What is a Character?

A Character is the standard container for managing people in image generation — it stores each person's appearance, outfit, and pose separately, decoupled from the main prompt. That way the main prompt only handles scene, style, and mood, and the Character card only handles the person — the two stop fighting each other.

Core idea

• Main prompt = scene, style, mood, composition direction
• Character = the person's appearance details (one card per person, reusable)

By "extracting" people out of the main prompt, the same character set can appear with different main prompts — change the scene, change the style, change the season; the character stays the same.

What a Character contains

Each Character has 4 core fields:

Field	Purpose	Example
Name	The label you give the character; only for your own reference	Whitey / Sakura Catgirl
Prompt	Describes the character's appearance, outfit, pose	1girl, cat ears, pink hair, school uniform, smile
Negative prompt	Things this character should not have	bad hands, extra fingers
Position	The character's target location on the canvas	C3 (center) / A1 (top-left), etc.

Relationship with the main prompt

The main prompt describes the overall scene (background, mood, composition, style); the Character describes the details of each person. They each handle their own domain, and the AI composes them onto the same canvas at the end.

Multi-character: up to 6 enabled at once

You can create any number of characters (saved in your Character library), but at most 6 can be enabled per generation.

Each character card has a ✓ enable toggle on the right:

• Enabled: participates in this generation, counts toward total tokens
• Disabled: stays in the library, doesn't participate, doesn't count tokens

Position control: 5×5 grid

Each character needs a "target position". The system offers two modes:

Auto mode (default, recommended)

Don't specify a position; let the AI decide each character's placement — natural composition, but who ends up where is random.

Good for scenes without strict layout constraints: group photos, ensemble shots, mood pieces.

Custom mode (5×5 precise grid)

The canvas is divided into 5 rows × 5 columns = 25 cells; tap the cell you want — good for scenes with explicit spatial arrangement like left/right dialogue, stacked layers, or specific compositions.

Default C3 center

New characters default to C3 (row=C middle, column=3 middle) — perfectly centered. If you're only drawing one character, don't touch it.

How tokens are counted

A character's prompt and negative prompt only count toward total tokens when the character is enabled:

• Main prompt cap: 512 tokens
• Negative prompt cap: 512 tokens
• Each enabled character's prompt → counts toward total prompt tokens
• Each enabled character's negativePrompt → counts toward total negative tokens

The "Field X · Total Y/512" indicator at the bottom-right of the input shows in real time:

• Field = the tokens in the currently-editing field
• Total = the sum across all enabled characters + main prompt + preset

Characters are reusable assets

Characters and Pre-configs are two fully independent systems — a Pre-config only saves the main prompt and parameters; it doesn't record which characters are enabled or their positions. When you switch Pre-configs, the character enable state / positions stay as they are; adjust manually if needed.

That independence is actually an advantage: your built characters always stand by in the library and can appear with any Pre-config.

Reuse: one character set, many scenes

Once "Sakura Catgirl" is built, switch the main prompt from spring sakura → snowy peak → Tokyo neon street — change only the scene, keep the character — and produce three completely different mood images in seconds.

No need to rewrite the appearance for every scene.

Combine: enable toggle + position

With the main prompt unchanged, just toggle enable + position to switch composition: two-person to three-person, swap left/right, single close-up — without rewriting any character description.

When you don't need Characters

In a few cases you genuinely don't need to build a Character:

• Pure background / landscape (no people)
• One-off quick tests (no plan to reuse the person)
• Abstract style art (pure geometry, pure texture, abstract art)

For most other scenes, if there's a "person" in the image, manage them as a Character.

Avoid color bleed between people

When drawing two visually distinct people (e.g., a black-haired guy + a blonde girl), hard-coding 1boy, black hair, 1girl, blonde hair, ... in the main prompt often "bleeds" — the guy also grows blonde hair. Split into two Characters + assign positions to fix this.

Tips

• Prefer Characters for people: even with a single person, build them as a Character card — easier to reuse across scenes later
• Keep people out of the main prompt: let the main prompt handle scene, style, mood; leave people entirely to Characters
• Start with auto position: when you're not sure, let the AI place them — composition is often more natural; refine later with the 5×5 grid
• Hitting the token cap: disable characters you don't need now, then consider trimming descriptions
• Character name doesn't go into generation: it's just a label for you — "Whitey" or "Sakura Catgirl" both work