Gradio

📖 Upload EPUB or TXT File

🤖 AI Model

🔑 AuthGPT Refresh Token

Browser login unavailable on HF Spaces. Paste your refresh token here.

📝 Translation Profile

🌐 Target Language

Replaces {target_lang} in prompts (used by Universal profile)

System Prompt (Translation Instructions)

Ready to translate

Upload an EPUB or TXT file and click 'Translate' to begin.

📋 Translation Logs

📥 Download Translated File

Extract character names and terms from EPUB files

Configure extraction settings below, then upload an EPUB file to extract a glossary.

📖 Upload EPUB File

🤖 AI Model

Min frequency

How many times a name must appear (lower = more terms)

1 10

Max titles

Limits to prevent huge glossaries

10 100

Max text size

Characters to analyze (0 = entire text)

Max sentences

Maximum sentences to send to AI (increase for more context)

50 500

Max names

Maximum number of character names to extract

10 200

Translation batch

Terms per API call (larger = faster but may reduce quality)

10 100

Chapter split threshold

Split large texts into chunks (0 = no splitting)

Target language

Language for translated glossary entries

Filter mode

What types of names to extract

all only_with_honorifics only_without_honorifics

Remove suffixes like '님', 'さん', '先生' from names

Remove honorifics from extracted names

Fuzzy threshold

How similar names must be to match (0.9 = 90% match, 1.0 = exact match)

0.5 1

System Prompt for Extraction

Customize how the AI extracts names and terms from your text.

Target language

Language for translated glossary entries

Extraction Template (Use placeholders: {language}, {min_frequency}, {max_names}, {max_titles})

You are a novel glossary extraction assistant.

You must strictly return ONLY CSV format with these columns and entry types in this exact order provided:

{fields}

For character entries, determine gender from context, leave empty if context is insufficient.
For non-character entries, leave gender empty.
The description column is mandatory and must be detailed

Critical Requirement: The translated name and description column must be in {language}.

For example:
character,ᫀ이히리ᄐ 나애,Dihirit Ade,female,The enigmatic guild leader of the Shadow Lotus who operates from the concealed backrooms of the capital, manipulating city politics through commerce and wielding dual daggers with lethal precision
character,ᫀ뢔사난,Kim Sang-hyu,male,A master swordsman from the Northern Sect known for his icy demeanor and unparalleled skill with the Frost Blade technique which he uses to defend the border fortress

CRITICAL EXTRACTION RULES:
- Extract All {entries}
- Do NOT extract sentences, dialogue, actions, questions, or statements as glossary entries
- REJECT entries that contain verbs or end with punctuation (?, !, .)
- REJECT entries starting with: "Me", "How", "What", "Why", "I", "He", "She", "They", "That's", "So", "Therefore", "Still", "But", "Protagonist". (The description column is excluded from this restriction)
- Do NOT output any entries that are rejected by the above rules; skip them entirely
- If unsure whether something is a proper noun/name, skip it
- The description column must contain detailed context/explanation
- You must include absolutely all characters found in the provided text in your glossary generation. Do not skip any character.

Ready to extract

Upload an EPUB file and click 'Extract Glossary' to begin.

📋 Extraction Logs

Quick Scan for Translation Quality

Scan translated content for common issues like untranslated text, formatting problems, and quality concerns.

Supported inputs:

📁 Output folder containing extracted HTML/XHTML files
📖 EPUB file (will be automatically extracted and scanned)
📦 ZIP file containing HTML/XHTML files

🤗 Hugging Face Spaces Mode Upload an EPUB or ZIP file containing the translated content. The scanner will extract and analyze the HTML/XHTML files inside.

📂 Upload EPUB or ZIP file

Quick Scan Mode (85% threshold, Speed optimized)

3-5x faster scanning
Checks consecutive chapters only
Simplified analysis
Good for large libraries
Minimal resource usage

Foreign Character Detection

Minimum foreign characters to flag

0 = always flag, higher = more tolerant

0 50

Detection Options

Check for excessive repetition

Check for glossary leakage (raw glossary entries in translation)

File Processing

Minimum file length (characters)

Skip files shorter than this

0 5000

Additional Checks

Identifies files that may have been incorrectly split or merged

Detect files with 2 or more headers (h1-h6 tags)

Checks if HTML files have proper structure

Flag HTML files with missing <html> tag

Check for insufficient paragraph tags

Minimum text in <p> tags (%)

Files with less than this percentage will be flagged

10 90

Report Settings

Report format

Summary = brief overview, Detailed = recommended, Verbose = all data

summary detailed verbose

Automatically save report after scan

Ready to scan

Enter the path to your output folder and click 'Quick Scan' to begin.

📋 Scan Logs

Configuration

Translation Profiles

Profiles are loaded from your config_web.json file. The web interface has its own separate configuration.

Universal:

You are a professional novel translator. You MUST translate the following text to {target_lang}.
- You MUST output ONLY in {target_lang}. No other languages are permitted.
- Preserve ALL HTML tags exa...

Korean_BeautifulSoup:

You are a professional Korean to English novel translator, you must strictly output only English text and HTML tags while following these rules:
- Use a natural, comedy-friendly English translation st...

Japanese_BeautifulSoup:

You are a professional Japanese to English novel translator, you must strictly output only English text and HTML tags while following these rules:
- Use a natural, comedy-friendly English translation ...

