Technical Breakdown: Building a 10‑Minute Cinematic Short With an AI‑First Pipeline

I spent a month building a 10‑minute cinematic sci‑fi drama using only AI. Here's what I learned.

I spent the last month building a 10‑minute cinematic sci‑fi drama and a trailer to go along with it — the short film Annex — using nothing but AI.

The project started as a simple question: can you make a realistic, emotionally grounded short film in 2026 using only AI tools, prompting, editing, and human creative direction?

Not a montage of pretty shots. Not a tech demo. A film. One that holds character, performance, staging, pacing, and tone across a full narrative arc, the way a real short is supposed to.

So I stopped theorizing about it and stress‑tested the entire process from end to end.

The film

The film opens in a rain‑soaked World War II trench at dawn. The world is grey. Sandbags sag overhead. Duckboards sink into black mud. The men along the walls are exhausted, soaked, injured, eating from tins, sleeping upright, and waiting for orders that already feel too late.

We follow Davis, a young American soldier, as the battlefield begins to go quiet. Then quieter. The enemy vanishes without explanation. Loaded machine guns are left behind. Coffee is still hot. Cigarettes are still burning. What begins as a grounded war drama slowly turns into something stranger, larger, and more impossible than anyone in that trench is prepared for.

I built two cuts: the full 10‑minute short film, and a 60‑second teaser trailer. The teaser is the hook. The 10‑minute film is the test.

I played Davis

I didn't just put myself somewhere in the film. I played the lead. Davis is the soldier the entire story follows, and that was deliberate. I wanted to point the experiment at the hardest possible target.

If AI filmmaking is going to mean anything for working directors, it cannot just produce a striking face for one shot and lose it in the next. It has to hold a specific person: the same face, the same build, the same emotional energy, and the same performance continuity across an entire story.

So I made the protagonist me. I wanted to see how close, how consistent, and how realistic one lead character could stay from the first frame to the last.

The toolchain

This was built with an AI‑first production pipeline. The pipeline was deliberately multi‑tool, with each platform doing the job it handled best.

• Writing and creative direction: ChatGPT, Claude, and Gemini — and myself.
• Still frames and keyframes: Nano Banana, GPT‑2 (OpenAI), Soul Cinema by Higgsfield.
• Image‑to‑video and motion: Kling AI, Seedance 2.0 via Higgsfield, OpenArt, CapCut Visual Studio.
• Voice: ElevenLabs.
• Music: Suno, YouTube, ElevenLabs.
• Sound design: Seed Audio 1.0 and ElevenLabs.
• Environments, previs, and virtual production: Unreal Engine.
• Titles, editing, and thumbnails: CapCut and Canva.
• Upscaling and restoration: Topaz Labs and DaVinci Resolve.
• Final color and assembly: DaVinci Resolve.

No crew. No set. No camera package. One creator, a stack of tools, and a lot of persistence. But here is the part the tool list never tells you: the tools were the easy part. The hard part was direction.

Read the full breakdown on LinkedIn using the button above.