Converting between different art styles using unsampler techniques

Other Questions workflow
1

I’m exploring ways to change images from one artistic style to another using an unsampler workflow. My focus is on transforming cartoon-style images into realistic ones, and vice versa.

I’ve tried various methods, but the results aren’t what I hoped for. The images often lose crucial details or fail to keep the original layout intact.

Has anyone managed to do effective style transfers with unsampler techniques? What settings work best for preserving image quality during the conversion? I’m especially looking for methods that can accurately represent facial features and intricate backgrounds without creating any unwanted effects.

Any advice on setting up workflows or specific configurations would be greatly appreciated. I’m eager to try different methods as long as they support effective two-way style conversion.

2

I’ve been experimenting with unsampler techniques for style transfer, and I found that preprocessing plays a crucial role. When converting cartoon images to realistic ones, I usually set the denoising strength between 0.6 and 0.75. Conversely, when going from realistic to cartoon, lowering it to around 0.4 to 0.5 yields better results. Additionally, using ControlNet with Canny edge detection helps maintain the structure, while adding depth estimation for complex backgrounds significantly enhances the quality of facial features and intricate details.