[Mobile] Sky; Children of the Light

[longbyte1]

Brilliant work!! Thank you so much!!
Next step is to crack into the .level files as I'm sure theres some tasty goodies in there

I'll have a looksie about matching up models with their textures if possible

---

Hi all,

Someone contacted me who was also wanting to extract mesh data for some personal research, and indeed we were met with success!

Here is Python code for Blender 3.2.2:

Code:

import ctypes
import struct
import io
from ctypes import *

f = open('/tmp/UnitCube.mesh', 'rb')

lz4 = CDLL('liblz4.so.1')

# read uncompressed size
f.seek(0x52)
uncompressed_size = struct.unpack('i', f.read(4))[0]

# read compressed size
f.seek(0x4e)
compressed_size = struct.unpack('i', f.read(4))[0]

# read num lods
f.seek(0x44)
num_lods = struct.unpack('i', f.read(4))[0]

print('compressed_size', compressed_size)
print('uncompressed_size', uncompressed_size)
print('num_lods', num_lods)

# get compressed content
f.seek(0x56)
src = f.read(compressed_size)

# get decompressed content
dest = ctypes.create_string_buffer(uncompressed_size)
ret = lz4.LZ4_decompress_safe(src, dest, compressed_size, uncompressed_size)
if ret <= 0:
    raise IOError('error decompressing mesh - file may not be valid')

buf = io.BytesIO(dest.raw)
buf.seek(0x74)
shared_vertex_count = struct.unpack('i', buf.read1(4))[0]
buf.seek(0x78)
total_vertex_count = struct.unpack('i', buf.read1(4))[0]
buf.seek(0x80)
point_count = struct.unpack('i', buf.read1(4))[0]
buf.seek(0x74)
uv_count = struct.unpack('i', buf.read1(4))[0]

print('shared_vertex_count', shared_vertex_count)
print('total_vertex_count', total_vertex_count)
print('point_count', point_count)
print('uv_count', uv_count)

# build vertex buffer
vertex_buffer = []
vertex_buffer_start = 0xb3
buf.seek(vertex_buffer_start)

for i in range(shared_vertex_count):
    # 3 floats
    x, y, z = struct.unpack('<fff4x', buf.read(16))
    vertex_buffer.append((x, y, z))

# build uv buffer
uv_buffer = []
uv_header_size = uv_count * 4 - 4
buf.read1(uv_header_size) # move caret to uv start
for i in range(uv_count):
    # 2 half-precision floats
    u, v = struct.unpack('<4xee8x', buf.read(16))
    uv_buffer.append((u, v))

# build index buffer
index_buffer = []
face_count = total_vertex_count // 3
buf.read1(4) # advance 4 bytes of padding
for i in range(face_count):
    # 3 shorts (each indexing into the vertex buffer)
    v1, v2, v3 = struct.unpack('<HHH', buf.read(6))
    index_buffer.append((v1, v2, v3))

f.close()
buf.close()

print(vertex_buffer)
print(index_buffer)

# build geometry from buffers
vertices = []
edges = []
faces = []
for face in index_buffer:
    v1i, v2i, v3i = face
    edges += [(v1i, v2i), (v2i, v3i), (v3i, v1i)]
    faces.append(face)

import bpy
mesh = bpy.data.meshes.new('created_mesh')
mesh.from_pydata(vertex_buffer, edges, faces)
mesh.update()

uvl = mesh.uv_layers.new()
uvl.data.foreach_set('uv', [uv for pair in [uv_buffer[l.vertex_index] for l in mesh.loops] for uv in pair])
mesh.uv_layers.active = uvl

obj = bpy.data.objects.new('created_object', mesh)
collection = bpy.data.collections.new('created_collection')
bpy.context.scene.collection.children.link(collection)
collection.objects.link(obj)

Just change the open() line to load the file you want. Note that this is only going to read the first LOD it finds, which should be the highest quality one.
For Windows users, you need to change the path in the CDLL() line to a path to a 64-bit LZ4 DLL (or 32-bit, depending on whether your Blender is 64- or 32-bit). You can find a working copy of LZ4 for Windows here, and you're looking for a DLL called either "msys-lz4-1.dll" or "liblz4.dll".

Many thanks to cobrakyle for their help.

Just another q, am I alright to add this work onto the first post for any others looking for this?

Sure! Once I have enough scripts, I will publish a repo on GitHub as well.