Filmmaker Turns 10-Year-Old Idea Into Reality Using AI Video Generation
Kévin Mendiboure wrote the concept for CATACOMBES a decade ago. No producer would fund it. The sci-fi horror short was too expensive and didn't fit French financing preferences for comedies and social dramas.
In April 2026, after three years testing AI video tools, Mendiboure saw a demo of Seedance 2 that changed his approach. The output looked like live-action footage. He rewrote the script as a short film and produced it solo in one month using 3,229 generations across 242 hours of work.
The result demonstrates what's becoming clear in professional filmmaking: AI video tools are moving from novelty to production method-but only for directors who understand traditional craft.
Script and Pre-Production Come First
Mendiboure's workflow mirrors conventional filmmaking. He started with a finished script, then moved through character development, location design, prop creation, and shot planning before touching any AI tool.
Character consistency required detailed reference sheets showing each character from multiple angles and in different states (injured, covered in alien substances). He used photographs of himself for the lead character François and his partner for General Moreau, with full rights cleared for their likenesses.
Location building consumed several days alone. Mendiboure generated hundreds of images to construct corridors, underground chambers, and combat spaces that would maintain visual coherence across the film.
"The most important part of making a movie, AI or not, is the script," Mendiboure said. "The camera or the AI is just a tool. The magic only happens when you have a good story to tell."
Dialogue Scenes Proved Harder Than Action
The technical challenge wasn't fight sequences or explosions. Calm dialogue scenes between multiple characters created continuity problems that required constant attention.
AI video generators don't follow cinematic grammar. They frequently violate the 180-degree rule in three-character conversations, breaking spatial continuity that viewers expect. Mendiboure solved this by editing in real-time alongside generation, building the cut simultaneously rather than shooting first and editing later.
During render waits-sometimes 8 to 18 minutes per shot-he reviewed continuity, refined upcoming prompts, and adjusted the shot list based on what the AI actually produced.
Iteration, Not Single Prompts
Mendiboure never got the right shot on the first generation. He developed a cost-control workflow: generate low-resolution fast versions first to test prompts, then switch to high quality once the direction was clear.
The helicopter transition sequence alone took nearly ten hours to produce ten seconds of film. That level of iteration is why AI filmmaking can become expensive quickly, despite the lower production costs compared to traditional shoots.
Some surprises came from deliberate ambiguity in prompts. By describing action without specifying exact shots, the tools suggested camera positions and staging he hadn't planned. A confession scene in the armoury showed this: Mendiboure intended a profile shot, but the AI positioned an extreme close-up from a high angle. The emotional impact was stronger, so he rewrote his shot list to match.
Professional Filmmaking Skills Remain Essential
Mendiboure emphasizes that AI filmmaking demands traditional director knowledge. Shot blocking, cinematography, color theory, and understanding how to compose a frame for emotional effect-these skills separate competent AI films from weak ones.
"Directing an AI film is a substantial amount of work," he said. "Prompting is just one part of it. If you have never directed films before, this part will be nearly impossible."
He compares the shift to how CGI became a new creative category after Toy Story, not a replacement for cinema itself. The tools don't eliminate the need for directorial vision; they expand what one person can produce independently.
What This Means for the Industry
Mendiboure sees AI becoming standard skill training in production companies within five years. Directors will use these tools to visualize shots before shooting, communicate vision to crews, and test complex sequences like car chases or fight choreography.
The barrier that budget imposed on imagination is lowering. Writers no longer need to unconsciously limit their scripts to what's financially feasible. A large-scale sci-fi short becomes possible without millions in visual effects budgets.
This creates space for emerging talent with original ideas. Mendiboure draws a parallel to Fede Alvarez, who made the low-budget short "Ataque de Pánico!" in 2009 and was hired by Sam Raimi after it circulated. Similar paths opened for David F. Sandberg and Michael Chaves.
"We will see new talents emerge who can now explore their full creative vision with these tools," Mendiboure said. "Filmmakers no longer have any budget excuse to hold back."
His advice to aspiring filmmakers: learn generative video tools after mastering traditional filmmaking fundamentals. Write stories without budget limitations. Build a portfolio that demonstrates what you can do. The window for being among the first is closing quickly.
For creatives considering these tools, the pattern is consistent: craft matters more than the medium. AI for creatives works best when the person holding the tool understands how to tell stories and compose images. The technology doesn't replace that knowledge-it just makes it possible to execute alone.
Your membership also unlocks:
