-
-
Notifications
You must be signed in to change notification settings - Fork 2
WorldEditor
A guide to each tab in the World Editor — what it does, why it exists, and the settings that aren't obvious from the screen.
💡 Every tab has a ? button beside its search box with a short version of this page. This is the long version.
Sections land here as each tab's help is written, so a tab missing below simply isn't documented yet.
The numbers that describe your player — health, coin, reputation, whatever your world needs. Each stat holds a value between a Min and a Max, and the AI sees them on every turn.
Prose alone drifts. A stat is a fact the AI has to write around: it can't narrate you sprinting across a rooftop while your Stamina reads 4/100 without the number contradicting it.
The narrator is explicitly told to let stats shape how an action turns out — a low stat as effort and cost, a high one as ease — and not to tabulate them or report their changes. A separate step handles the actual numbers. So stats steer the story without it reading like a character sheet.
Each stat's Name is always sent. The Stats chip in your prompt decides what accompanies it:
| Piece | Adds |
|---|---|
| Values | The current value and its ceiling — 62/100
|
| Status | The matching Stat Descriptor — a word for the current level |
| Meaning | The stat's Description — what it represents |
With no piece selected, the line is just the stat's name.
⚠️ Every stat is sent on every turn. Stats are one of the steadiest drains on your context budget — three meaningful stats beat a dozen decorative ones.
| Field | What it does |
|---|---|
| Name | Also how the AI refers to the stat, and how stat changes are matched back to it. |
| Type | Number. (List exists in the data format but isn't currently offered.) |
| Description | What the stat represents. Sent to the AI when the chip's Meaning piece is on. |
| Min / Max | The range. Values are always clamped to it. |
| Initial Value | Where the stat starts. Ignored when the stat has code. |
| Regen | Added to the value once per turn, then clamped. Positive heals over time; negative bleeds. |
| Body Sliders | Bind body morph sliders to this stat — its value, from Min to Max, drives each slider. Each slider belongs to only one stat. |
Descriptors turn a number into a word — Winded, Exhausted — which is what the AI receives when the chip's Status piece is on.
Each has a Threshold % and a Description. Two things trip authors up:
⚠️ Threshold % is a percentage of the way from Min to Max — not a raw value. With Min 0 / Max 200, a threshold of50means the value 100.
⚠️ The first descriptor at or below the threshold wins, in list order. List them low to high. If100 → Finesits first, it matches everything and nothing below it is ever reached.
Four checkboxes stop the AI moving a stat in one direction, while your world's own rules still can:
| Checkbox | Blocks |
|---|---|
| Don't increase | AI raising the value |
| Don't decrease | AI lowering the value |
| Don't increase max | AI raising the ceiling |
| Don't decrease Max | AI lowering the ceiling |
A stat can compute itself from the others. Write JavaScript that returns a number, and it recalculates each turn instead of using Initial Value. Test Code runs it immediately and shows the result or the error.
const health = stats.find(s => s.name === 'Health')?.value || 0;
const strength = stats.find(s => s.name === 'Strength')?.value || 0;
return (health + strength) / 2;Your code runs in an isolated sandbox with a time limit. It sees a read-only copy of every stat — id, name, type, description, min, max, value, regen — and nothing else: no other stat's code, no descriptors, no page, no network.
A calculated stat ignores the AI. Its value is recomputed every turn, so anything the AI writes is overwritten. The AI still reads the value and description normally.
📘 Full reference: Stat Code Guide.
Start with two or three stats the story would genuinely turn on, each with a Min, Max and a couple of descriptors. Add Regen, locks and code only when a stat needs behavior the AI shouldn't be inventing.
The people, creatures and things that populate your world — a ferryman, an eel-smoker, a barred door. An entity belongs to one or more Locations, and the AI is handed the ones that could turn up where the player is.
Left to itself, the narrator invents a stranger, names them, and forgets both by the next turn. An entity is a fixed, reusable character the story can return to — one the AI is reminded of every time the player is somewhere it lives.
