[USER INPUT — REQUIRED AT TOP]

DISH: ข้าวต้มมัด
IMAGE: {แนบภาพอาหารของคุณได้เลย — optional แต่ช่วยให้ label แม่นยำขึ้นมาก}

DISH_NAME_DISPLAY: {Optional}
TAGLINE: {Optional}
BACKGROUND_COLOR: {Optional}

LABEL_LANGUAGE:  Thai
LABEL_STYLE: {UPPERCASE SANS-SERIF (default) / Thai clean modern sans}
CUT_STYLE: Center
STEAM_EFFECT: Auto
GARNISH_MODE: Auto
SHADOW_INTENSITY: Auto

[PROMPT — NANO BANANA PRO]

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
STEP 0 — INPUT RESOLUTION (EXECUTE FIRST, BEFORE ALL OTHER STEPS)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Determine the active DISH identity using the following priority logic:

CASE A — IMAGE provided + DISH name provided (highest accuracy mode):
  → Use the uploaded image as the visual reference source of truth.
  → Use DISH name to confirm and disambiguate identity.
  → Cross-validate: if image and DISH name appear to conflict, 
     prefer the image and note the discrepancy internally.
  → Extract from image: exact layer structure, visible colors, 
     surface texture, translucency, and any garnish present.
  → Lock this as the definitive DISH_IDENTITY for all steps below.

CASE B — IMAGE provided, DISH name is empty (auto-identify mode):
  → Analyze the uploaded image carefully.
  → Identify the dish by: shape, color palette, visible layers, 
     texture, cooking style, and any culturally specific visual cues.
  → If the dish is Thai: cross-reference against known Thai desserts 
     and savory dishes — use the most confident match.
  → If confidence is high (>85%): proceed automatically with identified name.
  → If confidence is low: flag the top 2–3 candidates internally 
     and select the most visually consistent one; note uncertainty 
     in the auto-generated tagline if needed.
  → Extract from image: layer structure, colors, textures, component 
     boundaries — use these as the ground truth for labeling.
  → Lock identified dish as DISH_IDENTITY.

CASE C — DISH name provided, no IMAGE (name-only mode):
  → Use DISH name as the sole source of identity.
  → Reconstruct authentic layer structure from culinary knowledge 
     of {DISH} only.
  → Apply standard Pre-Render Label Audit (see below) based on 
     known recipe structure.
  → Lock DISH name as DISH_IDENTITY.

CASE D — Neither IMAGE nor DISH name provided:
  → Halt. Return message: 
     "กรุณาระบุชื่ออาหาร (DISH) หรือแนบภาพอาหาร (IMAGE) อย่างน้อยหนึ่งอย่าง"

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
STEP 1 — PRE-RENDER LABEL AUDIT
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Using DISH_IDENTITY resolved in Step 0, internally enumerate ALL 
visible layers and components that will appear in the cross-section.

Source of truth hierarchy (highest to lowest):
  1. Uploaded IMAGE (if provided) — visual evidence beats all
  2. Known authentic recipe structure of the identified dish
  3. General culinary logic for the dish category

Rules:
1. List only components that ARE part of the authentic recipe.
2. List only components that WILL be visually distinguishable in 
   cross-section (distinct color, texture, or clear boundary).
3. Assign each component a pixel-region: top / upper-mid / center / 
   lower-mid / bottom / outer shell / inner core / filling.
4. If two adjacent layers appear nearly identical in texture AND color, 
   merge into one label. Do not split into false sub-layers.
5. If a component exists in the recipe but is NOT visible from the 
   cut face (e.g., exterior glaze invisible from inside), do not 
   label it as an internal layer — only label it if visible on 
   the outer edge of the cross-section.
6. If IMAGE was provided: match observed layers in the image first, 
   then verify against recipe knowledge. Image takes priority.
7. Zero tolerance for invented components. No label may name 
   an ingredient not present in the authentic dish.

Output of this audit: a locked ingredient-layer map used for 
all labeling in the final render.

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
RENDER INSTRUCTIONS
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Create a perfect cinematic cross-section food photograph of 
DISH_IDENTITY, sliced cleanly and precisely to reveal the internal 
structure confirmed in Step 1. The cut face must face the camera 
straight-on, centered and symmetrical, like a scientific flavor 
diagram or architectural material study.

