Skip to main content
Wilds.aiCreator Studio
W
Editing worlds in the creator studio — wilds.ai Help

← Back to Wilds Handbook

Help & Guides

Editing worlds in the creator studio

The sections and panels of the wilds.ai creator studio, what each field controls, how the AI tools help, and how authored content flows into gameplay.

29 min read•Updated November 16, 2026

On this page

  • Opening the studio
  • The three-column layout
  • World section
  • World Overview
  • Genre & Themes
  • Writing Style
  • Lore
  • Regions
  • Narrative Style
  • Art Style
  • Content Policy
  • Scenes
  • Region Connections
  • Cross-Region Elements
  • Tags
  • Map Preview
  • Cutscenes
  • Custom / Advanced and Aesthetic
  • Monetization
  • Allowed Modes
  • Encounters
  • Dynamic Lore
  • Characters section
  • Player Seeds
  • NPCs
  • Mechanics section
  • Mechanics
  • Objectives
  • Endings
  • Consequences
  • Modules
  • Friendly-attack consequences
  • Story Threads
  • Choice Encounters
  • Living World
  • Iterate, Code and Hooks (AI-built games)
  • Items
  • Loot Tables
  • Crafting
  • Survival
  • Environment
  • Character Creation
  • Reputation
  • Merchants
  • Market
  • City plan (city builders)
  • Market and routes (trading sims)
  • Multiplayer
  • Game Master
  • Audio section
  • Audio
  • Assets section
  • Asset Browser
  • Asset Packs
  • Image Reuse
  • Cover & Screenshots
  • Forge tools
  • Asset privacy
  • Analytics section
  • Settings section
  • Game Shell
  • Turn Replay
  • VR / XR
  • Marketplace listing
  • Metaverse opt-in
  • Credits & Provenance
  • VN Config
  • Settings
  • Family-specific panels
  • Scene Painter
  • Character Designer
  • Route Graph
  • Audio Director
  • Roguelike Config
  • Dungeon Crawl
  • Audio Generator
  • Story Studio (Ink and Twine stories)
  • Write (left pane)
  • Play (right pane)
  • Toolbar overlays
  • Replace HTML (Twine)
  • Details
  • The right sidebar
  • Wilo
  • Inspector
  • Preview
  • AI Tools
  • Vibe Loop
  • How studio edits reach gameplay
  • Visibility
  • Publishing
  • Forking
  • References

Every field on your world lives in the creator studio. Scenes, NPCs, lore, merchants, tutorials, title screens, VN backgrounds: if it shapes how your world plays, you can edit it here. This guide walks through the studio's sections and panels and explains how studio edits flow into gameplay.

#Opening the studio

Open any world you own, press More beside Quickstart at the top of its lobby, and pick Edit. Or navigate directly to /app/play/<world-id>/edit. The studio requires a signed-up account; guests are redirected to the login page. You can only edit worlds where you are the owner.

The More menu holds the owner's other keys too: Analytics, Export (a .wilds.json file you can import again from /app/import) and Delete World, which asks you to type the world's name first. On a phone, 480 px wide and under, your world's Share key is in the menu as well, so the row stays on one line.

Staff who review content see a Staff view strip on other creators' public worlds, with an Open studio (view only) key and an Analytics key, and the same Open studio (view only) key on their public characters. The owner of a world sees the strip as Creator tools, with Studio and Analytics. For staff the key opens a read-only studio: the name of the work, a line naming its owner, a few facts (game type, visibility, how many regions, characters and factions), and keys to the public page, to play or chat, and to the owner's analytics. Nothing can be changed from it, because drafts and saves belong to the owner's account. A private world or character stays closed to staff.

The studio loads a draft: a mutable copy of your world that autosaves every 800ms as you type. When you're ready, click Publish in the top-right to create a new revision. Drafts persist across sessions, so you can close the studio and come back later. The draft is private to you: players keep playing the last published version until you click Publish.

Every edit stays on the page until the server has stored it. The status beside the world's name reads Saving… while an edit waits and Saved once it is stored. When a save fails (the connection drops, the server refuses a text), a Not saved line says why and offers Retry: under the header on a computer, and above the top bar on a phone, where the status reads Unsaved. Your changes stay in the fields, and your next edit or Retry sends them all. Play draft and Publish store your latest changes first, and neither starts while a save fails. Moving to another page of the app sends what is waiting, and your browser asks before you close or reload a tab that holds unsaved changes.

← PreviousWriting great promptsNext →WildWeave writing studio

Related guides

Getting started with wilds.ai

Create your first world from a single prompt.

Read →
Writing great prompts

Prompt patterns that produce the worlds you actually want.

Read →
Creator economy

How credits, earnings, and featuring work for published creators.

Read →
Online
|Persona: Default
Voice: Off

A refused edit and its reason stay on screen when the studio loads the world again while you work. If you keep typing through five saves in a row, Play draft and Publish stop and the line asks you to pause, then press Retry. If the save sent as you leave the studio fails, the studio shows it again with Retry the next time you open this world's studio in the same tab. Closing or reloading the tab ends that.

