Custom API: LLM (local or Cloud), Deepl(web), LibreTranslate, Yandex

[mastx911]

Hello, @killkimno!
I would like to express my sincere gratitude for the MORT Translate program. It is an incredibly useful and versatile tool for translating games in overlay mode. Thank you for your contribution!
I apologize if my formatting does not comply with GitHub standards, as I am not a programmer and have not previously used this resource for similar requests. I am pasting all the information as text for ease of reading.
You can do whatever you want with the instructions or files; I don't need them.

=============================================================================

Performance Comparison and Custom Translators
I tested the default (built-in) MORT translators and three custom solutions that I configured. I believe these configurations may be useful to the community, and I am happy to share them.
Comparative Test Results
The table below shows the quality score and translation speed for two types of text:

Important Note about Gemini
Gemini (free) showed excellent results in terms of quality and speed. However, due to its strict request limit (about 10), it is unfortunately unsuitable for continuous use in MORT.
Now let's move on to the instructions for setting up three custom translators.

Links for convenience
1 Custom Cloud Setup - below
2 Custom LibreTranslate Setup - #112 (comment)
3 Custom Local Ollama Setup - #112 (comment)
4 Custom Deepl (web) Setup (Recommended) - #112 (comment)
5 Custom Yandex Setup - #112 (comment)

=============================================================================

1 Custom Cloud Setup

Setup Steps in MORT
Follow these steps to set up the Custom Cloud API in MORT:

1. Open MORT and select Custom API from the list of translators.
2. Go to the Translation section and find Google Basic settings. This section is responsible for defining the source and target languages for your custom API.
   
3. Go to the Additional section and set it to Fast (this will help Mort not lose text when receiving it).
   
4. Go to the Additional section, then to Advanced Settings.
   Scroll down to the Translation subheading and scroll to the very bottom.
   
   Be sure to enter the following URL in the host field:

http://127.0.0.1:5000/translate

Important note: Use strictly 127.0.0.1 instead of localhost. Using localhost can add a delay of ~2 seconds to each request, which will critically affect the translation speed.
4) Don't forget to click Apply everywhere!
This completes the configuration of the MORT program.

=============================================================================

Custom Cloud

Obtaining an API key for Custom Cloud

To work with Custom Cloud, you will need an API key from the relevant service.

1. Registration: Go to the website and register (using your email or Google account):
   https://id.io.net/
2. Creating a Key: Go to the link below and click the “Create Key” button:
   https://id.io.net/ai/api-keys
   Name: You can enter any name for the key.
   Expiration date: Be sure to select the maximum expiration date (for example, 180 days) so that you don't have to repeat this procedure often.
3. Saving the Key (Critically Important!):
   Immediately copy the API key you received and save it in a secure location.
   Warning: The full key is only displayed once! If you lose it or fail to save it, you will have to create a new key in your profile settings.

=============================================================================

Configuring the Cloud Script (Cloud API Server)
Installing Python

1. Download Python: Follow the link and download the installer (python-3.10.11-amd64.exe):
   https://www.python.org/downloads/
   or quick download
   https://www.python.org/ftp/python/3.10.11/python-3.10.11-amd64.exe
2. Installation (Critically Important!): When installing, be sure to check the box Add python.exe to PATH. Leave all other settings at their default values.
   
3. Installing Libraries:
   Open the console (press Win + R and type cmd).
   Paste the following command
   python.exe -m pip install --upgrade pip && python -m pip install flask requests
   and press Enter. Wait for the installation to complete, then close the console:
   (This command will update the pip installer and install the necessary flask and requests libraries for the script to work.)
4. Download the Script: Get the script file itself ( Start Translation Cloud or Local.py).
   https://drive.google.com/drive/folders/1bUH4qvCLydyQqOsyi243_KuH6tXMlaau?usp=drive_link
5. Preliminary Launch: Launch the downloaded script (double-click on it) (Start Translation Cloud or Local.py). If Python was installed correctly, a console window should open. This confirms that the script is working. Close it.
6. Open the Start Translation Cloud or Local.py file via the context menu (Right click -> Open with -> Notepad or another editor).
   Find the place in the code where you need to insert your key. The script provides space for two cloud models (one main and one backup).
   
   Insert the API key you received between the quotation marks “”, without spaces.
   Example of key insertion:
   
   Saving: Be sure to save your changes to the file (Ctrl+S).