If IMAGE was provided: the cross-section render must be visually 
consistent with the uploaded reference — match the real color tones, 
proportions, surface finish, and layer arrangement observed in the image. 
Do not invent a different version of the dish.

Auto-fill logic (CRITICAL):
- DISH_NAME_DISPLAY: use user value if provided. If empty, 
  auto-generate from resolved DISH_IDENTITY.
  - LABEL_LANGUAGE = English: clean uppercase English display name.
  - LABEL_LANGUAGE = Thai: correct Thai name, perfectly spelled, 
    zero typo tolerance, correct tone marks and vowel placement.
- TAGLINE: use user value if provided. If empty, auto-generate 
  a short premium editorial subtitle matching dish type and structure.
- BACKGROUND_COLOR: use user value if provided. If empty, 
  auto-select a flat matte color with strong contrast against the dish.
  - Dark matte for pale dishes; lighter matte for dark dishes.
  - No neon. No colors too similar to the dish's dominant hue.

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
LABELING SYSTEM — ACCURACY-LOCKED
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

- Use ONLY the ingredient-layer map from Step 1. No additions.
- Thin horizontal label lines extend to the RIGHT of the cut face only.
- Each line originates from the exact visual center of its component's 
  region — not estimated, not floated.
- Lines must not cross each other.
- Lines must not touch or overlap the dish silhouette.
- Label count = exact number of confirmed distinct layers. No more, no less.
- Labels spaced evenly along right margin to prevent collision.
- LABEL_LANGUAGE = English: UPPERCASE name + smaller italic descriptor.
- LABEL_LANGUAGE = Thai: clean modern Thai sans-serif, legible, 
  correct spelling, correct tone marks, no broken characters, 
  no romanized substitution.
- Duplicate names forbidden — use positional qualifiers if same 
  ingredient appears at multiple levels 
  (e.g., ชั้นกะทิบน / ชั้นกะทิล่าง).

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
COMPOSITION, LIGHTING & TEXTURE
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Composition:
- Hero object centered in frame, cross-section facing camera directly.
- Natural food structure only — no invented impossible layers.
- Premium, editorial, highly intentional.

Lighting:
- Flat matte resolved background color.
- Dramatic side lighting from the left.
- Single warm spotlight focused on the cut face.
- Strong dimensionality and shadow separation.
- Shadow style follows {SHADOW_INTENSITY}.
- Mood: architectural food photography meets scientific illustration.

Texture realism (critical):
- Hyperreal textures appropriate to the dish: moisture, pores, 
  air pockets, glaze, gelatin translucency, coconut texture, 
  sticky rice grains, pandan layers, fibrous strands, cream layers, etc.
- Realistic density, softness, gloss, and translucency where appropriate.
- Steam only if natural to the dish AND STEAM_EFFECT ≠ "none".
- No plastic, fake CGI, or over-stylized appearance.
- No surreal or fantasy materials.

Authenticity:
- Full respect for Thai dessert identity if dish is Thai.
- Culturally accurate structure, color cues, and ingredient logic.
- No western substitution unless explicitly specified.

Style:
- Symmetrical, precise, controlled composition.
- Scientific illustration clarity + Michelin-star plating aesthetics.
- Subtle Wes Anderson-inspired symmetry (not cartoonish).
- 4K, tack sharp, ultra-detailed, clean premium food editorial render.
- Optimized for Nano Banana Pro image generation.

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
NEGATIVE CONSTRAINTS (HARD RULES)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
- No messy or torn cut edges.
- No melted/deformed structure unless dish naturally melts.
- No random garnish clutter.
- No incorrect, misspelled, or hallucinated text anywhere.
- No extra plates, hands, utensils, or props unless requested.
- No duplicate labels.
- No invented ingredients not in the authentic dish.
- No label pointing to a region not visually matching the named component.
- No floating label lines unanchored to a real visible layer.
- No more labels than visually confirmed distinct layers.
- If IMAGE provided: do not ignore the reference — 
  the render must be visually faithful to the uploaded photo.