Play draft plays the draft you are editing, unpublished changes included, in a new tab. On a computer it sits beside Playtest in the header; Playtest opens the lobby of the published world. On a phone it is the Play key in the top bar. The tab opens the moment you press the key and shows the game once your latest changes are stored and the session has started. If your browser does not open the tab, a line at the top of the studio says the preview is ready and offers Open the draft preview. If the save or the start fails, the tab closes and the line says the preview did not start, with Try again.

#The three-column layout

The studio is a three-column shell:

  • Left sidebar: tree of every panel, grouped into sections (World, Characters, Mechanics, Audio, Assets, Analytics, Settings). Search panels with the box at the top. Some panels appear only for some worlds: AI-built games get Iterate, Code and Hooks, retro-styled games get the Retro panels, and roguelikes get Procedural Layout.
  • Center: the active panel. Split the center with the toggle in the top-right to edit two panels side-by-side.
  • Right sidebar: five context tabs: Wilo (the in-app assistant), Inspector (stats and validation), Preview (world card and narration sample), AI Tools (context-sensitive generation) and Vibe Loop (the build record of an AI-built game). On a phone the tabs open as a sheet, without Inspector; a completeness badge takes its place.

#World section

#World Overview

Name, intro summary, genres, tone sliders (humor, darkness, formality, intensity), core loop description. These shape every AI response: the narrator reads them on every turn.

The intro summary is a rich text field. The intro card a strategy world opens with reads it as plain text: the editor's formatting is dropped, symbols such as & and < show as typed, and each paragraph keeps its own line. The session's opening narration is a separate text and this field does not edit it.

#Genre & Themes

The genre archetype drives default mechanics, presentation mode, and narrative profile. Themes are free-form strings that nudge the narrator and compiler toward specific aesthetics.

#Writing Style

Style card: the voice your narrator writes in. You can import a style from a corpus (paste text or a URL), extract key traits, and the narrator will match it.

#Lore

Free-form world lore + history + factions (with rank thresholds and cross-relations). Faction descriptions feed the reputation engine and the narrator's understanding of political dynamics.

#Regions

Add/remove regions, set descriptions, mark starting regions. The GM uses regions to decide where the player can travel and what content is accessible.

#Narrative Style

Controls for recap cadence, pacing, foreshadowing, point-of-no-return warnings, and chapter summaries. Also exposes Narration Perspective: the recommended default for sessions of this world (first person, second person, or third person by the player's persona name). Players see this as the world default in the lobby's Narration picker and can override it for their session; once they start a session, the perspective is locked for the entire run.

With point-of-no-return warnings on, the game asks "Are you sure?" only before an action the game cannot undo: crossing a one-way threshold, locking in an ending, abandoning a major objective, betraying, attacking or sacrificing an ally, or spending a unique resource. Words and affection never ask: flirting, an invitation, a kiss, a confession, or telling one character the player is choosing another go straight to the story. A warning names a character by the cast name, or by the player's own words for them.

#Art Style

Visual vocabulary for cover art, scene images, and NPC portraits. You can regenerate the cover from here.

#Content Policy

Set a hard ceiling on what content the narrator can generate (safe / standard / mature / private-adult). This caps the session tier: a standard-tier world cannot be played at mature tier.

#Scenes

Define scene contracts. In guided and infinite worlds, scenes can be fully authored, hybrid (partial contract + LLM gaps), or fully LLM-generated. In authored story mode, every scene must use authored content and publish validation rejects non-authored scene sources. A game that runs on its engine (an arcade game, a sim, a board game) gets no generated scenes: it plays only the scenes written here, in the turns that reach the narrator. A city builder's Studio leaves this panel out. Each scene has:

  • Type (exploration, combat, dialogue, boss, etc.)
  • Opening narration (rich text: what the player sees when the scene starts)
  • Scene image: prompt or URL for the background. Studio-authored images are used directly during gameplay without calling the AI. In authored story mode, background art must use a concrete image URL; prompt-only scene images are not valid there. During authored play, the active scene's image URL is stamped onto narrator turns, and scenes without a background explicitly clear the previous scene art instead of reusing stale imagery after a transition or reload.
  • Participants (NPCs pulled from your world)
  • Linked objectives and items
  • Transitions (condition-driven paths to other scenes)
  • Tone override (scene-specific tone sliders)
  • Entry conditions (freeform text)

#Region Connections

Travel routes between regions: distance, danger level, blocked/unlocked state, encounter tables.

#Cross-Region Elements

Multi-region dependencies: key locks, info gates, NPC referrals, item chains, lore reveals. Each connects a prerequisite region to a payoff region, with descriptions of what must happen in each. The Director uses these to pace progression across a multi-region campaign.

#Tags

Vibe, topic, mood, gameplay, audience and credit tags for the world. Tags the classifier added show as faded chips you can remove; editing one makes it yours. Staff tags are locked.

#Map Preview

The world's generated map as it will look once faction control settles, with faction markers. Worlds whose game type uses no generated map show a placeholder.

#Cutscenes

The world's level cutscenes, frame by frame, with image regeneration; on narrative worlds also the schedule of turn milestones.

#Custom / Advanced and Aesthetic

Custom / Advanced opens the full art style editor. Aesthetic sets the world's own colors and type through CSS variables and free CSS, with a live preview.

#Monetization

How players reach the world: free (the default) or, from Q3 2026, ad-supported or premium. Until then those two are shown but locked. See Free play, ads, and premium worlds.

