🧨 Dreambooth-Keras + WandB 🪄🐝

This notebook shows how to perform inference with a DreamBooth fine-tuned Stable Diffusion model.

🌈 Install Dreambooth-Keras

We would use soumik12345/dreambooth-keras which is a fork of sayakpaul/dreambooth-keras developed by Sayak Paul and Chansung Park.

🐝 Initialize WandB run

We initialize a Weights & Biases run for storing generated images to a Weights & Biases table.

🧑‍🎨 Perform Inference

First we load our model from Weights & Biases artifacts created using the dreambooth_keras.utils.DreamBoothCheckpointCallback which automatically logs model checkpoints as Weights & Biases artifacts at the end of each epoch during training. We load these checkpoint using the simple utility dreambooth_keras.utils.load_model_from_wandb_artifact.

Now, we perform inference on our dreamboothed stable-diffusion model.

Next we log our images to a Weights & Biases table that not only makes ut easier to visualize but also easily accessible for future reference.