Issue 00 — Spring 2025
/6
Maciej Żemojcin on the craft of AI filmmaking
Maciej discusses the practical and creative shifts introduced by AI in filmmaking — from mechanisms of pattern recognition to questions of authorship and aesthetics. Far from a solitary pursuit, he argues, AI-assisted filmmaking will continue to rely on community, dialogue, and shared practices.
SHARE THIS
Facebook / LinkedIn / X
GALLERY
Everyone wants to know: How do you make a film with AI?
MZ I’m afraid there’s no fixed answer for how to make a film with AI — or even how to start learning it. But like any film, it all begins with a story on a napkin — literally or figuratively. AI doesn’t change that: storytelling is always the foundation of filmmaking.
That said, we’re still in an experimental phase with AI in our field. What’s clear — even Disney’s CEO recently acknowledged it — is that this is one of the most powerful tools in cinema’s history.
So, how do we approach it? The first tip I offer when I run my workshops is: don’t jump from tool to tool in pursuit of a panacea. You’ll get lost. Start with the story and learn to write clearly — text is still your primary language. The more detail you give, the clearer the result. And once you realise AI can understand both language and images, it becomes a collaborative process, a multimedia conversation.
At its core, AI is a powerful pattern recognition tool. Like our own minds, it takes a shape and applies meaning. If I record a video of books on a shelf with my phone and ask AI to imagine them as skyscrapers, it works, because — like us — it can overlay one pattern onto another.
Storytelling is always the foundation of filmmaking.
AI doesn’t change that.
AI doesn’t change that.
Some of us are visual learners and creators. Are these tools making it possible to build a film idea not from text, but from images?
MZ Technically yes, but you can’t skip text entirely — at least not with the tools available now. A dialogue always takes place. This iterative process is the foundation of AI filmmaking. And nothing is perfect on the first try, we need multiple attempts to refine an idea. The progression typically follows three stages: text, still images, then animation. But the process flows both ways — so you can also start new iterations with images.
Tools like Midjourney, for instance, can describe images in words. You upload a picture, and it generates different prompts that try to capture what it sees. That’s a useful function you can use to enrich your ‘fluency’. In fact, it’s probably the best prompt engineering training you can get. AI can teach you its own language better than other humans can.
But to be clear: you always remain in control. We don’t collaborate with AI — we use it. The goal isn’t to replace the artist but to elevate creative practices. It’s still our story to tell.
We’re exploring, refining, growing ideas using the tools, just as we expanded the possibilities of artmaking with other tools in the past. And that authorship matters. Say I ask for a “samurai on a meadow” — AI gives me options. Maybe I like the atmosphere of one image but not the army in the background. So I refine it. Try again. Each choice I make becomes part of the creative process. It’s like painting — whether I add a single brushstroke or thousands, they all lead toward a final result that reflects my vision and expression.
Even the U.S. Copyright Office now recognises this. If you just type a generic prompt, that’s not your copyright. But if you go through an iterative process — if you direct it, shape it, and can demonstrate the significance of your input — then you can claim the product as yours. The authorship lies in the decisions you make along the way.
We don’t collaborate with AI — we use it.
Beyond the idea generation stage, how are AI tools supporting new productions?
MZ Traditional filmmaking follows a clear path — script, storyboard, shoot, edit. With AI, this can be reversed: you can start with generated images — like “dailies” — and build from there. You can interact with the tools through dialogue and even move between platforms — Midjourney, Claude, GPT — and let them interact.
Text, image, and movement are the ways we naturally express ourselves in our daily lives. AI guesses based on training, like we do. It assumes what’s behind your back even if it’s not shown — just as we fill in gaps with our imagination when we look at the world. That’s why we must remain the ones directing, choosing, maintaining our vision throughout the process.
With new creative options come new risks. If we rely on AI to generate final images without guiding it, everything starts to look the same. Working through pattern recognition, in plain terms, means getting statistically average results. So originality must always come from humans. Think of AI like a flail — you know, the medieval weapon. You are the counterweight. Without you balancing it, the tool is useless.
Are the aesthetics we see in AI-generated work — such as flawless skin or smooth colour transitions — driven by the training data and algorithms themselves, unless prompted otherwise?
MZ They are, to a degree. This ties back to pattern recognition. AI is great at combining similar patterns — like imagining books as skyscrapers. But the further apart the concepts are, the harder it is to combine them meaningfully — like connecting bananas and motorcycles. Still, AI allows us to play with these unusual pairings much faster than before, opening up a new kind of aesthetic based on the “energy” between distant ideas. That energy comes from the unexpected tension or harmony when two seemingly unrelated patterns meet. It can surprise us, amuse us, or evoke emotion — and that reaction is part of what defines the emerging AI aesthetic. Rather than just mimicking reality, AI-generated visuals often find their power in this strange in-between space.
Another emergent aesthetic language is that of hallucination — when the model doesn’t know something and invents it instead. Early models did this a lot, and some artists, like Pierre Zandrovich, embraced it. This style creates surreal, dreamlike images that once required complex and expensive VFX. Now you can generate wild visuals — like ballooning eyes or hybrid creatures — in minutes. This is a key part of AI’s emerging aesthetic: fast, imaginative, and sometimes very unpredictable.
Rather than just mimicking reality, AI‑generated visuals often find their power in this strange in‑between space.
Filmmaking is traditionally collaborative, but AI creation often seems like a solitary process — just one person and a machine. Is there still space for community in AI filmmaking?
MZ I come from the film industry, where collaboration is everything. Right now in the AI space, you often find two types: the hardcore tech nerds, and the one-person filmmaker armies. But neither reflects the collective spirit of filmmaking.
In traditional filmmaking, every role — director, DOP, production designer — talks, argues, and contributes. Even if AI radically redefines roles in filmmaking, that dialogue must remain. That’s why, when I run workshops, I bring the whole crew together. When people share tools, excitement builds — and that leads to mutual support.
I hope new AI-assisted collective endeavours emerge in our industry. Because great stories will not come from working alone with a machine. Films are made through thousands of shared decisions. That collective effort still matters — and it will define the future of AI filmmaking too.
Right now in the AI space, you often find two types:
the hardcore tech nerds, and the one‑person filmmaker armies.
But neither reflects the collective spirit of filmmaking.
But neither reflects the collective spirit of filmmaking.