A comprehensive browser-based tool that leverages Large Language Models (LLMs) to translate SRT subtitle files and automatically validate the output for common errors. Built specifically to create high-quality, script-accurate subtitles using your existing Gemini subscription via CLI Proxy API.
| Original TV Show in English | Output of translated SRT in Tenglish |
|---|---|
 |
 |
Note
Screenshots are taken from Psych, an American detective comedy-drama television series created by Steve Franks for USA Network.
My mom loves to watch movies and TV shows, but she isn't fluent in English. Properly translated subtitles are the perfect way for her to enjoy and understand the content. While Large Language Models (LLMs) are incredibly powerful and can translate subtitles in seconds, they often introduce subtle errors, such as:
- Timestamp Drift: The model may slightly alter timestamps, causing subtitles to appear too early or too late.
- Character "Hallucinations": When translating for a specific language (like Telugu), the model might mistakenly insert characters from a similar-looking but incorrect script (like Devanagari, Tamil, or Kannada).
- Incomplete Translations: Long subtitle files may be cut off mid-translation.
- Formatting Errors: Line breaks or subtitle numbering can sometimes be corrupted.
This tool automates the entire translation workflow, validates the output, and provides one-click fixes for common LLM errors—ensuring the subtitles are perfect.
This is the successor to my original SRT Subtitle Validator, which could only validate pre-translated files. The new SRT Translator & Validator combines translation and validation into a single, streamlined workflow with automatic error correction.
- Direct LLM Integration: Translate subtitles using your Gemini subscription (via CLI Proxy API)
- Customizable Prompts: Full markdown editor with live preview for translation instructions
- Partial Translation Recovery: Automatically detects incomplete translations and continues from where it stopped
- Language Detection: Automatically identifies the target language for proper filename generation
- Raw Output Logging: Saves all Gemini responses with timestamps for debugging
- Timestamp Synchronization: Compares timestamps between source and translated files, reporting any mismatches with detailed ranges
- Script Validation: Ensures translated text only contains characters from allowed Unicode ranges (e.g., English + Telugu)
- Configurable Character Sets: Search from 130+ language scripts or add custom Unicode ranges/characters
- Real-time Feedback: Color-coded error reports with subtitle indices and Unicode values
- Fix Timestamp Errors: One-click replacement of incorrect timestamps with source values
- Fix Script Errors: Sends problematic subtitles back to Gemini for automatic correction
- Combined Fixing: Handles both timestamp and script errors in a single operation
- Smart Validation: Re-checks after fixes and shows remaining issues
- Editable Output: View and manually edit translated subtitles before downloading
- Auto-save: Preserve manual edits while continuing the workflow
- Progress Tracking: Browser tab title shows current status (Translating, Translated, Fixing, Fixed, Error)
- Auto-scroll: Automatically scrolls to validation results
- Dark/Light Mode: Adapts to your system theme preference
- Configured Prompt Templates: Loads your saved configuration and prompt to minimize manual entry
- File Drag & Drop: Supports an alternative way to upload files
CLI Proxy API is required to access your Gemini subscription through an API interface.
For installation instructions, please visit: https://help.router-for.me/introduction/quick-start.html
Important: During login, you may need to choose or create a Google Cloud project. Select a project with no billing account linked to avoid unexpected charges. You can manage your projects at console.cloud.google.com.
- Run the login command:
cliproxyapi --loginIf you're an existing Gemini Code user:
cliproxyapi --login --project_id <your_project_id>- Follow the OAuth flow in your browser to authenticate
- The local OAuth callback uses port
8085
Create or edit your config.yaml file.
For more information, please visit: https://help.router-for.me/configuration/basic.html or use the provided example config.yaml.
Important
The api-keys value in config.yaml must match the API Key field in the web application. The key dummy is used by default if the field is left blank.
If installed via Homebrew:
/opt/homebrew/opt/cliproxyapi/bin/cliproxyapi --config ~/path/to/your/config.yamlIf built from source:
./cli-proxy-api --config /path/to/your/config.yamlThe server will start on http://localhost:8317 by default.
Open this link SRT Translator & Validator in any modern web browser.
- CLI Proxy API Endpoint: Default is
http://localhost:8317/v1/chat/completions - Model Name: Default is
gemini-2.5-pro - API Key: Enter the same key from your
config.yaml(e.g.,1234567)
By default, the tool allows:
- Basic Latin (English alphabet)
- Latin Supplement (é, ñ, etc.)
- Special characters (quotes, em-dash, line breaks)
To add your target language:
- Type the language name in the search bar (e.g., "Telugu", "Arabic", "Tamil")
- Click the suggested script to add it
- Or add custom Unicode ranges:
U+0C00-U+0C7For single characters:U+0C00
You can also remove any script (except special characters and line breaks) by clicking the × button.
Write your translation instructions in the prompt field. You can use markdown formatting and switch to Markdown View to preview how it renders.
