clip

Map images and text to shared embeddings for zero-shot classification and retrieval.

Updated Apr 19, 2026
One-click install
npx skills add https://github.com/gqf2008/hermez-ai --skill clip-gqf2008
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: clip
Source: https://github.com/gqf2008/hermez-ai/tree/main/skills/mlops/models/clip
Command: npx skills add https://github.com/gqf2008/hermez-ai --skill clip-gqf2008

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

OpenAI's CLIP model enables vision-language understanding by mapping images and text to a shared embedding space, allowing zero-shot image classification, image-text matching, and cross-modal retrieval without task-specific fine-tuning.

Core Features & Use Cases

  • Zero-shot image classification: classify images against arbitrary labels without labeled data.
  • Image-text similarity & retrieval: compute cross-modal embeddings for matching images and captions.
  • Content moderation, search & VQA: enable image search, safety moderation, and visual question answering workflows.

Quick Start

Load CLIP (ViT-B/32) and compare an image against a small set of text labels to obtain the top match.

Frequently Asked Questions about clip

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I perform zero-shot image classification without labeled data?

Zero-shot image classification maps images and text labels to a shared embedding space using a pretrained CLIP model, allowing you to classify images against arbitrary text labels without task-specific fine-tuning.

What is cross-modal retrieval and how does it work for image search?

Cross-modal retrieval computes embeddings for both images and text within a shared space, enabling image-text similarity matching so you can search and retrieve relevant images using natural language queries.

Can I use CLIP for content moderation and visual question answering?

Yes, CLIP supports content moderation and visual question answering workflows by leveraging its pretrained vision-language understanding to match visual data against safety or query text prompts without fine-tuning.

Do I need a pretrained model to perform image-text similarity matching?

Yes, performing image-text similarity matching requires a pretrained CLIP model like ViT-B/32, along with image preprocessing and text tokenization, to accurately map inputs into a shared embedding space.

When should I avoid using zero-shot classification for vision-language tasks?

Avoid zero-shot classification when your task requires highly specialized domain accuracy, as CLIP relies on generalized pretraining and skips task-specific fine-tuning, potentially limiting performance on niche visual datasets.