Setting-up access for using an AI Service

To use a certain AI model of one of the AI services you will need to:

• create an account with the AI service
• subscribe to a plan which will allow using the API of the AI service
• generate an “API Key” and use the API key in FAB Subtitler options when configuring the usage of the AI service

Configuration of AI Services in FAB Subtitler

• In FAB Subtitler Options select Specials / AI
• Add a new AI Service and enter all necessary settings as seen below

• You will have to use a different URL for every AI service. Please consult the AI service docs to find the correct URL.
• You will have to enter the AI Model that shall be used. Clicking on the button “…” will display a list of available models.
• The maximum number of reply tokens limits the size of the reply. You may require a higher number if you are analyzing a complete subtitle file.
• You can enter the approx. price for one million tokens (a sum for input and output tokens) and in that case FAB Subtitler will display the approx. price of the AI request and reply before executing the AI request.

AI Models with good quality

FAB has tested many AI models and the following the list of AI models that work well with FAB Subtitler:

• ChatGPT gpt-4.5-preview
• Anthropic claude-sonnet-4-20250514
• Google Gemini gemini-2.5-pro

Configuration of AI Requests in FAB Subtitler

For every task that will involve using AI in FAB Subtitler it is necessary to configure the AI Request very exactly. This way all AI tasks are preconfigured and can be executed very quickly in FAB Subtitler without a lot of thinking:

• In FAB Subtitler Options select Specials / AI
• Add a new AI Request and enter all necessary settings as seen below

Note that the AI Prompt is used to tell to the service what type of data is provided and the AI Query is the actual instruction what result should be provided by the AI service.

Example FAB queries are provided as examples when opening the “Analyze with AI” window. By clicking on “Edit AI Request” button you can see how the example queries were created so that you can create similar queries.

Requesting AI Replies in JSON Format

The AI Service can be instructed to return results in structured JSON format. In this case FAB Subtitler will not display the result as simple text in a separate window. FAB Subtitler can read results from the JSON structure, evaluate it and display the data in the “search result list”.

It is however important that the result JSON structure is a list which contains a record of results and every record must contain the field “number” indicating the number of the subtitle or the number of the phrase.

Example of JSON result that can be processed by FAB Subtitler:

[  
  {  
    "number": 10,  
    "previous_text": "Stabbing quadreds.",  
    "proposed_text": "Stabbing quadriceps.",  
    "message": "Fixed spelling of 'quadreds' to 'quadriceps'."  
  },  
  {  
    "number": 12,  
    "previous_text": "Everything pouncing through that door.",  
    "proposed_text": "Everything pouring through that door.",  
    "message": "Replaced incorrect word 'pouncing' with 'pouring' to convey people streaming in."  
  },  
  {  
    "number": 15,  
    "previous_text": "Can't wait until she comes home.",  
    "proposed_text": "I can't wait until she comes home.",  
    "message": "Added subject 'I' for grammatical completeness."  
  },  
  {  
    "number": 18,  
    "previous_text": "Oh, apart from having a brain injury, never better.",  
    "proposed_text": "Oh, apart from having a brain injury, I've never been better.",  
    "message": "Added subject and auxiliary verb for correct idiomatic expression."  
  }
]

The AI prompt that was used with ChatGPT o4-mini model and produces the above JSON result was:

You are a language editor. I will give you a list of subtitles in JSON format. Each subtitle has a sequence number and a short text. Your task is to:
 
- Analyze the text for spelling and syntax issues.

- Propose a corrected version if needed.

- Return **only** subtitles where you detected and fixed a problem.
 
For each returned subtitle, include:
 
- number: The subtitle's original number (from the input field `Number`, converted to an integer).

- previous_text: The original subtitle text.

- proposed_text: The corrected version of the text.

- message: A short explanation of the problem and how it was fixed (e.g., spelling correction, grammar improvement, punctuation fix).
 
Use this output format:
 
```json

{

  "number": 1,

  "previous_text": "Original subtitle text.",

  "proposed_text": "Corrected subtitle text.",

  "message": "Explanation of the correction."

}

```
 
Do not include subtitles where no problems were found.
 
Here is the input:

Creting efficient AI prompts

To create AI prompts which will return good results the best way is to ask the AI service directly how to write a prompt so that it will return the result that you require.

This page was last updated on 2025-07-02