For reusable prompts with dynamic values, create a template file with optional configuration:
- Create a
.txtfile with placeholders using{{Variable Name}}syntax - Click "📥 Import Template" or drag and drop the template file onto the prompt section
- Fill in the template variables that appear
- Click "Apply Template" to generate your final prompt
Basic Template Example:
You are translating subtitles for {{Show Name}}, Season {{Season}}, Episode {{Episode}}.
Translate from {{Source Language}} to {{Target Language}}.
**Style Guidelines:**
- Use {{Formality Level}} language
- Keep character names in {{Name Language}}
[Rest of your translation instructions...]When imported, you'll be prompted to fill in: Show Name, Season, Episode, Source Language, Target Language, Formality Level, and Name Language.
Configured Template Example (Advanced):
For even more automation, add a @Config header to pre-configure settings:
@Config
{{http://localhost:8317/v1/chat/completions}}
{{gemini-3-pro-preview}}
{{Basic Latin}} {{Latin Supplement}} {{Telugu}}
{{TV show}} {{Psych}} {{2006}} {{, S01E01}} {{, (Title)}}
You are translating subtitles for the {{Enter movie or TV show}} {{Movie/TV Show Name}} ({{Release Year}}){{Season/Episode Number}}{{Episode Title}}.
[Rest of your translation instructions...]Configuration Format:
- Line 1:
@Configmarker - Line 2: API endpoint URL
- Line 3: Model name
- Line 4: Allowed script ranges (language scripts or Unicode ranges)
- Line 5: Default values for placeholders in the prompt body (must match the order and number of unique placeholders)
- Line 6: Empty line (required separator)
- Lines 7+: Your prompt template with
{{placeholders}}
When a configured template is imported:
- API endpoint, model, and script ranges are automatically applied
- Template inputs are pre-filled with the default values
- You can still modify any values before applying
Tip
See Example Config Prompt Template.txt for a complete configured template example.
Example Non-Template Prompt:
You are a hyper-vigilant subtitle translator and formatter. Your task is to translate an English .srt file into casual "Tenglish" (a mix of Telugu and English). Your primary directive is 100% accuracy in timestamps and script usage. Failure to adhere to these rules is not an option.
**Goal:** Translate the attached English .srt file for the TV show Psych (2006), S02E02, (65 Million Years Off) into casual Telugu for a native Telugu speaker who is not fluent in English.
### Core Directives (Must be followed without exception):
**ABSOLUTE TIMESTAMP INTEGRITY:** This is the most critical rule. The timestamps in the output file MUST BE an exact, character-for-character, byte-for-byte copy of the timestamps in the source file. Treat the entire timestamp line (HH:MM:SS,ms --> HH:MM:SS,ms) as a single, unchangeable piece of data—a unique ID that must be copied exactly as it appears. Do not parse, interpret, round, or alter it in any way. A single missing or incorrect character is a total failure.
✅ **CORRECT:** 00:14:56,928 (From Input SRT)
❌ **INCORRECT:** 00:14:6,928 (Single Digit Missing)
❌ **INCORRECT:** 00:14:46,928 (Single Digit Mismatch)
❌ **INCORRECT:** 00:14:33,928 (Double Digit Mismatch)
**PERFECT LINE BREAKS:** Preserve the original line breaks within each subtitle entry. If the original English subtitle has two lines, the translated Tenglish subtitle must also have two lines. Do not merge lines.
**SCRIPT & LANGUAGE RULES (CRITICAL):**
- **Definition of "Tenglish":** This is a Mixed-Script translation.
- English Words: Keep common English words, names (Shawn Spencer), and simple phrases (Who's in there?) in English.
- Telugu Words: Translate complex concepts and words (requisitioning) into casual Telugu.
- **NO ROMANIZATION:** Do not transliterate Telugu words into English letters. Use the Telugu script.
- ❌ **INCORRECT:** "Nenu ready ga unnanu."
- ✅ **CORRECT:** "నేను ready గా ఉన్నాను."
- **ALLOWED CHARACTERS:** The text may ONLY contain Standard English (Latin) characters and Telugu script characters. ABSOLUTELY NO characters from any other script are allowed. This includes, but is not limited to, Devanagari (Hindi: नी), Gujarati (હ), Japanese (気), Tamil (த), Kannada (ಚಿ), or Malayalam (മ).
### Error Correction Examples (Pay close attention to these):
This table shows the exact type of script errors to avoid and their correct replacements.
| Original English | ❌ Incorrect Translation (Mistake) | ✅ Correct Translation (Goal) |
|---|---|---|
| - I can't see a thing in here.<br>- I got it. | ఇక్కడ నాకు ఏమీ కనపడట్లేదు. నా దగ్గర ఉంది. (Merged lines) | - ఇక్కడ నాకు ఏమీ కనపడట్లేదు.<br>- నా దగ్గర ఉంది. (Correctly kept two lines) |
| a killer's window | ఒక હત్యకుడి window (Uses Gujarati હ) | ఒక హత్యకుడి window (Uses correct Telugu హ) |
| if you don't mind | మీరు気に పట్టించుకోకపోతే (Uses Japanese 気) | మీరు పట్టించుకోకపోతే (Uses correct Telugu script only) |
| I am not a suspect | నేను suspect नी కాదు (Uses Devanagari नी) | నేను suspect ని కాదు (Uses correct Telugu ని) |
| How are you? | Ela unnavu? (Romanized - BANNED) | ఎలా ఉన్నావు? |
| Stop the car. | Car aapu. (Romanized - BANNED) | Car ఆపు. |
### Final Verification Protocol:
Before providing the final output, perform a self-correction check. Review your entire translated file one last time to confirm:
1. **Timestamp Check:** Is every single digit of every timestamp identical to the source?
