---
name: architecture-render-prompt
description: >-
  Build an architecture AI image prompt that actually holds, using a 9-part
  structure and word banks. Use whenever an architect or project manager wants
  to generate or improve a render/image prompt for a building — for GPT Image 2,
  Seedream 5, Nano Banana, Midjourney or any image model. Assembles shot, camera,
  subject, material, light, style, context, exclusions and technical finish into
  one clean natural-language prompt. Triggers on requests like "write me a render
  prompt", "improve this AI image prompt", "prompt for a photoreal exterior",
  "make a render of this building".
---

# Architecture Render Prompt

Turn a rough idea into a render prompt that holds together — where changing one
word does not turn the whole building into a different building.

The core idea: a strong architecture image prompt is not a longer prompt, it is a
**structured** one. It is nine decisions, made in order. Name each one deliberately
and the model stops guessing.

## When to use this

Use this skill when the user wants to:

- Write a new render/image prompt for a building (exterior, interior, aerial, detail).
- Improve or debug a prompt that keeps producing generic or wrong results.
- Understand *which* word to change when a render is close but not right.

## How to work

1. **Find out what you have.** Read whatever the user gives you — a sketch, a photo,
   a one-line brief, an existing prompt. Identify which of the nine parts are already
   decided and which are missing.
2. **Ask only for what's load-bearing and missing.** Usually: the building type
   (subject), the key materials, the shot, and the place. Do not interrogate — infer
   sensible defaults and state them.
3. **Assemble in plain descriptive sentences, NOT tag-soup.** GPT Image 2 and
   Seedream 5 read natural language best. Write it as a short paragraph a person could
   read aloud. (Only fall back to comma-tag style if the user names Midjourney.)
4. **Walk the nine parts in order** (below). Every prompt should touch all nine, even
   if briefly.
5. **Output the finished prompt in a copyable block**, then list the nine choices you
   made so the user can swap any one.
6. **Name one honest limitation** if the request implies the model will do real
   architecture (it won't — see the end).

## The nine parts (in order)

1. **Shot** — how you frame it. The single most important word.
2. **Camera** — the lens / focal length. Controls how honest the perspective feels.
3. **Subject** — what the building actually is. State the type first.
4. **Material** — name exactly what you want to see, never "wood" or "metal".
5. **Light** — time, quality, direction, and colour temperature (in Kelvin).
6. **Style** — the rendering look. Photoreal is only one option.
7. **Context** — the locality. Name the city; let the model fill the street with local life.
8. **Exclusion** — what to leave out, stated as plainly as what to include.
9. **Technical** — the closing words that push it toward a believable photograph.

## Anchor prompt template

Fill the [brackets] from the word banks. Keep it as flowing sentences.

```
A [style] architectural [exterior/interior] render in [shot], shot as if on a
[camera] lens at [eye level / low angle / aerial]. The subject is a [subject]:
[material 1], [material 2], and [material 3]. It is [light condition], [colour
temperature], with [light direction] casting [shadow quality]. Located in
[place], with [local element] and [local element] to ground it. [Exclusions].
[Technical finish].
```

**Worked example:**

```
A photoreal architectural exterior render in three-quarter street view, shot as
if on a 28mm lens at eye level. The subject is a 4-storey cross-laminated-timber
residential infill: a board-formed concrete base, blackbutt timber battens, and
anodised bronze window reveals. It is golden hour, warm 3200K light, with a low
sun from camera-left casting long soft shadows. Located in Melbourne, with a
laneway cafe and a passing tram to ground it. No cars, no warped glass. Natural
colour grade, full-frame quality, shot on camera.
```

## Word banks

### 1. Shot (how you frame it)
Three-quarter view (two facades, the default hero) · Head-on elevation (flat, material studies) ·
One-point perspective (depth down a street) · Two-point perspective (corner, natural 3D) ·
Aerial / bird's-eye (angled from above) · Top-down aerial (straight down, roof + site) ·
Worm's-eye / low angle (monumental) · Eye level (pedestrian, human scale) ·
Establishing shot (full context) · Wide shot (building + surroundings) ·
Detail / close-up (junctions, texture) · Section perspective (cut-through) ·
Axonometric (measured 3D, no perspective) · Isometric (equal axes, diagram) ·
Over-the-shoulder (foreground framing) · Corner approach (how you arrive) ·
Interior wide (room in full) · Ceiling-down (interior straight down) ·
Frame view (framed opening, inside-out) · Reflection (mirrored in glass/water)

