Repository navigation
Prompts
The Prompts tab in Settings holds the text the app sends to the AI for each request. A prompt preset is one full set of that text, with its per-prompt options. The app sends the prompts of the active preset.
The Prompts tab shows in Advanced mode only. To open it, open Settings, select Advanced in the switch next to the title, then open the Prompts tab.
A prompt sends only what it holds. Text you delete is gone from the request, and a chip you delete sends nothing. See The Chip Editor.
The built-in presets are read-only. To change a prompt, make your own preset first.
- Open the Prompts tab.
- Open the Preset list and select Add New Presetโฆ.
- Type a name in the New Preset dialog.
- Select Save.
The new preset is a copy of the preset that was active, and it is now the active preset. To copy a built-in preset in one step, select Duplicate & Edit on its read-only notice. The copy is named "preset name (copy)".
- Select your own preset in the Preset list.
- In the list of prompts, select the prompt, such as Choices.
- Under the prompt, select System Prompt, or User Message when the prompt has one.
- Type in the editor. To add a value from the game, select a chip in the bar above the editor. See The Chip Editor.
- Open the Preview tab to read the text with sample values in place of the chips.
The app saves each change at once. To go back to the shipped text, select Reset name Prompt under the editor, then confirm.
You can send one prompt to a different text endpoint, such as a small fast model for Choices. First add the endpoint as a preset on the Endpoints tab. See Text.
- Select your own preset in the Preset list.
- Select the prompt, then the Options row under it.
- Open the Endpoint list and select the endpoint preset.
- Read the line under the list. Reachable means the endpoint answers. Select Recheck to try again.
To send the prompt to the active endpoint again, select Use Active Endpoint. A prompt route stays on this device. It is never in an exported or published preset.
- Select the preset in the Preset list.
- Select the Export button.
- Select Copy code to copy a share code, or Export .json to save a file.
To add a preset that someone shared:
- Open the Preset list and select Import Presetโฆ.
- Paste the share code, or select Choose fileโฆ and pick the
.jsonfile. - Read the warnings and the preset's Overview, then set its Name.
- Select Import. When you selected the checkbox to overwrite a preset with the same name, the button reads Overwrite.
See Sharing a Preset for what a shared preset holds.
You must log in to Community Creations to publish. See Login and Register.
- Select your own preset in the Preset list.
- Select Overview at the top of the list of prompts.
- Add at least one model to Models.
- Fill in Author, Description and Tags.
- Select Publish, then follow the publish dialog.
Without a model in Models, Publish opens Add a Model. Select Open Overview to go to the field.
- On the main menu, select the world.
- In the world dialog, open the Prompts list.
- Select a preset. Use global preset goes back to the preset that Settings has active.
That world now runs on the preset you picked. Your other worlds do not change. A library Group can also carry a preset, and the list then shows Use group preset. The world's own pick comes first.
While you play a world with its own preset, the Prompts tab edits that world's pick. A note under the Preset list tells you so.
The Preset list holds four built-in presets, then your own presets.
| Preset | What it is |
|---|---|
| Default | The shipped prompts, with Markdown headings |
| Simple | The same prompts, with plain labels in place of headings, and chips set to plain text |
| XML | The same prompts, with each section in an XML tag, and chips set to XML |
| Experimental | A short narration prompt that sends entity summaries and lets the AI look up full entries. See The Experimental Preset. |
A built-in preset is read-only. Its editors show a lock notice with Duplicate & Edit. You can still read every prompt, and you can still switch Tools on and off for it in the Tools tab.
A preset holds:
- the text of every prompt and message
- each prompt's Options, except Endpoint, which only your own presets carry
- which Tools are on. See Tools.
- the Overview, on your own presets
The buttons next to the Preset list act on the active preset. On a narrow screen they are in the Preset Actions menu.
| Button | What it does |
|---|---|
| Rename | Changes your preset's name |
| Export | Opens the share code and the .json download. Built-in presets have it too. |
| Publish | Lists your preset in Community Creations. Shows while you are logged in. |
| Reset | Sets every prompt in your preset back to its shipped text |
| Delete | Removes your preset |
Reset and Delete ask you to confirm. Neither can be undone.
Experimental tests a different way to tell the AI about the entities. Its narration prompt sends a one-line summary of each entity, not the full description. The get_entity Tool is on, so the AI can fetch the full entry of an entity it writes about.
On an endpoint with no Tool support, the AI gets the summaries only. The narration then knows less about each entity than with Default. See Endpoints Without Tool Support.
A share code and a .json file hold the same preset:
- the prompt text and the section style
- the tuning: samplers, reasoning, Max Output, Include Attachments and Verbatim Turns
- the Overview
- which Tools are on, and a copy of each of your own Tools that is on
A shared preset never holds a prompt's Endpoint. Endpoint presets exist only on your device.
