image-search

Use when user wants to build image search or similar image finding. Triggers on: image search, similar image, visual search, image retrieval, CLIP, reverse image search, image matching, find similar photos.

Ir para a instalação

Repositório: zilliztech/milvus-marketplace
Última atividade na origem: 4 de fevereiro de 2026 às 02:42
Idioma detectado do SKILL.md: inglês
Estrelas: 3
Forks: 2

Opções de instalação

Por padrão, está selecionado o prompt que primeiro revisa a origem. Você pode mudar para um comando direto ou baixar uma cópia local.

Instalar com Codex ou Claude Copie este prompt, cole no Codex, Claude ou outro assistente e deixe que ele revise a página da skill e instale para você.

Um comando direto ignora o prompt de revisão. Verifique a origem antes de executá-lo.

npx skills add https://github.com/zilliztech/milvus-marketplace --skill image-search

O comando permanece em uma só linha. Role horizontalmente para revisá-lo antes de copiar.

Prefere uma cópia local? Baixe os arquivos disponíveis atualmente no SkillsMP.

Revise os arquivos de origem

Leia o SKILL.md e os arquivos complementares exibidos pelo SkillsMP antes de decidir se vai instalar.

Explorador de arquivos

5 arquivos

SKILL.md

Instruções da origem · Visualização somente leitura

Voltar à pesquisa Voltar à instalação

name	image-search
description	Use when user wants to build image search or similar image finding. Triggers on: image search, similar image, visual search, image retrieval, CLIP, reverse image search, image matching, find similar photos.

Image Search

Build image-to-image search systems that find visually similar images using deep learning embeddings.

When to Activate

Activate this skill when:

User wants to find similar images given an input image
User mentions "reverse image search", "visual similarity", "find similar"
User has an image database to search through
User wants to build product visual search, face matching, or duplicate detection

Do NOT activate when:

User wants to search images using text → use text-to-image-search
User has mixed text and image documents → use multimodal-rag
User needs video content search → use video-search

Interactive Flow

Step 1: Understand the Search Type

"What kind of image search do you need?"

A) Image-to-Image: Upload image, find similar ones

Product visual search (find similar products)
Duplicate/near-duplicate detection
Face recognition

B) Text-to-Image: Describe what you're looking for

Stock photo search
Descriptive queries ("red car on highway")

C) Both: Support both search modalities

E-commerce (upload photo OR describe)
Content management systems

Which do you need? (A/B/C)

Step 2: Determine Image Types

┌─────────────────────────────────────────────────────────┐ │ Image Search │ │ │ │ Query Image: [🐱 cat photo] │ │ │ │ │ ▼ │ │ ┌───────────────┐ │ │ │ CLIP │ Extract visual features │ │ │ Encoder │ (colors, shapes, objects) │ │ └───────┬───────┘ │ │ │ │ │ ▼ │ │ [0.23, -0.45, 0.12, ...] │ │ (512-dimensional vector) │ │ │ │ │ ▼ │ │ ┌────────────────────────┐ │ │ │ Vector Index │ Find similar vectors │ │ │ (1M image vectors) │ in milliseconds │ │ └────────────┬───────────┘ │ │ │ │ │ ▼ │ │ Results: [🐱] [🐱] [🐈] [🐕] │ │ (ranked by visual similarity) │ └─────────────────────────────────────────────────────────┘

