Skip to content
CONCEPTS
4 min readЧитать на русском

A.O.C. Application Skills — Three Composable Skills for Applied Anchor/Optics/Chemistry

A designer writes a generation prompt for the seventh visual in a campaign and notices it doesn't feel like the other six. The anchor wandered — the product's silhouette is slightly off. The lighting collapsed into "moody" again because that's the word the LLM reached for. There was no scaffolding telling the prompt what kind of decisions it was making, so every shot reinvents its own structure. By shot ten the campaign reads as ten different photographers.

What follows is the A.O.C. Application Skills pack: three composable Claude Code skills that apply the A.O.C. framework — Anchor / Optics / Chemistry — as the default spine of every generation prompt, with no opt-in required.

The three skills cover the full path from intent to production-ready prompt. aoc-auto is the safety belt — it fires on any prompt request and forces the three-axis layout even when the user doesn't ask. generation-prompts takes a shot from a shot-list and emits a model-specific prompt (Nano Banana 2, Seedance 2, MJ, Flux) with A.O.C. + typed references baked in. typed-key-visual is the heaviest of the three: a typed-reference composition pipeline that classifies reference images by role (identity / background / style / color) before assembling the A.O.C. prompt. Each is verbatim below.

How the three relate

| Stage in the pipeline | Primary skill | Reads concept | |---|---|---| | Any prompt request — default backbone | /aoc-auto | aoc-framework-paper | | Shot from shot-list → model-specific prompt | /generation-prompts | aoc-framework-paper + typed-reference-composition | | Reference images → key visual prompt | /typed-key-visual | typed-reference-composition + A.O.C. |

/aoc-auto is the floor — it can run on a one-line brief. /generation-prompts adds the seven-element shot structure on top. /typed-key-visual adds the reference-typing pipeline before the prompt is built. They compose downward: a typed-key-visual run includes A.O.C.; a generation-prompts run includes A.O.C.; aoc-auto is the standalone minimum.

Skill 1 — /aoc-auto

---
name: aoc-auto
description: Automatically apply the studio's A.O.C. framework (Anchor / Optics / Chemistry) to any generation prompt request. Triggers on any user request that asks for a generation prompt — for Nano Banana 2, Seedance 2, Midjourney, Flux — even if A.O.C. is not explicitly mentioned. The rule is "safety belt": the designer doesn't choose each time.
---

A.O.C. Auto

This is the default framework for any generative prompt in the studio. You apply it always, even when the user doesn't name it.

When to trigger

Any generation prompt request:

  • "Write a prompt for Nano Banana 2 …"
  • "Make a Seedance prompt for a clip"
  • "Build an MJ prompt: …"
  • "Describe the visual as … for the generator"

If the user explicitly asks to skip A.O.C. — yield, but ask the reason first; consistency is usually what gets lost.

Protocol

  1. Read the A.O.C. framework concept page. If missing — apply the three-axis base structure and at the end suggest reconstructing the concept page.

  2. Assemble the prompt on three axes:

    ANCHOR — what cannot change in frame:

    • Subject identity (face, product, brand elements)
    • Immutable geometry
    • Branded details

    OPTICS — how the shot is taken:

    • Light (source, direction, hardness, temperature)
    • Camera (focal length, position, movement)
    • Composition (rule of thirds, symmetry, leading lines)
    • Depth of field

    CHEMISTRY — how the shot feels:

    • Emotion (1–2 adjectives)
    • Palette (3 concrete colors)
    • Tonality (contrast, tone mapping)
    • Materiality (textures, surfaces)

  3. Emit two things:

    • Layout — table by three axes (for logic transparency)
    • Final prompt — flowing prose, ready to paste into the generator

Output format

## A.O.C. layout

**Anchor:** …
**Optics:** …
**Chemistry:** …

## Final prompt (Nano Banana 2)

<one flowing paragraph>

Live example

Request: "Prompt for a perfume visual: glass flacon, morning light"

Response:

A.O.C. layout

Anchor:    100ml clear glass perfume flacon, cylindrical, 
           black cap, gold foil label accent (immutable)
Optics:    soft diffuse morning sunlight from the right (5200K), 85mm lens, 
           f/2.8 aperture, shallow DOF, center-left composition
Chemistry: calm, premium, clean; palette: warm beige / 
           cool white / pearl gold; materials — glass, linen, marble

Final prompt
Premium perfume bottle product shot — clear cylindrical glass flacon, 
black cap, gold foil label accent, softly lit with morning sun coming 
from right side through sheer linen curtain, shallow depth of field at 
f/2.8, 85mm lens, composition favors center-left, subject resting on 
pale warm beige marble surface. Color palette: warm beige, cool white, 
pearl gold. Mood: calm, quiet luxury. Clean minimal still life.

