← Back to App

Character Expression Tags Guide

Learn how to express characters using appearance, pose, expression, and clothing tags.

1. The identity stack

Every well-rendered AI character starts with a clear identity stack: a short sequence of tokens at the front of the prompt that tells the model "this is the person." The rule of thumb is count + identity core + 3–5 appearance tokens + one pose + one expression + outfit group, in that order. Skip the count (1girl, solo or 1boy, solo) and the model will drift between one, two, or three characters depending on pose. Skip the outfit group and the model will improvise one that may not match your design.

Character rendering is attention-budget limited. A typical prompt has 40–75 usable tokens before features start to drop. Character identity alone should claim 15–20 of those — everything else (lighting, background, composition) shares the rest. If your character looks "almost right but different every time," you're usually running over budget.

2. Appearance — hair, eyes, body

Appearance is the most precise lever in character prompting. The subdivisions matter:

Hair

Length: very short / short / medium / long / very long. Style: ponytail, twin tails, braid, bun, drill hair. Color: blonde, silver, black, red, pink plus gradient variants (gradient hair, pink to blue).

Eyes

Color: blue / green / brown / hazel / violet / amber / red / golden. Effects: heterochromia (two colors), glowing eyes, slit pupils, sparkling eyes.

Body type

slim, athletic, muscular, petite, tall. Pair with age tag for intent.

Skin / features

freckles, mole, scar, tan skin, pale skin, glasses.

Token order matters: Length → style → color for hair, and color → effect for eyes. Putting color before length (silver long hair) works fine on anime checkpoints but can confuse SDXL.

👁️ Appearance/Face — Tag examples

small_face — Small Face

3. Pose — what the body is doing

Static poses

standing, sitting, lying down, kneeling, crouching

Action poses

running, jumping, dancing, fighting stance, dynamic pose

Hand gestures

hand on hip, arms crossed, peace sign, pointing, waving, holding

Gaze direction

looking at viewer, looking away, looking up, looking down, looking back

One pose rule: Two pose tokens (sitting, running) will confuse the model — it may half-sit while trying to run. If you want a specific compound action (sitting while reading), state it as a natural-language phrase: sitting, holding an open book.

🤸 Pose — Tag examples

standing — Standing

4. Expression — what the face conveys

Positive emotions

smile (subtle), gentle smile, open mouth smile (bigger), laughing, happy, blush

Negative emotions

crying, tears, sad, angry, scowling, frowning

Intensity / drama

serious, determined, confident, smug, shocked, embarrassed

Closed-eye states

closed eyes, half-closed eyes, winking, one eye closed

One expression rule: Mixing (smile, crying) creates unstable faces. If you want bittersweet (crying with a smile), phrase it precisely: teary eyes, gentle smile.

😊 Face/Expression — Tag examples

Happy_face — Happy

5. Clothing — wardrobe reference table

Outfit tags resolve more reliably when you pick one compound garment (e.g., school uniform) than when you assemble pieces. Use this table as a quick selector:

GenreCanonical tagCommon add-ons
Modern casualt-shirt, jeanssneakers, hoodie, backpack
Schoolschool uniformpleated skirt, ribbon, blazer, loafers
Formalevening dress / formal suitnecktie, bow tie, high heels
Fantasy warriorplate armorcape, gauntlets, sword, shield
Fantasy magemage robewizard hat, staff, spellbook, magical aura
East Asiankimono / hanbok / cheongsamobi, folding fan, hairpin
Cyberpunktechwearutility vest, goggles, LED accents, combat boots
Swimswimsuitbeach, sun hat, sunglasses
Sleepwearpajamasslippers, blanket, bed

A fuller wardrobe reference is in Clothing & Accessories Catalog.

👗 Clothing — Tag examples

bandeau — Bandeau

💎 Accessories — Tag examples

hairband — Hairband

6. Common mistakes

  • No character count. Always open with 1girl, solo or 1boy, solo. Without it, scenes with poses like running tend to clone the subject.
  • Too many appearance tags. Beyond 5–6 appearance tokens, features start to silently drop.
  • Color adjacency. red hair, red dress bleeds colors. Put a setting token between them.
  • Ambiguous pronouns. Relying on the model to infer sex from words like soldier produces mixed results. Always specify 1girl or 1boy.
  • Contradictory expressions. Don't mix smile with angry.

7. Related guides

For the full depth on each sub-area, read: Hair & Eyes Reference for color/style/effect options, and Clothing Catalog for a genre-by-genre wardrobe breakdown with color-leak workarounds.

Ready to build your prompt?

Click from 1,700+ tags to assemble your AI image prompt.

Start Prompt Builder →