Regularization images for Stable Diffusion v1.5
This dataset includes images generated automatically from various prompts (people, animals, objects, styles) in order to train DreamBooth models with regularization.
Approximately 60,000 512x512 PNG images in ZIP, 1.35 GB in Parquet format
MIT
Description
stablediffusion-v1-5-regularization-images is a massive collection of synthetic images generated with Stable Diffusion, structured for training with the loss of prior preservation of DreamBooth. The files are classified by theme (e.g. “guy”, “supermodel”, “supermodel”, “kitty”, “cyberpunk”) and organized in ZIP files containing images in 512x512 PNG format.
What is this dataset for?
- Train DreamBooth models with regularization to avoid overfitting
- Strengthen the robustness of models on common visual classes
- Test build performance on various styles or objects
Can it be enriched or improved?
Yes, it is possible to complete the dataset with other prompts or specific styles. Additional annotations by categories or quality levels can also be integrated for advanced uses (filtering, scoring, classification).
🔎 In summary
🧠 Recommended for
- DreamBooth model creators
- Broadcast researchers
- Custom avatar projects
🔧 Compatible tools
- Diffusers
- Dreambooth-Stable-Diffusion
- Fast-Stable-Diffusion
- AUTOMATIC1111
💡 Tip
For varied results, cross several sub-datasets by theme and style (e.g. people + landscapes + cyberpunk).
Frequently Asked Questions
Does this dataset contain real or generated data?
All images are synthetic, generated by Stable Diffusion with controlled parameters, making them ideal for training.
Can I use this dataset to create a custom avatar with DreamBooth?
Yes, these images are precisely used to regularize the model to prevent it from over-learning your avatar during training.
Is this dataset suitable for beginners in generative AI?
Yes, it's easy to use with tools like AUTOMATIC1111 or Diffusers, even without advanced technical expertise.
