Skip to main content

FC6 XBG Format — Reverse-Engineered Specification

Technology​

  • Engine: Dunia (Far Cry 6, codename "tetra")
  • Binary format: HSEM chunk-based (same lineage as FC2-FC5)
  • Version: 0x00130049 (u32 at file offset 4)
  • Verified against: 70,837/94,589 unpacked XBG files from real FC6 install

File Header (32 bytes)​

+ 0: u32 HSEM magic (0x4853454D)
+ 4: u32 version (0x00130049)
+ 8: u32 hash1 (zero observed)
+12: u32 hash2 (zero observed)
+16: u32 unknown (zero observed)
+20: u32 filesize_minus_12
+24: u32 reserved (zero)
+28: u32 n_chunks (UNRELIABLE — walk by fourcc+size instead)
+32: first chunk

Chunk Structure​

Each chunk: 4B name + 4B version + 4B ck_size + 4B ck_dsize + 4B reserved (20B header). ck_size includes the 20B header. Payload = ck_size - 20.

Chunk order (17 chunks observed):

LTMR → LEKS → EDON → MB2O → KSRP → DIKS → DNKS → ITOM → SDOL
→ DHRM → ZNRM → XOBB → HPSB → FIKS → SDPD → [PMCP] → [PMCU]

PMCP/PMCU may appear as trailing chunks after SDPD (not inline in the chain).

Chunk Contents​

LTMR — Material References​

u32 n_materials
u32 unknown
per material:
 u32 length, string path (e.g. "graphics\_materials\X.material.bin")
 u32 length, string name (e.g. "Vaas_head2")

EDON — Node/Bone Hierarchy​

u32 n_bones
per bone:
 i32 parent_id (-1 = root)
 f32[3] translation
 f32[4] quaternion (xyzw)
 u32 bone_id
 u32 length, string name

Uses same layout as FC5 _read_edon().

DNKS — Skin Cluster Data​

Contains SULC sub-chunk(s) for per-vertex palette indices, plus optional face-based weight clusters (buffer flag 0x8000) for vehicles.

SDOL — Mesh Geometry (bulk of file)​

Identical format to FC2-FC5 (verified: Noesis FC2/3 plugin, FC4 Blender importer by Szkaradek123, FC6 binary analysis).

u32 lod_count
per LOD:
 f32 lod_distance
 u32 vb_count
 per VB (16B):
 u32 flags — vertex format bitmask (D3D9-style, see below)
 u32 stride — bytes per vertex
 u32 vcount — number of vertices
 u32 offset — byte offset within LOD vertex data
 u32 sm_count — submesh count
 per submesh (28B):
 u32[7] — vb_index, material_index, sub_entry_index,
 idx_start, last_vert, vert_byte_off, reserved
 u32 vb_size — total vertex data bytes
 align(16)
 vertex_data — vcount × stride bytes per VB, concatenated
 align(16)
 u32 index_count
 align(16)
 index_data — index_count × u16 indices

PMCP — Position Compression​

f32 unknown
f32 pos_scale — typically 1/16384 (baked into vertex decode)

PMCU — UV Compression​

f32 uv_offset
f32 uv_scale

Vertex Format Flags (bitmask in VB descriptor flags field)​

Same flag system as FC2-FC5 (Gibbed Dunia Tools / Avatar module):

Bit Hex Name Size Description
0 0x0001 POS_FLOAT 12B Position: float32 × 3
1 0x0002 POS_INT16 8B Position: int16 × 4 (÷16383.5, most common)
2 0x0004 POS_HALF 8B Position: float16 × 4
3 0x0008 UV0 4B UV channel 0: int16 × 2
4 0x0010 BONE_WTS1 8B Bone weights: u8 × 4 + u8 × 4 (indices)
5 0x0020 BONE_WTS2 8B Bone weights set 2 (rare, >4 influences)
6 0x0040 NORMAL 4B Normal: R10G10B10A2 (FC5/FC6) or D3DCOLOR
7 0x0080 COLOR 4B Vertex color: D3DCOLOR (BGRA)
8 0x0100 TANGENT 4B Tangent: R10G10B10A2 or D3DCOLOR
9 0x0200 BINORMAL 4B Binormal: R10G10B10A2 or D3DCOLOR
10 0x0400 UNK_400 4B Unknown (sometimes stores tangent)
11 0x0800 UV1 4B UV channel 1: int16 × 2
12 0x1000 UV2 4B UV channel 2: int16 × 2

