Chinese Culture in Watercolor — AI Art Prompt Method

A complete system for creating and selling premium Chinese-watercolor AI art

—

CHAPTER 4 — THE PREMIUM PROMPT SKELETON, FIELD BY FIELD

4.1 Why skeleton beats inspiration

Chapter 2 introduced the six-part premium prompt; Chapter 3 gave you the cultural shelf. This chapter goes deep: we dissect each of the six slots, one at a time, and learn exactly what to put in it, what to leave out, and how the slots interact. When you finish this chapter you will be able to write a publishable prompt for any theme without staring at a blank page.

The core claim is simple and worth repeating: **reliability comes from structure, not from inspiration.** A model given one clever sentence produces a lottery ticket. A model given a six-slot paragraph produces a controlled illustration. The skeleton exists to remove the guesswork, so that your morning review session becomes *choosing* the best of five strong renders rather than *rescuing* one weak one.

A useful mental model: think of the image model as a very fast, very literal art assistant who has seen everything but understands nothing about your intent. Your job is not to be poetic — it is to be **specific in the slots that matter and silent in the slots that do not.** Every word you add is a constraint. Good prompt-craft is the discipline of adding only the constraints that move the picture.

Two rules govern the whole chapter:

• **One slot, one job.** Scene establishes composition; lighting establishes atmosphere; technique establishes the medium. Do not argue composition inside the lighting slot.

• **Order is a soft hierarchy.** Models weight earlier tokens slightly more heavily in most interfaces, so we lead with scene and end with mood. This also makes prompts easy to edit: if the render feels flat, you almost always fix it by rewriting the *lighting* and *mood* slots, not the scene.

4.2 Slot 1 — Scene: the spine of the image

The scene slot answers one question: **what is the single subject of this picture, and where is the camera?**

Write it as a noun phrase with a definite subject, not a list of ideas. “A quiet stone quay on Qixi night where a young woman in pale celadon hanfu releases a lantern” is a scene. “Chinese festival, lanterns, water, girl, romantic, night” is a search query, and it will render like one.

Three sub-decisions live in this slot:

1. **Subject specificity.** Name one primary subject (a figure, a boat, a mountain, a table). Prompts with two co-equal subjects usually split the model’s attention and yield mush. If you need a second subject — lovers, a boat and its crew — make one clearly primary and the other clearly secondary (“her lowered hands *and* the glowing lantern just touching the water”).

2. **Compositional anchor.** State the framing explicitly: “wide 3:2 landscape,” “vertical portrait,” “close foreground of the hands, background a receding skyline.” The model will otherwise default to a centered, mid-distance frame, which is the single most common reason a prompt feels generic.

3. **Time and place, briefly.** “On the fifteenth night of the first lunar month,” “at dawn on the river.” A few words of *when* transform the palette and costume choices the model makes downstream.

Do **not** put mood adjectives in this slot. Keep it factual and spatial. Save “hopeful,” “wistful,” and “hushed” for Slot 6.

4.3 Slot 2 — Layered depth: foreground, middle, background

This is the slot most sellers omit, and it is the difference between an illustration and a flat sticker. Explicitly assigning three depth planes forces the model to build **atmospheric recession** — the sense that the picture has air in it.

The pattern is always the same scaffold:

> Foreground: [a close, specific detail that frames the shot]. Middle: [the main action or the primary subject]. Background: [a wide, distant element that gives scale and openness].

Why this works: watercolor is a *depth-friendly* medium. Wet-on-wet washes in the background naturally read as distance; crisp dry-brush detail in the foreground naturally reads as nearness. When you name the three planes, you are prompting the model to use the medium’s own grammar of recession. You get mist, sky, receding rooftops, and an honest foreground — instead of everything competing at one focal length.

A worked scaffold for an ink-wash landscape:

> Foreground: a gnarled pine clinging to a cliff edge in crisp dry-brush. Middle: the layered mountain slope and a lone pavilion. Background: distant peaks dissolving into mist on the horizon.

Notice the foreground element is *close and detailed*, the middle is *the subject*, and the background is *large and soft*. Follow that rhythm every time and your images stop looking like they were printed on a postcard.

One caution: three planes is the sweet spot. Six planes becomes a cluttered list and the model starts dropping elements. If your scene is genuinely complex (a festival market), pick the three planes that matter and let the model fill the rest.

