Gustave Doré LORA for SDXL

Doré LoRa Model
Gustave Doré
Model Type LoRa (Low-Rank Adaptation)
Base Model Stable Diffusion XL
Links Download Model on HuggingFace

This article is about the LORA model trained on Gustave Doré's artwork. For the artist, see Gustave Doré.

The Gustave Doré LoRa is a fine-tuned adaptation for Stable Diffusion XL that was trained by Gabe. It captures the distinctive artistic style of 19th-century illustrator Gustave Doré. This project emerged from my fascination with the masterful wood engravings created by Dore. If you haven't seen his work, I highly recommend checking it out.

After collecting an extremely high quality dataset of Doré's illustrations, I set out to create a LoRa model that could capture his artistic style. I broke the dataset of high resolution images down into about 4000 1024x1024 images. I then used BLIP-2 to generate labels for the images. OneTrainer made it extremely simple to train, and it only took about 8 hours to train this LoRa locally on my GPU. The results have exceeded my expectations, successfully capturing many of the characteristics of Doré's style

▼ Style Comparison

The following images demonstrate the effectiveness of the LORA model. Both images were generated using the prompt "olympic snowboard trick, in the style of gustave dore":

Without LoRa


Generated without LORA

With Doré LoRa


Generated with LORA

▼ Usage

To use this LORA model with Stable Diffusion XL, follow these steps:

Setup

  1. Download the LORA model from HuggingFace
  2. Place the downloaded .safetensors file in your models/lora/ folder
  3. Ensure that you also have SDXL 1.0 installed
  4. Load the LoRa
  5. Use the term "in the style of Gustave Dore" to activate the LoRa model in your prompt

Recommended Settings

▼ Gallery