---
license: other
base_model: stable-diffusion-v1-5/stable-diffusion-v1-5
library_name: diffusers
tags:
  - stable-diffusion
  - stable-diffusion-1-5
  - lora
  - diffusers
  - hearthstone
  - card-art
pipeline_tag: text-to-image
---

# Hearthstone SD 1.5 LoRA

This repository contains a Stable Diffusion 1.5 LoRA adapter trained for Hearthstone-style card artwork generation as part of the COMP 646 HearthGen project.

## Files

- `pytorch_lora_weights.safetensors`: Diffusers-compatible LoRA weights.

## Intended Use

This model is intended for research and course-project reproduction. It is used by the HearthGen pipeline to compare:

- Stable Diffusion 1.5 text-only generation
- Stable Diffusion 1.5 with reference conditioning
- Stable Diffusion 1.5 + Hearthstone LoRA text-only generation
- Stable Diffusion 1.5 + Hearthstone LoRA with KG-retrieved reference conditioning

## Loading With Diffusers

```python
import torch
from diffusers import StableDiffusionPipeline

pipe = StableDiffusionPipeline.from_pretrained(
    "stable-diffusion-v1-5/stable-diffusion-v1-5",
    torch_dtype=torch.float16,
).to("cuda")

pipe.load_lora_weights("comp646/hearthstone-sd15-lora")

image = pipe(
    "hsart Hearthstone card art, Warrior minion, iron armor, glowing embers",
    num_inference_steps=30,
    guidance_scale=7.5,
).images[0]
```

## Dataset

The LoRA was trained from the project artwork dataset:

- `comp646/hearthstone-art-512`

The dataset contains Hearthstone artwork crops and metadata for project reproduction. Artwork and Hearthstone IP belong to Blizzard Entertainment; this model is distributed only for academic project use.

## Project

Repository:

- <https://github.com/TraceOnSnow/HearthstoneCardGenerator>