The default prompt introduces them under "Characters and things that may appear in this location". That hedge is a hint to the AI, not a rule the app enforces — nothing stops the narrator reaching for anyone on the list, and the wording is yours to change in the prompt editor.
Assignment to a location is the entire gate: an entity in no location never reaches the AI at all.
| Field | Sent? |
|---|---|
| Name | Always |
| AI-Facing Description | Yes — the main thing the AI knows |
| AI-Facing Summary | Only in prompt slots that ask for the short form |
| Type | Yes, as a plain field |
| Player-Facing Description | Never |
| Image, Image Tags, 3D model, group, order | Never |
💡 The two descriptions are disjoint, and that's the point. The player only ever sees the Player-Facing one; the AI only ever sees the AI-Facing one. So the AI-Facing Description is where a secret lives — the narrator can act on it while the player stays in the dark. The default prompt even asks the narrator to hold a name back until the player would plausibly have learned it, though that's a request to the AI rather than something the app enforces.
Three description fields, with distinct jobs:
- Player-Facing Description — what the player reads on the entity's card. Never sent to the AI, so it costs no context; whatever you write here, the player simply knows.
- AI-Facing Description — the full text the AI works from, and the player never sees it. This is the one that does the work, and the one a secret goes in.
- AI-Facing Summary — a one-line version for slots where the full description is too long. Blank is fine — it falls back to the full description. The default prompt uses summaries for entities in reachable locations, and full descriptions for the player's current location.
The ✨ toolbar beside AI-Facing Summary can draft it from your AI-Facing Description.
The link lives on the location, not the entity. The Locations picker here is a convenience — it writes the entity into each of those locations' own entity lists. You can set the same relationship from either end.
That storage detail has one consequence worth knowing:
⚠️ Deleting a location drops its entities from it, silently. The assignments live on the location, so they go when it does. The entities themselves survive — but one that was only in that location is now in no location, which means it never reaches the AI again. Nothing warns you, and the entity still looks fine in its own tab.
The default prompt feeds entities from three places, as separate blocks: the player's current location, its sub-locations, and reachable locations.
⚠️ Every entity at the player's location is sent every turn. A crowded location is a permanent context bill — two or three that the scene turns on beat a populated village.
Folders for your own sanity. Nesting and order are editor-only and never reach the AI — entities feed the AI exactly as if ungrouped, so grouping can never change the story.
The image and 3D model are for the player's screen. Image Tags are booru tags used only for AI image generation — the ✨ toolbar can draft them from the description, and uploading an image with an embedded prompt offers to use it. None of this reaches the narrator.
Write the AI-Facing Description first — it's the one that does the work, and it's safe to put things there the player shouldn't know yet. Add a Summary only if the entity turns up in reachable locations. Reach for the Player-Facing Description when you want the player to read something the AI has no use for.
The places your story happens. The player is always in exactly one, and it decides what the AI is told about the scene — the description, who's present, and where the story might go next.
Without a fixed place the narrator drifts: the tavern becomes a street becomes a forest, and nothing stays put. A location is an anchor the AI is handed again on every turn.
This is the part that surprises people.
💡 The player can travel anywhere, always. The in-game location list offers every location in your world, unfiltered, no matter how you've arranged them. Nesting never gates a player's choice, and no arrangement can strand them.
What nesting decides is where the story can take the player. When the AI reads an action as movement, it only considers places connected to where they already are:
| From | The story can move them |
|---|---|
| A top-level location | Down into its own sub-locations |
| A sub-location | Down into its children, up to its parent, and sideways to its siblings |
By default the story offers the move — a small Move to X? prompt with Go and Dismiss — rather than making it for you.
Two consequences worth knowing:
- A flat list of top-level locations gives the AI nothing to connect, so it never proposes a move. Travel becomes entirely player-driven — a legitimate design, just a deliberate one.
- A single-location world never runs the router at all.