2. **Line Break Check:** Does the line count of every subtitle entry match the source exactly?
3. **Script Check:** Have you scanned the entire text to ensure there are zero characters from any script other than English and Telugu? Verify common error characters like హ vs. હ.
Please begin the translation of the attached file, adhering strictly to these enhanced directives.- Click "Choose Input SRT file..." and select your English subtitle file
- Click "Translate with Gemini"
- The tool will:
- Detect the target language
- Send the file to Gemini for translation
- Automatically validate the output
- Show any errors found
Tip
You can drag and drop files directly onto the Prompt Template and Subtitle file upload buttons.
If Gemini stops mid-translation, you'll see:
- "
⚠️ Partial Translation Detected!" - Progress: "Translated: 250 / 500 subtitles"
- A "⏩ Continue Translating" button
Click the button to automatically resume translation from where it stopped. This process repeats until the file is complete.
- View the translated subtitles in the editable text area
- Make manual corrections if needed
- Click "💾 Save Edits" to preserve changes
If validation found errors, click "🔧 Fix All Errors" to:
- Automatically correct timestamp mismatches
- Send script errors back to Gemini for correction
- Re-validate and show remaining issues
- "💾 Download Translated SRT": Downloads the corrected subtitle file
- Filename format:
OriginalName(Language)(Model).srt - Example:
Psych.S04E05.1080p.BluRay.Remux.eng(Tenglish)(Gemini 2.5 Pro).srt
- Filename format:
- "📄 Download Raw Output": Downloads all Gemini responses with timestamps for debugging
The tool includes 130+ pre-configured language scripts:
- Indian: Telugu, Hindi, Tamil, Bengali, Gujarati, Kannada, Malayalam, Punjabi, Odia, Sinhala
- East Asian: Chinese (6 CJK blocks), Japanese (Hiragana, Katakana), Korean (Hangul)
- Southeast Asian: Thai, Lao, Burmese, Khmer, Tagalog, Balinese, Sundanese, Javanese
- Middle Eastern: Arabic (5 variants), Hebrew, Persian, Urdu, Syriac
- European: Greek, Cyrillic (4 variants), Armenian, Georgian
- African: Ethiopic, Tifinagh, Coptic, Vai, Bamum
- Historical: Egyptian Hieroglyphs, Cuneiform, Phoenician, Linear B, Runic, Ogham
- And many more...
You can also add any custom Unicode range or character.
- Ensure CLI Proxy API is running:
ps aux | grep cliproxyapi - Check the endpoint URL matches:
http://localhost:8317/v1/chat/completions - Verify the port in
config.yamlmatches (default: 8317)
- Check your Gemini quota/limits
- Try using
gemini-2.5-flashinstead ofgemini-2.5-pro - Verify your prompt is clear and includes the SRT file content
- Ensure the API Key in the web app matches
api-keysinconfig.yaml - Try re-logging in:
cliproxyapi --login
- The Gemini service may be temporarily overloaded or down
- Wait a few minutes and try again
- Check Gemini's status at status.cloud.google.com
- If persistent, try switching to a different model (e.g.,
gemini-2.5-flash) - Verify your CLI Proxy API connection: restart the server and check logs
- Some errors may require manual correction
- Check that you've added all necessary language scripts (e.g., Telugu)
- Review the raw output file for Gemini's explanation
- Gemini may have hit token limits; try splitting the file manually
- Check the raw output file to see where Gemini stopped
Edit the Model Name field to use other models:
gemini-2.5-flash: Faster, cheaper, good for simple translationsgemini-2.5-pro: More accurate, better for complex translationsgemini-3-pro-preview: Latest model, best quality and reasoning- See CLI Proxy API documentation for all supported models
Warning
The application has only been tested with gemini-3-pro-preview/gemini-2.5-pro and only uses gemini-2.5-flash to detect the target translation language for the purpose of naming the output SRT file.
Add specific character ranges for edge cases:
- Emoji:
U+1F600-U+1F64F - Mathematical symbols:
2200-22FF - Box drawing:
2500-257F - Any single character:
U+20AC(€ symbol)
Contributions are welcome! Some areas for improvement:
- Support for streaming responses (real-time translation preview)
- Batch file processing
- Export validation reports as JSON/CSV
- Integration with other LLM providers
Note
The Gemini LLM was chosen due to a comparatively higher output token limit for its web client.
- Built on top of CLI Proxy API
- Inspired by the need for accessible, quality subtitles for non-English speakers
MIT License - see LICENSE file for details
Made with ❤️ for my mom and anyone else who wants to enjoy content in their native language.