Skip to content

PROMPT_BASICS.md

Latest commit

 

History

History
120 lines (91 loc) · 3.83 KB

PROMPT_BASICS.md

File metadata and controls

120 lines (91 loc) · 3.83 KB

Prompt Basics Tutorial

This guide helps you write better comic prompts with three core blocks:

  • visual style
  • story plot
  • character description

You can use one long sentence, or compose a structured prompt from fields and then send it to llamagen.comic.create(...).

1) Basic Structure

Use this order for stable outputs:

  1. Visual style
  2. Story / scene goal
  3. Character details
  4. Composition / camera / mood
  5. Quality constraints

Template:

[Visual style], [Story plot], [Character description], [Composition and mood], [Quality constraints]

Example:

American comic illustration, bold ink lines and high-contrast colors, a rookie detective finds a hidden lab in rainy midnight Tokyo, main character: young female detective with short black hair and yellow raincoat, medium shot with cinematic rim light and tense atmosphere, clean anatomy, readable face, no text watermark

2) Prompt Writing Tips

  • Keep each block short but specific.
  • Put high-priority constraints early.
  • Keep character identity consistent across panels (same hair/clothes/color keywords).
  • Avoid conflicting styles in one prompt (for example: "minimal flat icon" and "hyper realistic oil painting").
  • If the result drifts, reduce abstraction and add concrete nouns (location, props, actions, lighting).

3) Reusable Prompt Script

Use a small helper to build prompts from structured inputs:

type PromptInput = {
  visualStyle: string;
  plot: string;
  characters: string;
  composition?: string;
  constraints?: string;
};

export function buildComicPrompt(input: PromptInput): string {
  return [
    input.visualStyle,
    input.plot,
    input.characters,
    input.composition ?? 'cinematic composition, clear focal point',
    input.constraints ?? 'clean line art, coherent anatomy, no watermark'
  ]
    .map((x) => x.trim())
    .filter(Boolean)
    .join(', ');
}

Usage:

import { LlamaGenClient } from 'comic';
import { buildComicPrompt } from './prompt-builder';

const client = new LlamaGenClient({ apiKey: process.env.LLAMAGEN_API_KEY! });

const prompt = buildComicPrompt({
  visualStyle: 'American comic illustration, bold outlines, vibrant high-contrast palette',
  plot: 'A resistance team escapes from a collapsing skybridge at sunset',
  characters: 'leader: tall woman with silver jacket; hacker: teenage boy with red visor',
  composition: 'wide shot, dynamic perspective, motion blur on debris',
  constraints: 'consistent character costume, sharp details, no text watermark'
});

const created = await client.comic.create({
  prompt,
  preset: 'neutral',
  size: '1024x1024'
});

Supported size values: 1024x1024, 512x768, 512x1024, 576x1024, 768x1024, 1024x768, 768x512, 1024x576, 1024x512.

4) Multi-panel Story Pattern

For multi-panel generation, keep visual style and character block fixed, only change panel action:

const style = 'American comic illustration, heavy ink lines, dramatic lighting';
const cast = 'hero: black trench coat and blue scarf; partner: orange bomber jacket';

const panelPrompts = [
  `${style}, ${cast}, panel 1: they enter the abandoned station, wide establishing shot`,
  `${style}, ${cast}, panel 2: they discover a glowing core room, medium shot with suspense`,
  `${style}, ${cast}, panel 3: explosion behind them while they jump to safety, dynamic action shot`
];

Then use createBatch(...) + waitForMany(...) from API Reference.

5) Troubleshooting

  • Problem: style is inconsistent
    • Fix: lock one style phrase and reuse it exactly in every panel prompt
  • Problem: character looks different each frame
    • Fix: repeat identity tokens (hair, outfit, age, accessories, color)
  • Problem: scene is too vague
    • Fix: add location + time + action verb + camera distance
  • Problem: cluttered output
    • Fix: remove extra adjectives and keep one main action per panel