image-generator | PHP 8.x to generate placeholder images | Data Manipulation library

I've also come across this problem. I don't think what you're trying to do is possible at the moment. There are not many ways for a plugin to hook into the build step directly:

• Transforms
• Linters
• Events

I think events would be the best solution. However, events also don't receive enough information to process a template in a structured way. I've opened an issue regarding this on Github. For your use-case, you'd need to get structured page data in this hook as well. Or eleventy would need to provide a build hook for each page. I suggest opening a new feature-request issue or adding a comment with your use-case to my issue above so those hooks can be implemented.

Another solution that requires a bit more setup for the users of your plugin would be to add your functionality as a filter instead of an automatic script that's applied to every template. This means that the users of your plugin would need to add their own template which passes the relevant data to your filter. Of course this also gives more fine-control to the user, which may be beneficial.

I use a similar approach for my site processwire.dev:

1. A special template loops over all posts and generates an HTML file which is used as a template for the generated preview images. This template is processed by eleventy. (source)
2. After the build step: I start a local server in the directory with the generated HTML files, open them using puppeteer and programmatically take a screenshot which is saved alongside the HTML templates. The HTML templates are then deleted.
3. This is integrated into the build step with a custom script that is executed after the eleventy build.

I've published the script used to take screenshots with Puppeteer as an NPM package (generate-preview-images), though it's very much still in alpha. But you can check the source code on Github to see how it works, maybe it helps with your plugin.

You don't need Invoke-Expression (and in fact, it's not recommended).

I think you can do this:

It's hard to find a model for each topic at a single place.

The general strategy could be searching in GitHub with the relevant tags ["tensorflow", "sport"].

You can generally find many models on model-zoo websites: https://modelzoo.co/

This is also useful: https://github.com/tensorflow/models

If you need code (probably with pre-trained weights): paperswithcode.com is a good place to search.