metadata
license: apache-2.0
new_version: nightknocker/contrastive-u-dae
Contrastive Booru-Image Embedding Model
The model was trained on masked pixiv images, tagged by CL tagger. The caption (prompt) was tokenized by ModernBERT.
Objectives
Image-Caption Similarity Scoring: The model assigns a score to each image based on how closely it is related to its associated caption.
Image Labeling: The image embedding is used to label images by character name.
Updates
This version contains up to 4200 characters extracted from public and private datasets.