UnityPy

UnityPy

A Unity asset extractor for Python based on AssetStudio.

Next to extraction, UnityPy also supports editing Unity assets. Via the typetree structure all object types can be edited in their native forms.

# modification via dict:
    raw_dict = obj.parse_as_dict()
    # modify raw dict
    obj.patch(raw_dict)
# modification via parsed class
    instance = obj.parse_as_object()
    # modify instance
    obj.patch(instance)

If you need advice or if you want to talk about (game) data-mining, feel free to join the UnityPy Discord.

If you're using UnityPy for a commercial project, a donation to a charitable cause or a sponsorship of this project is expected.

As UnityPy is still in active development, breaking changes can happen. These changes are usually limited to minor versions (x.y) and not to patch versions (x.y.z). So in case that you don't want to actively maintain your project, make sure to make a note of the used UnityPy version in your README or add a check in your code. e.g.

if UnityPy.__version__ != '1.9.6':
    raise ImportError("Invalid UnityPy version detected. Please use version 1.9.6")

1. Installation
2. Example
3. Important Classes
4. Important Object Types
5. Configurations
6. Credits

Installation

Python 3.8 or higher is required.

Install via PyPI:

pip install UnityPy

Install from source code:

git clone https://github.com/K0lb3/UnityPy.git
cd UnityPy
python -m pip install .

Notes

Windows

Visual C++ Redistributable is required for the brotli dependency. In case a new(ish) Python version is used, it can happen that the C-dependencies of UnityPy might not be precompiled for this version. In such cases the user either has to report this as issue or follow the steps of this issue to compile it oneself. Another option for the user is downgrading Python to the latest version supported by UnityPy. For this see the Python version badge at the top of the README.

Crash without warning/error

The C-implementation of the typetree reader can directly crash Python. In case this happens, the usage of the C-typetree reader can be disabled. Read this section for more details.

Example

The following is a simple example.

…

You probably have to read Important Classes and Important Object Types to understand how it works.

Users with slightly advanced Python skills should look at UnityPy/tools/extractor.py for a more advanced example. It can also be used as a general template or as an importable tool.

Important Classes

Environment

Environment loads and parses the given files. It can be initialized via:

• a file path - apk files can be loaded as well
• a folder path - loads all files in that folder (bad idea for folders with a lot of files)
• a stream - e.g., io.BytesIO, file stream,...
• a bytes object - will be loaded into a stream

UnityPy can detect if the file is a WebFile, BundleFile, Asset, or APK.

The unpacked assets will be loaded into .files, a dict consisting of asset-name : asset.

All objects of the loaded assets can be easily accessed via .objects, which itself is a simple recursive iterator.

import io
import UnityPy

# all of the following would work
src = "file_path"
src = b"bytes"
src = io.BytesIO(b"Streamable")

env = UnityPy.load(src)

for obj in env.objects:
    ...

# saving an edited file
    # apply modifications to the objects
    # don't forget to use data.save()
    ...
with open(dst, "wb") as f:
    f.write(env.file.save())

Asset

Assets (SerializedFile class) are a container that contains multiple objects. One of these objects can be an AssetBundle, which contains a file path for some of the objects in the same asset.

All objects can be found in the .objects dict - {ID : object}.

The objects with a file path can be found in the .container dict - {path : object}.

Object

Objects (ObjectReader class) contain the actual files, e.g., textures, text files, meshes, settings, ...

To acquire the actual data of an object it has to be parsed first. This happens via the parse functions mentioned below. This isn't done automatically to save time as only a small part of the objects are usually of interest. Serialized objects can be set with raw data using .set_raw_data(data) or modified with .save() function, if supported.

For object types with m_Name you can use .peek_name() to only read the name of the parsed object without parsing it completely, which is way faster.

There are two general parsing functions, .parse_as_object() and .parse_as_dict(). parse_as_dict parses the object data into a dict. parse_as_object parses the object data into a class. If the class is a Unity class, it's stub class from UnityPy.classes(.generated) will be used, if it's an unknown one, then it will be parsed into an UnknownObject, which simply acts as interface for the otherwise parsed dict. Some special classes, namely those below, have additional handlers added to their class for easier interaction with them.

The .patch(item) function can be used on all object (readers) to replace their data with the changed item, which has to be either a dict or of the class the object represents.

Example

for obj in env.objects:
    if obj.type.name == "the type you want":
        if obj.peek_name() != "the specific object you want":
            continue
        # parsing
        instance = obj.parse_as_object()
        dic = obj.parse_as_dict()

        # modifying
        instance.m_Name = "new name"
        dic["m_Name"] = "new name"

        # saving
        obj.patch(instance)
        obj.patch(dic)

Legacy

Following functions are legacy functions that will be removed in the future when major version 2 hits. The modern versions are equivalent to them and have a more correct type hints.

Important Object Types

Now UnityPy uses auto generated classes with some useful extension methods and properties defined in legacy_patch. You can search for a specific classes in the module UnityPy.classes with your IDE's autocompletion.

