clip

Zero-shot image classification categorizes images using natural language descriptions without prior training. This Skill leverages OpenAI's CLIP model to compare image and text embeddings via cosine similarity, matching visual content to descriptive text queries directly.

Semantic image search finds images based on descriptive text queries. This Skill computes cross-modal embeddings using the CLIP model and ranks images by their cosine similarity to the input text, enabling retrieval for collections like product photos.

No fine-tuning is required to classify images into custom categories. The Skill utilizes the Contrastive Language-Image Pre-training model to perform zero-shot classification, comparing images directly against your provided text labels.

Image-text matching for content moderation measures semantic similarity between visual content and safety descriptions. The Skill evaluates image and text embeddings using cosine similarity to flag inappropriate or violent images without task-specific training.

Running the CLIP model for vision-language tasks requires the transformers, torch, and pillow dependencies. These libraries handle the model inference, tensor operations, and image processing necessary to compute cross-modal embeddings.