=============================================================================

Launch and Operation (MORT + Script)
Launch Procedure

1. Launch the Script: Double-click on the script file (Start Translation Cloud or Local.py). A console should open. Do not close this window while you are using MORT! and do not click inside this window.
2. Launch MORT: Launch MORT as usual and configure it for the game.
   Done: MORT will now send requests to your locally running server (127.0.0.1:5000), which in turn uses the cloud API.
3. Additional Information about the Script and Tokens

1. Automatic Model Switching (Primary/Backup):
   The script uses two cloud models (primary and backup) to ensure uninterrupted operation.
   You always use the primary model. If it runs out of tokens, the script will automatically switch you to the backup model.
   There will be a 1-second delay when switching (only for the first request). There will be no further delays.
   The script will periodically check whether the main model's limits have been restored and will automatically switch you back at the first opportunity.
2. Translation Context:
   The script implements context for cloud models. You can leave this feature enabled or disabled at your discretion.
3. API Token Limits:
   In API (request) mode, each cloud model provides 500,000 tokens, which are refreshed every 24 hours.
   This volume is sufficient to send more than 1,000 requests with texts of 153 words each (plus instructions/prompts).
   Thus, one model should be enough for a whole day of active use. However, if the limit for one model is exhausted, the script activates a second model, which will also provide 500,000 tokens.
   Limits are updated every 24 hours. (There are 12 models available, but only the most suitable ones were selected for maximum speed and quality).

All available AI
Qwen:
Intel/Qwen3-Coder-480B-A35B-Instruct-int4-mixed-ar (9 out of 10 from Gemini), 20-40 seconds (sometimes drops out)
Qwen/Qwen3-235B-A22B-Thinking-2507 (0 out of 10 from Gemini),Think
Qwen/Qwen3-Next-80B-A3B-Instruct (9.3 out of 10 from Gemini) 4-5 sec
Qwen/Qwen2.5-VL-32B-Instruct (5.3 out of 10 from Gemini) 5.5-6.8 sec

Llama:
meta-llama/Llama-3.3-70B-Instruct (8.7 out of 10 from Gemini) 7.5-8.5 sec
meta-llama/Llama-3.2-90B-Vision-Instruct (9.3 out of 10 from Gemini) 9.5-15 sec
meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8 (9.2 out of 10 from Gemini), 3.2-4 sec

OpenAI:
openai/gpt-oss-120b (9.5 out of 10 from Gemini), 4-8 sec
openai/gpt-oss-20b (8.5 out of 10 from Gemini), 3-8 sec

Mistral:
mistralai/Mistral-Large-Instruct-2411 (5.9 out of 10 from Gemini) 15-16 sec (not compatible with temp 0.7, needs to be lower)
mistralai/Mistral-Nemo-Instruct-2407 (5.5 out of 10 from Gemini) 5-6.5 sec (not compatible with temp 0.7, needs to be lower)
mistralai/Devstral-Small-2505 (8.5 out of 10 from gemini) 7.5-8sec (not compatible with temp 0.7, needs to be lower)
mistralai/Magistral-Small-2506 (1.7 out of 10 from gemini) 9.5-10.5 sec (not compatible with 0.7 temp, needs to be lower)

Others:
swiss-ai/Apertus-70B-Instruct-2509 (9.2 out of 10 from Gemini) 7.5-10 seconds
deepseek-ai/DeepSeek-R1-0528 (9.5 out of 10 from Gemini), 60-80 seconds think
LLM360/K2-Think (3 out of 10 from Gemini) 8-50 sec think , (not compatible with temp 0.7, needs to be lower)

Possible useful links on IO Intelligence:

1. AI Models (Catalog and Chat for testing):
   https://ai.io.net/ai/models
2. Status of models and services:
   https://status.io.net/
3. Documentation and API Reference:
   https://docs.io.net/reference/get-started-with-io-intelligence-api

Below, I will add more API methods.

[mastx911]

2 Custom LibreTranslate Setup (You need 4 GB of hard disk space on drive C and 1 GB of RAM.)

Setup Steps in MORT
Follow these steps to set up the Custom API in MORT:

1. Open MORT and select Custom API from the list of translators.
2. Go to the Translation section and find Google Basic settings. This section is responsible for defining the source and target languages for your custom API.
   
3. Go to the Additional section and set it to Fast (this will help Mort not lose text when receiving it).
   
4. Go to the Additional section, then to Advanced Settings.
   Scroll down to the Translation subheading and scroll to the very bottom.
   