Chinese_BeautifulSoup:

You are a professional Chinese to English novel translator, you must strictly output only English text and HTML tags while following these rules:
- Use a natural, comedy-friendly English translation s...

Korean_html2text:

You are a professional Korean to English novel translator, you must strictly output only English text while following these rules:
- Use a natural, comedy-friendly English translation style that captu...

Japanese_html2text:

You are a professional Japanese to English novel translator, you must strictly output only English text while following these rules:
- Use a natural, comedy-friendly English translation style that cap...

Chinese_html2text:

You are a professional Chinese to English novel translator, you must strictly output only English text while following these rules:
- Use a natural, comedy-friendly English translation style that capt...

Manga_JP:

You are a professional Japanese to English Manga translator.
You have both the image of the Manga panel and the extracted text to work with.
Output only English text while following these rules: 

VIS...

Manga_KR:

You are a professional Korean to English Manhwa translator.
You have both the image of the Manhwa panel and the extracted text to work with.
Output only English text while following these rules: 

VIS...

Manga_CN:

You are a professional Chinese to English Manga translator.
You have both the image of the Manga panel and the extracted text to work with.
Output only English text while following these rules: 

VISU...

Original:

Return everything exactly as seen on the source....

Advanced Translation Settings

Threading delay (s)

0 5

API call delay (s) [SEND_INTERVAL_SECONDS]

Delay between API calls to avoid rate limits

0 10

Chapter range (e.g., 5-10)

Input Token limit

Disable Input Token Limit

Output Token limit

Contextual Translation

Translation History Limit

Rolling History Window

Batch Translation

Batch Size

Chapter Processing Options

Batch Translate Headers

Headers per batch

Use NCX-only Navigation (Compatibility Mode)

Attach CSS to Chapters (Fixes styling issues)

Retain source extension (no 'response_' prefix)

Groups chapters in batches of 3x batch size for memory management

Use Conservative Batching

⚠️ Disables ALL content safety filters for Gemini models (BLOCK_NONE)

Disable Gemini API Safety Filters

Direct HTTP POST with explicit headers

Use HTTP-only for OpenRouter (bypass SDK)

Sends Accept-Encoding: identity for uncompressed responses

Disable compression for OpenRouter (Accept-Encoding)

Chapter Extraction Settings

Text Extraction Method:

Standard uses BeautifulSoup, Enhanced uses html2text

standard enhanced

• Standard (BeautifulSoup) - Traditional HTML parsing, fast and reliable

• Enhanced (html2text) - Superior Unicode handling, cleaner text extraction

File Filtering Level:

Controls which files are extracted from EPUBs

smart comprehensive full

• Smart (Aggressive Filtering) - Skips navigation, TOC, copyright files

• Moderate - Only skips obvious navigation files

• Full (No Filtering) - Extracts ALL HTML/XHTML files

Response Handling & Retry Logic

GPT-5 Thinking (OpenRouter/OpenAI-style)

Controls GPT-5 and OpenRouter reasoning

Enable GPT / OR Thinking

Effort

OR Thinking Tokens

tokens

Provide Tokens to force a max token budget for other models; GPT-5 only uses Effort (low/medium/high)

Gemini Thinking Mode

Control Gemini's thinking process

Enable Gemini Thinking

Budget

tokens (0 = disabled)

0 = disabled, 512-24576 = limited thinking

🔒 API keys are encrypted when saved to config using AES encryption.

Save API Key (Encrypted)

Settings Status

How to Use Glossarion

Translation

Upload an EPUB file
Select AI model (GPT-4, Claude, etc.)
Enter your API key
Click "Translate"
Download the translated EPUB

Manga Translation

Upload manga image(s) (PNG, JPG, etc.)
Select AI model and enter API key
Choose translation profile (e.g., Manga_JP, Manga_KR)
Configure OCR settings (Google Cloud Vision recommended)
Enable bubble detection and inpainting for best results
Click "Translate Manga"

Glossary Extraction

Upload an EPUB file
Configure extraction settings
Click "Extract Glossary"
Use the CSV in future translations

API Keys

OpenAI: Get from https://platform.openai.com/api-keys
Anthropic: Get from https://console.anthropic.com/

Translation Profiles

Profiles contain detailed translation instructions and rules. Select a profile that matches your source language and style preferences.

You can create and edit profiles in the desktop application.

Tips

Use glossaries for consistent character name translation
Lower temperature (0.1-0.3) for more literal translations
Higher temperature (0.5-0.7) for more creative translations

Glossarion - AI-Powered Translation

Image Translation

Glossary Settings

Automatic Glossary Extraction Settings

Quality Assurance

Ready to translate

Parallel Panel Translation

Font Settings

Text Color

Shadow Settings

Background Settings

Ready to translate

Advanced Manga Translation Settings

Bubble Detection

Inpainting

Mask Dilation

HD Strategy for Inpainting

Image Tiling

Font Sizing Algorithm

Format Detection

Inpainting Performance

Performance

Model Optimization

Debug Options

Extract character names and terms from EPUB files

System Prompt for Extraction

Output Format Instructions

Ready to extract

Quick Scan for Translation Quality

Foreign Character Detection

Detection Options

File Processing

Additional Checks

Report Settings

Ready to scan

Configuration

Translation Profiles

Advanced Translation Settings

Chapter Processing Options

Chapter Extraction Settings

Response Handling & Retry Logic

How to Use Glossarion

Translation

Manga Translation

Glossary Extraction

API Keys

Translation Profiles

Tips