Load EPUB File

Every audiobook workflow in this pack starts here. LoadEPUB is the front door of comfyui_epub_tts: it reads a .epub from your ComfyUI input folder, strips out all the book scaffolding, and hands the real content to the rest of the graph as a SPINEWALKER object.

The part that matters more than you'd think: an EPUB isn't a text file. It's a zip of HTML documents plus a "spine" that says what order to read them in, and it's stuffed with copyright pages, dedications, tables of contents, and cover art. If you just dumped that at a TTS model you'd get a minute of "Nav. Table of contents. Copyright" before the book started - and the model would say all of it. This node is the filter that makes the difference between an audiobook and a robot reading a website.

What it actually does

Under the hood it uses ebooklib to read the book, then builds a SpineWalker that walks the spine and keeps only actual document content. The filtering is where you have control:

• skip_keywords - a comma or newline-separated list of keywords matched against each file's idref or filename: about, acknowledgement, colophon, copyright, cover, dedication, front, nav, praise, preface, title, toc by default. That's your front-and-back-matter filter.
• require_int_chapters - default true, and the sneaky one. It rejects any chapter whose filename has no digits in it. Most real EPUBs name chapters like ch_001.xhtml or 001.xhtml, and non-numeric files are usually navigational junk. If your book has chapter files named chapter-one.xhtml, you'll want this off.
• front_matter_keywords - chapter, part, prologue by default. Used to detect the first real chapter: anything before the first page that mentions one of those words gets rejected as front matter.
• min_text_length - drops near-empty pages (default 1, so basically only blank pages).

Then comes the part nobody sees: each kept page's HTML gets run through BeautifulSoup (which drops <details> blocks) and trafilatura (which extracts main content and throws away tables, comments, and images - tables sound like gibberish read aloud). URLs get stripped and whitespace collapsed, because a TTS model will cheerfully read out a 40-character web address. The output epub_data is that clean text, packaged as the pack's internal SPINEWALKER type.

The inputs that matter

The dropdown epub_file is the one you'll actually touch - it lists every .epub in your ComfyUI input folder (yes, that means the book has to live in ComfyUI/input, and you may need to refresh/restart for a new file to show up). skip_keywords is the second most useful knob when a book's front matter doesn't match the defaults. The epub_data output only plugs into PreviewEPUB or EPUBChapterSplitter - that custom type is deliberate, so don't expect it to connect to arbitrary text nodes.

Install

ComfyUI Manager (search comfyui_epub_tts) or:

cd ComfyUI/custom_nodes
git clone https://github.com/kallama/comfyui_epub_tts

Restart ComfyUI, then drop your book in ComfyUI/input. Dependencies are ebooklib, bs4, and trafilatura - all lightweight, no model files. Two honest warnings: it needs DRM-free EPUBs (ebooklib will choke on a locked Kindle-era file - it's not a DRM stripper), and the pack is a young v0.0.1 with a boilerplate README, so if the defaults misbehave, tweak the keyword lists rather than hunting for documentation.