A simple notebook that launches Stable Diffusion 2.0 Depth (depth-to-image) demonstration via Gradio UI
This stable-diffusion-2-depth model is resumed from stable-diffusion-2-base (512-base-ema.ckpt) and finetuned for 200k steps. Added an extra input channel to process the (relative) depth prediction produced by MiDaS (dpt_hybrid) which is used as an additional conditioning.
This notebook uses Gradio's share feature to serve the UI via a public URL, through a tunnel over HTTPs. To add security, this notebook forces you to set a user/password combo to access the UI