🖼️ CASTADI 🖋️

What is it?

Castadai is turns your text files in the castadai format into comic book pages with dialog bubbles, ai generated images (using stable diffusion) and automatic panel creation!

Example

-Page1

location: public park, yard, outside, park

[misty] weaving, saying hey, happy, smiling
"Heyy!!"

"This is an example"
"of what castadi can do!"

[misty] curious, surpries, happy, question
"What do you think?"

How can I help?

I have a lot of ideas for where to take this, but not enough time. If you think you can help with any of these it would be greatly appriciated!

• Controlnet(currently experimental) - use controlnet with panel creation to have a more consistant page!
• Better ui and extension integration
• Fully ai generated content - meaning some sort of an integration with an llm to basically have prompt-to-manga!

How is it done?

Castadi reads a text file and a settings file, and uses the stable diffusion webui local api to generate pictures based on your these files.

Settings.json

This is a settings file castadi reads for configuration. Here is an example one.

{
  "canvas_width": 1080,
  "canvas_height": 1920,
  "panel_min_width_percent": 0.21,
  "panel_min_height_percent": 0.117,
  "image_zoom": "4/9",
  "border_width": 7,
  "default_bubble": {
    "font_size": 24,
    "bubble_color": "(20, 20, 20)",
    "text_color": "white",
    "font": "C:\\Windows\\Fonts\\CascadiaMonoPL-ExtraLight.ttf"
  },

  "prompt_prefix": "(masterpiece, best quality:1.1)",
  "negative_prompt": "(bad quality, low quality:1.1), easynegative",

  "characters": {
    "misty": {
      "tags": "misty \\(pokemon\\), <lora:Misty:1>, yellow shirt, crop top, suspender shorts"
    },
    "popo": {
      "tags": "male, guy, dark hair, white suit, black pants",
      "bubble": {
        "bubble_color": "(180, 180, 180)"
      }
    }
  }
}

Syntax

Let's go through the example above with some changes:

-Page1(min: (0.3, 0.3))

location: public park, yard, outside, park

[misty] weaving, saying hey, happy, smiling
"Heyy!!"

"This is an example"
"of what castadi can do!"

split: h

[misty] curious, surpries, happy, question
"What do you think?"

-Page1(min: (0.3, 0.3))

We start with a page notation: Page1(min: (0.3, 0.3)). The parametes are optional. Currently only one exists: min. It is used to determine the minimum panel size as a percentage of the page size.

Each page is comprised of panels, which are separated by an empty line.

location: public park, yard, outside, park

Before starting the first panel, we specify a location. Each panel can have a different location that would be appended to the end of the prompt for stable diffusion. If no location is specified for each panel we take the previous one.

[misty] weaving, saying hey, happy, smiling
"Heyy!!"

Each panel is comprised of events, these can be scene descriptions or dialog. They are separated by new lines. Meaning each line is an event.

We start with a scene description, we use the character notation [misty], which will look inside our settings.json for a character named misty and replace the notation with the character's prompt tags.

We went down a line to signal a new event. This time it's a dialog. Each dialog needs to be said by a character, but we don't specify one in this case because it's the last character we mentioned, which would be misty.

"This is another example"
[popo] "yep, it sure is!"

If we look at then next panel, there are 2 more dialogs. The first is of misty because we didn't specify another. The second one would be of popo. These 2 notations are the only dialog notations we have.

split: h

[popo] curious, surpries, question, apathetic
"Who are we talking to tho"

we then specify the split between the previous panel and the next (as a horizontal split). Unlike the location, this resets every panel. Meaning if we don't specify it, it would be randomly generated.

That's basically it! This is our output: