The final film is 4 minutes and 4 seconds long. The project behind those four minutes was much larger: character studies, historical references, locations, armor, horses, props, generations that almost worked, generations that failed for one tiny reason, and a deadline that eventually forced me to stop improving the film and release it.
The Son He Did Not Know is based on one of the most painful episodes in Ferdowsi's Shahnameh: Rostam meets the young warrior Sohrab in combat without knowing that Sohrab is his son. By the time the truth is revealed, it is too late.
I did not want to make a simple “AI version of Rostam and Sohrab.” The story is already powerful. The more interesting question was how to frame it so that a modern viewer could feel the tragedy rather than simply receive a summary of it.
The film started with Ferdowsi, not with the battle
The decision that shaped everything was to begin outside the legend, with Ferdowsi writing it. The film moves between the poet and the world emerging from his manuscript. That gave me a visual grammar: ink can become earth, a pen stroke can become a battlefield, an eye can become another eye, and a father in one timeline can echo a father in another.
It also created a second father-son story. Ferdowsi's relationship with his own son Mehrak became an emotional frame around Rostam and Sohrab. I was not trying to claim that the two histories are identical. I wanted the film to live in the space between them: the private cost of distance between a father and a son.
Generating a shot is easy. Keeping a film alive across shots is not.
Generative video can give you a beautiful frame quickly. A film asks a different question: is this still the same person five shots later? Is the armband on the same arm? Is the horse still Rakhsh? Does the armor make sense after a cut? Is the sun coming from the same direction? Did a fortress suddenly appear on a battlefield where no fortress existed?
Those details became the real production work. We built recurring references for Rostam, Sohrab, Ferdowsi, Tahmineh, Rakhsh, the armband and locations. Prompts gradually stopped looking like image prompts and started looking like direction notes: camera position, lens feeling, blocking, continuity locks, sound, negative constraints, the exact state at the beginning of a shot and the exact state the following shot needed at the end.
The fight was especially unforgiving. Horses multiply continuity problems. Weapons multiply them again. Wrestling adds hands, body contact, weight, balance and gravity. If one element drifts, the audience may not know why a shot feels wrong — but they feel it.
The best effects were transitions with a reason to exist
I became less interested in spectacular transitions as the project went on. The useful ones were the transitions that carried meaning.
The ink transition works because Ferdowsi is writing the world we enter. The eye transition later works because it connects Rostam's grief to Ferdowsi's without a paragraph of explanation. A transition is strongest when removing it would damage the story, not just the style.
That became a useful rule for the entire edit: if a shot was only there because it looked expensive, it had to justify its time. Four minutes is not long. Every beautiful image competes with the story for seconds.
What the tools were actually good at
Higgsfield became the production environment, with Seedance doing much of the generative video work and Kling used for some tightly controlled continuations. Reference images and reusable character elements mattered more than I expected. They did not solve continuity automatically, but they made continuity something I could direct rather than simply hope for.
The final assembly happened in CapCut. That matters because generation and filmmaking are not the same step. The edit decides duration, rhythm, which take survives, where sound begins before an image, where silence is more useful than music, how long a reaction is allowed to live, and whether a transition is emotional or merely decorative.
The deadline was part of the filmmaking
The last stretch was messy in the way real deadlines are messy. There were still shots I wanted to improve. There were still planned seconds that could have been added. There were versions that were technically cleaner and versions that felt better. At some point the question changed from “Can this be better?” to “Is this a film I am willing to release?”
That was an important moment. Generative tools create an almost infinite temptation to regenerate. There is always another prompt, another seed, another model, another correction. Without a deadline, iteration can become avoidance.
The competition closing forced a final answer. The 4:04 export went out with its credits, watermarking, sound and the choices that survived the edit. I can already see things I would change. That is normal. The alternative is a perfect film that never leaves the timeline.
What I would carry into the next AI film
Characters, costume states, props, geography, light direction and the exact meaning of every recurring object should exist before expensive generation begins.
The strongest transitions in this film came from story structure, not from searching for an effect after the fact.
A shot needs handles, a readable beginning and an ending that can actually connect to something. Beautiful isolated clips are not enough.
When the story reaches grief, recognition or silence, a small eye movement can matter more than an army.
Infinite regeneration is not the same as refinement. A finished imperfect film teaches more than an unfinished perfect plan.
Now it is outside the edit
The competition is closed and the film is public. That changes the relationship to it. For weeks it was a collection of problems I could still solve. Now it is one 4:04 object that someone else can watch without knowing any of those problems.
That is probably the healthiest point in any creative project. The audience does not owe the production process sympathy. They only see what made it to the screen.
So now we hope it connects — and, yes, we can hope it wins.