AI filmmaking is now a thing, and it’s enabling creatives to share their ideas in new ways and with higher fidelity than ever before. As technology like Seedance gets into the hands of writers and directors, AI films are becoming more polished and watchable. New film CATACOMBS from filmmaker Kévin Mendiboure is one of those projects that may use AI to get the images up on screen, but like Kavan Cardoza’s The Chronicles of Bone before it, making this movie has one foot firmly in old-fashioned filmmaking techniques and artistry.
CATACOMBS starts with a script, pre-production and prop creation, but rather than turn to film, Mendiboure begins prompting. As he says below in my interview, “The most important part of making a movie, AI or not, is the script. The camera or the AI is just a tool. The magic only happens when you have a good story to tell”.
He adds: “From my perspective, AI will not replace cinema; I see it as a new creative category, the same way CGI was a new category when Toy Story arrived.”
Read insights on AI filmmaking and getting started in film, or catch up with what others have to say, in my interviews with Wonder Studio’s Justin Hackney and Autodesk Flow Studio’s Nikola Todorovic.
CB: What made you decide to build an entire short film using generative AI?
Kévin Mendiboure: I wrote the CATACOMBES concept 10 years ago, when I was developing ideas for my feature film The Follower, which I directed in 2017. The main problem was that the concept was far too expensive, and I had no producers willing to back it. In France, sci-fi horror is not particularly welcome; producers tend to prefer comedies and social dramas. Not really my genre.
From early 2023, I started learning AI tools, beginning with ChatGPT and then image generators like Midjourney. The possibilities were immediately impressive. Between 2023 and late 2025, I tested many AI video generators, but none convinced me; the results were too fake, too plastic, not live-action enough. Even Veo 3 by Google came close but still didn’t feel right.
Then on April 10th, 2026, I saw a video that changed everything for me: Zephyr. It was the first full AI original series produced by Higgsfield using Seedance 2. That was the quality I had been looking for during three years of learning: it genuinely looked like live-action footage. I decided that same day to rewrite CATACOMBES as a short film and test the technology. I loved every part of the process.
CB: How did you approach maintaining emotional performance and character consistency, given that everything is AI-generated?
KM: The most important part of making a movie, AI or not, is the script. The camera or the AI is just a tool. The magic only happens when you have a good story to tell. When I worked on CATACOMBES, I wrote every line of dialogue with a distinct personality for each character. My goal was simple: you should be able to guess who is speaking without reading the character names. When that works in a script, you’re on the right track.
Regarding consistency, working with AI requires creating character sheets and reference images for each character showing different angles: front, back, profile, close-up, with their costume and any important visual details. When a character changes during the story (for example, when François, Kane and ARES-7 are injured and covered in alien substances), you need to update all the character sheets before continuing with the next sequences.
Image 1 of 10
CB: What were the biggest creative surprises, moments where the AI output pushed the story or visuals in directions you didn’t originally plan?
KM: There were genuinely interesting surprises throughout the process: ideas that came from accidental generations. It’s similar to working with actors on set: sometimes an actor improvises something better than what you wrote, and you realise you should change the script.
One clear example: the confession scene in the armoury, where François speaks about his brother and the guilt he carries. The shot you see in the film was not the one I originally planned. I had intended a profile shot, but the AI placed the camera in a high-angle extreme close-up. It was far more emotionally powerful for that moment, so I kept it and rewrote my shot list accordingly.
CB: How did you balance “Hollywood-grade” cinematic ambition with the limitations or unpredictability of current generative tools?
KM: The secret is iteration. You never get the right shot on the first generation, which is also why AI filmmaking can become expensive quickly. To achieve a cinematic result, you need to generate many times, refining your prompt each time until you find the right shot. And you need a genuine passion for cinema, because you can sometimes spend ten hours on ten seconds of film. That was almost exactly the time I spent on the helicopter transition sequence.
My workflow to control costs was to use Seedance 2 fast, low-quality mode first (lower resolution, much cheaper), to find the right prompt. Once the result looks good in low quality, I switch to high quality and run multiple generations to find the best take. It is time-consuming for every single shot in the film.
Image 1 of 3
CB: In what ways did tools like Higgsfield Cinema Studio, Seedance 2, and Kling 3 shape the storytelling itself, rather than just execute it?
KM: The most interesting aspect of these tools is their capacity for improvisation. Sometimes I deliberately don’t give a precise shot briefing (I just describe the action) to see what the AI brings to the cinematography.
There are often good surprises, especially with multi-shot generation. In Seedance 2 and Kling 3, a multi-shot is a generation of up to 15 seconds that produces multiple connected shots based on your described actions and dialogue. Working this way, you can quickly discover camera positions, performance choices, and character staging you hadn’t planned. When that happens, you go back and rewrite your shot list.
CB: Is your workflow prompting and re-prompting, or do you create sets, think in 3D, plan shots and a script? Basically, is there more to this than people think?
KM: Directing an AI film is a substantial amount of work. Prompting is just one part of it. Once you have the script, you go through a full pre-production process, exactly like a traditional film. The first step is creating your characters (either through prompts, or as I did for François, using photographs of myself, and for General Moreau, photographs of my partner, both with full rights to use their likeness throughout the film).