Known format values​

  • 0x0BDA (3018) = 32B static: pos + 2 UVs + normal + color + fog + psize
  • 0x0BEA (3034) = 40B skinned: pos + 2 UVs + blendweight + blendindices + normal + color + fog + psize
  • 0x0BCA = 28B: pos + 2 UVs + normal + color (no fog/psize)
  • 0x0c7a = 40B: pos + 2 UVs + tangent + normal + color + fog + psize

Vertex component decode (per stride)​

StrideLayout
24pos_i16(8) + uv0_i16(4) + normal(4) + color(4) + unk(4)
28pos_i16(8) + uv0_i16(4) + uv1_i16(4) + normal(4) + color(4) + unk(4)
32pos_i16(8) + uv0_i16(4) + uv1_i16(4) + normal(4) + color(4) + fog(4) + psize(4)
36pos_i16(8) + uv0_i16(4) + uv1_i16(4) + unk(4) + normal(4) + color(4) + fog(4) + psize(4)
40pos_i16(8) + uv0_i16(4) + uv1_i16(4) + blendw(4) + blendi(4) + normal(4) + color(4) + fog(4) + psize(4)

Position decode​

x = posTrans + i16_x × posScale (posTrans=0, posScale from PMCP, typically 1/16384)
y = posTrans + i16_y × posScale
z = posTrans + i16_z × posScale

XZY order (not XYZ).

UV decode​

u = uvTrans + i16_u × uvScale (uvTrans=0, uvScale from PMCU, typically 1/16384)
v = uvTrans + i16_v × uvScale

Normal decode (FC5/FC6)​

R10G10B10A2_UNORM: 3 × 10-bit unsigned, normalized to [-1,1] + 2-bit A.

u = struct.unpack_from('<I', data, offset)[0]
x = (u & 0x3FF) / 1023.0 * 2.0 - 1.0
y = ((u >> 10) & 0x3FF) / 1023.0 * 2.0 - 1.0
z = ((u >> 20) & 0x3FF) / 1023.0 * 2.0 - 1.0

Index buffer​

index_count × u16, 16-byte aligned.
Face winding: reversed from file (game uses opposite front-face winding).

FC6 Vertex Element Types (DX12-style, from engine binary)​

FC6 also supports a DX12-style vertex element system (separate from the bitmask flags). Used by the shader system:

Types: Float1-4, UByte4/UByte4N, Short2/4/Short2N/Short4N, UShort2N/UShort4N, UInt1, Float_16_2/4, UNorm_10_10_10_2

Semantics (CS_ in binary, S_ in shadersobj.dat XML): CS_Position, CS_Normal, CS_Tangent, CS_Binormal, CS_TangentSign, CS_TexCoord, CS_Color, CS_BlendWeights, CS_BlendIndices

Sources​

  • Noesis plugin: github.com/alrusdi/noesis_farcry (FC2/3)
  • FC4 Blender importer: Blender249[FarCry4] by Szkaradek123
  • Avatar SDOL parser: blender-io-dunia/modules/Avatar/sdol_avatar.py
  • Vertex flags: Gibbed Dunia Tools + blender-io-dunia VertexFlags class
  • FC6 types/semantics: FC6 engine binary string dump (FC6 Strings.csv)
  • XBG format notes: docs/farcry/xbg-format.md
  • Vertex buffer formats: docs/farcry/vertex-buffer-formats.md