Worked Example — Anime / 2D → Live-Action Real

Original worked example by jnMetaCode (MIT). Applies the 5-stage structure to the "anime character as live-action" translation — a hugely searched genre that fails in two opposite ways: cheap cosplay, or plastic CG. The trick is translating the medium (cel shading → real materials + physics) while preserving the silhouette that makes the character readable. This template leans hardest on Rule 7 — the recognisable trait-bundle is exactly what IP filters catch.

Concept: a stylised "wind swordsman" concept rendered as a real person in a real world. Describe traits, never an IP name.

Further reading (inspiration, not copied — all rewritten in our structure): style-transfer / anime categories in zhangchenchen/awesome_sora2_prompt (MIT).

中文版 →

Variables you need to define first

Variable	This example	Swap for…
`{{silhouette}}`	tall, lean, long coat, single shoulder guard	the readable shape that IDs the character
`{{signature}}`	a streak of pale-blue hair, a chipped jade pendant	the 2–3 traits that say "it's them" — no IP name
`{{palette}}`	indigo + worn steel + pale blue	the character's colour identity
`{{world}}`	a rain-wet stone bridge at dusk	the scene translated to a real location
`{{prop}}`	a worn, real-steel curved blade	the iconic item, made physical

The complete prompt (copy-paste ready)

1 · Core theme

anime character reimagined as live-action | photoreal materials and physics | preserved silhouette + signature traits | cinematic grade | no cosplay cheapness, no CG plastic, no cel shading

2 · Character & scene

Character (translate the medium, keep the silhouette): a real human with the readable {{silhouette}} and {{signature}} that identify the character — but rendered fully real: real skin with pores and faint texture, real hair (individual strands, not a helmet), real woven fabric with weight and wrinkles, {{prop}} as actual worn metal. Drop every 2D cue — no cel shading, no flat anime eyes, no impossible hair gravity. Eyes and proportions human, not enlarged.

Imperfections (this is what separates "real" from "cosplay"): skin sheen and stray flyaway hairs, fabric creased and a little worn, a real scuff on the {{prop}}, faint sweat. Cosplay looks new and posed; real looks lived-in.

Material-translation checklist (swap each 2D cue for its real-world equivalent — this is the core move):

cel-shaded skin → real skin with pores + subsurface scattering
flat anime hair → individual strands that move with the air
enlarged flat eyes → human-proportioned eyes with real catchlights
drawn ink outline → no outline; real depth, shadow and contact
flat colour fill → woven fabric with weight, real metal with wear

Scene: {{world}}, {{palette}} carried into the real lighting.

3 · Atmosphere & quality

Shot on ARRI Alexa with vintage primes, cinematic grade in {{palette}}. Real-world physics on everything — cloth swings and settles with weight, hair moves with air, light wraps real skin. Filmic grain. The goal is "a still from a live-action film adaptation," not a render or a con photo.

4 · Camera rules

A slow, deliberate reveal — push or arc that lets the real materials read.

Breathing: "Handheld shot. Throughout, maintain an extremely subtle, breath-like camera float to enhance presence."
Sound: No score. Production audio only — rain on stone, fabric shift, the scrape of {{prop}}, ambient world tone.

5 · Storyboard (3 beats, ~10s)

0–3s · Detail (sell the realism in close-up)
Action: Macro on real skin / a flyaway hair / the woven fabric / the
        scuffed {{prop}} — the textures that say "this is real."
Camera: Slow push on the detail, shallow focus.

3–7s · The silhouette (now we recognise them)
Action: Pull to reveal the full {{silhouette}} in {{world}} — the shape
        and {{signature}} make the character readable, but every surface
        is real.
Camera: Slow arc, breath-float, rain and air moving cloth and hair.

7–10s · A held look (restrained close)
Action: A small, human expression — a breath, a glance, no anime pose.
Close:  No power-up glow, no logo, no flashy hero stance. Just a real
        person who happens to be that character, standing in real rain.

Negative prompt (Seedance / Kling — paste into the dedicated field)

cel shading, anime shading, flat 2D look, cartoon, enlarged anime eyes, impossible hair gravity, plastic CG skin, doll-like face, cosplay wig look, cheap costume, brand-new unworn fabric, oversaturated colors, glossy render, video-game look, watermark, text overlay, logo, distorted face, extra fingers, melting/morphing geometry, frame flicker, jarring hard cuts, lifeless locked-off camera

Why it's built this way

Translate the medium, keep the silhouette. The character reads because of shape + 2–3 signature traits, not because of cel shading. Swap every 2D cue for a real-material equivalent (cel → skin pores, flat hair → strands, anime eyes → human eyes) and keep the silhouette — that's the whole craft of "2D → real."
Imperfection is the cosplay/real divide (Rule 5). New, clean, posed = cosplay. Worn fabric, scuffed prop, flyaway hair, real skin = live-action film. The wear is what sells it.
Heavy on Rule 7 — describe traits, never the IP. A recognisable trait-bundle (the name, the exact iconic outfit) is exactly what Sora's lookalike filter and Seedance's IP filter catch. Use {{silhouette}} + {{signature}} as descriptions, not the character's name.
Restrained close (Rule 6). A human breath, not a power-up pose. Keeps it grounded as "film adaptation," not fan-art animation.

Usage: generate the detail beat (0–3s) first — if the skin, hair and fabric read as genuinely real there, the full reveal will hold. If it comes out plasticky or cel-shaded, push the negative field hard. Sanitise any IP-recognisable wording before running on Seedance/Sora.

Model: Veo 3 and Sora 2 give the best photoreal skin/material translation; both have strict lookalike filters, so keep it trait-described, not named. Kling and Seedance also work — Seedance especially will reject a named IP outright, so describe only.