Audiovisual · 2 of 3  ·  4:14 · 2026

The Big Tree

Three weeks of my own screen data, retold as a medieval anime battle.

Narrative direction AI-generated Coursework

The Big Tree — poster frame Watch on YouTube — no autoplay

4:14 · 2026 · 1080×1444 vertical · 30 fps

Author
Story, script, direction, edit — Eduardo Huanqui. Opening scene drafted by Claude, discarded.
Key frames
ChatGPT
Motion
Higgsfield (Seedance 2.0), then MidJourney
Score & audio
Suno and Higgsfield
Edit
CapCut
Data
Claude Code, local

Every image in this film is AI-generated. University coursework, MA Generative Design & AI, University of Europe. Not released commercially.

The idea

I didn't know I consume way more than I create — almost all my screen time is watching, scrolling. The one habit nobody's algorithm scheduled is the train ride, Duolingo and a book. And I fell for “the next tool will finally fix it.”

The constraint I set myself

The first scene came back from Claude as a cliché. From there I wrote every scene and every line of dialogue myself; the tools only rendered what was already written.

Three decision shots

01The characters crossed over

Rejected generation — the line landed on the wrong character Rejected
Final — the line on the boss Final

I wanted the boss's line on the boss. Higgsfield put it on my protagonist, in the enemy's gloves. Names in the prompt do nothing — it assigns roles by upload order. Third attempt, tested by hand.

02The crawlers work by number, not size

Rejected generation — one crawler the size of a building Rejected
Final — a swarm at true scale Final

The tool returned one creature the size of a building. Scale had to go into the prompt explicitly. From the version I am rebuilding now.

03Nothing symbolic was invented

Words surfaced from the data, rendered in the frame

Every symbol had to come from the conclusions of my own data, not from my taste — these words, and the colour of every weapon. The AI generated inside that rule.

AI in the workflow

Used

Higgsfield animated every key frame; when its credits ran out, MidJourney finished the motion, without sound.

Not used

My eight platform archives were consolidated by Claude Code on my own machine, not pasted into a chat model. The raw exports never left the laptop.

What it transfers to product work

Understand a human behaviour or circumstance, and with that information use my storytelling techniques to create a coherent system.