Tutorial

1. How to Play

FicMachine is not a game in the traditional sense. There are no mechanics, no stats, and no game master. The fun lies in setting up interesting characters and situations and seeing where the AI takes the story next.

1.1 The Loop

Everything revolves around this loop: you read what's there, you edit it, and then press Continue to see what happens next.

Each time you edit the text — even small changes, like tweaking a line of dialogue or adding a detail about how a character reacts — you're steering the tone and the content of the story. The AI picks up on what's there and follows along.

So the loop isn't just "generate, read, generate." It's more like a conversation between you and the AI, happening inside the story itself. You nudge the story and the AI responds, then you nudge again.

The best moments come when the AI comes up with something great you didn't expect. When it picks up on something you set up and does something interesting and unexpected. Those moments are why you come back to play.

1.2 Getting Started

Here's how you get the most out of FicMachine.

Pick a scenario with a strong premise. Browse the scenarios or make up your own. Choose one that sounds like a story you'd want to read, one that features your favourite characters, or one that lets you roleplay a crazy situation. Don't plan every detail; you'll make it up as you go.

Read the opening, then make it yours. When you start a scenario, you'll get an opening scene. Read it. If something doesn't fit — a character's name, a detail about the setting, the tone of a line — just edit it. The whole text is yours from the moment you start. Nothing is locked.

Press Continue and see what happens. The AI will write the next piece of the story based on everything that came before it. Read what it gives you and tweak it. If something feels off or flat, don't leave it there, the AI will tend to follow the patterns that are in the recent text. Adjust it, and the AI will adjust to your changes.

Think in Scenes. Ask yourself: What is this scene about? Who does what? And what changes at the end? When you start a new scene, I recommend you use an Author's Note to let the AI know where to go.

Don't plan everything on the scene, let yourself be surprised. Your author's note can be vague, like [Author's Note: Sarah and Sam meet for coffee, and Sarah tells an uncomfortable truth.] You don't know what Sarah will say, or how Sam will react, which is part of the fun.

If you still have questions, come by the Discord. We'll be happy to answer them.

2. Context

Large language models are great improv actors, but they have zero long-term memory. Every time you click Continue, the AI has to re-read the entire script from scratch to understand what it needs to do.

This script is called the Context, and it has a strict size limit measured in Tokens.

2.1. What are Tokens?

Tokens are chunks of words that the AI uses to process text.

• In the English language, one token is roughly 3/4 of a word.
• The AI can only hold a certain amount of tokens at once. If the story gets too long, the earliest parts fall off, and the AI forgets them.

2.2. How Context is built

We can't fit an entire novel into the Context. Instead, we assemble a "cheat sheet" for the AI. The Context is made up of three parts:

• AI Instructions: These tell the AI who they are and what their job is.

  For example: "You are the narrator of a gritty cyberpunk noir. Focus on cynical characters and punchy banter."

  This is how you tell the AI to write a story instead of behaving like a helpful chatbot assistant.

• Story Cards These are character and setting information that the AI needs to know to keep the story consistent.

  Story cards have a set of trigger words. When any of the triggers appear in the story, the card is added to the Context.

  Story cards can be pinned, meaning that they will be added to the context even if the triggers do not appear in the recent story.

• Recent Story

  The rest of the Context is made up of the most recent story.

3. AI Instructions

This is where you tell the AI who they are, the tone of the scene, and the rules of the world.

The AI has a limited "cognitive bandwidth" to follow instructions. To get the best results, you need to be strategic about what you include here.

3.1. Don't State the Obvious

Every word in your AI Instructions costs Tokens, which eats into your story's budget.

If you add an instruction that specifies the setting as "Cyberpunk," the AI will already know that cars fly, corporations control the world, there are neon lights everywhere, and people pay in credits.

Use the AI Instructions to specify things that the AI can't figure out on its own.

3.2. Less is More

You may be tempted to add more and more instructions to craft your perfect writing partner. However, each instruction that you add dilutes the strength of all the other ones.

If you give the AI three instructions, it will likely follow all of them. But if you give the AI thirty instructions, it will either start ignoring them or start producing nonsense, trying to follow all instructions at once.

Focus on the essentials. What are the 3-5 things that must happen for your story to work?

3.3. Don't Fight the Model's Nature

AI Instructions are powerful, but they are not magic. They cannot fundamentally change how the underlying AI model was trained.

If a model was trained to be polite and helpful, writing an instruction like "Use rude and offensive language" will result in weird, passive-aggressive slop.

