An open API service indexing awesome lists of open source software.

https://github.com/stablecanvas/sd-webui-a1111-prompt-parser

sd a1111 prompt parser for javascript.
https://github.com/stablecanvas/sd-webui-a1111-prompt-parser

a1111-stable-diffusion-webui a1111-webui parser prompt sd sdxl stable-diffusion stable-diffusion-webui stablediffusion

Last synced: about 1 month ago
JSON representation

sd a1111 prompt parser for javascript.

Awesome Lists containing this project

README

          

# 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

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

const parser = new PromptParser();

const prompt = `masterpiece, 1girl, blonde hair, , (chromatic aberration:0.7), sharp focus, hyper detailed, (fog:0.7), , [real photo], [highlight:dark:0.9], (((good anatomy)))`;

const output = parser.parse(prompt);

console.log(output);
```

Output:

```json
[
{ "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.

```typescript
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

```typescript
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

```typescript
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

```typescript
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

```typescript
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

```typescript
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