4.4 Slot 3 — Character and costume detail

For any prompt with a human subject, this slot determines whether you get a person or a mannequin. The rule is **specific garments, specific gestures, specific materials.**

Instead of “a woman in traditional dress,” write “a graceful young woman in pale celadon hanfu with a wide embroidered sash, a loose outer robe, and jade hairpins, her sleeves spilled across the table.” Three layers of clothing, one accessory family, one telling gesture. That is a costume description, not a costume label.

Three reliable levers:

• **Layer count.** Hanfu, hanbok, and period garments are *layered* by design. Naming two or three layers (inner robe, outer robe, sash or collar) instantly reads as authentic and gives the model rich material to paint.

• **Fabric and finish.** Silk, gauze, brocade, linen, fur collar, gilt thread. Materials catch light differently, and naming them sharpens the lighting slot that follows.

• **A gesture that shows character.** Lowered eyes, a hand resting on a rail, sleeves spilled, a needle lifted. Gesture is where mood begins, and it lets the human read as *present* rather than posed.

For non-human subjects — a mountain, a teapot, a lantern — repurpose this slot as **object and texture detail**: the glaze, the weave, the patina, the ink grain. A lone pavilion becomes “a wooden pavilion with grey-tiled eaves and weathered red rails.” Same slot, same job: specificity.

Keep it tasteful and non-explicit. You are painting grace, posture, and fabric — poise, not provocation. This both respects the culture and keeps your products sellable on mainstream platforms.

4.5 Slot 4 — Lighting: the atmosphere engine

If you only ever master one slot, master this one. Lighting is what separates a pretty composition from a *moving* image, and it is the fastest fix for a flat render.

Every lighting description needs two halves:

• **A source and direction.** “Warm single lantern light from the left.” “Cool moonlight from behind the loom.” “Low golden dawn raking across the rooftops.”

• **A temperature contrast.** Warm against cool. This is the classic trick of every great watercolor night scene: candle-gold and lantern-amber pressed against indigo, blue-violet, and moonlight silver. The contrast does the emotional work for free.

Some reliable kitchens, ready to drop into a prompt:

• **Festival night:** warm candle-gold glow everywhere against deep cool blue-violet, cool moonlight overhead.

• **Intimate interior:** a warm single lamp between two figures, cool dusk seeping in at the window edges.

• **Dawn river:** low rose-gold sun on the water, cool mist still holding the far bank.

• **Autumn garden:** warm afternoon sun through the canopy, dappling the scene; cool long shadows in the grass.

• **Snow interior:** cool snowy light outside against a warm interior wash of red and gold.

Two refinements worth internalizing. First, **backlit figures** (rim light) are almost always more beautiful than front-lit ones — “cool moonlight silvering the outline of the figures” reads as luminous. Second, **let the light fall on one thing.** A picture in which *everything* is lit is a picture with no emphasis. Say where the light lands: “warm gold dancing on their faces and the rim of the cup.”

4.6 Slot 5 — Watercolor technique: naming the medium

This slot is what earns you “I can’t believe this is AI — it looks painted.” Most prompt writers neglect it entirely, then wonder why their “watercolor” output looks like a slick digital render with a paper texture stapled on.

Watercolor has a vocabulary. Use it with intent, because each term produces a distinct, recognizable effect:

• **Wet-on-wet** — pigment blooming into damp paper. Use it for *sky, mist, water, and atmosphere*. It is what makes distance feel soft and alive.

• **Translucent washes** — thin layers of pigment that glow from within. Use it for *silks, robes, lanterns, and any luminous surface*. Transparency is the medium’s signature, and naming it prevents opaque, poster-like color.

• **Dry-brush** — a broken, scratchy stroke of little water. Use it for *bark, stone, thatch, lattice, rigging, hair, and branch detail*. It is your foreground and your texture.

• **Bleeding edges and soft blooms** — the fuzzy, spreading boundaries where pigment wanders. Use it for *haze and emotional softness*.

• **Negative space** — the bare paper left unpainted, standing in for light and emptiness. Use it for *stars, dew, glow, and the unmistakable discipline of Chinese ink painting*.

• **Texture surfacing** — “cold-pressed cotton paper,” “visible paper grain,” “tooth of the paper.”

