Headcanon Prompts: 25 Paste-Ready Templates

The four slots are not a theory of prompt writing. They are the four controls that exist on the generator page: a name box, five category chips, six tone chips and five count buttons. Nothing else on the page takes input. So a headcanon prompt can only specify those four things, and everything a prompt asks for beyond them — no family, one person per line, no dialogue — has to be enforced by you when you read the results. This page gives you twenty-five prompts written that way, and it gives you the counts behind them, because the pool here is fixed and countable rather than generated fresh each time.

Every number below comes from scanning the file that serves the tool — a bank of five categories by six tones by six lines, 180 lines in total — and from replaying the shuffle offline. Nothing is quoted from another site and nothing is estimated. Where a count contradicts something you would expect, the count is what I wrote.

Averages over the whole pool: each slot was changed on its own and the resulting batch compared with the batch from the unchanged prompt, five lines each. The two zeros at the bottom are structural rather than a finding — two different categories never share a line, and a changed count re-shuffles the batch instead of extending it — which is why the practical rule is to move one slot at a time and expect a fresh five.

One line that holds all four slots

Every prompt on this page is a version of this sentence:

{count} {category} headcanons for {name}, {tone} — plus one constraint the tool cannot enforce.

Four things are open and one thing is fixed. The four open slots are what the generator reads; the constraint after the dash is what you read. That split is the reason these prompts are short. A prompt that tries to specify the constraint through the slots — by asking for a tone that supposedly produces family-free backstory, or a category that supposedly produces dialogue — is asking the pool for something it does not hold, and the counts later on this page show how that ends.

The order matters more than it looks. Fill the name first, because it is the one slot that cannot be left empty if you want the same set twice. Then category, so you know which sixth of the pool you are reading. Then tone, because it is the slot that most often changes what a line is for rather than which line you get. Then count last, because it is the only slot that changes the shape of the answer rather than its content. If you fill them in the other order you will spend two rolls fixing a batch that was always going to be replaced.

What each slot changes, measured

One thousand nine hundred and eighty-four combinations are legal: any subset of the six tones, including none, against any subset of the five categories except the empty one. The pool behind them runs from 6 lines to 180, in seventeen different sizes. So the four slots are not decorative — they really do point at different amounts of material. What they do not do is mix.

• Name. Swapping the character and leaving everything else alone keeps an average of 0.25 of the five lines; 34 of the 40 built-in names return a batch with nothing in common with the original. Across all 780 pairs of built-in names, 664 pairs (85.1%) share no line at all. Case counts too: Marlow and marlow at the same settings share nothing.
• Tone. Swapping one tone for another keeps 0.33 of five on average — five of the six single tones return a completely different batch, and one returns two of the same lines by coincidence. Narrowing to three tones rather than none keeps 0.00 of five. The tone slot also never changes the category of a line: the categories you get back are decided by the category slot, and the tone only decides which of the six lines inside each category you see.
• Category. Two categories share no text at all, so the survival rate is 0.00 by construction. One category holds 36 lines, and one category at one tone holds 6 — that is the floor, and it is also the ceiling on how much a maximally narrow prompt can ever show you.
• Count. This is the one that surprises people. Asking for 3 instead of 5 does not return the first three of the same five; the two batches share nothing. Change it and you have a new set, not a longer one. What count does control is breadth: at every count from 1 to 5, with all five categories on, you get exactly that many different categories back — 2,000 rolls per count, 100% of them, and not one repeated line inside a batch. So the count slot means “how many of the five buckets do you want to hear from”, not “how much do you want”.

There is a fifth input that no prompt formula mentions, and it is the order you press the buttons. The shuffle takes its starting number from a string that includes the selected tones and categories in the order they were switched on. Turning all five categories on in the default order and then turning Personality off and back on again — which leaves the screen identical — produces a batch sharing 0 of 5 lines with the first one. Reversing the five category chips gives 0 of 5. Selecting two tones in the opposite order gives 1 of 5. This is the honest answer to “why did I get different lines than my friend with the same prompt”: the prompt includes the history of the clicks, not only their end state.

Which is why the empty name box is a trap. If the name slot is empty, the tool picks one of 40 placeholder names at random, so two people who both leave it blank are not running the same prompt. A sample of 200 empty-box rolls reached all 40 of them. Type the name, even if it is a placeholder for a character you have not named yet.