#Allowed Modes

Which conversation modes chatters can use with a character: Free RP, Grounded RP (intimacy earned through trust) and Virtual Assistant. Free RP is never available for Safe-tier characters or characters tagged as minors.

#Encounters

Weighted wandering-monster tables per region. A region without a table is peaceful.

#Dynamic Lore

Discoverable lore entries revealed during gameplay when trigger conditions are met (keyword match, location entry, NPC encounter, objective complete). When discovered, the entry is injected into the world_canon RAG cell and appears in the player's codex.

#Characters section

#Player Seeds

The characters, goals and themes players are asked to bring to a session.

#NPCs

Every NPC has 17 fields:

  • Basics: name, narrative role (quest giver, ally, antagonist, mentor, etc.), faction
  • Demographics: gender, age range, tier (important / combatant / background), HUD visibility
  • Personality: rich-text personality description, HEXACO sliders (6 axes: honesty-humility, emotionality, extraversion, agreeableness, conscientiousness, openness)
  • Narrative: backstory (2-3 sentences fed to the narrator), goals (concrete behavior drivers), narrative weight slider, secrets, quest hooks
  • Relationships: links to other NPCs with type (ally, rival, mentor, lover, etc.) and description
  • Appearance: a canonical physical description (hair, build, age, attire, one distinguishing feature) plus the image-only avatarPrompt and voice profile ID. The narrator, opening prose, lore, and cover all read this description as canon, and the portrait is rendered from it.
  • Content policy cap: hard ceiling on interactions with this NPC (auto-defaults based on age range)
Generate avatar prompts from profile
Click Generate from profile on any NPC's avatar prompt field and the AI will write a detailed portrait prompt from the NPC's role, personality, gender, and world style.

#Mechanics section

#Mechanics

Plays as. A narrative world's Mechanics panel opens with three cards, Story, Adventure and Tabletop, the same three the create page offers, with the world's own picked. A world made before presets shows the one its systems read as, with the line "Read from this world's systems." under the cards; so does a Tabletop world whose character sheet never built, as its lobby and its page read it, and picking Tabletop there builds the sheet.

Picking another card prepares the world before anything changes. The card shows "Preparing items…" while the item catalog gets the entries that fit the world's setting (a premodern or future world gets no pistol, a world with no recorded era only the entries that fit any era); a switch to Tabletop shows "Preparing the character sheet…" while the world's classes and ancestries are built, and what you save in other panels meanwhile is kept. The other cards and every toggle wait meanwhile. When it is done the card is picked and "Items ready: 14 in the catalog." gives the count the world got. If the preparation fails, the earlier card stays picked, nothing is changed, and Try again runs it again; offline, nothing is sent.

A switch puts the world's systems where the card puts them. Change a toggle afterwards and the panel reads "Custom: Adventure with Survival on", with Reset to Adventure to put the card's systems back. See the difference opens the create page's sample turn, three ways. Tabletop is not offered in Character mode or while the tabletop view is switched off; on a narrative world it takes the place of the TTRPG Mode toggle, and under Tabletop the Dice select cannot be set to Off. How a world plays is part of what the content rating reads, so a public or unlisted world is rated again when it is next published after a switch.

Master toggle for combat, puzzles, progression, items, dice, objectives, survival, consequences, GM, market, TTRPG, and environment. Disabling a system at the blueprint level turns off the corresponding runtime engine.

A narrative or TTRPG world's session keeps the mechanics it started with until it restarts. When the session is created it pins how the world plays (Story, Adventure or Tabletop), whether combat, puzzles, progression, items, dice, objectives, survival, TTRPG and environment are on, and whether its dice show; every turn plays by that pin, and loading a saved checkpoint keeps it. A change to those reaches every session started after you publish it, and never a game in progress, even with live editing on.

#Objectives

Quest DAG: each objective has type (main/sub/optional/hidden/timed), description (rich text), prerequisite objectives, visibility (visible/hidden/discovered), completion + failure conditions (visual predicate builder), and rewards (XP, flags, items).

In a narrative session a story flag in a completion condition (found-hidden-altar-idol) is the milestone the story must show. Each turn a judge reads the narration and the player's action and is asked, for each objective in play, whether its milestone happened: the flag's words, not the title. A yes sets the flag only when it quotes the words of the narration that show it (this turn's, or one of the last three replies). An answer with no quote, a quote the narration does not hold, or a quote from the player's typed action alone sets nothing: the action is what the player tried, and the narration says what happened. The objective then completes on the next turn once its other conditions (a region visited, an NPC met) also hold. Name a flag for the thing that happens (cave-troll-slain, not troll-quest-done), because the judge reads those words. The quote is matched without regard to case, quotes, punctuation or dashes: "idol — cold" matches "idol—cold", and "well worn" matches "well-worn". An answer refused for its quote is logged with the objective's id and the reason (no quote, too short, not in the narration, or no player in it), so a judge that paraphrases shows in the logs.

The player's own objectives are the player's doing. The paragraph a quote comes from must have the player in it: the narration addressing the player ("you"), the player's name, in a first-person session the narration's "I", "me" or "my" outside quoted dialogue, or a word of the player's action that turn. A quote that runs across paragraphs is read at the paragraph where it ends. A milestone another character reaches alone, while the player's action failed or went elsewhere, sets nothing. The judge also reads the last three narrator replies, and may quote them, so a milestone the narration showed on a turn the judge missed is still credited on the next turns.

