Transform Any Image into a Masterpiece
Unlock the beauty of expressive brushstrokes, vibrant colors, and rich textures with Digital Impressions. This LoRA reimagines any image in the style of classic Impressionism, merging timeless artistry with the versatility of digital media. Whether you’re working with portraits, landscapes, or abstract concepts, Digital Impressions brings your creations to life with a true masterpiece's lush color and textured brushwork.
Digital Impressions adds depth, atmosphere, and a painterly quality to each image, reminiscent of the works of the Impressionist masters. It’s perfect for artists, storytellers, and fantasy creators who seek to infuse their visuals with the charm and emotional energy of traditional oil painting, all with a modern, digital twist.
Usage
To use the most recent version of the LoRA, use the following settings:
Trigger word: dibrshstrk; however, using a trigger may be unnecessary. Try it both ways, and let me know what works best.
Other tokens that work well: No tokens are needed to trigger the style and transform most images, but you could add words like "brushstroke" and "impressionist style" to get Flux moving in the right direction.
Lora Strength: A strength between 0.8 and 1.2 is recommended for Flux prompts written in natural language, but pull back to 0.6 to 1.0 for Danbooru tag Flux prompts that usually name more anime-style images.
For differences in previous versions, see the version notes below.
Version Notes
Version 2.0 is not necessarily an improvement over Version 1.0, but it does differ in several ways. So, try them both out; each could have its uses depending on your personal preferences and goals.
Version 2.0
Version 2.0 was trained on the same dataset and with the same settings, but the dataset was captioned first but with captions that followed the guidelines in the dataset preparation guide. Adding captions resulted in a LoRA that lost some of its unique styles but showed better prompt adherence and generalizability, as well as subjects with a more varied array of characteristics.
Version 1.0
Version 1.0 was trained on several paintings with heavy, visible brushstrokes. The training images had no captions other than the trigger word. This allowed the style to be emphasized, but also a bit of overfitting on the subjects and composition. The style was captured using a selection of parameters to highlight the art style. Once I do more testing and ensure I can fix a few mistakes in Version 1.0, I'll write up the new technique so others can add it to their LoRA training toolkit.