After characters, you define your locations. For CATACOMBES, I used hundreds of images and generations to build all the sets: corridors, large underground chambers, combat spaces. That process alone took several days. Then come the props: weapons, the helicopter, objects that appear throughout the film.
After pre-production, you build your shot list. Like in traditional filmmaking, you need to plan every shot for the film to maintain the right immersion and emotion. This requires real cinematography knowledge. If you have never directed films before, this part will be nearly impossible. AI filmmaking demands filmmaker skills, and with AI, those skills will be pushed further than ever.
The final pre-production step is creating start frames. These are the reference images you feed into the video generation tools to produce each shot. They are the single most important factor in achieving a photorealistic result with Seedance 2 or Kling. A weak start frame means a weak shot: characters will look artificial and unconvincing.
CB: How do you think AI changes the role of the filmmaker? Are you still a director or a curator of the generated outcomes?
KM: AI will change the role of the filmmaker permanently. With these tools, a filmmaker can now be assisted at every stage of the directing process, including in live-action productions.
For example, you can use AI to build and test your shot lists before a single day of shooting. You can pre-visualise complex sequences like car chases or fight scenes. You can generate visual references for your entire crew. You can explore visual effects in ways that would previously have required a full VFX department.
Artists will save enormous amounts of time and will be able to push their imagination further. There are no longer any budget-imposed limits on what you can envision. I am still a director. I am currently developing a project in discussion with Netflix. But in my free time, I can now work on new concepts quickly and independently. And some of those projects might serve as proof of concept to bring to producers for live-action development.
From my perspective, AI will not replace cinema; I see it as a new creative category, the same way CGI was a new category when Toy Story arrived.
CB: What was the most technically difficult part of producing CATACOMBES solo in the time frame, and how did you solve it across the pipeline?
KM: Surprisingly, the most technically difficult sequences in CATACOMBES were not the fight scenes. By far the hardest parts were the calm dialogue scenes.
AI is not a natural director. It does not respect cinematic grammar. When you need a reverse angle for a three-character dialogue, AI will frequently break the 180-degree rule, which is a serious continuity problem.
The other major difficulty is shot continuity. The AI pipeline does not work like traditional filmmaking. In live-action, you shoot your material first, then enter post-production. In an AI pipeline, you have to run your editing software simultaneously with your generation software, building the cut in real time. It is the only way to maintain visual continuity throughout the film.
At certain points, a Seedance 2 generation takes between eight and eighteen minutes to render. I used that waiting time to edit, review continuity, and refine upcoming shots. I optimised every minute to complete the film within one month. In total: 3,229 image and video generations, 242 hours of work.
CB: Do you see AI-generated filmmaking as a threat to traditional crews, or as a parallel creative lane that expands what’s possible on smaller budgets?
KM: AI filmmaking will break down the barriers that have always existed in cinema. Before AI, when you wrote a script, your imagination was unconsciously limited by what was financially feasible. A large-scale sci-fi short film was simply not possible; the visual effects alone would cost millions.
Now, filmmakers can write whatever they can imagine without budget constraints. I believe we will see emerging talents with genuinely original ideas and spectacular cinematography that would previously never have existed.
A good historical parallel: Fede Alvarez made “Ataque de Pánico!” in 2009, an impressive short film showing giant robots invading Montevideo, made with almost no budget. Sam Raimi saw it and hired him. The same path happened for David F. Sandberg and Michael Chaves.
AI will not kill cinema or film crews. It will create new jobs in a parallel creative genre, and it will assist traditional productions at every level.
Image 1 of 4
CB: Are we at a point where professional knowledge of filmmaking – shot blocking, colour grading, cinematography – will set some AI films apart from others?
KM: Absolutely. The best AI films will come from skilled directors; that will not change. It is the same in traditional cinema: give a 100-million-dollar budget and a full crew to two different directors, and you will not necessarily get two good films. The result always depends on the director’s vision and skills.
To develop those skills, you have to shoot films in real life, work with teams, make mistakes, start again, attend film festivals, receive feedback, and keep going. Filmmaking is a lifelong learning process, and genuine craft only comes from real experience.
CB: Looking ahead, what does a “normal” filmmaking workflow look like in five years if tools like these continue to evolve at the same pace?
KM: I imagine AI will become a standard skill required to work in a production company. Directors will use AI to visualise what they have in mind and to communicate their vision precisely to their team before and during shooting.
I remember two years ago, when I directed my short film MASK (Screamfest Official Selection 2025), AI tools were just good enough to generate rough storyboards. I used that to show my team the frame compositions and camera movements I wanted.
At that time, people said AI would never replace live-action shooting because the quality was too low. Now, AI films are photorealistic enough that within a few months it may genuinely become impossible to tell whether a film on Netflix is fully AI-generated or live-action.
CB: Finally, Tarantino famously started out making low-budget movies while working at a film rental store. Are we on the cusp of the next Tarantino making their movie breakout using AI?
KM: Absolutely. We will see new talents emerge who can now explore their full creative vision with these tools. We are at a moment where everything is possible. Filmmakers no longer have any budget excuse to hold back. And the window of opportunity to be among the first is closing quickly.
My advice to aspiring filmmakers: once you have learned the fundamentals of filmmaking and directed a few short films, write stories without budget limitations and use AI to build your portfolio, develop your cinematography, and show the world what you are capable of.
