media-toolkit 0.2.12

Web-ready standardized file processing and serialization. Read, load and convert to standard file types with a common interface.

MediaToolkit

Web-ready standardized file processing and serialization

Features

Read, load and convert to standard file types with a common interface. Especially useful for code that works with multiple file types like images, audio, video, etc.

Load and convert from and to common data types:

• numpy arrays
• file paths
• bytes
• base64
• json
• urls
• etc.

Transmit files between services with a common interface

• Native FastSDK and FastTaskAPI integration
• Supports httpx, requests

Work with native python libs like BytesIO.

Only use the file types you need, no unnecessary dependencies.

Installation

You can install the package with PIP, or clone the repository.

# install from pypi
pip install media-toolkit
# install without dependencies: this is useful if you only need the basic functionality (working with files)
pip install media-toolkit --no-deps
# if you want to use certain file types, and convenience functions
pip install media-toolkit[VideoFile]  # or [AudioFile, VideoFile, ...]
# install from github for newest release
pip install git+git://github.com/SocAIty/media-toolkit

The package checks if you have missing dependencies for certain file types while using. Use the --no-deps flag for a minimal tiny pure python installation. The package with dependencies is quite small < 39kb itself.

Note: for VideoFile you will also need to install ffmpeg

Usage

Create a media-file from any data type

The library automatically detects the data type and loads it correctly.

from media_toolkit import MediaFile, ImageFile, AudioFile, VideoFile

# could be a path, url, base64, bytesio, file_handle, numpy array ...
arbitrary_data = "...."
# Instantiate an image file
new_file = ImageFile().from_any(arbitrary_data)

All files (ImageFile, AudioFile, VideoFile) types support the same interface / methods.

Explicitly load from a certain type.

This method is more secure than from_any, because it definitely uses the correct method to load the file.

new_file = MediaFile()

new_file.from_file("path/to/file")
new_file.from_file(open("path/to/file", "rb"))
new_file.from_numpy_array(my_array)
new_file.from_bytes(b'bytes')
new_file.from_base64('base64string')
new_file.from_starlette_upload_file(starlette_upload_file)

Convert to any format or write to file

Supports common serialization methods like bytes(), np.array(), dict()

my_file = ImageFile().from_file("path/to/my_image.png")

my_file.save("path/to/new_file.png")  
as_numpy_array = my_file.to_numpy_array()
as_numpy_array = np.array(my_file)

as_bytes = my_file.to_bytes()
as_bytes = bytes(my_file)
as_base64 = my_file.to_base64()
as_json = my_file.to_json()

Working with Collections of Files

MediaList

A flexible list that can handle multiple media files with type safety:

from media_toolkit import MediaList, AudioFile

# Create a list that only accepts AudioFiles
audio_list = MediaList[AudioFile]()

# Add files to the list
audio_list.append("path/to/audio.mp3")
audio_list.extend(["url1", "url2"])

# Process all files
for audio in audio_list:
    print(audio.file_size())

# Convert all files to base64
base64_files = audio_list.to_base64()

MediaDict

A dictionary for organizing media files with keys:

from media_toolkit import MediaDict, ImageFile

# Create a dictionary that only accepts ImageFiles
image_dict = MediaDict[ImageFile]()

# Add files with keys
image_dict["profile"] = "path/to/profile.jpg"
image_dict["banner"] = "https://example.com/banner.png"

# Process files
for key, image in image_dict.items():
    print(f"{key}: {image.file_size()}")

# Convert to JSON
json_data = image_dict.to_json()

Both MediaList and MediaDict support:

• Type safety with generic types (e.g., MediaList[AudioFile])
• Lazy loading of files
• Batch processing
• Common operations (to_base64, to_bytes, etc.)
• Nested structures (MediaDict inside MediaList and vice versa)

Working with VideoFiles

The VideoFiles use PyAV together with pydub. VideoFiles support extra methods like audio extraction and combining video and audio. PyAV is a powerful Pythonic binding for FFmpeg libraries that supports many video formats and codecs and is known for robust, efficient processing.

# load the video file
vf = VideoFile().from_file("test_files/test_vid_1.mp4")

# extract audio_file
yf = vf.extract_audio("extracted_audio.mp3")

# stream the video
for img, audio in vf.to_video_stream(include_audio=True):
    cv2.imwrite("outtest.png", img)

# add audio to an videofile (supports files and numpy.array)
vf.add_audio("path/to/audio.mp3")

# create a video from a folder
VideoFile().from_dir("path/to/image_folder", audio=f"extracted_audio.mp3", frame_rate=30)

# create a video from a video stream
fromstream = VideoFile().from_video_stream(vf.to_video_stream(include_audio=True))

Web-features

We intent to make transmitting files between services as easy as possible. Here are some examples for services and clients.

FastTaskAPI - Services

The library supports the FastTaskAPI and FastSDK for easy file transmission between services. Simply use the files in the task_endpoint function definition and transmitted data will be converted. Check out the FastTaskAPI documentation for more information.

from fast_task_api import ImageFile, AudioFile, VideoFile

@app.task_endpoint("/my_file_upload")
def my_upload_image(image: ImageFile, audio: AudioFile, video: VideoFile):
    image_as_np_array = np.array(image)

fastAPI - services

You can use the files in fastapi and transform the starlette upload file to a MediaFile.

@app.post("/upload")
async def upload_file(file: UploadFile = File(...)):
    mf = ImageFile().from_any(file)
    return {"filename": file.filename}

Client with: requests, httpx

To send a MediaFile to an openapi endpoint you can use the following method:

import httpx

my_media_file = ImageFile().from_file("path/to/my_image.png")
my_files = {
  "param_name": my_media_file.to_httpx_send_able_tuple()
  ...
}
response = httpx.Client().post(url, files=my_files)

How it works

If media-file is instantiated with from_* it converts it to an intermediate representation. The to_* methods then convert it to the desired format.

Currently the intermediate representation is supported in memory with (BytesIO) or on disk with temporary files.

ToDo:

• decreasing redundancies for _file_info() method