Skill 2 — /generation-prompts

---
name: generation-prompts
description: Produce production-ready image and video generation prompts from a shot list, applying the studio's A.O.C. framework automatically. Use when the user has a shot list or scene description and asks for "prompts for Nano Banana 2", "Seedance prompt", "build a prod-ready prompt" or similar.
---

Generation Prompts

Turn a shot into a prod-ready prompt for the generator. The studio frameworks (A.O.C., typed-reference) should always be used — always lean on them.

Sources — MUST read

  1. The A.O.C. framework concept page — Anchor / Optics / Chemistry
  2. The typed-reference composition concept page — reference typing (identity / bg / style / color)
  3. The shot from the active project (7-element structure)

If a file is missing — don't guess. Ask: "Do you want me to stub the A.O.C. concept first, or skip it and apply the base structure?"

For image (Nano Banana 2 / MJ / Flux)

Three sections in sequence:

1. A.O.C. layout

ANCHOR:   <identity: who/what, immutable — face, product, brand>
OPTICS:   <light, camera, composition, focal length, depth>
CHEMISTRY:<emotion, palette, tone, atmosphere, materiality>

2. Typed references (if any)

identity ref: <path/description>
bg ref:       <...>
style ref:    <...>
color ref:    <...>

3. Final prompt

One flowing paragraph, optimized for the specific engine (Nano Banana 2 loves descriptive adjectives; Flux prefers shorter phrases).

For video (Seedance 2)

Motion-only format. Don't describe shot content — it's already set by first/last frame. Motion only:

  • Camera: <type + speed + direction>
  • Subject motion: <one clear action>
  • Environmental motion: <wind / smoke / reflections — optional>
  • Duration: <seconds>

Emit as compact JSON (following the Seedance 2 cinematic video prompt engineer format if available).

Censorship hygiene (Seedance 2)

If the shot contains people:

  • No age words (teenager / young / mature)
  • Role over age ("the dancer", not "the young dancer")
  • Tag refs via @Image if present

Saving

Ask: "Save the prompts to the project folder or file as a finding in a shared prompt library?"

Skill 3 — /typed-key-visual

---
name: typed-key-visual
description: Build a production-ready key visual through a typed-reference composition pipeline. Use when the design lead wants to create a key visual, poster, or moodboard frame from reference images. Classifies references by role (identity / background / style / color), assembles an A.O.C. prompt, and produces a final prompt ready for Nano Banana 2 or similar generator.
---

Typed Key Visual

Pipeline: scattered refs → typed composition → A.O.C. prompt → final visual.

Sources

  1. The typed-reference composition concept page — typing rules
  2. The A.O.C. framework concept page — prompt structure
  3. Reference images supplied by the user or from the project's asset library

Step 1 — Classify refs

For each image, assign a role:

  • identity — who/what is in frame (face, product, object with recognizable geometry)
  • background — environment, scene, space
  • style — artistic manner, render, era
  • color — palette and light temperature

Emit a table: | File | Type | Why | |---|---|---| | ref-01.jpg | identity | … | | ... | ... | ... |

If a type is absent — ask whether it's needed or can be skipped.

Step 2 — A.O.C. layout

Based on the active project context and typed refs, assemble:

ANCHOR:   <from identity ref + project subject>
OPTICS:   <lighting and composition from background + style ref>
CHEMISTRY:<palette from color ref + emotion from brief>

Step 3 — Final prompt

One flowing paragraph, 80–120 words. Include:

  • Anchor subject (immutable)
  • Environment and light (optics)
  • Palette and tonality (chemistry)
  • Output format (aspect ratio, resolution if applicable)

End with explicit ref tags in the generator's format (e.g. @identity.jpg, @bg.jpg).

Step 4 — Visualization

If Claude Design mode is available — hand the prompt to it. Otherwise — format for manual paste into Nano Banana 2 / Flux / MJ.

Step 5 — Fixation

Ask: "Save this pipeline as a finding?" If yes:

  • Prompt → project prompt library
  • Ref typing → short section in the project's key-visuals document
  • Entry in the project log

What you DON'T do

  • Don't mix refs "Pinterest-style". Each ref plays one specific role.
  • Don't invent ref types beyond the four canonical ones. If a 5th is needed (e.g. texture) — propose extending the concept first.

See also