Stable Diffusion Subject Tuning
Teaching a diffusion model a subject and then using it in new settings.
Fine tuned Stable Diffusion with DreamBooth and LoRA on custom people and objects, then connected the result to the Automatic1111 API for text to image and image to image generation.
The practical question was not only whether an image could look good once. It was whether a subject could remain recognizable while the style, scene, and prompt changed.
The experiments
The work covered subject preparation, training, prompt design, and generation across several conditions. Metadata and prompt parsing utilities made it possible to compare outputs instead of losing the settings behind each image.
The application layer
The Automatic1111 API turned the experiments into a repeatable generation flow that another script or interface could call.
selected tools
- Python
- Stable Diffusion
- DreamBooth
- LoRA
- diffusers