Verified against ChatGPT · 2026-08-09
Generate story premises from a vague spark idea that skip past the first, most obvious version of themselves
Takes a rough idea and produces several genuinely distinct premises built from it, each deliberately steering away from the most predictable take so you're choosing between real options, not near-duplicates.
The prompt
Ready to copy — highlighted parts are example details you can swap.
You are generating distinct story premises from a rough spark idea I'll give you — not variations on one premise, genuinely different directions a writer could take this idea. SPARK IDEA A lighthouse keeper who starts receiving messages meant for a lighthouse that was decommissioned decades ago. GENRE Literary speculative fiction, quiet and melancholic rather than horror. WHAT I'VE ALREADY SEEN DONE WITH THIS IDEA (avoid these) Ghost story where the keeper realizes they're actually dead; time-loop story where the messages are from their own future. INTENDED AUDIENCE Adult literary fiction readers, the kind who'd read Klara and the Sun. NUMBER OF PREMISES 4 RULES Before generating anything, identify what the single most obvious, most-written version of this spark idea would be — the version most people would land on first — and do not include it among your answers; use it only as a marker of what to steer clear of. For each premise, change a different structural element from the obvious version: who the protagonist is, what they actually want versus what they think they want, what the central obstacle really is, or where in the timeline the story starts. Each premise must be specific enough to write from immediately — a named-enough protagonist, a concrete situation, a clear central tension — not a one-line pitch so abstract it could describe five different stories. State explicitly, for each premise, what makes it different from the obvious version, so I can tell the variation is deliberate rather than cosmetic. WHAT NOT TO DO Do not produce premises that differ only in surface details like character names or settings while keeping the same underlying conflict and shape — that's not a distinct premise, it's a reskin. Do not include any premise that matches something I listed under 'already seen done.' OUTPUT FORMAT For each premise: a title, a 2-3 sentence pitch, and one line naming the specific structural element it varies from the obvious version. Present 4 of them, ranked by how much story potential you think each has for the stated audience.
Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
Asked to brainstorm premises from a spark idea, GPT-5.1's first few outputs reliably cluster around whatever the single most statistically common treatment of that idea is in its training distribution, because that's the lowest-perplexity, most reinforced completion for the prompt — which is exactly the version a writer wants to skip past, since if it's the first thing the model reaches for, it's likely also the first thing most other writers reach for. Explicitly naming the obvious version and instructing the model to treat it as a boundary rather than an option forces it to search a different, less-traveled part of its own output space for the actual answers, which produces measurably more distinct results than simply asking for 'creative' or 'unexpected' premises in the abstract — vague creativity instructions don't give the model anything concrete to diverge from. Requiring each premise to name the specific structural element it varies (protagonist, want-versus-need, obstacle, or timeline entry point) prevents the common failure mode of 'reskinned' premises that swap surface details like names and settings while leaving the same underlying shape intact, which happens because changing surface details is the cheapest way to appear to have generated something new without doing the harder work of restructuring. The audience field matters because premise quality is not audience-neutral — a premise with real potential for literary readers may be a poor fit for a middle-grade audience, and without that constraint the model tends to average toward a generic 'broadly appealing' pitch that doesn't strongly serve anyone.
What you get back
1. The Wrong Frequency — A retired keeper's estranged daughter, now a radio engineer, starts receiving the messages herself after inheriting the lighthouse, and has to decide whether decoding them means reopening a relationship she'd rather leave closed. Varies: protagonist is the inheritor, not the original keeper, shifting the story from haunting to inheritance.
Verified against
ChatGPT GPT-5.1 · 2026-08-09
Changelog
- 2026-08-09 — Initial publish, verified against ChatGPT GPT-5.1.
Need this built into your business?
If a prompt isn't enough — what Scult builds, built and maintained for you — that's Scult's day job.
EXPLORE WHAT SCULT BUILDS