The Import Preset dialog shows warnings before you import. A preset with a Script Tool warns you, because a script runs code when the AI calls it. Clear the tuning checkbox to import the prompt text only. When you already have a preset with that name, a checkbox offers to overwrite it. Without it, the import adds a separate copy.
Overview is the first row in the list of prompts. It shows on your own presets only. It describes the preset for other players when you publish it.
| Field | What it holds |
|---|---|
| Author | Who wrote the preset |
| Description | What the preset changes and how to use it |
| Tags | The preset's style and purpose |
| Models | The models the preset works well with. Publishing needs at least one, because players filter prompts by model. |
The list on the left groups the prompts by the job they do. A prompt shows only while its feature is on. When you turn a feature off, its prompt is not in the list, and the app sends nothing for it.
| Prompt | Group | What it does in a turn | Shows while |
|---|---|---|---|
| Narration | Story | Writes the story text you read each turn | Always |
| Planning | Story | Plans the turn before narration writes it: who is present and what happens next | Thinking is Planning |
| Director | Story | Describes the scene first: who is here and what each entity is doing | Thinking is Staged |
| Character | Story | Runs once for each staged entity, after the director. States what that entity wants this turn, in the first person. | Thinking is Staged |
| Discover Entity | Story | Writes a lasting note for each new entity the story names. Runs after narration names a new entity. | Describe New Characters is on |
| Storyboard | Story | Merges the intentions of every staged entity into one plan, before narration | Thinking is Staged |
| Choices | Story | Writes your choices for the next turn, after narration | Choices is on |
| Stat Updates | Trackers | Reads the turn after narration and records which stats changed | Stat Updates is on |
| Location Change | Trackers | Decides after narration if your action takes you to a new location | Location Change is on |
| Clock | Trackers | Measures how much story time the turn took. Runs after narration. | Measured Clock is on |
| Opening | Trackers | Reads the opening scene once, at the start, to set the time of day | Measured Clock is on |
| Summaries | Memory | Condenses an older turn into one line that the AI can still read later. Runs when a turn is too old to go word for word. | Memory Summaries is on |
| Milestone Select | Memory | Decides between turns which summaries stay in long-term memory | Memory Summaries is on |
| Diary | Memory | Writes a private, first-person note from each present entity. Runs when a turn is too old to go word for word. | Character Diaries is on and Thinking is Staged |
| Scene Tags | Images | Tags the action of the scene for the scene image, after narration | Enable Image Generation is on |
The settings in the last column are in Settings โ Output, and Enable Image Generation is in Settings โ Endpoints โ Image. See Settings. A line above each editor repeats what the prompt does.
Select a prompt to open its Anatomy. The rows under the prompt open its other surfaces. On a narrow screen, one list at the top holds the prompts and the surfaces.
| Surface | What it holds | Which prompts have it |
|---|---|---|
| Anatomy | The request this prompt sends, built from your current settings | Every prompt |
| System Prompt | The instructions the AI reads first | Every prompt |
| User Message | The message that asks for the reply, such as your action | Narration with Thinking set to Native, Choices, Stat Updates, Location Change, Clock, Opening, Summaries, Milestone Select, Director, Discover Entity, Scene Tags |
| Messages | Extra lines in the narration history | Narration, while one of its messages is live |
| Options | Per-prompt settings | Every prompt |
The View full screen button opens the whole panel, with the list, in a large window.
Anatomy shows the Request Anatomy: a labeled map of every message in the request this prompt sends. It uses your live settings. In a game, it uses that game's values. Outside a game, it uses sample values.
| Tab | What it shows |
|---|---|
| Chips | Your template, with each value as its chip. Dashed chips are parts the app fills in. Select one to open the prompt that writes it. |
| Preview | The full text of the request. The highlighted parts are text you typed. Select one to open the editor you typed it in. |
On a wide screen, a button shows Chips and Preview side by side. When your settings never send this request, Anatomy says so and shows nothing.
The System Prompt is the main text of the prompt. It tells the AI its job and holds the chips for the world, the scene and the story so far. Edit it in the chip editor.
The User Message is the last message of the request. It usually holds your action or the narration, and the task for this reply. Narration has it only while Thinking is Native. In the other thinking modes, the app writes that message itself.
Messages holds the extra lines that Narration puts in its history. Each field shows when the app sends it, and has its own Reset.
| Message | Sent when |
|---|---|
| Recap Message | Memory Summaries has condensed older turns. It asks for the story so far, and the summaries answer it. |
| Now Message | At the end of the recap. It says where things stand now. |
| Recall Message | Scene Recall brings back an old turn. It marks the scene as the past. |
| Direction Message | Your action has text in [square brackets], while Thinking is Native. It tells the AI the brackets are your direction as the author. |
A field shows only while its feature can send it.
