Facial Vision Agent

Facial Vision Agent is an AI-powered library for analyzing facial morphology and hair characteristics from images using vision models and vision-capable LLMs.

This package focuses on visual feature extraction — it identifies facial proportions and prominent features and analyzes hair type, length, density and condition. It is designed to be used as a Python package and integrated into larger systems as an agent that processes image analysis tasks.

Key features

• Facial feature detection: face shape, proportions, forehead, eyebrows, eyes, nose, mouth, chin and jawline.
• Hair analysis: type (straight/wavy/curly/coily), length, approximate color, density and condition.
• Confidence metrics for each analysis stage.
• Lightweight API designed for automated testing and CI-driven publishing.

Installation

Developer / editable install:

python -m pip install -e .[dev]

Install from PyPI (when published):

pip install facial-vision-agent

Quick start

from facial_vision_agent import FacialVisionAgent
from agent_core_framework import AgentTask

agent = FacialVisionAgent(openrouter_api_key="YOUR_API_KEY")

# Full image analysis
task = AgentTask(type="analyze_image", payload={"image_path": "photo.jpg"})
response = agent.process(task)
print(response.data)

Payload formats

The agent accepts two ways to provide the image in the AgentTask payload:

• image_path: filesystem path to an image file (e.g. "photo.jpg").
• base64_image: a Base64-encoded string of the image (e.g. when the image is uploaded from a frontend).

Priority and behavior:

• If base64_image is present in the payload, the agent will use that string directly.
• If only image_path is provided, the agent will read the file and convert it to Base64 internally.
• If neither field is present, the agent will return an error (the current tests expect the message: "Image path is required").

Examples:

# Using filesystem path (legacy)
task = AgentTask(type="analyze_image", payload={"image_path": "photo.jpg"})

# Using Base64 directly
task = AgentTask(type="analyze_image", payload={"base64_image": "<BASE64_STRING_HERE>"})

Note on Data URIs:

• We recommend sending only the Base64 portion (without the data:image/…;base64, prefix). If you send a full data URI, strip the prefix before calling the agent.

Testing

Run the test suite locally:

python -m pytest

Publishing to TestPyPI (manual)

1. Bump the version in pyproject.toml (e.g. to 0.2.0).
2. Build distributions:

python -m build

3. Upload to TestPyPI (create a TestPyPI API token first):

python -m twine upload --repository-url https://test.pypi.org/legacy/ dist/*

Notes on CI

A GitHub Actions workflow (.github/workflows/publish-testpypi.yml) is included in this repository. To enable automated publishing to TestPyPI, add your TestPyPI token as a repository secret named TEST_PYPI_API_TOKEN and run the workflow manually or by pushing a test tag.

Contributing

Contributions are welcome. Please open an issue or a pull request. Follow the existing code style, add tests for new behavior, and keep changes focused and small.

License

MIT — see the LICENSE file for details.