The narrator is told each objective's milestone beside its title ("You Help B Find His Melody (done when: inspired b song)"), and that it lands only through the player's own part. A flag that only this judge sets is the one way such an objective completes, so the milestone has to happen in the narration; an umbrella main quest, which completes from its sub-objectives, carries no milestone line. The line names every gate the narrator can write toward, not the flags alone: a character to meet by roster name, a region to reach by name, and an item to hold by its id's words ("Trace the Sealed Crate (done when: the player meets Mara Quill; the player holds the courier sealed ledger)"). A character already met, a region already visited and an item the player already holds drop out of it, and a gate on a place or a character the world does not have is never in it, because the session drops that gate when it starts. An item the narration then hands the player in those words is stored under the item's id, so the has_item gate passes. Name a quest item's id for the thing itself (courier-sealed-ledger), because the narrator writes those words.

An NPC condition matches the character's roster name, title and all: the target rowan-hale is met by "Chief Engineer Rowan Hale". A target of one word (guard) needs the whole name, so "Head Guard" does not meet it, and an article is not a word of the target: the-guard is met by "The Guard", not by "Captain of the Guard". When a session starts it drops a completion gate on a character the world does not have; a gate on a roster name without its title stays, and the objective waits for that meeting.

A hidden objective (type hidden, or visibility hidden on the objective or on the session's copy of it) stays off the player's objective list and the World Codex until the story reveals it or the player completes it. Narrative sessions start it as available so the judge can still read its milestone, and the list hides it all the same: its title would spoil the secret ending it gates. The narrator and the GM never see it either until then: neither its title nor its id is in their prompts, so the narration cannot plant its secret as a stray prop. A hint and the objective reminder never name it either.

An objective's reward XP lands on the turn it completes, at the value set here. A main quest that completes with its last milestone adds its own reward beside the milestone's. Only the party's objectives pay the player: the player's own, a companion's, and one with no owner set. An objective owned by an NPC, a faction or the world (a rival's scheme that completes when its timer runs out) sets its reward flags and pays no XP.

A save from the play screen never undoes the story's progress. The page saves its objectives after board events, and in a narrative session it can still hold the objectives from before the last turn. The session's objectives win: every flag keeps the session's value, and every objective keeps its status, its timer and its progress, so a milestone the judge set stays set, a completed objective is never completed, or paid, a second time, a failed objective does not fail again, and a timed objective's clock (a rival's scheme included) never runs back. The save adds only what the session does not hold yet.

#Endings

Branching campaign conclusions. Each ending has a title, type, priority, conditions (predicate builder), and epilogue variants with rich-text prose and tone selection (triumphant, bittersweet, tragic, mysterious, open-ended).

Once an ending fires, a save from the play screen keeps its epilogue variant and the turn it fired on, so the ending card shows the right epilogue and turn count after a reload.

An ending other than a defeat waits for the world's main objective: it fires only once a terminal main objective (the umbrella quest, or the last main of a chain) and every objective marked required for ending are complete. An ending that names a main objective in its own required objectives, or sets a minimum turn count, is checked on its own condition. A secret ending with a higher priority outranks the canonical one only once both are eligible, so a side objective alone never ends the story. Generated endings get the main objective added to their requirements the same way.

When an ending fires, its card opens as a dialog over the session. Keep reading (or Escape) shrinks it to an "Ending reached" chip that reopens it, so the transcript stays readable. The card's Turns stat is the player turn the ending landed on, counting the first player turn as 1.

The ending is stored with the session, so the card opens again when the session is reopened, on any device. It shows the epilogue written here, so the epilogue appears even when the turn's narration does not tell it.

#Consequences

Long-tail effects that mature over N turns. Author a narrative seed, set the conditions, and the Director will surface it at the right moment.

#Modules

A library of the building blocks the game engine offers: gameplay systems, behavior scripts, widgets, templates and primitives, with filters, search and per-entry help. Gameplay systems can be pinned, with settings, so the AI build uses them; the other kinds are read-only.

#Friendly-attack consequences

Rules for what happens when the player attacks an ally. Add a rule inline; each saves with the draft.

#Story Threads

A read-only list of the promises and questions the world raises, written when the world was compiled. It never shows how a thread pays off. To change a thread, compile the world again.

#Choice Encounters

Authored decision moments: a title, a prompt, a weight, the regions they can appear in, and whether they happen once, each with choices gated on stats and weighted outcomes (health, gold, XP, conditions). The on/off switch is on the Mechanics panel; this panel holds the encounters and how often they appear.

#Living World

The world simulation for strategy, survival, narrative and tabletop worlds. A change here overrides the compiler's choice and survives a recompile. Strategy and survival worlds rebuild the simulation on the next compile, charged to you at what it costs to run; narrative and tabletop worlds take a step every few turns as people play, and each player pays for the steps in their own game. The panel says who pays before you turn it on. What players see and pay: Living Worlds.

