ReX_Upscale (Hugging Face link) is a fine-tuned super-resolution model based on the Real-ESRGAN architecture. It was trained on the high-quality DIV2K dataset, which includes 800 high-resolution training images and their corresponding low-resolution counterparts, along with 100 validation images.

The training was conducted on an NVIDIA Tesla T4 GPU. The training epoch was not able to be completed as my free usage limited. Although the training was not fully completed, the model achieved moderately good output quality. A more refined version of ReX_Upscale, trained with additional epochs and an expanded dataset, will be released in the future for improved performance

the training was done on some anime and cartoon type simple images as the training was too much harsh thus realistic images may not perform as well as others ,if anyone wants i can train a new model that is fine tuned on real images with lots of data