Opera
How It Was Made
From a feeling to a visual system
Natural-language prompts rarely reproduce the image in your head. For Opera, I turned one emotional reference into a Skill — a reusable set of visual rules — so every new story comes out in the same visual language.
01
Start from a feeling, not a frame
Chen Kaige’s Farewell My Concubine was the starting point, but I didn’t want to recreate its shots. What I wanted was its emotion: grief, hysteria, a performer losing control.
02
Explore with abstract words
In Midjourney, using my personal style profile, I prompted with abstract ideas like life and death and parting instead of literal descriptions. Abstract words lead to images you would never think to ask for.
Narrowing hundreds of generations down to the one closest to the image in my head.
03
Turn one image into rules
I gave that frame to Codex to break down its color, lighting, composition, costume and set. The result was a fixed vocabulary: the style’s DNA. Extreme close-ups set against extreme wides were chosen on purpose, to fit my fast-cut editing.
04
Package it as a Skill, then stress-test it
The rules became one Skill. Each test starts with a spiritual core, such as “a person’s greatest enemy is himself.” Codex writes the prompt, Seedance 2.5 renders it on IMA Studio’s infinite canvas, and I revise the Skill wherever the result drifts. It stabilizes after two or three rounds.
05
New story, same visual DNA
Once the Skill is stable, I only tell the AI what to shoot. What it should look like is already built into the Skill. The stage, the tragedy and the nightclub are three worlds with one visual language.
Columns: stage · tragedy · nightclub. All generated from the same Skill.
← Back to AI Work