livepng

LivePNG

LivePNG is a format to create avatars based on PNG images.

Demo

https://github.com/user-attachments/assets/efdddbcf-2850-4325-ab2c-0381f46072f1

Examples

The easiest way to understand the format is by looking at the examples folder.

Models structure

Every model is described by a JSON file called model.json inside its folder. Its properties are:

• name: Name of the model
• version: Version of LivePNG (for now only 1 is available)
• styles: The list of styles a model has.

Styles

A style is supposed to be a different style for a model, for example a clothes change or a different state. Every style has a name style_name, and is contained in the assets/style_name folder. Every style has at least one expression.

Expressions

An expression is supposed to be a different expression for a model, for example a different face expression based on the mood or the emotions of a character. Every expression has an expression_name, and is contained in the assets/style_name/expression_name. Expressions might be different for each style. If you want a default expression, name it idle, otherwise the first expression in alphabetic order is taken. Every expression has at least one variant

Variants

Variants are different variants of an expression. Their objective is to make the character appear more lively, and show different images for each expression. Every variant has a variant_name and is contained in the assets/style_name/expression_name/variant_name. The variant_name folder must contain the image files that show different states of the lips.

Python Examples

Installing

To install and update the package, you can user pip:

pip install -U livepng

Quick start

from livepng import LivePNG

model = LivePNG("path/to/model.json")
# Setting the second style
styles = model.get_styles()
model.set_current_style(styles[list(styles.keys())[1]])
# Setting an happy expression
model.set_current_expression("happy")
# Add a function as a callback
# we are setting a stub function `update_image` as callback
# this function will be called every time there is a frame update
# You can subscribe to more events implementing th Live2DModelObserver interface
model.subscribe_callback(update_iamge)
# Start speaking with lipsync
# It will play audio and set a random variant before starting to speak
# It will call update_image every 0.1s passing the frame to show
model.speak("file.wav", random_variant=True, play_audio=True, frame_rate=10, interrupt_others=True, start_thread=True)