   Be sure to enter the following URL in the host field:

http://127.0.0.1:5000/translate

Important note: Use strictly 127.0.0.1 instead of localhost. Using localhost can add a delay of ~2 seconds to each request, which will critically affect the translation speed.
4) Don't forget to click Apply everywhere!
This completes the configuration of the MORT program.

=============================================================================

LibreTranslate Setup
Python Installation

1. Download Python: Follow the link and download the installer (python-3.10.11-amd64.exe):
   https://www.python.org/downloads/
   or quick download
   https://www.python.org/ftp/python/3.10.11/python-3.10.11-amd64.exe
2. Installation (Critically Important!): When installing, be sure to check the box Add python.exe to PATH. Leave all other settings at their default values.
   
3. Installing Libraries and LibreTranslate:
   Open the console (press Win + R and type cmd).
   Paste the following command
   pip install libretranslate && python.exe -m pip install --upgrade pip && python -m pip install flask requests
   and press Enter. Wait for the installation to complete, then close the console:
   (This command will update the pip installer and install the necessary flask and requests libraries for the script to work, and install LibreTranslate )
4. Download 2 files: Get the script and batch files (LibreTranslate).
   https://drive.google.com/drive/folders/1ULR_AOqBgGF0UotplMafmXORpCaNLybM?usp=sharing
5. Copying app.py: Copy the app.py file to the LibreTranslate system folder.
   Example path (for standard Python installation):
   C:\Users\YOUR_USER\AppData\Local\Programs\Python\Python310\Lib\site-packages\libretranslate
   Note: Make sure you replace the existing app.py file in this directory.
6. Editing the Startup File (.bat): Open the START_libretranslate.bat file (right-click -> Edit).
   Language Settings:
   In this file, you can specify which language models to load.
   List the desired language pairs separated by commas (e.g., uk, en, ja, ko):
   Example:...
   --load-only uk,en,ja
   Save the changes to the file.
7. Core Settings (Optional): In the same file, you can configure how many processor cores LibreTranslate will use for its work.
8. First Launch and Model Download:
   Double-click on the START_libretranslate.bat file.
   The console will start and begin downloading the necessary language models. Each model is approximately 150-200 MB in size.
   Be sure to wait until the download is complete.
   Successful completion will be confirmed by the line:
   Running on http://127.0.0.1:5000

=============================================================================

Launch and Operation (MORT + LibreTranslate)
Once you have completed the configuration, you can launch MORT and LibreTranslate in the following order:

1. Launching the LibreTranslate Server:
   Double-click on the START_libretranslate.bat file. A console window should open.
   Important: DO NOT close this console window and DO NOT click inside it while you are using MORT, otherwise the server will stop!
2. Launch MORT: Launch MORT as usual and configure it for the game you want to use it with.
   Done: MORT will send requests to your local LibreTranslate server using the language models you have configured.

Below, I will add more API methods.

[mastx911]

3 Custom Local Ollama Setup (You need 1.5-8 GB of VRAM.)

Setup Steps in MORT
Follow these steps to set up the Custom API in MORT:

1. Open MORT and select Custom API from the list of translators.
2. Go to the Translation section and find Google Basic settings. This section is responsible for defining the source and target languages for your custom API.
   
3. Go to the Additional section and set it to Fast (this will help Mort not lose text when receiving it).
   
4. Go to the Additional section, then to Advanced Settings.
   Scroll down to the Translation subheading and scroll to the very bottom.
   
   Be sure to enter the following URL in the host field:

http://127.0.0.1:5000/translate

Important note: Use strictly 127.0.0.1 instead of localhost. Using localhost can add a delay of ~2 seconds to each request, which will critically affect the translation speed.
4) Don't forget to click Apply everywhere!
This completes the configuration of the MORT program.

=============================================================================

Configuring the Local Script (Local API Server)
Installing Python

1. Download Python: Follow the link and download the installer (python-3.10.11-amd64.exe):
   https://www.python.org/downloads/
   or quick download
   https://www.python.org/ftp/python/3.10.11/python-3.10.11-amd64.exe
2. Installation (Critically Important!): When installing, be sure to check the box Add python.exe to PATH. Leave all other settings at their default values.
   
3. Installing Libraries:
   Open the console (press Win + R and type cmd).
   Paste the following command
   python.exe -m pip install --upgrade pip && python -m pip install flask requests
   and press Enter. Wait for the installation to complete, then close the console:
   (This command will update the pip installer and install the necessary flask and requests libraries for the script to work.)
4. Download the Script: Get the script file itself ( Start Translation Local.py).
   https://drive.google.com/drive/folders/1zo-QOt-xyS3jb17k6EtstGNFdHO08LPz?usp=sharing
5. Preliminary Launch: Launch the downloaded script (double-click on it) (Start Translation Local.py). If Python was installed correctly, a console window should open. This confirms that the script is working. Close it.
6. Changing the Model Name: In the configuration or launch file, you can change the name of the local model being used.
   
