The VN Theater playbook: before the curtain, during the show, after

The builder’s note answered what Theater mode is and why it refuses to guess who’s speaking. This is the other document: the player’s manual. It assumes you already know the pitch — any novel on your shelf can perform itself as a visual novel, in place, no conversion (the Theater page has a demo you can poke at in the browser) — and it walks the parts you actually operate, in show order: before the curtain, during, after. Everything below is shipping behavior in the current build. No futures.

Before the curtain: which book, and what to prepare

Almost nothing. The entrance lives in the reading page’s menu, and any book can walk through it, because compiling prose into a stage script calls no AI, costs nothing, and works offline. Trying it on a whim is free in every sense.

What actually shapes the evening is the book itself. Dialogue-dense fiction — romance, ensemble casts, light novels — fills the stage with nameplates and dialogue boxes; prose-heavy literary work performs too, just with more of its lines delivered as narration. It’s worth saying what the theater is not: the open-source tradition for novel-to-VN is offline conversion, a pipeline that outputs a separate game project for an engine like Ren’Py, divorced from the book forever after. The theater skips that step entirely, which is also why “conversion failed” is not an error it can produce.

One preparation that pays: give your recurring characters profiles. The theater assigns speakers only on strong structural evidence and sends every uncertain line to the narrator, so the more the cast is described, the more lines earn a nameplate. Skippable. Worth it.

During the show: pacing, speeds, and AUTO’s one rule

Tap the dialogue box to advance — that’s the whole core loop. The typewriter has three speeds (20, 45, and 80 characters per second; the impatient go straight to the top). AUTO mode takes over pacing when you let it: it dwells on each line in proportion to length, rolls from scene to scene on its own, and obeys one hard rule — a choice panel always stops it. Watching a chapter on AUTO feels close to watching subtitled animation, hands nowhere near the screen.

Three small moves, easy to miss: swipe up for the backlog (closing it backs out of that layer only), long-press to hide the whole interface, and don’t worry about the black title card between chapters — that’s a scene change, not a hang.

During the show: how backdrops spend money, and how they don’t

Backdrops are one of exactly two actions in the theater that cost anything (the other is confirming a story choice). Both show you the bill before any work starts. A single scene: tap Generate this scene, read the cost confirmation, and on approval the app renders a backdrop from the scene’s actual prose — our first test run measured about 35 seconds — after which that scene reuses its image on every future visit, zero further calls. Skip all of it and the show performs on the default gradient.

Whole-chapter batching is the advanced move. When the current chapter is missing two or more backdrops, a batch entry appears in the top bar, and it opens with a quote sheet rather than a spinner: how many scenes, which image model, and where this month’s media budget stands. Approve it and generation walks scene by scene, each backdrop fading in the moment it’s done; a single failure stops the run in place instead of billing the remainder. That monthly budget line is a shared ceiling across batch features — the same cap covers serial comic generation — built against the “left it running overnight, spent a month’s budget” failure. Re-batching a chapter you’ve already dressed skips every reused scene at zero cost. On a BYOK key all of this bills at your provider’s list price, no markup.

During the show: standing at a fork

A choice panel at the frontier of the text is standard visual-novel furniture, with one extra beat: the first tap expands a direction’s outline, and only the second tap confirms. It’s the same anti-fat-finger design the app puts in front of everything that costs money — before that second tap, no continuation call exists. After it, the AI starts writing the story in that direction, and you can stay on stage and watch it perform, or drop back to the reader and watch the text stream in.

The other kind of fork is historical: replay far enough and the panel reappears at a spot where you’ve already chosen once. Here the two taps mean different things. Empty space continues along the current worldline. A direction card means regret — rewrite from this point. The distinction deserves half a second of thought the first time you meet it, though the stakes are low: the direction you walk away from stays in the book as a branch.

After the curtain: the form your choices take

Branches. Every direction you confirmed becomes, once its continuation finishes, a real branch of the book — carrying its full continuation history, tagged with the direction you picked, growing a fresh set of candidates at its tail. Back in the reader you can keep reading it; re-enter the theater and the stage recompiles around the new text and keeps performing. Don’t want it? Switch back to the canon line. The original text stays read-only throughout — the old promise of the branching system, with the theater as one more door into it. And progress remains singular: exit the theater and the reader is open to the page your performance reached.

A sequencing tip from our own play sessions: learn the room before you decorate it. Run a dialogue-heavy book on the default backdrop until AUTO and the two-tap choice feel natural, then decide whether the scenes deserve art. And if you’re budgeting, note that a choice confirmed on stage is billed as ordinary AI story continuation — same pipeline, same per-segment arithmetic as continuing in the reader, which the cost breakdown prices to the cent.