Prompt Crafting Guide

Getting Started

Welcome to Anime AI Gen! This manual provides essential techniques for producing premium anime-style artwork with the Animagine 4 model. These recommendations will help you achieve both visually appealing and technically correct results.

Prompt Construction

First and most importantly: prompts must be written in English only - yes, exclusively in English. For best results, you should input your prompts using a tag-based approach, adding elements one by one.

Constructive Prompts

Implement an organized prompt structure: 'character type (1girl/1boy/1other), character identity, source material, rating, supplementary information, quality modifiers'. Sample: 'coffee shop setting, female character, golden hair, azure eyes, sophisticated, rabbit-themed costume, server'.

Exclusion Prompts

Utilize negative prompts to eliminate unwanted elements. Example exclusion list: 'low resolution, anatomical inaccuracies, hand defects, written text, errors, digit abnormalities, cropping artifacts, inferior quality, signatures, watermarks, blurriness'.

Tag Implementation

Quality Modifiers

Regulate overall image excellence and detail precision. Available options: 'exceptional work', 'premium quality', 'substandard quality', 'poorest quality'.

Scoring Tags

Offer refined control over output quality. Implementable tags: 'top-tier score', 'excellent score', 'acceptable score', 'mediocre score', 'poor score', 'minimal score'.

Era-Specific Tags

Affect artistic approach based on chronological periods. Compatible tags: '2005 style', 'specific year {n}', '2025 style'.

Content Rating Tags

Manage appropriateness level. Supported classifications: 'family-friendly', 'discretion advised', 'adult content', 'graphic material'.

Configuration Settings

CFG Adjustment

Configure CFG scale within 4-7 range for optimal output. Suggested value: 5.

Sampling Iterations

Employ 25-28 processing steps. Recommended: 28 steps.

Sampling Method

Preferred algorithm: Euler Ancestral (Euler a variant).

Dimension Selection

Select from standard resolutions: 1024x1024 (square), 1152x896 (9:7 ratio), 1216x832 (3:2), 1344x768 (7:4), 1536x640 (12:5), 896x1152 (7:9), 832x1216 (2:3), 768x1344 (4:7), 640x1536 (5:12).

Expert Configurations

Guidance Calibration

Adjust to 0.5 value to prevent excessive brightness.

PAG Configuration

Set at 3 for improved generation outcomes.

Automatic Prompt Enhancement

Active by default. Appends quality improvement tags to user prompts.

Generation Seed

Use -1 for randomized seeds or input specific values for reproducible results.

Technical Constraints

Input Requirements

Exclusively supports tag-based text input; natural language processing is limited. English language input mandatory.

Anatomical Rendering

Challenges with intricate body structures, especially complex hand positions and finger articulation.

Text Integration

Current version cannot reliably generate legible text within images.

Contemporary Characters

Newer fictional characters may exhibit reduced accuracy due to insufficient training data.

Multi-Character Scenes

Compositions featuring multiple subjects often require meticulous prompt engineering.

Resolution Limits

Higher dimensions (e.g., 1536x1536) may display quality degradation as training utilized standard SDXL resolution.

Stylistic Consistency

May necessitate explicit style tags as training prioritized character fidelity over stylistic uniformity.