7. List of available models that I have tested:
   gemma3n:e4b (10/10 quality assessment) 7.8GB Vram 5.8 sec
   gemma3n:e2b (8/10 quality assessment) 6.0gb Vram 4.8 sec
   zongwei/gemma3-translator:4b (7 out of 10 quality assessment) 5.0gb Vram 4.3sec
   gemma2:2b (5 out of 10 quality assessment) 2.8gb Vram 4.2sec
   gemma3:1b (4 out of 10 quality assessment) 1.2gb Vram 3.3sec

=============================================================================

1. Installing and Downloading Models
   Download Ollama: Follow the link and download the program:
   https://ollama.com/
2. Install the program.
   Important: In the Ollama settings, select a storage location for local models. (They range in size from 2 GB to 8 GB.)
3. Downloading Language Models (LLM):
   Open Ollama and use its web interface.
   Select the desired model using the name from your list and start the download by requesting it in the chat.

=============================================================================

1. Launch the Script: Double-click on the script file (Start Translation Local.py). A console should open. Do not close this window while you are using MORT! and do not click inside this window.
2. Launch Ollama and do not close it.
3. Launch MORT: Launch MORT as usual and configure it for the game.

Tested on i5 12600k / 32GB RAM / RTX 3080

[killkimno]

Thank you! ill research it

[mastx911]

Hello again!

I would like to share some ideas and observations about possible improvements to your wonderful program:

1. Text recognition (OCR): I was looking for alternatives to WindowsOCR and found that Snipping Tool demonstrates significantly better recognition quality. It might be worth considering integrating its methods to improve accuracy. I admit that I am not an expert in this field, unlike you, who created such a functional tool, but the difference in results seemed significant to me.

— Note on additional free OCR engines:
I also looked for other, more accurate engines that can be used offline and for free (as an alternative to WindowsOCR):
PaddleOCR: It is said to be one of the most accurate for complex text (on a background or at an angle).
RapidOCR: This is a faster version of PaddleOCR, which is better suited for real-time translation because it is well optimized to run on a regular processor (CPU).

2. Text-to-speech (TTS) conversion: I found the following alternatives for voiceovers:
   Online: Edge TTS (for high quality).
   Offline: Voicevox (for offline use).

I think adding these features would make your versatile program even more valuable, and the community would be very grateful to you.

Thank you for your work!

[killkimno]

Thanks to share information

Snipping Tool is currently under application research.

[mastx911]

4 Custom Deepl (web) Setup (Recommended) (Requires 1.2 GB of RAM.)
Note: Bypasses most DeepL restrictions, but not all.

Setup Steps in MORT
Follow these steps to set up the Custom API in MORT:

1. Open MORT and select Custom API from the list of translators.
2. Go to the Translation section and find Google Basic settings. This section is responsible for defining the source and target languages for your custom API.
   
3. Go to the Additional section and set it to Fast (this will help Mort not lose text when receiving it).
   
4. Go to the Additional section, then to Advanced Settings.
   Scroll down to the Translation subheading and scroll to the very bottom.
   
   Be sure to enter the following URL in the host field:

http://127.0.0.1:5000/translate

Important note: Use strictly 127.0.0.1 instead of localhost. Using localhost can add a delay of ~2 seconds to each request, which will critically affect the translation speed.
5) Don't forget to click Apply everywhere!
This completes the configuration of the MORT program.

=============================================================================

Deepl (web) Setup
Python Installation

1. Download Python: Follow the link and download the installer (python-3.10.11-amd64.exe):
   https://www.python.org/downloads/
   or quick download
   https://www.python.org/ftp/python/3.10.11/python-3.10.11-amd64.exe
2. Installation (Critically Important!): When installing, be sure to check the box Add python.exe to PATH. Leave all other settings at their default values.
   
