New to Claude Skills? Learn how to install them →

nousresearch on GitHub

CLIP

Free

Zero-shot image classification and image-text search.

Get this skill

Free · Opens the source repo

What CLIP does

CLIP, developed by OpenAI, is a powerful model designed for understanding images in relation to natural language. It excels in zero-shot image classification, meaning it can categorize images without needing training data specific to those categories. This capability is particularly useful for developers and designers looking to implement image recognition features without the overhead of dataset preparation and model training. CLIP can also perform image-text matching, enabling semantic searches where users can find images based on descriptive text queries.

The model has been trained on a vast dataset of 400 million image-text pairs, allowing it to generalize well across various contexts. This makes it suitable for applications such as content moderation, where it can detect inappropriate content, and visual question answering, where it can provide answers based on image content. The versatility of CLIP extends to cross-modal retrieval, allowing for seamless interaction between text and images, which is beneficial for creating more intuitive user experiences in applications.

Installation of CLIP is straightforward, requiring a few lines of Python code to set up the environment and load the model. Once operational, users can leverage its capabilities for tasks like semantic image search or batch processing of images, enhancing efficiency in workflows that involve large datasets. The model supports various architectures, including ResNet and Vision Transformer, allowing users to choose based on their specific performance needs.

While CLIP is a robust tool for many applications, it is important to note its limitations. It is not designed for fine-grained classification tasks and may not perform well with vague or poorly defined labels. Additionally, users should be aware of potential biases in the model due to its training data sourced from the web. Overall, CLIP is a valuable asset for developers and designers aiming to integrate advanced image understanding features into their projects.

When to use it

Use CLIP when you need to classify images without training data or perform image-text similarity searches.

When not to use it

CLIP is not suitable for tasks requiring fine-grained classification or when working with vague image descriptions.

What you can build with it

Image Classification without Training Data

Utilize CLIP to classify images into broad categories without the need for labeled training datasets, streamlining the deployment of image recognition features.

Content Moderation for Applications

Implement CLIP to automatically detect NSFW or violent content in user-uploaded images, enhancing the safety of your platform.

Semantic Search for Image Retrieval

Leverage CLIP's capabilities to enable users to search for images using natural language queries, improving the user experience in applications.

How to install CLIP

View source

1. Install with the skills CLI

npx skills add nousresearch/hermes-agent/clip --agent claude-code

2. Or install it manually

Download the skill folder and drop it into ~/.claude/skills/ for all projects, or .claude/skills/ to scope it to one repo. Restart Claude Code so it picks up the new skill.

Anthropic's agentic coding CLI, and the reference implementation of Agent Skills. Drop a skill folder into ~/.claude/skills and Claude Code loads it automatically whenever a task matches the skill's description. Claude Code docs

Inside SKILL.md

Written by nousresearch

CLIP - Contrastive Language-Image Pre-Training

OpenAI's model that understands images from natural language.

When to use CLIP

Use when:

  • Zero-shot image classification (no training data needed)
  • Image-text similarity/matching
  • Semantic image search
  • Content moderation (detect NSFW, violence)
  • Visual question answering
  • Cross-modal retrieval (image→text, text→image)

Metrics:

  • 25,300+ GitHub stars
  • Trained on 400M image-text pairs
  • Matches ResNet-50 on ImageNet (zero-shot)
  • MIT License

Use alternatives instead:

  • BLIP-2: Better captioning
  • LLaVA: Vision-language chat
  • Segment Anything: Image segmentation

Quick start

Installation

pip install git+https://github.com/openai/CLIP.git
pip install torch torchvision ftfy regex tqdm

Zero-shot classification

import torch
import clip
from PIL import Image

# Load model
device = "cuda" if torch.cuda.is_available() else "cpu"
model, preprocess = clip.load("ViT-B/32", device=device)

# Load image
image = preprocess(Image.open("photo.jpg")).unsqueeze(0).to(device)

# Define possible labels
text = clip.tokenize(["a dog", "a cat", "a bird", "a car"]).to(device)

# Compute similarity
with torch.no_grad():
    image_features = model.encode_image(image)
    text_features = model.encode_text(text)

    # Cosine similarity
    logits_per_image, logits_per_text = model(image, text)
    probs = logits_per_image.softmax(dim=-1).cpu().numpy()

# Print results
labels = ["a dog", "a cat", "a bird", "a car"]
for label, prob in zip(labels, probs[0]):
    print(f"{label}: {prob:.2%}")