Options holds the settings of one prompt. On a built-in preset they are read-only.
| Option | What it does | Shows on |
|---|---|---|
| Endpoint | Sends this prompt to an endpoint preset. Use Active Endpoint follows the Endpoints tab. See How to Route a Prompt to Another Endpoint. | Every prompt |
| Include Attachments | Sends the images attached to your action with this prompt. On for Narration only, by default. | Prompts that send your action, while Image Attachments is on |
| Max Output | Caps the answer length of this prompt. Off shows Auto and the shipped cap. It does not cap reasoning. | Planning, Director, Character, Storyboard, Choices, Discover Entity, Summaries, Milestone Select, Diary, Scene Tags |
| Verbatim Turns | Sets how many recent turns go word for word. Older turns go as summaries. | Narration, Planning, Choices, Stat Updates, Location Change, Summaries, while Memory Summaries is on |
| Native Reasoning | Sets whether this prompt reasons, and how hard. Clear the checkbox and the prompt does no reasoning. Global follows Native Reasoning in Settings โ Output. Model Default sends no hint. | Endpoints whose model reasons. Not on Narration with Thinking set to Inline. |
| Reasoning Budget | Sets the reasoning tokens as a percent of the endpoint's Max Output Tokens. The answer keeps its full length. | Endpoints that take a token budget, such as the desktop engine |
| Custom Temperature | Sets this prompt's temperature. Off shows the value the prompt sends now. | Every prompt |
| Custom Repetition Penalty | Sets this prompt's repetition penalty. Off shows the value the prompt sends now. | Every prompt |
Reasoning Budget needs Max Output Tokens on the prompt's endpoint. Without it, the slider is off, and a note tells you to set it. A model that always reasons keeps the Native Reasoning checkbox on, with a note.
Each prompt editor is a chip editor. A chip is a box in the text. When the app sends the request, it puts a value in each chip, such as the world description, the stats or your action.
- A chip sends text, and no chip sends nothing. The app adds nothing to a prompt that you did not place. Delete the Notes chip, and the AI never reads your notes.
- The bar above the editor lists the chips this prompt can use. Select a chip to put it at the cursor, or drag it into the text.
- Select a placed chip to open its pop-out. The pop-out sets what the chip sends. Some chips have a Scope or Content choice, such as Full or Summary. The Stats, Traits, Location, Entities and Persona chips have a Format of Simple, Markdown or XML. Some chips also have Header, Prepend and Append text.
- The editor has two tabs. Edit holds the template. Preview shows the text with each chip filled in. Outside a game, a Sample data tag marks the sample values. Preview is off while the prompt has no chips.
The editor has Undo and Redo buttons. Edit full screen opens the editor in a large window. For the chips that carry the persona and names, see Personas for Authors.
A world can bring its own Narration, Choices or Stats prompt. An author writes them under Custom Prompts in the World Editor's Overview tab. See World Editor Overview. A world prompt takes the place of your preset's prompt for that pass.
When you select such a world on the main menu, a notice under its details names the custom prompts, such as "This world uses a custom narration prompt." Select View on the notice to open Custom Prompts.
Custom Prompts compares the world's prompt with the prompt Formamorph ships for that pass. It does not compare with your own preset.
| Control | What it does |
|---|---|
| Changes | Shows one text with the changes marked. Added text is tinted green. Removed text is struck through in red. |
| Raw | Shows the world's prompt as the author wrote it |
| Pass tabs | Narration, Choices and Stats. Show when the world rewrites more than one pass. |
Chips show as their raw text, such as <WORLD DESCRIPTION>. To play the world with your own prompts, clear Use this world's prompt under the world's buttons on the main menu. See The World Dialog.
- ๐งฐ Tools: the functions the AI can call, and which preset turns them on
- โ๏ธ Settings: the features that add or remove prompts, and the text endpoints
- ๐ง Story Memory: what Summaries, Milestone Select and Diary keep
Playing
- ๐ Connect Your Own AI
- ๐ฑ Install on Android
- ๐ Library
- ๐ง Avatars
- ๐ช Starting a Game
- ๐ฎ How to Play
- โ Formaquestion
- ๐พ Saves and Backup
- ๐ง Story Memory
- ๐ญ Entities in Game
- ๐ชช Personas
- ๐ Community Creations
- โ๏ธ Settings
Reference
- ๐ค Glossary
- โ๏ธ Text Formatting
- ๐ ๏ธ World Editor
- ๐งช Test Bench
- ๐ชช Personas for Authors
- ๐ World Format
- ๐งฎ Scripts Guide
- ๐ Linked Content
Project