The twenty-five prompts

Five per category. Each one fixes a different thing the generator cannot fix for you, and each one is a sentence you can paste as-is after replacing the four slots. The constraint is deliberately the kind you can check at a glance, because on a fixed pool the prompts do not filter the output — you do. One pass through all twenty-five at a budget of two to five lines each draws 95 lines and touches 95 different ones, which is 52.8% of the pool; running the same twenty-five again reaches 123 of 180, or 68.3%. That gap is the point of keeping the list short: the same twenty-five prompts, aimed at six different names, will not run dry quickly.

Personality — prompts 1 to 5

Personality is the category people reach for first and the hardest one to keep interesting, because it is the one that most easily turns into a list of nice adjectives. Every prompt here forces a behaviour instead.

• 1 · {count} Personality headcanons for {name}, {tone}. Each line has to be a pattern rather than an event — true on an ordinary Tuesday, not only on the day it happened.
• 2 · {count} Personality lines for {name}, {tone}, and make one of them contradict the rest. The contradiction is what turns five separate facts into one person with a fault line.
• 3 · List {count} Personality headcanons for {name}, {tone}. Nothing about feelings — only habits another person could name from across the table. This is the fastest repair for a batch that reads like a horoscope.
• 4 · {name}, {count} Personality lines, {tone}. Every line has to survive being said out loud to {name}'s face. Say them out loud. The ones that make you wince are usually the true ones.
• 5 · {count} Personality headcanons for {name}, {tone} — then add one sentence to each saying why it would be annoying to live with. The second sentence is the one that goes into your draft; the first is scaffolding you delete.

Backstory — prompts 6 to 10

Backstory prompts fail in one direction, which is turning into a chronology. The fix is to ask for decisions rather than dates. It is also the category where the pool is most cooperative: across all 180 lines, no line contains a year, and exactly one line in the whole bank contains an Arabic digit at all — it is a line about deciding at 2 a.m. and re-deciding at noon. There is nothing here to date-stamp, so a prompt that asks for dates is asking you to write the dates yourself.

• 6 · {count} Backstory headcanons for {name}, {tone}. Each line names something that happened, not a way they felt about it. Events survive editing; feelings about events rarely do.
• 7 · Backstory for {name}: {count} lines, {tone}. Skip childhood — start at the first job. Nearly free on this pool, and it kills the most common cliche in one instruction.
• 8 · {count} Backstory lines for {name}, {tone}. Every line is a decision that is still costing them something. Cost is what makes backstory matter to the present scene.
• 9 · {name}'s Backstory, {count} lines, {tone}. No line may mention a family member. Two of the 180 lines mention family at all, so this constraint costs you almost nothing and removes the default answer.
• 10 · Write {count} Backstory headcanons for {name}, {tone}, each one explaining a habit I already know about them. Paste the group 1 lines in front of this prompt and the backstory has a job to do.

Relationships — prompts 11 to 15

A relationship prompt is the category slot pointed at Relationships, which is 36 of the 180 lines. Most of the way people ask for these goes wrong in the same way: they ask for two characters, and the tool writes about one. Placeholder counts confirm it — every line in the pool is written around a single name, used exactly once, and no line names a second person. So the useful move is to specify what the missing person is doing, usually by disappearing. Prompts 12 and 15 do that.

• 11 · {count} Relationships headcanons for {name}, {tone}. One line per person, no two lines about the same person. Forces breadth, which is what a relationship web actually needs.
• 12 · Relationships for {name}: {count} lines, {tone}. The other person never appears on the page — they can only be referred to. This is the one that matches how the pool is built, and it produces lines you can drop into a scene unchanged.
• 13 · {count} Relationships lines for {name}, {tone}. Each line has to say what {name} still owes someone. Debt is a relationship stated as an obligation, which is easier to stage than “a close bond”.
• 14 · Relationships for {name}, {count} lines, {tone}. After each line, add one sentence for what the other person would say back. The reply is where the relationship stops being a description and becomes a scene. If you want the full version of that step, the relationship piece takes one line all the way into a scene.
• 15 · {count} Relationships headcanons for {name}, {tone}, with exactly one other person on the page per line. One is the number the pool can actually hold; anything more and you are writing the second character yourself.

Quirks — prompts 16 to 20

Quirks is the category most likely to produce something you can use immediately, and the one most likely to produce a list of unrelated oddities. The prompts here push towards objects and towards restraint, because a quirk that has a prop in it can be filmed, and a quirk that has no explanation is usually the one readers keep.

