Make new photos from your existing ones
This is a demo of using Dreambooth & Stable Diffusion to generate images using your own photos. The demo runs on Google Colab's free tier, on machines with T4 GPUs (thanks, Google!)
- Open the Google Colab notebook from here (part of this very repo)
- Follow instructions at the top of the notebook to prepare your photos
- Run the Colab step by step, and in the end play around with various prompts to create new images
- Detects & extracts face info from all the given photos, using the excellent Deepface library
- Generates embedding vector using DeepFace for each face
- Groups similar faces together using DBSCAN algorithm from the scikit-learn library. This identifies a set of unique people in the photos
- You select a person/group of photos and Dreambooth is fine-tuned with those photos
- You play around with various prompts to create new images
Steps 4 & 5 above reuse a lot from ShriramShrirao's excellent Dreambooth tutorial. Thank you!