Available models

# Models (sorted by size)
models = [
    "RN50",           # ResNet-50
    "RN101",          # ResNet-101
    "ViT-B/32",       # Vision Transformer (recommended)
    "ViT-B/16",       # Better quality, slower
    "ViT-L/14",       # Best quality, slowest
]

model, preprocess = clip.load("ViT-B/32")
ModelParametersSpeedQuality
RN50102MFastGood
ViT-B/32151MMediumBetter
ViT-L/14428MSlowBest

Image-text similarity

# Compute embeddings
image_features = model.encode_image(image)
text_features = model.encode_text(text)

# Normalize
image_features /= image_features.norm(dim=-1, keepdim=True)
text_features /= text_features.norm(dim=-1, keepdim=True)

# Cosine similarity
similarity = (image_features @ text_features.T).item()
print(f"Similarity: {similarity:.4f}")

Semantic image search

# Index images
image_paths = ["img1.jpg", "img2.jpg", "img3.jpg"]
image_embeddings = []

for img_path in image_paths:
    image = preprocess(Image.open(img_path)).unsqueeze(0).to(device)
    with torch.no_grad():
        embedding = model.encode_image(image)
        embedding /= embedding.norm(dim=-1, keepdim=True)
    image_embeddings.append(embedding)

image_embeddings = torch.cat(image_embeddings)

# Search with text query
query = "a sunset over the ocean"
text_input = clip.tokenize([query]).to(device)
with torch.no_grad():
    text_embedding = model.encode_text(text_input)
    text_embedding /= text_embedding.norm(dim=-1, keepdim=True)

# Find most similar images
similarities = (text_embedding @ image_embeddings.T).squeeze(0)
top_k = similarities.topk(3)

for idx, score in zip(top_k.indices, top_k.values):
    print(f"{image_paths[idx]}: {score:.3f}")

Content moderation

# Define categories
categories = [
    "safe for work",
    "not safe for work",
    "violent content",
    "graphic content"
]

text = clip.tokenize(categories).to(device)

# Check image
with torch.no_grad():
    logits_per_image, _ = model(image, text)
    probs = logits_per_image.softmax(dim=-1)

# Get classification
max_idx = probs.argmax().item()
max_prob = probs[0, max_idx].item()

print(f"Category: {categories[max_idx]} ({max_prob:.2%})")

Batch processing

# Process multiple images
images = [preprocess(Image.open(f"img{i}.jpg")) for i in range(10)]
images = torch.stack(images).to(device)

with torch.no_grad():
    image_features = model.encode_image(images)
    image_features /= image_features.norm(dim=-1, keepdim=True)

# Batch text
texts = ["a dog", "a cat", "a bird"]
text_tokens = clip.tokenize(texts).to(device)

with torch.no_grad():
    text_features = model.encode_text(text_tokens)
    text_features /= text_features.norm(dim=-1, keepdim=True)

# Similarity matrix (10 images × 3 texts)
similarities = image_features @ text_features.T
print(similarities.shape)  # (10, 3)

Integration with vector databases

# Store CLIP embeddings in Chroma/FAISS
import chromadb

client = chromadb.Client()
collection = client.create_collection("image_embeddings")

# Add image embeddings
for img_path, embedding in zip(image_paths, image_embeddings):
    collection.add(
        embeddings=[embedding.cpu().numpy().tolist()],
        metadatas=[{"path": img_path}],
        ids=[img_path]
    )

# Query with text
query = "a sunset"
text_embedding = model.encode_text(clip.tokenize([query]))
results = collection.query(
    query_embeddings=[text_embedding.cpu().numpy().tolist()],
    n_results=5
)

Best practices

  1. Use ViT-B/32 for most cases - Good balance
  2. Normalize embeddings - Required for cosine similarity
  3. Batch processing - More efficient
  4. Cache embeddings - Expensive to recompute
  5. Use descriptive labels - Better zero-shot performance
  6. GPU recommended - 10-50× faster
  7. Preprocess images - Use provided preprocess function

Performance

OperationCPUGPU (V100)
Image encoding~200ms~20ms
Text encoding~50ms~5ms
Similarity compute<1ms<1ms

Limitations

  1. Not for fine-grained tasks - Best for broad categories
  2. Requires descriptive text - Vague labels perform poorly
  3. Biased on web data - May have dataset biases
  4. No bounding boxes - Whole image only
  5. Limited spatial understanding - Position/counting weak

Resources

Frequently asked questions about CLIP

Similar skills