### 2. Camera (focal length)
14mm ultra-wide, dramatic, heavy distortion · 16mm very wide, whole interiors ·
18mm wide establishing, full facade + context · 20mm wide, big scenes, mild distortion ·
24mm the wide workhorse, interiors & exteriors · 28mm natural wide, pedestrian street view ·
35mm reportage, natural, environmental · 50mm human eye, honest proportion ·
70mm short tele, slight compression · 85mm portrait length, material close-ups ·
100mm fine detail, sample-board level · 135mm strong compression, stacks facades ·
200mm heavy compression, layers a streetscape · Tilt-shift corrects verticals (the architecture lens) ·
Fisheye 180°, experimental only

### 3. Subject (the building type)
Low-rise infill · Mid-rise apartments · Tower · Adaptive reuse · Heritage facade ·
Civic building (library, hall, gallery) · Cultural pavilion · Education campus ·
Workplace · Retail / mixed-use · Hospitality · Masterplan

### 4. Material (name exactly what you want)
Board-formed concrete · Off-form concrete · Blackbutt battens · Spotted gum ·
Recycled brick · Face brick · Anodised bronze · Weathering steel · Zinc standing seam ·
Glazed terracotta · Rammed earth · Sandstone · Powder-coated aluminium · Fluted glass ·
Timber screens · Perforated metal · Travertine · Polished plaster · Charred timber (shou sugi ban) ·
Terrazzo
Rule: name the species, the finish, the profile. Never "wood" or "metal".

### 5. Light (time, quality, direction, temperature)
Golden hour (warm low sun, long shadows) · Blue hour (cool dusk, warm interior glow) ·
Overcast (soft, flat, true colour) · Midday sun (hard, crisp shadows) ·
Night, floodlit · Backlit (glowing edges) · Side light (rakes texture) · Long shadows ·
Interior warm · Neon glow · Moonlight
Colour temperature: 3000K warm (tungsten) · 4000K neutral · 5000K+ cool (daylight).
Trick: use a fill light to open the shadows, e.g. "a soft blue fill".

### 6. Style (the rendering look)
Photorealistic (the default) · Post-digital collage · Paper craft · Nordic minimalist ·
Watercolour · Line drawing · Diagram / axonometric · Clay / white model (massing) ·
Concept sketch · Photomontage · Cutaway · Blueprint

### 7. Context (ground it in a place)
Name the locality and let the model add fitting local life:
Melbourne (laneway cafe, tram) · Sydney (harbour light, ferries) ·
India (festival crowd, colour, parade) · Tokyo (neon signs, dense foot traffic) ·
London (red bus, black cab, drizzle) · Paris (cafe chairs, plane trees) ·
New York (yellow cab, steam, crowds) · Bali (tropical planting, scooters) ·
Dubai (heat haze) · Marrakech (market stalls, warm dust)

### 8. Exclusion (what to leave out)
No cars · No people · No signage · No warped glass (windows stay flat and true) ·
No clutter (bins, junk) · No lens flare · No HDR halo · No watermark / text ·
No distortion · No oversaturation

### 9. Technical (the finish)
Photoreal · Natural colour grade · Full-frame · Subtle 35mm film grain ·
Shallow depth of field (subject pops) · Deep depth of field (sharp front to back) ·
True white balance · HDR (detail in shadow and sky) · Soft shadows ·
PBR / physically-based lighting · 8K resolution · Shot on camera (reads real, not CG) ·
RAW capture (flat, full latitude to grade)

## Output format

Always return:

1. **The prompt**, in one copyable block, as flowing natural-language sentences.
2. **The nine choices** as a short list (Shot: … · Camera: … · Subject: … etc.) so the
   user can swap any single word.
3. **One swap suggestion** — the single change most likely to improve the result.

## Honest limitation (always include when relevant)

These words control the *picture*, not the architecture. AI will not read the site,
know the planning scheme, size a member, or defend a decision in review. It melts
geometry and invents junctions if unchecked. Use it to communicate and to test
directions fast — then check every frame like a junior's drawing. You direct; it draws.

---

By Chiang Ning · chiangning.net
Copyright 2026. Chiang Ning. All Rights Reserved.
