STYLE‑FRAME PROTOTYPEExploratory style‑frames and short motion tests for a hybrid live‑action/animation series, achieving a painterly, “Flee‑like” realism that can be re‑generated from a single prompt —no reference image required—
for clear legal ownership.
When the U.S.‑based AI studio offered me to help shape the visual identity of a new upcoming animated docu‑series, the brief sounded simple:
“To make it look more like the FLEE reference, unique look away from anime.
The man look less Asian and more Russian”
The main challenge was keeping the characters grounded in realism while every diffusion model kept drifting into anime. The solution was to embed explicit negative cues in the prompt (“no manga, no hard anime eyes”) and iterate until the output stayed consistent across shots.
Workflow
This exercise proves that a distinct, emotionally grounded look can be locked in by prompt alone —
a crucial precedent for studios navigating AI IP constraints.
I needed to test whether this style could be recreated without any reference images—using only a text prompt. Here’s the pipeline of the “reverse process”:
Prompt reconstruction with GPT-o3 — I asked the model to write a prompt that perfectly described my finished style-frame, keeping the exact Midjourney seed.
Midjourney refinement — I iterated on that prompt inside MJ until I arrived at a stable, final version.
Photoshop finishing pass — I applied Color Transfer and subtle Adjustment Layers in Adobe Photoshop to polish the frame.
Midjourney
Photoshop finishing
Result:
The style can be regenerated end-to-end from text + seed alone,
no image reference required — a clean, studio-friendly IP pipeline.
