<aside> 🎭
How the look was found, and how the cast was locked before a single shot was generated. Everything on this page was generated by us. The moodboard paintings that guided it belong to their artists and are not reproduced; they are credited on the main page.
</aside>
One scene held constant: the Austrian holding the amber bottle to a window, warm gold. Only the method changes.
A · named the artists, no references
B · described the medium in detail, no references
C · framed as a film still, no references
D · hybrid style words, no references
E · the same words plus three reference paintings · picked as the target
F · minimal words plus the same references
What it proved. Every frame without references fell into a generic preset: too realistic (A), hard outlined TV animation (C), muddy impasto (D). The references carried the style, and words about the medium fought them. Naming the artists while their work was attached pulled toward realistic concept art.
A words only recipe. We measured frame E and wrote its look down as text: line weight, fills, blush, grain, a precise colour ramp. It held consistency across scenes, and every frame came back as a coloured storyboard instead of a painting.
words only, gold
words only, cold
Using the target frame itself as the reference. It taught composition, not style. Every new shot collapsed into a person against a flat wall, and one even came back with a painted border.
target frame as reference, magenta
target frame as reference, Bangkok
Frame E's real recipe (the same three references and its prompt structure) on the opposite scene: a night stall in 1970s Bangkok instead of a sunlit office, two men instead of one, amber instead of gold. The paint held.
The transfer test: same references, same frozen blocks, a different world
Then this frame came back with identical paint and was called AI slop.
Centred, flat, a bulb in the middle of the frame, a man stirring a pot
Everything that differed from E was camera. Face size, placement, depth, where the light comes from, the value range, and whether the character is thinking or doing a task. That is why the prompt recipe has a frozen camera block.
Three sheets were approved before any shot was generated. Three views each on warm cream: full length, chest up front, chest up profile. Flat, even sheet light.