Motivation:
SDXL is a great model, especially for fine-tuning, but fine-tuning on SDXL's standard size is expensive, so i made a model which can generate images of 256x256px size.
It allows faster and cheaper training experimentation and can improve a creative process through much faster inference time.
Huuuuge thanks to Microsoft for providing GPU for training ❤️
Trained on LAION Improved Aesthetics 6plus.
Want to collaborate on open source together with me? Send me a message:
kopyloleh@gmail.com / telegram @kopyl
