An aesthetic predictor that takes anime images and gives it a score out of 10. Ideally you take this model and fine-tune it to your own preferences. There's a simple gradio demo and full training script here: https://github.com/CoffeeVampir3/easy-aesthetic-predictor
This model is based on ConVneXt: https://github.com/facebookresearch/ConvNeXt
This model was trained for 3 epochs at a learning rate of 1e-6 and then 9 epochs at 1e-5, training took about 6 hours on a 3090 TI with a dataset of 20,000 labeled images.
Inference Providers
NEW
This model is not currently available via any of the supported third-party Inference Providers, and
HF Inference API was unable to determine this model's library.