• 16 · {count} Quirks for {name}, {tone}. Each quirk has to be visible from across a room. Turns an internal habit into staging, in one instruction.
• 17 · Quirks for {name}: {count} lines, {tone}. Every line contains one physical object with a place in the room. The object is the thing you will actually write about; the habit is only its cause.
• 18 · {count} Quirks for {name}, {tone} — small things they refuse to explain. The refusal is what makes a quirk feel like a person rather than a gimmick.
• 19 · {name}, {count} Quirks, {tone}. No line may use the words always or never. Cheap to check, and it removes the two words that make quirks sound like summaries.
• 20 · Write {count} Quirks for {name}, {tone} — things a rival would copy on purpose to annoy them. Using a quirk as ammunition proves it is noticeable, which is the only test that matters.

Secrets — prompts 21 to 25

Secrets is where the pool is at its most useful and also where a prompt most easily overshoots, because “a secret” invites melodrama. The prompts below keep the scale down on purpose: a secret small enough to keep forever is worth more in a draft than a confession, and the lines in this category tend to be statements the character has never said out loud rather than revelations about their past.

• 21 · {count} Secrets for {name}, {tone}. Every line is something they have never said out loud. The most reliable of the five, because it is the definition most of the pool already satisfies.
• 22 · Secrets for {name}: {count} lines, {tone}. Each secret has to be small enough to keep forever. This is the anti-melodrama switch.
• 23 · {count} Secrets for {name}, {tone} — and make one of them not really a secret. The odd one out tells you which of the others is doing the work.
• 24 · {name}, {count} Secrets, {tone}. No line may name the person who knows. Keeping the knower off the page is what makes it a secret rather than an announcement.
• 25 · Write {count} Secrets for {name}, {tone} — secrets only because nobody ever asked. A secret with no drama in it is the hardest kind to write and the best kind to have.

One prompt, run for real

Here is prompt 12 with the slots filled in and the settings written out, so you can run it and compare. Name slot: Odette (one of the 40 built-in names, so there is no typing to get wrong). Category slot: Relationships only. Tone slot: Angsty only. Count slot: 5. The batch you get the moment you finish setting the controls is this one, word for word:

• Odette loves people at a distance, convinced that closer would be worse for everyone involved.
• Odette keeps offering people the version of themselves that is easiest to leave.
• Odette says I'm fine to the one person who can tell, because that person is the only one it would hurt.
• Odette has an apology drafted for someone they may never see again.
• Odette keeps one friendship alive mostly out of loyalty to who the two of them used to be.

Nothing was edited, including the missing quotation marks around the third line — no line in the pool contains a quotation mark, which is why every one of these reads as a statement about speech rather than speech. Four of the five lines fit prompt 12 as written: the other person is referred to and never appears. The fifth, about the apology, names no one at all, which is the constraint failing on one line out of five and is exactly the kind of thing you learn to expect from a fixed pool instead of blaming the prompt.

Now press the roll button once. You get a second five, and it overlaps the first by four lines — keeps offering people, says I'm fine, loves people at a distance and keeps one friendship alive all come back — and the one line that changes is this: “Odette is the one everyone calls at two in the morning, and the one nobody thinks to call at noon.” That overlap is not a coincidence, it is arithmetic. That combination of slots holds exactly six lines, so a five-line batch can only swap one of them out. Rolling that cell 200 times returns those same six lines and nothing else, in six possible arrangements, which is the ceiling of a narrow prompt and the reason the count slot matters: widening the category or tone slot is the only way to get material you have not already seen.

The same six-and-six ceiling has one useful side effect. Because a batch never repeats a line inside itself, and never repeats a category, a five-line roll with all five categories on is guaranteed to be five different buckets — one category each, no exceptions in 2,000 rolls per count. If your draft needs five different angles on a character and you keep getting three variations of the same one, the problem is that you narrowed the category slot. Open all five, roll once, and you have one line from each bucket to work with.

What a prompt cannot ask for

This is the part that saves time. A prompt is a request, and a request can only be answered from what the pool holds. Scanning all 180 lines gives a hard list of what is not in there:

• Dialogue. Zero lines contain a quotation mark, so no prompt can return a line of speech. Six lines contain an apostrophe and all six are contractions or possessives. If you want a character to say something, the pool will give you a statement about their speech and you write the words.
• Questions and exclamations. Zero lines end in a question mark, zero end in an exclamation mark, and all 180 end in a full stop. A prompt that asks for “a shouted line” or “something they would ask” will come back as a calm statement, every time.
• World facts. No line contains a year. No line contains a currency symbol or a word for money. Exactly one line in 180 contains an Arabic digit. No line contains a capitalised proper noun anywhere except at the very start, where the name slot sits. A prompt asking for a salary, a hometown, a school or a date has nothing to draw on — those are all slots you fill yourself.
• A second character. Every one of the 180 lines is built around a single name and uses it exactly once. There is no line that can describe two people at the same time.
• An empty result. The narrowest legal prompt — one category, one tone — still has six lines behind it, and the count buttons stop at five, so no combination of the four slots can return nothing. The tool carries a message for the empty case anyway; it is unreachable. If a prompt comes back blank on another generator, that generator is doing something this one does not: calling a model.
• A chosen archetype. Every roll also picks one of 24 archetype labels, which is printed above the batch. It is not one of the four slots and there is no control for it: it is drawn by the same shuffle, so changing any slot changes it, and 500 rolls of one prompt eventually reach all 24. It is a label to describe the batch with, not something to ask for.

