# Manga Generator

Turn your story into a manga: the model adapts parts of the book into a script of pages and panels, and then writes image
generation tags for every panel - ready for Stable Diffusion, Forge, Midjourney and the like. It is a port of the
[MangaGenerator](https://github.com/Enerccio/SillyTavern-MangaGenerator) extension of SillyTavern.

## Opening it

Choose **Manga Generator** in the settings menu (⚙) of the [story editor](../books/story-editor.md). One large window opens;
nothing is added to the story screen itself. You can drag the window and resize it.

## Mangas

A book can have several mangas, shown as tabs at the top of the window. The buttons next to the tabs add a manga (+),
rename it, move it to the left or right and delete it. A manga is kept with the book, so it is in the book's backups and
copies.

Below the tabs are the settings of the open manga:

| Setting | What it does |
|---|---|
| Model and protocol | What the requests go to. The book's own by default; the last choice is remembered. |
| Token budget | How long a script the model is asked for (`` in the prompt). |
| Parts per section | How many parts of the story a new section adapts. |
| Context before / after | How many more parts before and after those a new section gives the model as the background. |
| Script prompt, Tags prompt | Replace the built-in prompts of this manga. **Reset to default** brings the built-in text back. |
| Additional rules | Put in place of `` in the script prompt, for rules of your own (no sound effects, a style...). |
| Copy all | Copies every script of the manga, then every tags text, to the clipboard. |
| Continuous mode | Generates the manga on its own, see [Continuous mode](#continuous-mode). |

## Settings

*Settings → Extension Settings → Manga Generator Settings* holds what a **new** manga starts with: the token budget, parts per
section, the context before and after, the script prompt, the tags prompt and the additional rules - and a text put before every
request, for instructions that hold for all of your mangas. A prompt left empty, or unchanged, is the built-in one. A manga you made
already keeps its own settings. The panel is saved with the **Save** button of the *Settings* page.

## Sections

**Add section** adds a section for the parts of the story that follow the last section (the first one starts with part 1).
Parts are numbered from 1 like in the outline. A section shows:

- **Adapt from part / to part** - the parts the model turns into panels,
- **Context from part / to part** - the parts the model reads; the ones outside the adapted range are only the background
  (earlier events, or what comes next), and the model is told not to adapt them. Change any of the four numbers whenever
  you want a different slice.

Each section has two views, **Script** and **Tags**:

1. **Generate script** streams the script into the section. A reasoning model's thinking is shown above it
   (*Thought for 12.3 s*). The script continues the page numbers of the section before it and does not repeat what that
   script already covers.
2. Switch to **Tags** and **Generate tags**: for each panel of the script, tags for the characters and the scene.

![The Manga Generator window with the script of a section](../../images/plugin-manga.png)

Either text can be **edited** by hand (the pen), **copied**, and a generation can be stopped with **Abort** - what the model
wrote so far is kept. Sections generate independently, so you can start several; while any model is writing, sections
cannot be deleted and mangas cannot be changed.

![The same section on the Tags view](../../images/plugin-manga-tags.png)

**Override** gives one section its own script prompt, tags prompt or additional rules, or a **Rules addendum** that is added to
the rules of the manga instead of replacing them. The eye shows the exact request the model would get, with its size.

## Continuous mode

**Continuous mode** (next to *Add section*) lets the manga write itself, one kind of text at a time, so the model gets the same prompt
again and again (and a server that keeps a prompt cache can reuse it). Choose what it generates in the box next to the button:

- **Scripts only** - it goes through the sections in order and generates the script of every section that has none (a section with a
  script is left as it is). When all of them have one it adds a section for the next parts of the story - as *Add section* does, with
  the *Parts per section* and the context of the manga - generates its script, and goes on, until the story has no part left for another
  section.
- **Tags only** - it generates the tags of every section that has a script but no tags. It adds no section; it ends when every script has
  its tags.

In both it ends when you press **Stop continuous mode** (the button turns into it; what the model has written so far is kept). A failure
of the model, a section with no text in its parts, or an empty answer stops it too, with a message. While it runs the sections cannot
be generated or changed by hand, and the mangas cannot be changed; closing the window stops it. Each section it works on is scrolled
into view. To get both, run *Scripts only* first and *Tags only* after it.

## The lore and the size of the context

The model gets the text of the section's parts, and with it the lore the text calls for: entries of the book's lorebook (and its
sub lorebooks) whose filters appear in that text, found like in a generation, including the scan depth. Entries that are
always on come too. So the characters and places of those parts reach the model with their descriptions, which the image tags
need.

All of it has to fit the **context size** of the model - the context of the AI and the protocol, minus the room for the
reply. When the entries do not fit, the later ones are left out (a notice tells how many; the eye shows it too) - make the section
smaller, or the lore shorter. A request that does not fit even without the lore is still sent, with a warning.

## What is different from SillyTavern

- A chat message is a **part** of the story; there are no names in front of messages, no persona and no character card -
  characters come from the lorebook.
- The world info "trigger" of the extension is not needed: the lore is activated by the text itself.
- The context size is taken from the AI and the protocol, there is no setting for it.
- A new section takes 3 parts with 1 part of context on each side, as a part of a story is far longer than a chat message.
  Change this in the settings (below).
- Tabs are moved with the arrow buttons, not by dragging.