Upload a Book, Get a Cast

You can now upload a manuscript and get a cast list back. Chapters are detected automatically, and every line in the book is assigned to a character: dialogue attributed to whoever spoke it, narration left with the narrator. You cast a voice per character and generate.

This is the part of turning a book into a multi-voice audiobook that nobody talks about, because until now it was the part you did by hand. If you have a full-length ebook and just want it read in one voice, our general ebook-to-audiobook guide covers that flow. This post is about the new casting pass specifically: what it detects, what it gets right, what it gets wrong, and where you fix it.

Upload a manuscript in EPUB, TXT, PDF, or DOCX. The system splits it into chapters, reads through every line, and decides who is speaking: the narrator, or a character. Lines of dialogue get attributed to a name where the text makes that clear; everything else stays with the narrator. What comes back is a cast list built from your own book, not a generic template, ready for you to assign a voice from the library to each name.

The whole flow runs in four stages, and you can stop and check the work at any of them before spending a single point on generation.

The upload accepts EPUB, TXT, PDF, and DOCX. It does not accept MOBI; convert that file to EPUB first with any free converter before you upload it. The output is per-chapter audio, not a packaged M4B file, and there is no built-in pipeline to an audiobook distributor. You assemble and submit the final files yourself.

The old way to make a multi-voice audiobook meant reading through your own manuscript first, marking every line as narration or dialogue, and noting which character said what, all before you could touch a single voice setting. That step alone could take longer than recording the book. The casting pass exists to do that marking for you, so the first thing you see after upload is a list of characters and their lines, not a blank page waiting for you to start tagging.

This is a first pass over a real manuscript, and dialogue attribution has genuinely ambiguous cases: an unmarked line after a group scene, a name that only appears three chapters later, a narrator who briefly speaks as a character. Do not expect a perfect cast list on the first read. Expect a cast list that has already done the tedious 90% of the work, so you spend your time on the interesting 10%: the handful of lines that need a human decision.

Think of the analysis as a draft you check, not a final answer you publish. Skim the cast list chapter by chapter, listen for anything that reads oddly once it is voiced, and correct the attribution in the editor. That correction takes seconds per line; doing the attribution from scratch across a full novel does not.

Once the cast list is confirmed, casting is just picking a voice per name from the library. Aim for contrast between characters who share scenes, and keep your narrator distinct from every character so listeners always know when the narration has handed off to dialogue. The six voices below are a reasonable starting cast for a book with a narrator, two or three principal characters, and one gruffer supporting voice.

The analysis itself is a normal part of the audiobook workflow. Generating the audio costs points the same way any narration does, charged by character count for the chapter you render, not by the length of the finished audio. A free account gets a one-time trial allowance, enough to run the analysis and generate a sample chapter so you can judge the cast list on your own book before you commit to a paid plan. See pricing for the plan breakdown if you plan to narrate a full-length manuscript.