What a read-through tells you
- Dialogue that reads well but sounds stiff. On the page you supply the rhythm; out loud, you hear what's actually written.
- Characters who sound alike. If you can't tell two characters apart with your eyes closed, their dialogue may be too similar too.
- Exposition that stops a scene dead.
- Runtime. Hearing it is far more reliable than estimating; the old rule of thumb is about a minute per screenplay page.
- Jokes: whether the rhythm of a joke works when it's said, not read.
How to run one
- A project per screenplay or play, a script per scene or act.
- Convert the draft to
Name: linedialogue. See turning a screenplay into a voiceover script. - Cast distinct voices: your own samples in different characters, or friends who lend their voices. Distinctness matters more than resemblance to your dream cast.
- Decide on a narrator for action lines. A light touch, only what listeners need to follow, keeps the read moving.
- Generate every scene with Generate all new lines.
- Listen through in order, with the script open, making notes.
Revising
Rewrite a line, generate it again, and listen to the scene once more. Only the lines you changed are re-made, so draft two of a scene costs a few takes. Keep earlier drafts as separate scripts if you want to compare versions.
Sharing the read
Share the project with co-writers, a producer or a writing group. With Can view they can listen and download; with Can edit they can rewrite lines themselves. Everyone can leave notes in the comments, by scene and line number. See collaborating on a project.
What it doesn't replace
A read with real actors gives you their instincts: the line they find a better way to say, the moment they question. An AI read gives you the draft exactly as written, as often as you like, at no cost to anyone's time. Many writers use the AI read to get the draft ready for the human one. See AI voices vs voice actors.
Keep it private, or label it
Table reads are usually just for the team. If you publish one, for example as a pitch, say the voices are AI-generated, and only use voices cloned with permission.