suriyagunasekar/midjourney-text-prompts

# Midjourney Text Prompts Dataset

This dataset contains extracted and cleaned Midjourney text prompts from Discord message JSON files.

## Dataset Description

- **Source**: Midjourney Discord server message data
- **Task**: Text-to-image prompt extraction
- **Message Types**: Filtered for INITIAL_OR_VARIATION (type 0) and UPSCALE (type 19) messages

## Data Processing

1. Extracted prompts from messages with double-star formatting (`**prompt**`)
2. Removed embedded image URLs (e.g., `https://s.mj.run/xxx`)
3. Removed duplicate and invalid prompts
4. Split into train/validation/test sets

## Dataset Structure

- `train.jsonl`: 53,610 prompts (80%)
- `validation.jsonl`: 6,701 prompts (10%)
- `test.jsonl`: 6,702 prompts (10%)

## Data Format

Each line is a JSON object with a `prompt` field:
```json
{"prompt": "a beautiful sunset over the ocean, photorealistic, dramatic lighting --ar 16:9"}
```

## Use Cases

This dataset is suitable for:
- Language model fine-tuning on text-to-image prompts
- Prompt generation training
- AI art prompt analysis

## Statistics

- Total unique prompts: 67,013
- Average prompt length: ~119 characters
- Median prompt length: 86 characters