moondream 0.0.3

Python client library for moondream

Moondream Python Client Library

Official Python client library for Moondream, a tiny vision language model that can analyze images and answer questions about them. This library supports both local inference and cloud-based API access.

Features

• Local Inference: Run the model directly on your machine using CPU
• Cloud API: Access Moondream's hosted service for faster inference
• Streaming: Stream responses token by token for real-time output
• Multiple Model Sizes: Choose between 0.5B and 2B parameter models
• Multiple Tasks: Caption images, answer questions, detect objects, and locate points

Installation

Install the package from PyPI:

pip install moondream==0.0.2

Quick Start

Using Cloud API

To use Moondream's cloud API, you'll first need an API key. Sign up for a free account at console.moondream.ai to get your key. Once you have your key, you can use it to initialize the client as shown below.

import moondream as md
from PIL import Image

# Initialize with API key
model = md.vl(api_key="your-api-key")

# Load an image
image = Image.open("path/to/image.jpg")

# Generate a caption
caption = model.caption(image)["caption"]
print("Caption:", caption)

# Ask a question
answer = model.query(image, "What's in this image?")["answer"]
print("Answer:", answer)

# Stream the response
for chunk in model.caption(image, stream=True)["caption"]:
    print(chunk, end="", flush=True)

Using Local Inference

First, download the model weights. We recommend the int8 weights for most applications:

Model	Precision	Download Size	Memory Usage	Download Link
Moondream 2B	int8	1,733 MiB	2,624 MiB	Download
Moondream 2B	int4	1,167 MiB	2,002 MiB	Download
Moondream 0.5B	int8	593 MiB	996 MiB	Download
Moondream 0.5B	int4	422 MiB	816 MiB	Download

Then use the model locally:

import moondream as md
from PIL import Image

# Initialize with local model path
model = md.vl(model="path/to/moondream-2b-int8.bin")

# Load and encode image
image = Image.open("path/to/image.jpg")

# Since encoding an image is computationally expensive, you can encode it once
# and reuse the encoded version for multiple queries/captions/etc. This avoids
# having to re-encode the same image multiple times.
encoded_image = model.encode_image(image)

# Generate caption
caption = model.caption(encoded_image)["caption"]
print("Caption:", caption)

# Ask questions
answer = model.query(encoded_image, "What's in this image?")["answer"]
print("Answer:", answer)

API Reference

Constructor

model = md.vl(
    model="path/to/model.bin",  # For local inference
    api_key="your-api-key"      # For cloud API access
)

Methods

caption(image, length="normal", stream=False, settings=None)

Generate a caption for an image.

result = model.caption(image)
# or with streaming
for chunk in model.caption(image, stream=True)["caption"]:
    print(chunk, end="")

query(image, question, stream=False, settings=None)

Ask a question about an image.

result = model.query(image, "What's in this image?")
# or with streaming
for chunk in model.query(image, "What's in this image?", stream=True)["answer"]:
    print(chunk, end="")

detect(image, object)

Detect and locate specific objects in an image.

result = model.detect(image, "car")

point(image, object)

Get coordinates of specific objects in an image.

result = model.point(image, "person")

Input Types

• Images can be provided as:
  • PIL.Image.Image objects
  • Encoded image objects (from model.encode_image())

Response Types

All methods return typed dictionaries:

• CaptionOutput: {"caption": str | Generator}
• QueryOutput: {"answer": str | Generator}
• DetectOutput: {"objects": List[Region]}
• PointOutput: {"points": List[Point]}

Performance Notes

• Local inference currently only supports CPU execution
• CUDA (GPU) and MPS (Apple Silicon) support coming soon
• For optimal performance with GPU/MPS, use the PyTorch implementation for now

Links

• Website
• Demo