Texture2D

• .m_Name
• .image converts the texture into a PIL.Image
• .m_Width - texture width (int)
• .m_Height - texture height (int)

Export

from PIL import Image
for obj in env.objects:
    if obj.type.name == "Texture2D":
        # export texture
        tex = obj.parse_as_object()
        path = os.path.join(export_dir, f"{tex.m_Name}.png")
        tex.image.save(path)
        # edit texture
        fp = os.path.join(replace_dir, f"{tex.m_Name}.png")
        pil_img = Image.open(fp)
        tex.image = pil_img
        tex.save()

Sprite

Sprites are part of a texture and can have a separate alpha-image as well. Unlike most other extractors (including AssetStudio), UnityPy merges those two images by itself.

• .m_Name
• .image - converts the merged texture part into a PIL.Image
• .m_Width - sprite width (int)
• .m_Height - sprite height (int)

Export

for obj in env.objects:
    if obj.type.name == "Sprite":
        sprite = obj.parse_as_object()
        path = os.path.join(export_dir, f"{sprite.m_Name}.png")
        sprite.image.save(path)

TextAsset

TextAssets are usually normal text files.

• .m_Name
• .m_Script - str

Some games save binary data as TextAssets. As m_Script gets handled as str by default, use m_Script.encode("utf-8", "surrogateescape") to retrieve the original binary data.

Export

for obj in env.objects:
    if obj.type.name == "TextAsset":
        # export asset
        txt = obj.parse_as_object()
        path = os.path.join(export_dir, f"{txt.m_Name}.txt")
        with open(path, "wb") as f:
            f.write(txt.m_Script.encode("utf-8", "surrogateescape"))
        # edit asset
        fp = os.path.join(replace_dir, f"{txt.m_Name}.txt")
        with open(fp, "rb") as f:
            txt.m_Script = f.read().decode("utf-8", "surrogateescape")
        txt.save()

MonoBehaviour

MonoBehaviour assets are usually used to save the class instances with their values. The structure/typetree for these classes might not be contained in the asset files. In such cases see the 2nd example (TypeTreeGenerator) below.

• .m_Name
• .m_Script
• custom data

Export

import json

for obj in env.objects:
    if obj.type.name == "MonoBehaviour":
        # export
        # save decoded data
        tree = obj.parse_as_dict()
        fp = os.path.join(extract_dir, f"{tree['m_Name']}.json")
        with open(fp, "wt", encoding = "utf8") as f:
            json.dump(tree, f, ensure_ascii = False, indent = 4)

        # edit
        tree = obj.parse_as_dict()
        # apply modifications to the data within the tree
        obj.patch(tree)

TypeTreeGenerator

UnityPy can generate the typetrees of MonoBehaviours from the game assemblies using an optional package, TypeTreeGeneratorAPI, which has to be installed via pip. UnityPy will automatically try to generate the typetree of MonoBehaviours if the typetree is missing in the assets and env.typetree_generator is set.

…

AudioClip

• .samples - {sample-name : sample-data}

The samples are converted into the .wav format. The sample data is a .wav file in bytes.

clip: AudioClip
for name, data in clip.samples.items():
    with open(name, "wb") as f:
        f.write(data)

Font

Export

if obj.type.name == "Font":
    font: Font = obj.parse_as_object()
    if font.m_FontData:
        extension = ".ttf"
        if font.m_FontData[0:4] == b"OTTO":
            extension = ".otf"

    with open(os.path.join(path, font.m_Name+extension), "wb") as f:
        f.write(font.m_FontData)

Mesh

• .export() - mesh exported as .obj (str)

The mesh will be converted to the Wavefront .obj file format.

mesh: Mesh
with open(f"{mesh.m_Name}.obj", "wt", newline = "") as f:
    # newline = "" is important
    f.write(mesh.export())

Renderer, MeshRenderer, SkinnedMeshRenderer

ALPHA-VERSION

• .export(export_dir) - exports the associated mesh, materials, and textures into the given directory

The mesh and materials will be in the Wavefront formats.

mesh_renderer: Renderer
export_dir: str

if mesh_renderer.m_GameObject:
    # get the name of the model
    game_obj_reader = mesh_renderer.m_GameObject.deref()
    game_obj_name = game_obj_reader.peek_name()
    export_dir = os.path.join(export_dir, game_obj_name)
mesh_renderer.export(export_dir)

Texture2DArray

WARNING - not well tested

• .m_Name
• .image converts the texture2darray into a PIL.Image
• .m_Width - texture width (int)
• .m_Height - texture height (int)

Export

import os
from PIL import Image
for obj in env.objects:
    if obj.type.name == "Texture2DArray":
        # export texture
        tex_arr = obj.parse_as_object()
        for i, image in enumerate(tex_arr.images):
            image.save(os.path.join(path, f"{tex_arr.m_Name}_{i}.png"))
        # editing isn't supported yet!

Configurations

There're several configurations and interfaces that provide the c