Without realizing it, a full year has passed since I started creating AI videos. Although most of the time was spent mastering the technology rather than producing a sufficient number of works, and I didn’t successfully form a team, starting from this new year, I will focus wholeheartedly on creation and continuously output works.
What does it feel like to be an AI creator? I can only say that every time I open my editing software and start working on a new segment, my heart feels uncertain, worried that I won’t be able to execute what’s already written in the script. Sometimes this even triggers procrastination, and I’ll put off working on the next segment for several days.
This is truly a feeling that combines both creative freedom and creative difficulty. Sometimes I struggle endlessly to produce a single shot, while other times there are pleasant surprises. My mood constantly swings between disappointment and celebration. The current state of AI development has numerous limitations and is unable to accomplish many things.
My first encounter with AI was during Midjourney’s beta testing, when video tools weren’t as advanced as they are today. Midjourney’s hands were still frequently corrupted. Later, when Stable Diffusion 1.5 came out, I developed an obsessive-compulsive disorder about checking hands. The first thing I would do when an image was generated was count exactly how many fingers there were. Using Photoshop to fix bad hands would generate even more bad hands.
Then came endless gacha-style generation. When the tools weren’t sophisticated enough, the number of attempts was significantly higher than it is today. Some creators believe this gacha mechanism corrupts their creative process and choose to quit. I think it’s manageable, but I know directors who would generate over 80 attempts for a single shot—they certainly have enough patience.
When Kling came out around this time last year, I was extremely excited and immediately made a Batman video. That was my first AI video, called “Batman vs. AI.” It was purely a PowerPoint-style video, but it already had a certain Gotham tone. It also featured Kevin Conroy’s voice cloning. Even looking at it now, it remains an interesting work.
Initially, it was just playing with AI without any pressure. Unlike now, where every work must be better than the previous one, and every dialogue scene or action sequence must showcase the most advanced technology, even though the most advanced technology is still far from film industry standards. Sometimes, I do miss the time when I would simply make my favorite Batman video just because it was fun.
I still have a Red Hood script that I’ve always wanted to film, but now my Computational Universe story schedule is fully booked, leaving no opportunity to work on it. There’s too little time left for playing, and too much time needed to become professional. However, I will still create stories that emerge from the depths of my heart; I won’t craft soulless stories. We’re in the AI era now—I don’t want to make bad films.
I’m primarily technology-focused because I used to be a programmer, and I always have excessive confidence in technology. Therefore, in the past few months, I’ve been researching technology. The most pressing issue was character consistency. Now, a year later, character consistency problems have been largely solved because we have multiple multi-reference tools. But I still find basic LoRA training very useful, even after multi-reference tools emerged. So I won’t abandon LoRA training for now.
However, I’m gradually suppressing my excessive pursuit of technology and putting more time into creation rather than obsessing over technical details. AI technology evolves so rapidly that content researched a few months ago is often replaced by various updates. But I’ve found that the series of image processing methods I developed through this process remains consistently useful. Technology is the foundation.
In the first few months of 2025, I was obsessed with seeking team collaboration, wanting to scale up my AI creation, but I failed. This took about four months of my time. Later, I realized I still just want to focus on creation. Creation is truly a process where pain and joy coexist.
Every morning, I wake up, write a blog post, then spend several hours writing novels, and dedicate all remaining time to creating AI videos. At the end of each day, I watch films and TV shows to increase my viewing repertoire. Every day feels fulfilling, as if I’ve made a little progress.
I still believe AI can empower everyone to become a director. All we need is more professional training, building a solid foundation in audiovisual language, and carefully observing how other directors work. As technology matures, the quality of our work will become increasingly higher.
Looking back a year later, I realize I took quite a few detours, but my passion for AI has only grown stronger. Because of this passion, I pay attention to details and refuse to lower my standards. I’ve been creating the Computational Universe for over three years now—it’s time to seriously fill in the gaps.
In the new year, you’ll see me producing many short films, as well as writing short stories and novels. As for visualizing novels through AI, I believe it will take another one to two years for the technology to mature to a stable plateau. We are currently in a period of rapid development, still at the cutting edge of this era.
Follow me as we explore AI filmmaking together.
