Étincel

Philosophy

Why Étincel exists

I built this after watching thousands of people fight AI to write fiction well. The fight taught me something that turned out to have nothing to do with fiction.

green This page scored 0/100 on Étincel's own audit.

Where this started

I run AIStoryHub, a tool for people writing fiction with AI: novels, story worlds, characters who have to stay consistent across three hundred pages. Thousands of people used it to write real, workshop-quality fiction. Most of them hit the same wall. The writing was fluent and it read like nobody in particular had written it. Correct sentences, no voice. Fiction is unforgiving about that in a way most writing isn't. A reader feels a flat scene even when they can't say why.

What fixing that taught me

Getting AI-assisted fiction to read like a specific writer wrote it took more than a good prompt. It took measuring what that writer's voice actually does: sentence rhythm, how paragraphs vary in length, the words they reach for without noticing. And it took a real, specific list of the tells that give AI prose away regardless of who's supposedly writing it: the em dash where a comma would do, the paragraph that resolves too neatly, the case study with no flaws. None of that turned out to be fiction-specific. It's what makes any writing sound like a person instead of an average of everyone.

Why non-fiction, why now

Almost nobody writes novels. Almost everybody writes emails, Slack messages, PR descriptions, blog posts, the ordinary writing that makes up nearly all AI-assisted prose today. The same measurement and the same tell list, pointed at that instead of fiction, is Étincel. Not a new idea, the same one, aimed at a much bigger audience with a lot less patience for reading something fake.

Why it's open

Nothing Étincel decides is a model call making a judgment in the dark. It's a rule you can read: the audit corpus, the scoring, the way a voice gets measured, all of it lives in a public repository, not behind an API you have to trust blindly. Disagree with a rule, and you can see the exact line of code that's wrong instead of guessing. Fork it and run your own copy if you'd rather not depend on mine. A tool that asks you to hand over your writing voice shouldn't also ask you to take it on faith.

Read the code on GitHubSee how it works