If the AI's answer doesn't match a connected place, it's discarded and nothing is offered — the story can never teleport the player somewhere unconnected.
| Field | Sent? |
|---|---|
| Name | Always |
| AI-Facing Description | Yes — the main thing the AI knows |
| AI-Facing Summary | Only in prompt slots that ask for the short form |
| Player-Facing Description | Never |
| Background image, Image Tags, ambient sound, starting flag, nesting | Never |
The default prompt gives the AI the current location in full, and its sub-locations and reachable places as summaries.
💡 The two descriptions are disjoint. The player only ever sees the Player-Facing one; the AI only ever sees the AI-Facing one — so the AI-Facing Description is where a secret lives.
The ✨ toolbar beside AI-Facing Summary can draft it from your AI-Facing Description. Blank is fine — it falls back to the full description.
The Entities picker lists who's at this location. It's the same link an entity's own Locations picker writes, seen from the other end — set it wherever you prefer.
The checkbox marks a place a new game can begin. It does more than it looks:
| Ticked | Result |
|---|---|
| None | The game starts at a random location — any of them. Rarely what you want. |
| One | Every game starts there. |
| Several | The player chooses between them before starting. |
Write the AI-Facing Description first — it's the one doing the work. Reach for nesting when you want the story to move the player on its own, and tick at least one starting location so new games don't begin somewhere arbitrary.
📎 The world file also supports a
connectionslist for wiring locations together by name, but there's no editor field for it — it can only be set by hand-editing the world JSON.
The choices that make one playthrough different from the next — Scarred, Silver-Tongued, Afraid of Water. The player picks their traits before the story starts, and the ones they take are described to the AI on every turn.
A trait is a durable fact about the character. Stats move constantly and the story moves with them; a trait stays put, so the narrator is handed the same truth on turn one and turn ninety. A stat says how much; a trait says who you are.
💡 Only chosen traits reach the AI. A trait the player didn't take is sent nowhere and does nothing. Everything below applies to the ones they picked.
| Field | Sent? |
|---|---|
| Name | Always (it's the fallback when the AI description is blank) |
| AI-Facing Description | Yes — what the AI is told the trait means |
| Player-Facing Description | Never |
| Stat Changes | Never |
A blank AI-Facing Description falls back to just the trait's name — often enough for something self-explanatory like Left-Handed.
💡 Stat Changes are invisible to the AI. It's told the player is Sickly; it's never told that cost them 20 Vigor. The number does its work through the stat itself, so write the description as a fact the narrator can act on — "Flinches at open water", not "-20 swimming".
Pre-checks the trait on the selection screen. The player can still untick it — it's a default, not a requirement.
Each row adjusts one stat when the trait is taken: a stat, a number, and what to change.
⚠️ Every type is an adjustment, not a setting.+20on a stat that starts at 50 gives 70, not 20.
| Type | Effect |
|---|---|
| Starting Value | Shifts where the stat begins. |
| Min | Raises the floor, pulling the value up with it if it's below. Can be lowered only far enough to undo another trait's raise — never below the floor the stat was authored with. |
| Max | Moves the ceiling either way. Lowering it below the current value drags the value down too. |
| Regen | Adds to what the stat recovers each turn. Negative bleeds. |
Min and Max are deliberately asymmetric: a trait can take the ceiling anywhere, but can never push a stat below the range its author designed.
Groups organize the list — and unlike organizational folders elsewhere, a trait group also speaks to the AI. Give a group an AI-Facing Description and it becomes a header above its chosen traits, letting you frame a whole set at once ("Origin: where this life began"). A group with no chosen traits inside it is skipped entirely.
Name the trait, write one line of AI-Facing Description that reads as character rather than mechanics, and add Stat Changes only when the trait should also move a number. Leave the description blank for anything the name already says.
Your world's lorebook. Each book holds entries; an entry injects its content into the AI's prompt whenever one of its keywords appears in the text being scanned.
The AI can't hold your whole world in mind at once — everything it knows on a given turn has to fit in a limited context window. Writing every detail into your world description spends that budget on every turn, whether it's relevant or not.
The Dictionary is the alternative: lore sits on standby and costs nothing until something brings it up. Mention the Gloamwater, and the AI suddenly knows what it is.
