sd-webui-a1111-prompt-parser

Introduction

sd-webui-a1111-prompt-parser is a Stable Diffusion webUI (A1111) prompt parser for JavaScript. It parses Stable Diffusion model prompts into structured data for easy analysis and manipulation by developers.

Features

• Parses A1111 format prompts, supporting the following syntax:
  • Plain text
  • Emphasis (parentheses)
  • Weight (brackets)
  • Lora models
  • Hypernetwork models
  • Negative prompts (square brackets)
  • Step Control (scheduling)
• Converts parsed results into JavaScript objects for easy manipulation and use
• Supports regenerating A1111 format prompts from JavaScript objects

Installation

pnpm add @stable-canvas/sd-webui-a1111-prompt-parser

Usage

import { PromptParser } from "@stable-canvas/sd-webui-a1111-prompt-parser";

const parser = new PromptParser();

const prompt = `masterpiece, 1girl, blonde hair, <lora:Zelda_v1:0.5>, (chromatic aberration:0.7), sharp focus, hyper detailed, (fog:0.7), <hypernet:sxz-bloom:0.5>, [real photo], [highlight:dark:0.9], (((good anatomy)))`;

const output = parser.parse(prompt);

console.log(output);

Output:

[
  { "type": "plain", "value": "masterpiece" },
  { "type": "plain", "value": "1girl" },
  { "type": "plain", "value": "blonde hair" },
  { "type": "extra_networks", "value": "lora", "args": ["Zelda_v1", "0.5"] },
  {
    "type": "weighted",
    "value": 0.7,
    "args": [{ "type": "plain", "value": "chromatic aberration" }]
  },
  { "type": "plain", "value": "sharp focus" },
  { "type": "plain", "value": "hyper detailed" },
  {
    "type": "weighted",
    "value": 0.7,
    "args": [{ "type": "plain", "value": "fog" }]
  },
  {
    "type": "extra_networks",
    "value": "hypernet",
    "args": ["sxz-bloom", "0.5"]
  },
  {
    "type": "negative",
    "value": 1,
    "args": [{ "type": "plain", "value": "real photo" }]
  },
  {
    "type": "scheduled_full",
    "value": 0.9,
    "args": [
      [{ "type": "plain", "value": "highlight" }],
      [{ "type": "plain", "value": "dark" }]
    ]
  },
  {
    "type": "positive",
    "value": 3,
    "args": [{ "type": "plain", "value": "good anatomy" }]
  }
]

Usage Case: de-SD-styled (remove stable-diffusion styled prompt)

This function is used to clear all SD-WebUI style prompts, which is necessary when training new models based on DiT.

Theoretically, it should iterate through the tokens (this is a high level AST structure), but generally, no one writes prompt control syntax beyond two levels.

import {
  PromptParser,
  generation_str,
} from "@stable-canvas/sd-webui-a1111-prompt-parser";

const parser = new PromptParser();

function cleanPrompt(prompt: string): string {
  const tokens = parser
    .parse(prompt)
    .filter((x) => x.type !== "extra_networks")
    .flatMap((x) => {
      switch (x.type) {
        case "weighted":
        case "negative":
        case "positive":
        case "scheduled_full":
          return x.args;
        default:
          return x;
      }
    });

  return generation_str(tokens).trim();
}

API

PromptParser Class

Constructor

new PromptParser(options?: SDPromptParser.ILarkOptions);

• options: Optional parameters for configuring the Lark parser. For detailed parameter descriptions, refer to the [Lark documentation](https://lark-parser.readthedocs.io/en/latest/classes/ Lark.html#lark.Lark).

parse Method

parse(text: string, options?: ParseOptions): SDPromptParser.PromptNode[];

• text: The prompt string to be parsed.
• options: Optional parameters for configuring parsing behavior.
  • force: When set to true, parsing is forced even if there are syntax errors in the prompt. The parsing results may be incomplete. The default is false.
• Returns: An array of parsed prompt nodes, node type definitions refer to SDPromptParser.PromptNode.

compilation Function

compilation(node: SDPromptParser.IPromptASTNode): SDPromptParser.PromptNode[];

• node: The root node of the abstract syntax tree (AST) generated by the Lark parser.
• Returns: An array of parsed prompt nodes, node type definitions refer to SDPromptParser.PromptNode.

generation_token Function

generation_token(nodes: SDPromptParser.PromptNode[], options?: GenerationOptions): string[];

• nodes: An array of nodes for which to generate prompt strings.
• options: Optional parameters for configuring generation behavior.
  • remove_1_weighted: When set to true, nodes with a weight of 1 are removed. The default is false.
• Returns: An array of generated prompt strings.

generation_str Function

generation_str(nodes: SDPromptParser.PromptNode[], options?: GenerationOptions): string;

• nodes: An array of nodes for which to generate the prompt string.
• options: Optional parameters for configuring generation behavior.
  • remove_1_weighted: When set to true, nodes with a weight of 1 are removed. The default is false.
• Returns: The generated prompt string.

Build

1. Build the Parser

1.1 Install Lark.js

pip install lark-js

1.2 Build

pnpm build-lark

2. Build the Package

pnpm build

Test

pnpm test

License

MIT