Anime Image Quality rating model for Datasets

Aesthetic Shadow is a sophisticated visual assessment tool with a whopping 1.1 billion parameters. Its primary purpose is to gauge the artistic quality of anime images. This means it can analyze high-resolution anime art. The model takes in a 1024x1024 resized picture, making it compatible with standard Stable Diffusion XL output resolution. The result of the model is a numerical score between 0 and 1 that measures just how visually captivating the artwork is.

Aesthetic Shadow was trained in recognizing small details, body proportions and assessing the overall quality of anime illustrations.

Refer to the Jupyter Notebook provided in the files on Huggingface for a simple to use Colab notebook that lets you run batch inference on your own Datasets.

It's important to note that the model may not always provide a completely accurate evaluation and does not represent the actual quality. The main application for this model is filtering out low-quality images from large-scale datasets like Danbooru.

You can find the model and demo here on Huggingface: https://huggingface.co/shadowlilac/aesthetic-shadow

If you have any questions, feel free to ask.