Markup is breaking due to HTML entities generated in the alt text

[CjRootStudio]

Responses from GPT sometimes includes words or phrases surrounded by quote marks (or other HTML special characters/entities). This is causing issues where the generated alt text breaks the HTML markup.

Example

<img src="example.jpg" alt="Stacked coins of increasing height next to wooden blocks spelling "HEALTH." The arrangement symbolizes the connection between health and wealth.">

Notice how HEALTH is surrounded by quote marks. This is breaking the markup and causing multiple knock-on effects.

Suggested Fix

Could the returned response be further sanitised to escape special characters? There are cases where they might be relevant to the description so removing them entirely wouldn't be ideal. For example:

$result = trim($response->choices[0]->message->content);

// Escape the special characters
$sanitisedResult = htmlspecialchars($result, ENT_QUOTES, 'UTF-8');

Environment

Statamic: 4.58.2
PHP: 8.2.17
A.I.D.A: 1.0.1

[Sm1lEE]

Hi @CjRootStudio ,

thanks for your report. We included your suggested fix in the newest version and also provided an alternative solution for custom generators using Statamic's sanitize modifier.

[CjRootStudio]

Thanks! 🙏