#Iterate, Code and Hooks (AI-built games)

Iterate is a chat over the game's source: describe a change and each message runs one edit, shows what changed and any build errors, and saves the draft. Its credit cost shows beside Send. Code opens the source in a code editor (entry, game script, styles, hooks and each custom component) with Save; add a component by name. Hooks lists the game's lifecycle events; pick one to edit its function and Test it to see what it dispatches. See Customizing an AI-built game.

#Items

Item catalog: weapons, armor, consumables, quest items. Each has stats, rarity, slot assignments, and optional lore.

#Loot Tables

Drop tables for enemies, chests, and rewards. Link items by ID with weighted probabilities.

#Crafting

Recipes: inputs → outputs with discovery rules and skill requirements.

#Survival

Hunger, thirst, fatigue, temperature: pressure profiles with decay rates and severity thresholds.

#Environment

Day/night cycles, weather patterns, environmental modifiers that affect combat and dialogue.

#Character Creation

Define archetypes players can choose at session start. Each archetype has a description, avatar prompt seed, and starting stat bundle.

#Reputation

Faction reputation engine: standing tiers, cross-faction effects, moral alignment tracking.

#Merchants

Shops with type (fixed / traveling / event), region assignment, NPC link, faction, base stock (pulled from item catalog), sell discount, restock trigger, currency.

A world with items and no merchants of its own gets one shop per region at compile: up to six items from the catalog (three of each, consumables unlimited), priced from each item's value with a 1.2 markup, bound to the region's first NPC whose role is a trader's (merchant, shopkeeper, innkeeper and the like). Story objects (quest and key items) and items worth 0 are never stock. The restock trigger decides when a shop's stock comes back to its base stock: per rest (the default) after a rest in the story, the story moving into a camp or downtime scene, or the board's Long rest; per scene each time a new scene starts; never not at all. A restock never lowers a count and keeps what players sold to the shop. What players see in a shop is in Inventory, shops and loot.

#Market

Deterministic market simulation: price curves, supply/demand, regional price differences. This is the market the narrator runs in a story world. A strategy sim plays a market of its own, so its Studio shows Market and routes in place of this panel.

#City plan (city builders)

The plan a city builder plays. A session started after a save plays the saved plan.

  • Map and terrain: the map's width and height in tiles (4 to 256), the share of water, forest and rough ground (together at most 100%), and the terrain seed. Another seed draws another map with the same shares.
  • Opening city: a picture of the city a run opens on, drawn by the game's own engine, with its counts of land, water, buildings, roads and power plants. The engine lays the starter district out in the middle of the map. Starting funds is the treasury a run opens with.
  • Goal: how a run is won (reach a population, reach a score, keep the city running for a number of years, or free build with no goal), the target, and how many broke years in a row lose the city. The panel shows the goal line players see.

A value the game would refuse, such as a map 2 tiles wide, is not saved, and the field says why. Nor is a number typed on the way to it: a map width typed as 300 keeps the width it had, not 30.

#Market and routes (trading sims)

The season a trading sim plays. A session started after a save plays the saved values.

  • Season: the season's length in days. The game ends when the last day is over; 0 is a season with no last day.
  • Goods: each good's name and its lowest, base and highest price.
  • Routes: for each place, the routes that leave it, each with its cost, its risk and its trip time in minutes. A route runs one way. You can add a route to a place that has none from here, and remove one. A trip takes at least a quarter of a day, however short its time.

#Multiplayer

Max players, supported modes (solo, co-op, competitive), turn mode (sequential, simultaneous, adaptive).

#Game Master

GM profile: style, verbosity, fairness, improvisational flexibility, ruling preferences.

#Audio section

#Audio

The world's sound: global music (scale, tempo, track), each scene's soundscape (mood, volume, playlist, ambient and effects), the sound-effect palette and ambient queries. Retro-styled games add Retro Audio.

#Assets section

#Asset Browser

Browse, search, and manage the game art for your world. The browser shows assets bound to your blueprint alongside the full public catalog, filtered by type (tilesheet, spritesheet, icon set, environment, UI element), genre, art style, and free-text tags.

The catalog draws from two sources:

  • Stock art. CC0-licensed professional game art from Kenney (60,000+ assets), OpenGameArt (community-contributed pixel art, tiles, and sprites), and curated itch.io packs. Stock assets are grouped into visually coherent packs where every image shares the same palette, outline weight, and lighting. During catalog search, stock assets receive a 1.2x relevance boost because human-curated art coheres better within a set than independent AI generations.
  • AI-generated art. Created by the Asset Forge during compilation when the catalog has no match. Each generated asset enters the public catalog (unless marked private) so future compilations can reuse it.

When a stock pack matches your world's genre and art style, the forge pulls the entire pack, so floor tiles, character sprites, and item icons all share the same visual family. Missing assets get AI-generated with the pack's style anchor as a reference image, ensuring the generated art inherits the same lighting, palette, and texture.

#Asset Packs

Packs are complete sets of visually consistent assets grouped by genre and art style. Each pack carries two metadata artifacts the forge uses for cohesion:

  • Style anchor. The most representative image. All AI-generated gap-fillers reference this image so they match the pack's visual DNA.
  • Cohesion description. A structured profile of the pack's texture, lighting direction, dominant palette (hex values), outline characteristics, shading technique, scale, and perspective.

