Skip to content

PyPI Downloads PyPI

Tagging media with Langchain

If you're working with images & texts you can use the multimodal LLMs supported by Langchain.

Prerequisites

  • A concrete Langchain library installed and configured

Supported media

  • TEXT
  • IMAGE

Installation

pip install media-tagging-langchain

uv pip install media-tagging-langchain

If you want to use server capabilities of media-tagger you need to install additional dependencies:

pip install media-tagging[server]

Usage

media-tagger tag MEDIA_PATHs \
  --media-type IMAGE \
  --tagger langchain \
  --tagger.llm_class_name=<FULLY_QUALIFIED_CLASS_NAME>

Where

  • <FULLY_QUALIFIED_CLASS_NAME> - fully path to the LLM class (i.e. langchain_google_genai.ChatGoogleGenerativeAI). Corresponding library should be installed before calling the media-tagger.
import media_tagging

media_tagger = media_tagging.MediaTaggingService()

request = media_tagging.MediaTaggingRequest(
  media_type='IMAGE',
  media_paths=['image1.png', 'image2.png'],
  tagger_type='langchain',
  tagging_options={
    'llm_class_name': 'langchain_google_genai.ChatGoogleGenerativeAI',
  }
)

result = media_tagger.tag_media(request)

result.save(output='tagging_results', writer='csv')

curl -X 'POST' \
  'http://127.0.0.1:8000/api/tag' \
  -H 'accept: application/json' \
  -H 'Content-Type: application/json' \
  -d '{
    "tagger_type": "langchain",
    "media_type": "IMAGE",
    "media_paths": [
      "image1.png",
      "image2.png"
    ],
    "tagging_options": {
      "llm_class_name": "langchain_google_genai.ChatGoogleGenerativeAI"
    }
  }'

Customization

Depending on selected LLM you may pass additional parameters to configure it. For example, for ChatOllama you can setup over 20 parameters of your choice such as base_url, temperature, reasoning, etc.

model

media-tagger tag MEDIA_PATHs \
  --media-type IMAGE \
  --tagger langchain \
  --tagger.llm_class_name=<FULLY_QUALIFIED_CLASS_NAME> \
  --tagger.model=MODEL_NAME

Where

  • MODEL_NAME - name of the models supported by a concrete LLM (i.e. gemma4:e2b supported by langchain_ollama.ChatOllama).
import media_tagging

media_tagger = media_tagging.MediaTaggingService()

request = media_tagging.MediaTaggingRequest(
  media_type='IMAGE',
  media_paths=['image1.png', 'image2.png'],
  tagger_type='langchain',
  tagging_options={
    'llm_class_name': 'langchain_ollama.ChatOllama',
    'model': 'gemma4:e2b',
  }
)

result = media_tagger.tag_media(request)

result.save(output='tagging_results', writer='csv')

curl -X 'POST' \
  'http://127.0.0.1:8000/api/tag' \
  -H 'accept: application/json' \
  -H 'Content-Type: application/json' \
  -d '{
    "tagger_type": "langchain",
    "media_type": "IMAGE",
    "media_paths": [
      "image1.png",
      "image2.png"
    ],
    "tagging_options": {
      "llm_class_name": "langchain_google_genai.ChatGoogleGenerativeAI",
      "model": "gemma4:e2b"
    }
  }'