Models

In brief, models are how you can customize the output of the generated stories. You can link a model to your source, and the settings of that model are used to rewrite the articles from your source into actual news stories that can be read aloud by your newscaster.

A model can be of the type News, Weather, or Traffic. This type is crucial because the handling of the output will vary depending on it. Additionally, the settings you can modify on the source depend on the model type.

You may have already noticed that RadioNewsAI has some built-in models. In most cases, these built-in models are perfectly fine, meaning there’s no need for you to create your custom models. If you're daring and wish to build a custom model, please see the instructions below.

Content Model Settings

When creating a model, you can choose from "News," "Weather," or "Traffic." This selection is made because the output varies depending on the content type. Once a content model is created, its type cannot be changed.

Model settings form

  • Base Model - You can base your content model on another (built-in) model. This can be useful when you only want to make minor adjustments. As long as you do not add any examples or a custom script, the examples and scripts from the base model are used.
  • Use Source Content As-Is - Check this setting if you do not want to rewrite the source to which this model is attached. This can be useful if you want to use the content from an RSS feed exactly as it appears in the feed when read aloud.
  • Sentence Count - Set a limit on the number of sentences in the output.
  • Word Count - Set a limit on the number of words in the output.

The above settings only function if you include examples that implements these limits. Without appropriate examples, the AI may produce unexpected results when you modify sentence and word counts without providing custom examples.

Content Model Examples

The models tab on the model settings page

By "examples," we mean actual instances of news stories (or weather updates or traffic information) that adhere to how your model is set up. Each example should contain one news story, so ensure you create an example for each story separately, not adding an entire newscast to one example.

It's advisable to always include examples that comply with the model settings and the directions from the scripts. This assists the AI model in comprehending how to interpret all the settings. For instance, if you set the sentence limit to 3 and the word count limit to 50, provide examples that contain 3 sentences and approximately 50 words.

In addition to settings for length, the AI is trained to recognize and implement the tone and style present in your examples when rewriting your news stories. This enables you to have entirely custom news stories that incorporate your unique style.

It is recommended to add a minimum of 3 examples and no more than 6 examples.

For a "Weather" type model, it's necessary to provide 3 to 6 examples for each version of the weather forecast: morning, afternoon, evening, and night.

Content Model Script

The custom script you supply to your model is utilized as a "format" in the prompt to generate the rewritten version of an article from a source connected to your model. Therefore, it's your opportunity to add some additional instructions to the AI to personalize the output. Technically, you can add any custom parameter you desire here.

How to Set Up a Model Script

To implement a custom script in a content model:

How to setup a model script

  1. Select the Use Custom Script option in a content model.
  2. Save the model after enabling the custom script option.
  3. The script field will emerge below the content model form. Here, you can input your custom instructions.

In the case of a "Weather" type model, a script is necessary for generating weather forecasts for the morning, afternoon, evening, and night, as the format of these forecasts can vary significantly.

Scripts with Examples

Scripts function optimally when paired with well-constructed examples that implement these instructions. For instance, you might add a script for a weather model as follows:

{pick something from the data and summarize the weather in max 10 words, do not use temperatures}

This proves particularly effective when you supply examples like the ones below:

Cloudy with some sunshine tomorrow.

Rain expected later in the afternoon.

Rain expected later on tonight.

Custom stories

Applying the concepts from the above sections to the real world? It's pretty straightforward! See this quick start about how to customize stories by creating your own custom model.