Browse other creators' public packs from the Asset Browser. Click Use this pack to import all assets into your current blueprint.

Reuse saves credits
Pulling a catalog asset costs 0.5 credits. Generating a new one costs 5 (public) or 10 (private). For a typical roguelike needing 40-60 assets, catalog coverage cuts the total cost by roughly 70% compared to pure AI generation.

#Image Reuse

Two tabs: Outgoing lists this world's images that other worlds reused, and Incoming lists images from other worlds served into this one. Unlink an outgoing image to make it private and stop its reuse.

#Cover & Screenshots

The screenshot gallery and the featured cover: pick any screenshot to feature it.

#Forge tools

The Forge's tools, the Audio Generator among them, also open here as panels and work on this world.

#Asset privacy

Catalog assets generated for your games (sprites, tiles, icons, music, SFX, ambient loops, NPC voices) carry a visibility: public rows join the shared catalog and can be reused by other creators' games; private rows stay yours. The rules by plan:

  • Free plan. Every generated and uploaded asset publishes to the public catalog. The visibility controls are disabled with an upgrade hint; this is how the shared catalog grows.
  • Paid plans (Plus, Pro, Forge). You control visibility at three scopes:

    • Per asset. Flip any asset you authored between public and private from the Asset Browser or your account catalog pages (/app/account/music-catalog and siblings).
    • Per game. "Keep this game's assets private" in the Asset Browser or the Settings panel makes new generations for that game default private, with a one-click sweep over the game's existing assets.
    • All games. The account-wide default in /app/settings (next to the remix defaults) sets the visibility for newly generated assets everywhere, with an optional apply-to-existing sweep that can target only AI-generated rows so uploads stay put.

Resolution order for a new asset: per-game flag, then your account default, then the plan baseline (Forge defaults private; Plus and Pro default public). Generating public costs 5 credits per asset, private costs 10. Making a row private removes it from catalog search; games that already bound it keep working. This is separate from Visual Style Privacy (the Forge-only toggle in Settings that keeps painted scene images out of the cross-world reuse pool) and from the world's own visibility.

#Analytics section

The world's analytics screen, the same one its world page shows. See Publish and share your worlds for what it tracks.

#Settings section

#Game Shell

Seven tabbed sub-sections controlling the non-diegetic wrapper around gameplay:

  • Story Mode: authored, guided, or infinite. This is the top-level narrative generation policy for the world. authored locks the experience to creator-defined scenes and disables prompt-only scene painting. guided is the default: creators define anchors and the AI fills gaps between them. infinite gives the Director full authority to generate arcs, scenes, and pacing on demand.
  • Title Screen: mode (none / minimal / cinematic / interactive), world title, tagline, logo style, duration, skippable toggle, cover image URL, and five composable visual effects layers (background animation, particles, text treatment, color palette, timing curve. 30,000+ unique combinations).
  • Dialogue: response mode (freeform / choice-only / hybrid / choice-with-custom / voice-hybrid), presentation style (chat bubbles / classic box / visual novel / cinematic / minimal), choice count (3-5), tone chips, gated choices.
  • Tutorial: mode (none / first-scene-guided / contextual-tooltips / interactive-walkthrough / calibration-first / replayable-help), skip enabled, replay enabled, calibration required.
  • Hints: master toggle, mode (off / manual / stall-triggered / progressive / adaptive / character-whisper / adaptive-puzzle), delivery channel (narrator / character / UI overlay / codex nudge), spoiler budget slider (1-5), stall detection (turns + minutes + failures), cooldown, max per objective.
  • Story UX: recap on resume, recap minutes, point-of-no-return warnings, chapter summaries, objective reminder cadence, ending foreshadowing.
  • Advanced: anti-cliche level (off / standard / aggressive), contextual action suggestions, strategic analysis.

#Turn Replay

Pick a session of your world and step through it turn by turn: the player's action, the reply, state changes, memory traces, judge verdicts and raw state. Read-only.

#VR / XR

Per-world VR settings: mode (auto, cinema, native or off), character presence, locomotion, vignette, snap turn, sitting mode and VR tier. Anything left unset follows the game type's defaults. See VR mode for creators.

#Marketplace listing

Sell this world: set a fixed price or pay what you want, see what you keep and what the platform takes at that price, then publish, pause or retire the listing. Free accounts see an upgrade note: selling needs Plus or higher. See Publish and share your worlds.

#Metaverse opt-in

Submit a published world to the $WILD metaverse program, which launches in Q3 2026. See Submitting a world to the metaverse.

#Credits & Provenance

For classic recreations: the original author, publication year and license. Interactive-fiction stories edit in the Story Studio below.

#VN Config

Only relevant for visual-novel family worlds:

  • Dialogue mode: ADV (text box at bottom) vs NVL (full-screen scrolling text)
  • Default theme: preset name
  • Active extensions: extension IDs to enable
  • Theme overrides: colors (dialog bg, text, speaker name, choice bg/hover, inner thought, narration), typography (font family, size, line height), animation (typewriter speed, cursor character, transition duration), layout (dialog min/max height, sprite max height), custom CSS injection
  • CG Gallery: event illustrations unlocked during gameplay. Each entry has a scene ID, prompt, image URL, and starting unlock state.

