What are Prompts?

The quality of your prompt directly determines the quality of your output. Mastering a few core principles is more effective than piling up a hundred decorative words.

Two writing styles

Tag-style (recommended for beginners)

Comma-separated keywords, each describing a single feature:

1girl, long silver hair, blue eyes, school uniform, cherry blossom tree, smile, sunny day

Pros:

• Highly controllable — modifying a feature only requires finding the right keyword
• Order conveys weight — earlier keywords carry more importance

Natural language

Construct the scene with full descriptive sentences:

A black cat sitting on a windowsill at sunset, gazing out at a city skyline, warm orange light flooding through the window, oil painting style

Pros: natural, evocative, capable of expressing coherent stories or actions. However, sentence comprehension is less stable across models compared to tag-style.

TIP

The two styles can be mixed — sentences for the main scene, tags for details. This is a common pattern among creators.

❌ vs ✅

❌ Pile-up style

beautiful gorgeous stunning amazing pretty cute lovely girl with hair and eyes wearing clothes

Too many empty modifiers, lacking specific features. The AI doesn't know what to draw — outcome is pure luck.

✅ Specific style

1girl, twin tails, red ribbon, white blouse, navy skirt, holding a book, library background

Each keyword corresponds to a concretely visualizable element so the AI has anchors to rely on.

Rule of thumb

If removing a word changes nothing in the picture, it's redundant — replace it with something more specific.

Four-element breakdown

A good prompt should cover the following four layers — missing any will cause issues:

Layer	Content	Example	What happens if missing
Subject	who / what	1girl, cat, robot	No clear subject; AI guesses
Appearance	hair, outfit, expression, pose	long red hair, school uniform, smiling	Inconsistent appearance across runs
Scene	location, background, lighting	library, sunset, soft light	Empty or random backgrounds
Style	art style, quality	anime style, detailed, masterpiece	Style drifts every run

Useful tag cheat sheet

When you don't know what to write, pick combinations from the lists below.

Subject (put at the front)

Type	Common tags
Single person	1girl, 1boy, 1other
Multiple people	2girls, multiple girls, group
Animals	cat, dog, dragon, animal focus

View & framing

You want	Tag
Looking up	from below
Looking down	from above
First-person	pov
Half / full body	portrait, full body, cowboy shot

Style / medium

Style	Tag
Anime	anime style, anime coloring
Oil painting	oil painting (medium)
Watercolor	watercolor (medium)
Ink	ink (medium)
Photoreal	photorealistic, realistic
Retro era	1990s (style), retro

Quality boost (auto-appended by the system)

The system automatically appends:

very aesthetic, masterpiece, no text

These three serve as baseline quality. You don't need to write them manually. If you want to emphasize quality further, you can add:

best quality, detailed, ultra detailed

TIP

More quality words isn't better — 2–3 is enough. Over-stacking dilutes AI attention.

Weight control (advanced)

You can give a keyword extra weight to emphasize or de-emphasize its impact:

{red hair}        // strengthen
{{red hair}}      // strengthen further (double layer)
[blue eyes]       // weaken
[[blue eyes]]     // weaken further (double layer)

WARNING

Nesting beyond 3 layers tends to produce strange results — keep it to two layers max.

Beyond brackets, adjusting order is another common "soft weight" — keywords closer to the front carry more weight. If you're unsure whether to use {}, try moving the word to the front first.

Tips

• Simple to complex: start with 5–6 core keywords, see the result, then add more
• One change at a time: modify only one keyword per iteration to identify what works