A short film can take a long time
I made Terminal Velocity in my own time, partly to see how far I could take an idea using the AI tools available to me. I expected a lot of experimentation. I underestimated how much work would follow the first convincing clip.
A ten-second video can look extraordinary on its own. That says little about whether it belongs in a film. Put two impressive shots together and the lighting, scale, character or movement may change without explanation. The audience notices the break even if each clip looks polished.
The difficult part was continuity
I needed each scene to feel as though it came from the same world. That meant deciding what had to remain consistent before generating more material: the atmosphere, the visual rules, the characters and the way one scene should lead into the next.
I built a small prompt-development app inside Google Omni to help prepare scenes against that vision. It did not solve the creative problem for me. It gave me a way to make the instructions more consistent, review what I was trying to create and correct the direction when a result wandered.
Most of the work did not make the film
The project took roughly 100 hours and involved around 500 video generations. About 10 per cent made the finished cut. The other 90 per cent was not simply wasted material. It was the cost of finding out which images worked together and which ones only looked good in isolation.
Rejecting a clip was sometimes harder than making it. There were shots I liked, but they changed the visual language, weakened the rhythm or made the next scene impossible to believe. Keeping them would have made the film longer and less coherent.
I used several tools, not one
No single application handled every step well. I used Nano Banana for storyboards, ChatGPT to help develop the script, ElevenLabs for voiceover, Omni and Veo for video generation and sequencing, and CapCut to shape the final edit.
The tools supplied possibilities, not a finished film. The decisions about what to keep, what to regenerate, how long a moment should last and how the story should progress still had to be made. That was the part I found most demanding.
Editing gave the experiment a shape
The turning point was treating the clips as raw material rather than completed scenes. Once I stopped trying to preserve every successful generation, the edit became more disciplined. Pacing, transitions and sound mattered as much as the individual images.
It also changed how I think about creative AI. The ability to produce something plausible quickly is useful. It does not remove the need for taste, judgement and the patience to work through what is not yet right.
A recognition I did not set out to chase
In October 2026, Terminal Velocity received Outstanding AI Film recognition at the first edition of the New World AI Film Festival. I was pleased to see a personal experiment recognised for its artistic expression, originality and creative use of AI.
The recognition matters to me, but the more lasting lesson is how much of the creative work remains invisible. A few minutes on screen can represent hundreds of choices, rejected versions and corrections. AI changed how I could make this film. It did not make those choices for me.