#Settings

World-level settings: tagline, visibility (private / unlisted / public), content policy tier, death mode (soft / hard / permadeath / custom), co-op mode (solo / simultaneous / hybrid), allow live editing, allow cloning, linked characters.

#Family-specific panels

Visual novel worlds get four additional panels from the VN family. A picture drawn for the world, in these panels or by a backfill, is stored in your draft as well as in the world when the draft's slot is empty or holds the picture the drawing replaced, so the panels show the art the world plays with. A slot you filled yourself keeps your picture. Play draft on a visual novel plays your draft's backgrounds, characters and routes.

#Scene Painter

Gallery of background locations with time-of-day variants. Edit prompts, paste custom image URLs, regenerate via AI, add or remove locations.

#Character Designer

Expression sets per character. Edit display name, base portrait prompt, and each expression slot (neutral, happy, sad, angry, surprised, embarrassed, thinking). Paste custom URLs or regenerate.

#Route Graph

Visual route/ending editor. Add/remove routes, edit labels, mark secret routes, link to characters, manage unlock conditions (affection threshold, story flag, route complete, choice made, objective complete), link endings.

#Audio Director

BGM and SFX mapping tables. Assign track IDs to scene type + mood combinations, and provide a playable track URL for each BGM you want the VN viewport to actually play. For SFX, actionType is the semantic slot the narrator sees in the prompt, and sfxId is the concrete ID it emits in VN sfx actions. This lets creators keep readable semantic labels like door_open while still mapping them to custom runtime IDs like door_creak.

Roguelike worlds get panels from the roguelike family beside Procedural Layout, among them the two below. A saved change in them plays in the next new session: Play draft starts one from the draft, and players get it in their new sessions once the world is published.

#Roguelike Config

The viewport mode: top down (a 2D grid), first person (a raycast view from the player) or hybrid (top down in the dungeon, switched with V or the view button). The choice is saved to familyConfig.viewportMode, the field the game reads, and to the top-level viewportMode the compiler builds familyConfig from. The size and shape of each floor come from the world's layout in Procedural Layout.

#Dungeon Crawl

The per-floor plan every descent replays. Each floor's theme and enemy counts are edited here; its enemy labels, loot and boss mark are shown. Regenerate from progression rebuilds the plan from the world's progression curve with no model call. It keeps the landmark names of the progression's landmark floors and the world's name in the floor themes, and it asks first, since every floor edit made here is lost.

#Audio Generator

The Audio Generator (in the Forge at /app/forge/audio-gen, and as a Studio panel) makes one sound effect, music loop or ambient bed from a short description. Each type's cost shows in the Type list, read from the same table the charge uses, and you are charged only when a clip is generated; the balance is checked first, so a clip you cannot pay for is never started.

While it works, a line under Generate shows the stage and a clock: "Generating…" with the type's usual wait (sound effects take a few seconds, ambient beds about half a minute, music loops one to two minutes), "Taking longer than usual…" past it, then "Saving the clip…". The clip plays the moment it is saved, and the form is free again so you can make another while it is checked for the catalog ("Checking it for the catalog…"). The verdict follows: "Saved to your private catalog." or "Saved to your public catalog." (a free plan's clips are published, and the line says so), "Not added to the catalog: under the quality bar.", or "The catalog check didn't run." The download works in every case. Under the player, a line shows the credits charged and what is left.

A refusal says what to do next and never charges: "Not enough credits. This costs 3; you have 1." with Get credits, "The audio service is busy. Try again in a minute.", "The generation took too long and was stopped.", "This audio type is unavailable.", or "Audio generation failed." with Try again.

#Story Studio (Ink and Twine stories)

Interactive-fiction stories open in their own editor at /app/story/<id>/edit, not the game studio. Stories created from scratch at Create a Story or imported from an .ink / .ink.json / Twine HTML file land there directly, and old game-studio edit links redirect.

The Story Studio is a two-pane writing app:

#Write (left pane)

A full-height code editor with syntax highlighting for both engines. Ink stories edit the canonical .ink text: what you write is exactly what plays and exactly what exports. Twine stories edit twee source derived from your published HTML; saves swap your passages back into the original file, so the story format engine you published with keeps rendering, untouched. Validation problems (a missing start passage, duplicate passage names) appear in a slim strip under the editor with clickable line jumps; they're an expected mid-edit state, not a failure.

#Play (right pane)

The live story. Ink recompiles as you type, replays your recorded choice path, and lets you click through choices or restart. Twine reloads the sandboxed story after each save so your edit plays. Drag the divider (or use arrow keys on it) to resize the split.

#Toolbar overlays

Map opens your story's structure: for Ink, an editable graph of knots and stitches where changes write back into the script; for Twine, the passage map. Variables and Enhance (per-passage scene art prompts and ambient moods) open the same way for both engines.

#Replace HTML (Twine)

Twine stories carry a Replace HTML button in the top bar. Publish a new build from Twine, pick the file, confirm, and the story updates in place: every passage is replaced, while the cover, credits, reviews, play stats, and your Enhance work stay. The same checks as import apply (self-contained HTML with its story format engine embedded).

#Details

One slide-over holds everything else: story info, tags, content rating, cover and screenshots, audio, credits and provenance (author byline, publication year, license), publishing, and analytics. The two story checks (title and setting summary) sit at the top. Publish from the top bar when they're green.