3. Installing Libraries:
   Open the console (press Win + R and type cmd).
   Paste the following command
   python -m pip install --upgrade pip && python -m pip install flask requests websocket-client pywin32
   and press Enter. Wait for the installation to complete, then close the console:
   (This command will update the pip installer and install the necessary flask and requests libraries for the script to work )
4. Preparing the working environment
   Script: Download the Mort+Deepl (web).py file from Google Drive.
   https://drive.google.com/drive/folders/1TyeoIrNz7gOtAVOVZOH9SfXrv1rXNjM5
   Browser: For stability, we recommend using the portable version of Chrome. Download it here: Google Chrome Portable 64-bit. (300mb)
   https://portableapps.com/apps/internet/google-chrome-portable-64
   Unzip it to any convenient location.
5. Copy the Mort+Deepl (web).py script file to the folder with the unpacked browser at the following path:
   GoogleChromePortable64\App\Chrome-bin
   For convenience, create a shortcut to this script on your desktop.
6. Run the script via the shortcut.
   Wait: Wait until the browser has fully loaded (this may take up to 60 seconds).
   Successful completion will be confirmed by the line:
   BROWSERS ARE READY TO GO. API GATEWAY IS RUNNING ON 127.0.0.1:5000

=============================================================================

Startup and operation (MORT + Deepl (web))
Once you have completed the configuration, you can start MORT and Deepl in the following order:

1. Start the Deepl server (web):
   Double-click your shortcut and wait about 60 seconds. A console window will open.
   Important: DO NOT close this console window or click on it while using MORT, otherwise the server will stop!
2. Launch MORT: Launch MORT as usual and configure it for the game you want to use it with.
   Done: MORT will send requests to your local Deepl server (web) using the language models you have configured.

=============================================================================

Configurable parameters (Buttons):
ENABLE_CONTROL_LOG: Enables/disables detailed logging to the console and Log.txt file.

HEADLESS_MODE_ENABLED: if True, browsers run in hidden mode (Faster translation by ~0.5s). If False, windows are visible (Slower due to interface rendering).

CONTEXT_ENABLED: Improves translation quality by remembering the previous phrase. (Note: highly recommended for better accuracy, though it may rarely cause context-related glitches).

TRANSLATION_TIMEOUT: The time in seconds that the script waits for a response from DeepL before switching to another pool.

CHROME_LANG_CODE: Allows you to set the browser interface language (default is en). (Do not change unless absolutely necessary.)

REQUEST_RESTART_LIMIT: The script switches the active browser to a fresh one and sends the old one to restart and clear. This solves the problem of memory leaks and DeepL freezes during prolonged operation. It also bypasses blocking.

[mastx911]

5 Custom Yandex Setup
(well suited for translation into Cyrillic)

Setup Steps in MORT
Follow these steps to set up the Custom API in MORT:

1. Open MORT and select Custom API from the list of translators.
2. Go to the Translation section and find Google Basic settings. This section is responsible for defining the source and target languages for your custom API.
   
3. Go to the Additional section and set it to Fast (this will help Mort not lose text when receiving it).
   
4. Go to the Additional section, then to Advanced Settings.
   Scroll down to the Translation subheading and scroll to the very bottom.
   
   Be sure to enter the following URL in the host field:

http://127.0.0.1:5000/translate

Important note: Use strictly 127.0.0.1 instead of localhost. Using localhost can add a delay of ~2 seconds to each request, which will critically affect the translation speed.
5) Don't forget to click Apply everywhere!
This completes the configuration of the MORT program.

=============================================================================

Yandex Setup
Python Installation

1. Download Python: Follow the link and download the installer (python-3.10.11-amd64.exe):
   https://www.python.org/downloads/
   or quick download
   https://www.python.org/ftp/python/3.10.11/python-3.10.11-amd64.exe
2. Installation (Critically Important!): When installing, be sure to check the box Add python.exe to PATH. Leave all other settings at their default values.
   
3. Installing Libraries:
   Open the console (press Win + R and type cmd).
   Paste the following command
   python.exe -m pip install --upgrade pip && python -m pip install flask requests
   and press Enter. Wait for the installation to complete, then close the console:
   (This command will update the pip installer and install the necessary flask and requests libraries for the script to work )
4. Preparing the working environment
   Script: Download the Mort+yandex.py file from Google Drive.
   https://drive.google.com/drive/folders/1s81lqufbTCCnrUEbZ7VbYAzknSoxDcD5

=============================================================================

Startup and operation (Mort+yandex)
Once you have completed the configuration, you can start MORT and Yandex in the following order:

1. Start the Yandex server (web):
   Double-click your Mort+yandex.py. A console window will open.
   Important: DO NOT close this console window or click on it while using MORT, otherwise the server will stop!
2. Launch MORT: Launch MORT as usual and configure it for the game you want to use it with.
   Done: MORT will send requests to your local Yandex server using the language models you have configured.