos
There was also far more editing involved than I expected. People sometimes imagine AI filmmaking consists of typing a prompt and receiving a finished scene. What they don't see are the dozens of clips that never make it into the final cut. For every scene that worked reasonably well, there were several that didn't. The finished film contains only a fraction of the footage that was actually generated, with the rest quietly disappearing onto the digital equivalent of the cutting-room floor.
Everything was assembled in Shotcut, where clips were trimmed, rearranged and occasionally rescued from disasters of AI's own making. One solution I discovered was the use of black title screens. Originally they were there to disguise continuity issues and break the story into manageable sections. Ironically, they ended up improving the film. Instead of feeling like a limitation, they helped create a more deliberate pace and gave the story room to breathe.
Sometimes necessity accidentally becomes style.
Will AI Replace Filmmakers?
The experience also changed my opinion of AI itself. Before starting the project I had read plenty of articles claiming artificial intelligence would soon replace filmmakers. After spending a couple of weeks trying to persuade a computer to keep the same samurai wearing the same clothes in the same forest, I remain unconvinced.
The technology is remarkable. What would have required actors, costumes, locations and a substantial budget only a few years ago can now be attempted by one person sitting at home. That is genuinely revolutionary. Yet the technology still has obvious limitations. It struggles with consistency. It struggles with movement. It struggles with continuity. Most importantly, it struggles with understanding why a scene matters in the first place.
AI can generate images. It can generate video. It can generate possibilities. What it cannot generate is judgement.
The decisions still belong to the creator. Which shot stays? Which shot goes? Which scene works? Which scene feels wrong? Those choices remain stubbornly human. The biggest surprise of the entire project wasn't how good AI has become. It was how much work was still required. The software may have helped create the images, but it certainly didn't make the film on its own.
Would I do it again? Absolutely.
Would it be easier the second time? Almost certainly.
Would I trust AI to make the whole thing without me?
Not a chance.
At least not until it learns which way round a samurai is supposed to hold a sword.
Watch The Keepsake (2026)
Set in Early Edo-period Japan, The Keepsake follows a wandering ronin whose journey becomes entwined with a mysterious samurai and a painful memory from his past.
This is Part One of the story.
If you enjoy samurai films, historical drama or are simply curious about the current state of AI filmmaking, I'd be delighted if you gave it a watch. And if you enjoy it, please consider subscribing to my new YouTube channel. It's completely free and helps support future projects.
Thank you for reading, and thank you for watching.