On phones the panes become Write / Play tabs at the bottom of the screen.

#The right sidebar

#Wilo

The in-app assistant, inline. It knows which world you are editing and shares one conversation with the Wilo drawer, so reopening the world resumes it. See How to use Wilo.

#Inspector

Three cards:

  • Validation: checklist of world completeness (world name, setting summary, regions, NPCs, endings for campaigns, item catalog if combat enabled, title screen configured, VN characters/backgrounds for VN worlds, cross-region elements for 3+ region worlds)
  • World Stats: 12 counters (regions, NPCs, factions, scenes, items, objectives, endings, loot tables, recipes, dynamic lore, merchants, cross-region elements)
  • Active Mechanics: pills showing which engines are enabled
  • Completeness: percentage gauge based on validation checks

#Preview

Three modes:

  • World Card: mockup showing how your world appears in browse and search. Cover image pulled from gameShell.title.coverImageUrl.
  • Narration: click Generate Sample to get a live AI-generated paragraph in your current tone and style.
  • Tone Radar: SVG visualization of your four tone sliders.

#AI Tools

Context-sensitive generation tools that change based on the active panel:

  • World Overview → Generate Setting Summary, Generate Regions
  • NPC Editor → Generate NPCs, Generate NPC Personality, Generate Avatar Prompt
  • Lore Editor → Generate Lore, Generate History, Generate Factions
  • Item Editor → Generate Items
  • Endings Editor → Generate Endings
  • Art Style → Generate Art Style

Generated content is automatically merged into your draft: regions are appended to world.regions, NPCs to world.npcs, art style to artStyleVocabulary, etc.

Suggestions appear below the tools, flagging missing content (no endings defined, combat enabled but no items, no merchants despite item catalog, etc.). Click Generate to run the matching AI tool, or Go to panel for suggestions without a fix tool.

Research search box lets you look up reference material that gets indexed into your world_canon RAG cell.

#Vibe Loop

For AI-built games only; other worlds show nothing here. It records the game's last build: the game architecture, a loop summary (exit reason, cost, duration, revision, custom components), one chip per quality track with its verdict, and the list of build steps. Three controls rerun it: Polish this game, Re-iterate stuck tracks (20% of a full run's credits, for tracks marked yellow or red) and Revert to prior (free, back to the previous revision).

#How studio edits reach gameplay

Authored content flows directly into the runtime:

  • Scene images: when a scene contract has sceneImageUrl set, the turn pipeline uses it directly and skips AI generation, saving credits. In guided and infinite story modes, sceneImagePrompt without a URL can still drive runtime scene painting. In authored story mode, prompt-only scene images do not render: authored scenes that want background art must provide a concrete sceneImageUrl, Studio warns about that requirement, and the runtime now clears old authored backgrounds when the next authored scene intentionally has no image.
  • NPC avatars: avatarPrompt is consumed by the enrichment pipeline (generates portraits at compile time) and by the runtime async portrait step (when new NPCs appear mid-session).
  • Dynamic lore: discovered entries are injected into the world_canon RAG cell so the narrator references them when relevant.
  • Dialogue mode, title effects, tutorial config, hint policy: read directly by the shell runtime.
  • Merchants: every session opens with your shops (or the ones made for each region), and each restocks by its trigger.
  • Endings: the EndingEngine evaluates conditions every turn.

#Visibility

Every world has one of three visibility levels, set in the studio Settings panel and changeable anytime:

  • Private (only you). Only you can open it. It never appears in Browse or search, and the direct link will not work for anyone else. New worlds start here.
  • Unlisted (link-only). Anyone with the direct link can open and play it, but it stays out of Browse, search, and your public profile. Use it to share a work in progress with specific people without publishing to the catalog.
  • Public (browsable). Listed in Browse and search and discoverable by everyone. Public worlds can be featured on your profile and, with Allow cloning on, remixed by other creators.

Setting a published world back to Private hides it from the catalog without losing your work or its revision history.

#Publishing

Click Publish in the top-right to open the publish modal:

  1. Review your changes
  2. Add an optional publish note
  3. Confirm

Publishing creates a new immutable revision of the same world: the listing, URL, reviews, and stats stay; only the version advances. Live editing is on by default, so the change reaches active sessions on the player's next turn and each player sees a brief, dismissible "World updated" toast carrying your publish note. The mechanics a session pins (see Mechanics) and structural changes apply to new sessions only. Turn live editing off to keep in-progress players pinned to the revision they started on. New sessions always start on the latest revision. When Allow cloning is on, that toast offers Snapshot previous version so a player can keep the version they were on.

#Forking

Worlds marked with Allow cloning can be forked by other players. Forks inherit your content but become independent drafts owned by the forker. Clone attribution is shown in browse cards and in the studio header.


#References

  1. Getting Started: Create your first world before opening the studio.
  2. Writing Great Prompts: Better prompts produce better starting blueprints.
  3. Creator Economy: Publishing, monetization, and the Forge plan.
  4. HEXACO Personality Model: The 6-factor model behind NPC personality sliders.
  5. Character Relationships: How trust and personality affect NPC behavior.
  6. wilds.ai FAQ: Studio-related questions and troubleshooting.