A turn is one action from you and the AI's reply. On each turn, Formamorph scans:
| Scanned | Always? |
|---|---|
| Current scene — your location, the characters present, and the action you just took | Always |
| Earlier turns — your actions and the AI's replies | Up to the entry's Scan depth |
Your world description is deliberately excluded. It's present every turn, so terms inside it would fire their entries constantly and defeat the point.
Select an entry to open it. Trigger Keywords (Key) and Value (injected on keyword match) are the whole feature — everything else is there for a specific problem, and is safe to ignore until you hit one.
Options
| Checkbox | What it does |
|---|---|
| Always inject | Skip the scan; send this entry every turn. Costs context every turn, so use sparingly. |
| Regex | Treat keywords as regular expressions instead of plain text. |
| Whole words | Match on word boundaries, so art stops firing inside cart. |
| Case-sensitive | Off by default. |
| Recursive | Lets the entry be fired by the content of entries that already activated, not just by the scene. |
Scan depth (messages) — how many earlier messages to search. Leave it blank (all history) to search everything; 0 searches only the current scene.
Secondary Keywords — an extra condition on top of the trigger. bridge fires only if toll also appears in the scanned text.
| Checkbox | What it does |
|---|---|
| Require all | Every secondary keyword must appear, not just one of them. |
| Exclude (activate when absent) | Inverts the test — the entry fires only when the secondary keywords are missing. |
Each book splits its entries into two collapsible groups, BACKGROUND and FOREGROUND. They're two separate lore blocks in the system prompt, and an entry's group decides which one it joins. New entries land in Foreground.
To move an entry between them, drag it from one group into the other. There's no dropdown — the groups are drop zones.
By default Background sits earlier in the prompt than Foreground, but you control placement: both blocks are filled by prompt chips you can move in the prompt editor. If your prompt has no Background chip, those entries fall into Foreground instead.
Books group related entries. Their order sets the order entries are injected, and disabling a book mutes everything inside it at once.
A book's enabled state is a default, not a lock. Before starting, players may see a step where they can toggle and reorder your books — alongside any dictionaries from their own library. That step only appears when there's a real choice to make: more than one book in the world, or at least one dictionary saved in the player's library.
Start with one book and a handful of plain keyword entries. Everything above exists for a specific problem — reach for it only when an entry fires when it shouldn't, or fails to fire when it should.
Reusable bits of world text you define once and drop into your writing as chips — an eye color, a street name, a deity. Each has a Name and a list of Values, and everywhere its chip appears, it resolves to one of those values when the story runs.
So a world can vary without being rewritten. Author "the {{Eye Color}} stranger" once and it reads as a real detail every playthrough — the same detail on purpose, or a fresh one each time.
The kind is inferred from how many values you give it — there's no separate switch.
| Values | Kind | Resolves to |
|---|---|---|
| One | Variable | Always that value. Edit it here once and every chip updates. |
| Two or more | Wildcard | A random one of the values. |
| None | — | Nothing (empty). Give it at least one value. |
A Wildcard chip chooses how its roll is shared, per placement:
| Scope | Behavior |
|---|---|
| World | Every World chip of this placeholder shows the same rolled value everywhere — one town name, used consistently. |
| Unique | Each placement rolls on its own — ten Unique Eye Color chips give ten independent eyes. |
A Variable ignores this — it's the same single value no matter what.
💡 A Wildcard rolls once, when a game begins, and the result is stored in that save. The stranger with gray eyes on turn one still has them on turn ninety, and reloading the save changes nothing. A new game rolls fresh.
Placeholders resolve both in what the AI reads and in what the player sees, in any field with the chip picker:
- entity, location and dictionary descriptions
- the readme
- the system prompt addition
⚠️ The World Description is the exception. It's shown in the library before a playthrough exists, so there are no rolls to resolve yet — that field takes no chips at all.
Define the placeholder here, then place its chip from any field that offers them. Reach for a Wildcard when you want variety, a Variable when you want one editable fact in many places.
Playing
Reference
Project