A full technique clause might read: *”Painted in loose translucent watercolor: wet-on-wet night sky, luminous washes on each glowing lantern, fine dry-brush on the strings and pagoda lattice, star-like negative-space highlights on bare cold-pressed cotton paper.”*

Note the pattern: **technique term → what it is applied to.** Never just say “watercolor painting.” Always bind each technique to a specific element. That binding is the whole skill.

4.7 Slot 6 — Mood: the closing chord

The last slot is three to four adjectives that name the emotional target. It is short on purpose. “Joyous, hopeful, hushed-wonder.” “Refined, quiet, elegant.” “Earthy, tender, longing-at-dusk.”

Two rules. First, **keep every adjective non-visual.** “Hopeful” and “wistful” steer feeling; “beautiful” and “amazing” steer nothing and waste tokens. Second, **stay coherent.** Do not stack contradictory moods; “cheerful and melancholic and epic” gives the model no direction. Pick one emotional center and let two or three adjectives orbit it.

Mood is your final tuning dial. When a render is technically correct but emotionally inert, you fix it here and in the lighting slot — never by adding more scene.

4.8 Putting it together: the seven-move assembly

Here is the full method as a repeatable recipe. Write your prompt in this order:

1. **Name the scene** — subject, camera, time and place.

2. **Assign three planes** — foreground / middle / background.

3. **Dress the subject** — layered garments or object texture, plus one gesture.

4. **Light it** — one source, one direction, warm-against-cool.

5. **Paint it** — bind each watercolor technique to a specific element.

6. **Name the mood** — three coherent adjectives.

7. **Close the file** — aspect ratio + “print-quality, tasteful and elegant, no text or calligraphy.”

That last move is a professional habit. Stating the aspect ratio prevents surprise crops; “no text or calligraphy” prevents the model from inventing garbled characters — a common and product-killing artifact. Write these on autopilot every single time.

Worked example, Mid-Autumn, assembled in one pass:

> *A moonlit Mid-Autumn courtyard where a family gathers beneath a great osmanthus tree to share mooncakes and gaze at the full moon. Foreground: a low round table of mooncakes, pomelos, and a teapot in crisp detail. Middle: three generations seated on stools, the eldest lifting a cup toward the sky. Background: a tiled roofline and an enormous luminous full moon. Warm lantern glow on the faces against cool silver moonlight on the garden. Painted loose with wet-on-wet night air, translucent washes on the silks and mooncakes, delicate dry-brush on the branches, generous negative space around the moon. Gentle, grateful, reunion-warm mood. Wide 3:2 landscape, print-quality, tasteful and elegant, no text or calligraphy.*

Read it aloud once and you can *see* the picture. That is the test: if the prompt does not conjure a specific image in your own mind, it will not conjure one in the model’s either.

4.9 Common failures and their slot-level fixes

• **Flat and poster-like** → the technique slot is missing or generic. Add wet-on-wet, translucent washes, and negative space, each bound to an element.

• **Muddy, undifferentiated focal plane** → the scene slot has no compositional anchor. Add framing and foreground/middle/background.

• **Generic faces and clothing** → the costume slot is a label, not a description. Add layer count, fabric, and a gesture.

• **Beautiful but emotionally dead** → lighting lacks a temperature contrast, or mood is missing. Add warm-against-cool and three coherent adjectives.

• **Garbled pseudo-characters in the art** → you forgot “no text or calligraphy.”

• **Everything competes for attention** → you lit everything. Say where the light lands.

Notice that every diagnosis names a *slot*. That is the point of the skeleton: it converts the fuzzy complaint “this doesn’t feel right” into a specific, fixable edit. Prompt engineering, done this way, becomes less like gambling and more like tuning an instrument.

Chapter 4 — Key takeaways

• Reliability comes from structure, not inspiration: one slot, one job, ordered scene → depth → costume → lighting → technique → mood.

• The layered-depth slot (foreground / middle / background) is the single most neglected lever and the fastest route to illustration-grade images.

• Lighting is the atmosphere engine: always give one source, one direction, and a warm-against-cool contrast, and always say where the light lands.

• Bind every watercolor technique term to a specific element; never just say “watercolor painting.”

• Close every prompt with aspect ratio plus “no text or calligraphy” — a professional habit that prevents surprise crops and garbled artifacts.

• Diagnose failures by slot. “This feels flat” is not a note; “the technique slot is generic” is.