Alexandre Garnier

Please read https://docs.gitlab.com/ee/ci/docker/using_docker_images.html#override-the-entrypoint-of-an-image The image's entrypoint is used and as pointed in earlier comment, the entrypoint is to launch the chromedriver server. So just reset the entrypoint in the gitlab-ci...

This is quite logic as neither java nor maven are provided in the image `zenika/alpine-chrome:with-chromedriver`. You need to install them in your CI script or to extend the image first...

Some trainings use multiple title levels in same slide on purpose, so titles (at least more than 3 `###`) are not an indicator of slide change.

Wouldn't it be more an issue of font? Like depending on the build process it's using different alternatives of fonts (between CSS declared fonts, fonts available in container and fonts...

When checking the font actually used through the font info of the inspect tool of Firefox: - Web view:  - H2: Montserrat Bold https://github.com/Zenika/sensei/blob/08945d91491f8733e10e98dbb1f82c10a7d2f14c/src/slides/slides.css#L23 - text: Source Sans Pro...

@hgwood If you look closely, the boldness of the title is different between web view and PDF. Also in code bloc we can see that the first line is splitted...

When launching decktape with option `--screenshot`, which is doing a screenshot from pupeeter page object, the screenshot result is the same as in the PDF viewers.

Also I suspect some of the effects are from image sizing and placement, which can be subtly different in PDF than in browser. But I think most of this issue...

Useful reading abour PDF, fonts and reader interpretation: https://linuxfr.org/users/pied/journaux/pdf-mais-que-fait-la-police

No idea, it was not in framework...