---
reliability: 4.4 # 5 − .3 stills gen − .6 i2v pan wave (optional but default) − .2 music + .5 proven E2E showcase (THE GATE, 17/17 jobs first try, ~$6.56); finishing chain deterministic ffmpeg; stills-only mode scores higher in practice
---

# Fusion architecture walk — agent playbook ⭐ flagship

> Paste this whole file (with the BRIEF filled in) into Claude cowork, chat, or Code. **One building footprint. N cross-cultural fusions.** Build a game-design / concept-art location bible where the SAME architectural object (a temple, a palace gate, a town square, a citadel) is rendered through N curated east-meets-west cultural fusion lenses — silhouette held constant, ornament + material + setting variable. Useful for level design, location pitches, and alt-history world-building reels.

## What you'll get

A location bible structured the way a AAA game's concept-art lead presents a hero location for executive approval — but with cross-cultural breadth no concept artist could deliver in a sprint:

- **A single architectural object held constant** — same silhouette, same scale, same camera angle, same time-of-day across every cell. The building footprint is the through-line; the cultural fusion is the variable that swaps ornament + material + adjoining setting.
- **N fusion cells** (default 8 — heavier per-cell visual budget than character bibles, so 8 is the sweet spot) — each cell pairs one Eastern architectural tradition with one Western, with a defensible fusion grammar (which element gets blended where: silhouette vs roofline vs ornament vs material vs surrounding setting).
- **Cinema 16:9 widescreen** (1920×1080) — establishing-shot framing, the format a game pitch deck or art-bible PDF uses.
- **Optional pan-across reel** — each location gets a slow horizontal pan via `kling-v3-i2v` (8s, `resolution:"auto"`, "slow horizontal camera pan left-to-right across the structure, the rest of the scene held still"), stitched into a 64s walkthrough reel with atmospheric score.
- **One unified score** — atmospheric instrumental (organ pad + soft percussion + distant bell), the score of arrival at a sacred / civic / monumental place.
- **Showcase HTML page** — single shareable page with the 8-cell grid + (optional) reel + cell-by-cell fusion breakdown + pipeline trace + cost summary.

**Time**: ~12–18 minutes wall clock. **Your attention**: ~7 minutes across 2 gates. **Cost**: $2 stills-only / $9 stills + reel + score for 8 cells. *This is the location-bible flagship* — the playbook for executive-grade game / film alt-history concept-art decks.

## Tell the agent about the building + the fusions

```yaml
building_archetype:        # e.g. "ceremonial gateway" / "harbor lighthouse" / "town square + clocktower" / "citadel approach" / "monastery cloister"
building_silhouette:       # The verbatim block repeated in every cell prompt to hold silhouette + scale constant. Example:
                           #   "A monumental ceremonial gateway / arch, three stories tall, central archway 8 meters wide flanked by
                           #    two solid pylons. Stone construction with raised platform base. Viewer stands 30 meters back at
                           #    ground level, looking up at slight 5° angle. Time of day: golden afternoon."
composition_constant:      # Verbatim composition rule across all cells. Example:
                           #   "Cinematic establishing shot, 35mm equivalent, building occupies 60% of frame width. Golden hour
                           #    light from upper-right at 35°, long warm shadows pooling toward viewer. Atmospheric haze in distance.
                           #    Painterly photoreal museum-grade architectural rendering. NO readable text anywhere in frame."
cell_count:                # 6 (tight) / 8 (default) / 12 (extensive). 1 cell per fusion pairing.
cells:                     # Ordered list of N east-meets-west fusion pairings. Each item:
                           #   - east_lens:        Forbidden City ceremonial gate (yellow glazed tile, red lacquer columns, cloud-motif eaves)
                           #     west_lens:        Roman triumphal arch (carved relief panels, marble inscriptions)
                           #     fusion_grammar:   Chinese gate silhouette × Roman relief carving + marble pylon facing
fusion_quality_guards:     # Default ON:
                           #   - No readable text on inscriptions / banners / plaques (text rendering failure)
                           #   - No "cyberpunk samurai temple" stock-mashup phrasing
                           #   - Same camera angle + time of day + atmospheric density in every cell
                           #   - Single-building focus (no companion structures, no crowd extras)
                           #   - Architecturally plausible (load-bearing arch where an arch goes; no floating wood beams)
include_reel:              # true (default) — fire N × kling-v3-i2v slow pans + 1 minimax-music, total ~$9 for 8 cells
                           # false           — stills-only, total ~$2
music_brief:               # Score mood arc. Example:
                           #   "Atmospheric instrumental, slow organ pad + soft percussion + distant single-bell tolls,
                           #    building from contemplative arrival to civic majesty across 64s. No vocals."
walk_slug:                 # kebab-case for output filenames + showcase page. e.g. the-gate / the-citadel / the-cloister.
```

