AI filmmaking is an entirely new field that has only existed for one year. I participated throughout this entire year. This article is my record; I won’t go into detailed technical specifics. Finding AI was like finding true love for me—I went all in and switched careers entirely.
I graduated from New York University with a degree in philosophy. Then, I spent over a decade in tech entrepreneurship, writing code and novels for over ten years. I no longer write code and have gone all-in on AI. I have wanted to work in this field for years; the past year was my first. Like “Batman: Year One,” I experienced a long rookie period.
I was already playing with AI art generation the year before last. I subscribed to Midjourney in its first month and used it to create countless World of Warcraft avatars, achieving illustration freedom. I used Midjourney continuously for days, even forgetting to sleep. But at that time, its prompt adherence wasn’t good enough, and the aesthetic quality of the images was far from what it is now. But I was captivated by it—it was all about pure passion.
As of today, I’ve generated 13,584 images with Midjourney. Not too many, but I’ve passed the 10,000-image rookie threshold. Midjourney has always been my primary image-generation tool when making AI videos. Currently, nothing exists that can surpass it artistically—it remains the king. So, if you’re just starting with AI filmmaking, getting familiar with Midjourney first is a good strategy.
After playing with Midjourney the year before last, I started experimenting with Stable Diffusion’s WebUI. I originally wanted to use it for clothing model commissions, but the limitations of the SD1.5 model were too significant—clothing companies said the generated images weren’t usable. I also tried making AI comics but didn’t achieve much.
Later, when Runway’s Gen2 came out, I created a series called “The Computational Star Map.” Video generation was still unstable then, and the results were quite surreal. I’ve since hidden the “The Computational Star Map” I posted on WeChat Channels because the quality was terrible, but it remains a memory.
I’m disappointed that I didn’t touch AI for several months. Everything develops in twists and turns—there’s no way to achieve success overnight.
Until June 2024, when I first saw Kling, Kling was still in the queue, but because I had over 6,000 followers on Bilibili, I shortened the queue process. The moment I saw Kling, I knew this was it. The future had arrived. I immediately made my first real AI short film, “Batman vs. AI.”
At that time, lip-sync could only be done with Runway, and it would fail with even slight lighting changes. So when Batman spoke, it could only be still images and in relatively bright environments. This wasn’t very Batman-like. But Kling’s magic was captivating even when it first launched.
Then, I began intensive research. In the first week, I got sick from overwork and had to rest for several days. I wrote “Siyuan AI 2024 Magic Book v1,” a technical research notebook of over 60,000 words—though compared to the various AI technical documents I wrote later, this was actually just the beginning.
I have always had a habit of over-preparation. Writing the magic book was also before I started making AI films—I explored technology first. Since I was once a programmer, technology has brought me joy. Although many people making AI videos think technology isn’t essential, focusing on art and storytelling is enough.
But AI is the ultimate fusion of technology, art, and literature. Each component is indispensable. I’m exceptionally geeky and love this hardcore technical vibe because delving into technology gives me a sense of security. Dealing with people is actually more complex while interacting with technology is relaxing.
At this point, I had chosen my technical route: using open-source model LoRAs combined with OpenPose to control character dynamics, then combining with Midjourney. Many people opposed this technical route, thinking Midjourney was the primary tool while open-source models were of too poor quality. I knew Flux’s lighting effects weren’t good enough, but I believed open-source models would keep improving.
As I researched SDXL model LoRA training, Flux burst onto the scene. From its appearance to ecosystem maturity, Flux actually took half a year. But when it first came out, its prompt adherence and new possibilities honestly overshadowed its overly greasy character rendering.
The time came to September 2024. I tried making some films and participating in Bilibili’s Vacat Awards and Kling’s AI short drama competition. The former gave me a consolation prize; the latter, nothing. At that time, I made a series called “Memory of Distant Stars,” driven purely by passion but with no foundation in audiovisual language. Looking back now, it’s unwatchable.
This was the first time I realized the technical approach couldn’t work completely. So, I began studying directing knowledge and watching various tutorials. But while learning AI production is easy, developing directorial thinking is much harder. This involves both an aesthetic foundation and proficiency. Every shot needs design intention, not randomness.
At this point, creators originally from the film industry who transitioned to AI filmmaking had considerable advantages. Moreover, many platforms haven’t yet limited AI film content. It was truly a pioneering period. Many people started forming teams then. And I… spent several more months diving into technology, unable to leave my comfort zone.
The effect I wanted to achieve was my designated characters in my designated scenes, wearing exotic costumes and acting in my designated drama. Each of these was a challenge and needed a cinematic atmosphere. At that time, I wanted to film “The Computational Wandering” as a long-form series.
But to this day, I’ve found that overly long AI productions still have problems. People might not even persist through the first 10 minutes. This can only wait for technological advancement.
Since I failed to form a team last year, and I really wanted to team up after making the “The Computational Wandering” pilot in early 2025, I spent various efforts seeking team collaboration—but all failed, and I wasted at least four months without focusing on creation. This was also demoralizing.
But entering May 2025, I finally picked up AI filmmaking again. Nearly a year had passed. Everyone had run ahead of me—forming teams, shooting short dramas, and entering combat mode. I was still accumulating some foundational works. Was I afraid? A little. Anxious? Also, a little.
But I was only more certain that AI is my true love. In the first half of 2025, I accumulated 25 short story concepts, each perfect for writing as a novel and adapting into AI short films. Without selection, I’ll shoot them in order of the concepts because I want to challenge myself to express any content I write through AI.
Meanwhile, I’ll continue writing long-form novels. Once the books are ready, AI technology will have made great strides and be capable of supporting long-form series production. Throughout this AI wave, I took many detours before finally determining that creation brings me the most joy.
My original intention in making AIGC videos was to visualize my novels. I won’t forget this original aspiration, regardless of what I experience. AI represents a cyber renaissance. And now is my first year of AI creation. In “Batman: Year One,” after Batman is injured, he sits in his study, almost ready to give up. But at that moment, a bat breaks through the window, and he becomes Batman.
After experiencing various delays, I discovered that only continuously creating the computational universe is what I truly want. For the first time in three years, creating the computational universe will become where I invest my primary energy. Jiaozi went into seclusion for 5 years to make “Nezha 2″—I can’t compare to him, but I’ll gradually create this universe bit by bit.
The first of those 25 short stories is “The Resolve.” This title also reflects my current state of mind. There’s nothing to fear—charge forward until the world’s end.