The pattern behind that list is worth stating plainly, because it is the difference between a prompt you can trust and one you cannot: on a fixed pool, the four slots pick lines and the constraint decides which of them you keep. A prompt that tries to make the constraint do the picking will hand you five lines that quietly ignore it, and the fifth line of the Odette batch above is what that looks like.

Where the twenty-five land, and where they miss

Twenty-five prompts is not a full sweep of the pool, and the gap is measurable. Each prompt above pins one category and one tone, which means it points at one of the 30 cells in the grid, each holding six lines. The twenty-five prompts as written land on 25 of those 30 cells. Five cells are never touched by the list: Personality at Melancholy, Backstory at Chaotic, Relationships at Angsty, Quirks at Romantic and Secrets at Funny. That is not an oversight in the list so much as a demonstration of how a prompt library fails — it is always short in a pattern, and the pattern is easiest to see in a grid. If you are writing one character rather than many, the fix is to walk the 30-cell matrix deliberately instead of working down a list.

Run once with a budget of two to five lines each, the twenty-five draw 95 lines and all 95 are different: 52.8% of the pool in one pass, 19 lines from each category. Run the same twenty-five a second time and 28 of the 95 are new, which brings two passes to 123 of 180, or 68.3%. The remaining 57 lines are not hidden or locked, they are simply in cells the list does not point at, and a name change on any single prompt returns a different set drawn from the same cell. So the honest summary of the shelf life is this: twenty-five prompts are enough for a cast of two or three, and for anything longer the prompt list is a starting position rather than a supply.

How to check any prompt, on any generator

The four slots are not specific to this tool. Any headcanon generator that takes a name, a category or type, a tone and a count is asking for the same four things, so the same three checks apply before you trust a prompt:

• Ask what the pool holds before you ask for something. If the tool has no control for the thing you want, you are not going to get it by phrasing. On this site the list above is the honest inventory; on another tool, run the same prompt with two different names and see whether the difference is real.
• Move one slot per roll. The measurements above are the reason: moving four slots at once on a fixed pool gives you a batch you cannot compare to anything, including the one you liked.
• Write down the four settings before you roll, not after. Name spelled exactly, category chips that were on, tones that were on, count. On this tool clicking any chip resets the shuffle to its first position, so the batch on screen right after you finish setting the controls is always the same batch for those settings — which is what makes a set reproducible, and what makes the empty name box the one thing to avoid. If you want that mechanism control by control, the walkthrough takes it apart; if you want to know why two characters end up sharing lines even when you do everything right, the repeat piece has the numbers.

Quick answers

• Do I have to use these exact words? The four slots are the part that has to be exact, because they are the part the tool reads. The constraint after the dash is the part worth rewording — it is a filter you apply by hand, and it should say whatever you actually need to keep out of the batch.
• What if I only want one line? Set count to 1 and use prompt 1, 16 or 21, which produce the most immediately usable single lines. One line at one category is a sixth of a cell, so at count 1 you are being shown one of the six candidates at random; if the first one is wrong, roll again rather than rewriting the prompt, because the prompt is not the bottleneck at that point.
• Should I paste these into an AI instead? You can, and the four slots are the right four to give it. The difference is what comes back: an AI will happily write you dialogue, a year and a salary, because it has a whole language behind it rather than 180 lines. This generator will not, and the exchange is that every line here is one you can re-find later. Prompts 7 and 19 are worth keeping in the list either way, since “skip childhood” and “no always or never” are useful instructions for any writer, human or otherwise.
• How is this page different from the other prompt page on this site? The quote piece starts with one line of dialogue and works backwards into prose, with four slots of its own for staging it. This page stops one step earlier and covers the request itself: what to type, what each of the four typed things changes, and which of the twenty-five to reach for. If you are looking for a finished set of lines rather than a way to ask for them, the annotated examples page has twenty, with the reason each one works.

More notes on how the bank is laid out, what it does with a name you type, and how to get a set you can reproduce, are collected on the guides page.