---

## Quality controls (auto-applied)

The closed quality loop that separates this from "a folder of fantasy architecture":

- **Brief approval first.** Before any MCP fires, the agent shows you the building silhouette spec + the N cell pairings + the constants for review.
- **Silhouette + scale constant.** The `building_silhouette` block is repeated verbatim in every cell prompt, biasing gpt-image toward shape-stability. Without it, each cell builds a different structure.
- **Camera + light constant.** Same eye-level, same focal length, same golden-hour direction across all cells. The viewer reads the gallery as one building seen through 8 cultural lenses, not 8 unrelated buildings.
- **Curated pairings.** Each cell uses "X silhouette × Y ornament + Z setting" fusion grammar. "Forbidden City × Roman triumphal arch" is curated; "Asian × European" is kitsch.
- **No readable text guard.** Every prompt explicitly says "NO readable text" because misspelled fake-Chinese / fake-Latin inscriptions would break the architectural credibility.
- **Architectural plausibility.** Prompts request load-bearing elements where load-bearing elements go (no floating wood beams under a stone roof; no glass dome on a wattle-and-daub frame).
- **gpt-image as the renderer.** Best stylization fidelity for named architectural traditions — recraft would render generic fantasy-castle for any prompt.
- **Mid-render check.** After Wave 1 lands, the agent surfaces a contact-sheet for your review. You can re-fire any off-silhouette or implausible cell before the i2v wave fires.
- **Slow pan only on i2v.** Camera pans say "horizontal pan only, structure held still, no roof tile or banner motion". Aggressive zoom or weather animation breaks the establishing-shot aesthetic.

> **Why one-building-eight-worlds works:** game design + film concept art often need "what if this culture had built it" alt-history pitches. Doing this by hand requires deep architectural training in every culture — months of study per cell. The fusion playbook collapses it to one afternoon: pick the silhouette, pick the pairings, run. The same building reads as the same place across radically different ornament + material vocabularies, which makes the pitch defensible to a creative director rather than read as random AI fantasy.

## Output structure

```
public/chapters/<walk_slug>/
  cell-01-{east}-{west}.png            ← 8 keyframe establishing shots
  ...
  cell-01.mp4                          ← (optional) 8 i2v slow pans
  ...
  music.mp3                            ← (optional) unified atmospheric score
  <walk_slug>-walkthrough.mp4          ← (optional) 64s pan reel
public/chapters/<walk_slug>-example.html  ← showcase page
public/case-studies/<walk_slug>.html      ← optional showcase entry
```

## Reproduce via MCP

1. `brand_kit_create({ palette, keywords })` — optional; keeps the cells on one palette.
2. **Wave 1 — N stills (parallel):** N × `create_media({ action: "generate", model_override: "gpt-image", aspect_ratio: "16:9", prompt: building_silhouette + composition_constant + cell[i].fusion_grammar })`. The silhouette + composition blocks repeat verbatim in every cell — that's what holds the building constant.
3. **GATE — contact-sheet check.** Re-fire any off-silhouette / implausible cell with `create_media action:"restyle"` before the i2v wave.
4. **Wave 2 — i2v (optional, parallel):** N × `create_media({ action: "animate", model_override: "kling-v3-i2v", source_url: cell[i]_url, resolution: "auto", duration: 8, prompt: "slow horizontal pan only, structure held still, no tile/banner motion" })`.
5. **Score:** `create_media({ action: "music", prompt: music_brief, lyrics_prompt: "[Instrumental]" })`. ALWAYS pass `lyrics_prompt:"[Instrumental]"` — else the music cap sings the brief text.
6. **Reel:** `create_media model_override:"ffmpeg-concat"` (stitch pans) → `create_media model_override:"ffmpeg-mux"` with `audio_fill:"loop"` (score under the FULL 64s reel) → `create_media model_override:"ffmpeg-export"` (16:9, 1920×1080).

The example for this playbook is **THE GATE** — a single ceremonial gateway rendered through 8 east-meets-west alt-history cultural fusions. **Live example shipped 2026-05-30**: [`/chapters/the-gate-example.html`](https://storyboard.daydream.monster/chapters/the-gate-example.html) — 8 cinema 16:9 establishing shots + 40s slow-pan walkthrough reel with civic-monumental organ-bell score, 17/17 jobs succeeded first try, ~$6.56 end-to-end.