Pipeline · Scenes · Step 3

Text — the finished prose

The last step, and the only one a reader ever sees. The Text is the scene written out in full, from the Contract, in your book’s voice. Here you read it, make small edits by hand, and run it through the de-AI suite — the tools that pull machine tells out of the prose until it reads like it was written, not generated.

The Text tab — the finished scene prose with AI-tic spans underlined inline, a right panel showing a ten-axis Analyze scorecard and a Hunt-Tics tag histogram with per-finding suggestions, and a bottom bar of Analyze / Polish / Hunt tics / Fix AI tics actions each with a tier picker — Text: the finished prose (with tics highlighted inline), the Analyze scorecard and tic hunter (right), and the QA action bar (bottom).

The prose & its versions

The scene reads top to bottom in book typography. Every render is kept as a numbered version (the latest few) — the small numbered buttons let you view an earlier one and activate it to make it the one in the book. Edit opens the whole scene as plain text for a manual rewrite, saved as a new version. A word count sits in the header.

Editing by selection

Select any span of the prose and a small toolbar appears — the way to fix one sentence without re-rendering the scene. You can Rewrite, Enrich, Expand, Shorten, Fix or Remove just that fragment, or give a custom instruction. Precise, cheap surgery on the parts that need it.

The QA suite — four passes

The bar along the bottom holds four passes; each runs over the finished prose and each has its own quality tier. Results appear in the panel on the right. Run them in whatever order suits, but the natural flow is Analyze → Hunt → Fix → Polish:

Analyze

Scores the scene on ten axes and gives an overall mark, flags the weakest ones, and lists any machine-cliché phrases it finds. The read-out, not a rewrite — it tells you where the prose is thin.

Hunt tics

The AI-tic hunter. It scans the prose for the tells of machine writing and tags each one — dense similes, over-explanations, triads, machine metaphors, filtered perception, abstraction. Tick “Highlight in text” and every hit is underlined in place; each carries a concrete fix.

Fix AI tics

One batch pass that applies the hunter’s fixes across the whole scene at once — cut the similes, humanize the machine language, prune the over-explanations. Available once a hunt has run.

Polish

A targeted rewrite that lifts the weakest axes. It writes a new version and re-scores it — and if the score drops instead of rising, it reverts automatically, so a polish can only help.

What Analyze measures

The scorecard rates the scene on ten axes, each 0–10, and marks the weakest so you know where to spend a Polish:

Plot Logic — do the events follow
Characters — do they act in character
Style — is the prose controlled
Pacing — does it move right
Emotions — does the feeling land
Worldbuilding — is the setting present
Dialogue — does speech carry weight
Continuity — does it fit what came before
AI cliché — is it free of machine phrasing
Narrator voice — an aspirational reach — a wry, self-aware narrator (✦ doesn’t count against the score)

Why this step matters

This is where prose stops reading like a machine. Storyum’s edge is the structure — the plot the whole book is built on — but a reader still meets it as sentences. The hunter and its fixes target the exact tells that give AI prose away (endless similes, hedged “not X, but Y” phrasing, everything explained twice), and Polish lifts the axes that came back weak. Spend Goldhere, especially on Fix and Polish — it’s the last mile between generated and written.

When the Text reads clean — the score where you want it, the tic list empty — the scene is done. Move to the next one, and when the chapters fill in, head to Book to assemble and export the whole thing.