Images are fixed by fixing source document.

The generation tool also should be fixed so that it stops with an error if it encounters a figure in an unsupported format rather than silently creating useless output with mixed up figures.

The tool has to parse an HTML document to extract the images. This is only practical if the document follows the convention when it comes to figures. In some cases, the documents had 2 images overlay-ed on one another. The format was understood but it is impossible to know which image was the "correct" image.

That said, the tool is much better at reporting errors so these issues can be caught during a manual review of output.