DeepFakeDetector/manual-gen-images

Manual Prompt-Aligned AI Image Dataset

Description

This dataset contains manually generated AI images created from a fixed, predefined set of prompts. Each image is directly aligned to its corresponding prompt file and generated across multiple image generation models.

Prompt Alignment

Each image is generated directly from a corresponding prompt file found in ./prompts.

Directory Structure

images/
├── dalle/
│   ├── p001.png
│   ├── p002.png
│   └── ...
├── midjourney/
│   ├── p001.png
│   └── ...
├── bing/
│   ├── p001.jpg
│   └── ...
├── stable_diffusion/
│   ├── p001.png
│   └── ...
├── ideogram/
│   ├── p001.png
│   └── ...
└── nanobanana/
    ├── p001.png
    └── ...
prompts/
├── p001.txt
└── ...

Generators Used

• DALL-E
• Midjourney (In Progress)
• Bing (MAI-Image-1 Model)
• Stable Diffusion (In Progress)
• Ideogram (In Progress)
• FLUX (skipped due to required paid credits)
• Nano Banana

Metadata Inference

Metadata is derived implicitly from file paths:

• generator → parent directory name under images/
• prompt_id → filename (e.g., p042)
• prompt_text → contents of prompts/p042.txt

Example derived record:

{
  "image_path": ".../images/dalle/p042.png", # note that all images in bing folder are jpg
  "generator": "dalle",
  "prompt_id": "p042",
  "prompt_text": "...contents of prompts/p042.txt..."
}

Limitations

This dataset was manually generated and is therefore limited in scale and generator coverage.