fcy322/314554031-lab2-lora

NYCU Lab 2 — 314554031

Two trained UNet LoRA adapters for the course assignment. The base model is CompVis/stable-diffusion-v1-4. These are adapter weights, not a full base model.

• style/: four provided artistic-custom images, 100 epochs / 400 updates.
• identity/: four provided dreambooth-cat images, 500 updates.

Both use rank 4, learning rate 1e-4, resolution 512, and fp16 mixed precision. See training_config.json for exact settings and dataset_manifest.json for the supplied dataset paths, captions, and hashes.

Starter code and dataset source: https://github.com/Jing-En-Huang/Lab2-DDIM-LoRA

Download for grading

Download the corresponding 314554031_task2-1_lora.zip or 314554031_task2-2_lora.zip. Unzip, then use:

pipe.load_lora_weights("path/to/unzipped/folder")

With the Hub, load a specific subfolder:

pipe.load_lora_weights("fcy322/314554031-lab2-lora", subfolder="style")
# or subfolder="identity"

The instance prompt for the identity adapter is a photo of a sks cat.