If you told the model to "focus on sensory details", it may start describing random smells every few paragraphs, as it doesn't know what else to do with those instructions.

Instead of trying to force a model to do something it can't via instructions, it is better to edit the output directly, or switch the underlying model in the settings.

4. Story Cards

Story Cards are the encyclopedia of your world. They hold the details about characters, locations, and lore that the AI needs to keep the story consistent over time.

4.1. Pinned and Triggered Cards

There are two ways to get a card into the AI's Context:

• Pinned Cards: These are always added to the context, no matter what. Use them for your main character or to provide a summary of what has happened so far.

• Triggered Cards: These stay hidden until a specific word is mentioned in the story. Best for side characters and world elements that don't always need to be in memory.

4.2. Keywords

Triggered cards rely on keywords to know when to show up. The matching logic is based on partial word matches.

• Case-Insensitive: Capitalization doesn't matter. Excalibur matches excalibur.
• Partial matches: The keyword "age" would trigger a card if the story contains the word "teenager".

Note: The system does not smartly detect plurals. Elves will not trigger the keyword elf.

4.3. Story Card Limits

The system runs three checks before adding a Triggered Card:

• Is the exact trigger in the recent story text?
• Will adding this card push the actual trigger word out of the story? (If the AI can't see the word that triggered the card, we don't include it.)
• The 50% Budget Rule: If the AI Instructions plus the Story Cards take up more than 50% of the total memory. We don't add any more story cards to save space for the story.

5. Author's Notes: Guiding the AI

Sometimes, you might find yourself stuck in a loop where the AI just isn't getting it. You hit the retry button five times, but the AI keeps writing the same things.

Instead of hoping for the best, you can use an Author's Note. Type it directly into the story using square brackets.

[Author's note: instruction goes here]

Examples:

• [Author's note: The next scene should be tense and silent. No dialogue.]
• [Author's note: A sudden explosion interrupts the conversation.]
• [Author's note: Sarah reveals an uncomfortable truth.]

When the AI sees the [Author's note: ...] format, it understands that this is not part of the story, but a meta-instruction for how to continue the story.

You can use [Note: ], [Director's Note: ], [Author's Note: ], etc. As long as it's distinct from the main text, the AI should understand it.

Author's notes are often much more effective than editing the AI Instructions because they are immediate and specific to the current moment.

• Keep it focused: Use an Author's Note for the immediate next scene or tone shift.
• Clean up later: After the AI follows your direction, you can delete the [Author's note: ...] line from the story text if you want to keep it clean. (Though leaving it in doesn't hurt.)

Author's notes are not only for troubleshooting, but they can also be a lot of fun. If you get bored with a slice-of-life summer vacation story that's going nowhere, try this:

[Author’s Note: The return home is eerie and quiet. There are signs of a zombie apocalypse.]

6. Troubleshooting

Or how to stop the AI from doing X

The AI may get stuck in a loop of bad habits. Maybe it keeps describing a "musty smell" in every paragraph, or perhaps every character is unexplicably rude.

The AI is a pattern matcher. If it sees a theme in the recent story (e.g., "Main character gets insulted"), it will think that this is what the story is about, and do that again.

Here is how you fix it

6.1. Nip it in the Bud

The most powerful tool you have is your keyboard.

It's a matter of time until the AI mentions the "smell of ozone" in your cyberpunk scenario. If you leave it there, the AI sees it as a valid pattern. A few paragraphs later, it will mention it again. By the tenth paragraph, it's a recurring motif.

The fix is simple: Remove it the first time it appears.

If the AI makes a character oddly annoying, edit that sentence to make them neutral or friendly before you continue.

If you remove the "bad behavior" from the text, it no longer exists in the Context. If it’s not in the context, the AI has no pattern to copy, and it will be less likely to generate the same thing again.

6.2. AI Instruction audit

Sometimes, the problem is more insidious.

You might have an instruction like "describe scenes with rich sensory details" or "create dramatic conflict."

As the AI tries to follow the instructions, it may start describing random smells or making characters antagonistic in arbitrary points in the story.

Check your AI Instructions. If the AI is doing something annoying, temporarily delete any instructions related to that topic.

6.3. The "Negative Instruction" Trap

You may be tempted to add a new instruction: "Do not describe smells."

This often fails, as AI models sometimes struggle with negatives. Telling it "Don't think of a pink elephant" makes it focus on "Pink Elephant."

The Better Way is to give a positive redirection.

Bad: "Do not make characters mean."

Good: "Characters should be cooperative and friendly unless provoked."