from pymilvus import MilvusClient, DataType from sentence_transformers import SentenceTransformer from PIL import Image class ImageSearch: def __init__(self, uri: str = "./milvus.db"): self.client = MilvusClient(uri=uri) self.model = SentenceTransformer('clip-ViT-B-32') self.dim = 512 self.collection_name = "image_search" self._init_collection() def _init_collection(self): if self.client.has_collection(self.collection_name): return schema = self.client.create_schema() schema.add_field("id", DataType.INT64, is_primary=True, auto_id=True) schema.add_field("image_path", DataType.VARCHAR, max_length=512) schema.add_field("embedding", DataType.FLOAT_VECTOR, dim=self.dim) index_params = self.client.prepare_index_params() index_params.add_index("embedding", index_type="AUTOINDEX", metric_type="COSINE") self.client.create_collection( collection_name=self.collection_name, schema=schema, index_params=index_params ) def add_images(self, image_paths: list): """Add images to index""" images = [Image.open(p).convert('RGB') for p in image_paths] embeddings = self.model.encode(images).tolist() data = [{"image_path": path, "embedding": emb} for path, emb in zip(image_paths, embeddings)] self.client.insert(collection_name=self.collection_name, data=data) def search_by_image(self, image_path: str, limit: int = 10): """Image-to-image search""" image = Image.open(image_path).convert('RGB') embedding = self.model.encode(image).tolist() results = self.client.search( collection_name=self.collection_name, data=[embedding], limit=limit, output_fields=["image_path"] ) return [{"path": hit["entity"]["image_path"], "score": hit["distance"]} for hit in results[0]] def search_by_text(self, text: str, limit: int = 10): """Text-to-image search (CLIP enables this!)""" embedding = self.model.encode(text).tolist() results = self.client.search( collection_name=self.collection_name, data=[embedding], limit=limit, output_fields=["image_path"] ) return [{"path": hit["entity"]["image_path"], "score": hit["distance"]} for hit in results[0]] # Usage search = ImageSearch() # Index images search.add_images(["cat1.jpg", "cat2.jpg", "dog1.jpg"]) # Search by image results = search.search_by_image("query.jpg") # Search by text results = search.search_by_text("a fluffy orange cat")

Type	Characteristics	Model Recommendation
General photos	Diverse subjects	CLIP ViT-B-32
Product images	Clean backgrounds	CLIP ViT-L-14
Faces	Portrait photos	FaceNet or CLIP
Domain-specific	Medical, satellite, etc.	Domain fine-tuned

Need	Best Solution
Find visually similar images	✅ Image Search (this skill)
Search images by description	`text-to-image-search`
Q&A on documents with images	`multimodal-rag`
Search video content	`video-search`

Model	Dimensions	Speed	Quality	Best For
clip-ViT-B-32	512	★★★★★	★★★	General, fast
clip-ViT-L-14	768	★★★	★★★★★	High accuracy
clip-ViT-B-16	512	★★★★	★★★★	Balanced
chinese-clip	512	★★★★	★★★★	Chinese text queries
SigLIP	768	★★★	★★★★★	Latest, best quality

Need	Upgrade To
Better text search results	`text-to-image-search` with VLM
Filter by metadata (category, date)	Add `filtered-search` pattern
Q&A on image content	`multimodal-rag`
Video content search	`video-search`

image-search

Opções de instalação

Revise os arquivos de origem

Image Search

When to Activate

Interactive Flow

Step 1: Understand the Search Type

Step 2: Determine Image Types

Image Search

When to Activate

Interactive Flow

Step 1: Understand the Search Type

Step 2: Determine Image Types

Step 3: Confirm Configuration

Core Concepts

Mental Model: Visual Fingerprint

How CLIP Works

Why Image Search Over Alternatives

Implementation

Model Selection Guide

Using Chinese-CLIP

Image Preprocessing

Common Pitfalls

❌ Pitfall 1: Not Converting Image Mode

❌ Pitfall 2: Memory Issues with Large Images

❌ Pitfall 3: Expecting Text-Quality Results

❌ Pitfall 4: Searching with Very Different Images

When to Level Up

References

image-search

Informações da origem

Opções de instalação

Revise os arquivos de origem

Mais deste repositório

Mais deste repositório

Image Search

When to Activate

Interactive Flow

Step 1: Understand the Search Type

Step 2: Determine Image Types

Image Search

When to Activate

Interactive Flow

Step 1: Understand the Search Type

Step 2: Determine Image Types

Step 3: Confirm Configuration

Core Concepts

Mental Model: Visual Fingerprint

How CLIP Works

Why Image Search Over Alternatives

Implementation

Model Selection Guide

Using Chinese-CLIP

Image Preprocessing

Common Pitfalls

❌ Pitfall 1: Not Converting Image Mode

❌ Pitfall 2: Memory Issues with Large Images

❌ Pitfall 3: Expecting Text-Quality Results

❌ Pitfall 4: Searching with Very Different Images

When to Level Up

References