Lab 2 Task 2-2 Pikmin DreamBooth LoRA (sks fairy)

This LoRA was trained for NYCU Image and Video Generation Lab 2 Task 2-2 on 14 self-photographed images of one red Pikmin toy.

Training configuration

Base model: CompVis/stable-diffusion-v1-4
Instance prompt: a photo of a sks fairy
Validation prompt: A photo of a sks fairy in a bucket
Resolution: 512
Batch size: 1 per GPU
Gradient accumulation steps: 1
Maximum training steps: 500
Learning rate: 1e-4 with constant scheduler
Mixed precision: no
Seed: 0

Usage

Load pytorch_lora_weights.safetensors with a Diffusers StableDiffusionPipeline and include sks fairy in the prompt, for example a photo of a sks fairy on the beach.

SHA-256 of pytorch_lora_weights.safetensors: abc963aecf3c2d2d459d98ad092fa6f8b14ee1e3e0ec1a22f5115b3eb795c0be

An earlier version trained with a photo of a sks toy is kept in this repository's commit history for comparison.

This repository is provided for coursework and evaluation.

Training images

The training_images/ directory contains the 14 self-photographed images used for DreamBooth LoRA training. The uploaded copies have been re-encoded without EXIF metadata, including GPS coordinates, timestamps, and camera information.

Downloads last month: 35

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for Yue-2003/lab2-task2-2-pikmin-lora

Base model

CompVis/stable-diffusion-v1-4

Adapter

(617)

this model