This article's documentation is for anything that uses the Source engine. Click here for more information.

MDL

From Valve Developer Community
Jump to navigation Jump to search
English (en)Translate (Translate)
edit

MDL is the extension for Source's proprietary model format. It defines the structure of the model along with animation, bounding box, hit box, materials, mesh and LOD information. It does not, however, contain all the information needed for the model. Additional data is stored in PHY, ANI, VTX and VVD files, and sometimes, usually for shared animations, other .mdl files.

File format

Some details of the file format may be gleaned from the source code in Valve's studio.h, specifically the struct studiohdr_t. The early header defines a series of offsets and lengths for various sub-sections within the file, along with some key scalar information. The MDL also contains the names of materials (VMT), which may be used and referenced in various ways.

Main header

To get the latest header for specific game, please use the studio.h file in the Valve's SDK instead.

struct studiohdr_t
{
    int         id;             // Model format ID, such as "IDST" (0x49 0x44 0x53 0x54)
    int         version;        // Format version number, such as 48 (0x30,0x00,0x00,0x00)
    int         checksum;       // This has to be the same in the phy and vtx files to load!
    char        name[64];       // The internal name of the model, padding with null bytes.
                                // Typically "my_model.mdl" will have an internal name of "my_model"
    int         dataLength;     // Data size of MDL file in bytes.
 
    // A vector is 12 bytes, three 4-byte float-values in a row.
    Vector      eyeposition;    // Position of player viewpoint relative to model origin
    Vector      illumposition;  // Position (relative to model origin) used to calculate ambient light contribution and cubemap reflections for the entire model.
    Vector      hull_min;       // Corner of model hull box with the least X/Y/Z values
    Vector      hull_max;       // Opposite corner of model hull box
    Vector      view_bbmin;     // Same, but for bounding box,
    Vector      view_bbmax;     // which is used for view culling
 
    int         flags;          // Binary flags in little-endian order. 
                                // ex (0x010000C0) means flags for position 0, 30, and 31 are set. 
                                // Set model flags section for more information
 
    /*
     * After this point, the header contains many references to offsets
     * within the MDL file and the number of items at those offsets.
     *
     * Offsets are from the very beginning of the file.
     * 
     * Note that indexes/counts are not always paired and ordered consistently.
     */    
 
    // mstudiobone_t
    int        bone_count;    // Number of data sections (of type mstudiobone_t)
    int        bone_offset;   // Offset of first data section
 
    // mstudiobonecontroller_t
    int        bonecontroller_count;
    int        bonecontroller_offset;
 
    // mstudiohitboxset_t
    int        hitbox_count;
    int        hitbox_offset;
 
    // mstudioanimdesc_t
    int        localanim_count;
    int        localanim_offset;
 
    // mstudioseqdesc_t
    int        localseq_count;
    int        localseq_offset;
 
    int        activitylistversion; // ??
    int        eventsindexed;       // ??
 
    // VMT texture filenames
    // mstudiotexture_t
    int        texture_count;
    int        texture_offset;
 
    // This offset points to a series of ints.
    // Each int value, in turn, is an offset relative to the start of this header/the-file,
    // At which there is a null-terminated string.
    int        texturedir_count;
    int        texturedir_offset;
 
    // Each skin-family assigns a texture-id to a skin location
    int        skinreference_count;
    int        skinrfamily_count;
    int        skinreference_index;
 
    // mstudiobodyparts_t
    int        bodypart_count;
    int        bodypart_offset;
 
    // Local attachment points        
    // mstudioattachment_t
    int        attachment_count;
    int        attachment_offset;
 
    // Node values appear to be single bytes, while their names are null-terminated strings.
    int        localnode_count;
    int        localnode_index;
    int        localnode_name_index;
 
    // mstudioflexdesc_t
    int        flexdesc_count;
    int        flexdesc_index;
 
    // mstudioflexcontroller_t
    int        flexcontroller_count;
    int        flexcontroller_index;
 
    // mstudioflexrule_t
    int        flexrules_count;
    int        flexrules_index;
 
    // IK probably referse to inverse kinematics
    // mstudioikchain_t
    int        ikchain_count;
    int        ikchain_index;
 
    // Information about any "mouth" on the model for speech animation
    // More than one sounds pretty creepy.
    // mstudiomouth_t
    int        mouths_count; 
    int        mouths_index;
 
    // mstudioposeparamdesc_t
    int        localposeparam_count;
    int        localposeparam_index;
 
    /*
     * For anyone trying to follow along, as of this writing,
     * the next "surfaceprop_index" value is at position 0x0134 (308)
     * from the start of the file.
     */
 
    // Surface property value (single null-terminated string)
    int        surfaceprop_index;
 
    // Unusual: In this one index comes first, then count.
    // Key-value data is a series of strings. If you can't find
    // what you're interested in, check the associated PHY file as well.
    int        keyvalue_index;
    int        keyvalue_count;    
 
    // More inverse-kinematics
    // mstudioiklock_t
    int        iklock_count;
    int        iklock_index;
 
 
    float      mass;      // Mass of object (4-bytes) in kilograms

    int        contents;    // contents flag, as defined in bspflags.h
                            // not all content types are valid; see 
                            // documentation on $contents QC command
 
    // Other models can be referenced for re-used sequences and animations
    // (See also: The $includemodel QC option.)
    // mstudiomodelgroup_t
    int        includemodel_count;
    int        includemodel_index;
    
    int        virtualModel;    // Placeholder for mutable-void*
    // Note that the SDK only compiles as 32-bit, so an int and a pointer are the same size (4 bytes)
 
    // mstudioanimblock_t
    int        animblocks_name_index;
    int        animblocks_count;
    int        animblocks_index;
    
    int        animblockModel; // Placeholder for mutable-void*

    // Points to a series of bytes?
    int        bonetablename_index;
    
    int        vertex_base;    // Placeholder for void*
    int        offset_base;    // Placeholder for void*
    
    // Used with $constantdirectionallight from the QC 
    // Model should have flag #13 set if enabled
    byte        directionaldotproduct;
    
    byte        rootLod;    // Preferred rather than clamped
    
    // 0 means any allowed, N means Lod 0 -> (N-1)
    byte        numAllowedRootLods;    
    
    byte        unused0; // ??
    int         unused1; // ??
    
    // mstudioflexcontrollerui_t
    int         flexcontrollerui_count;
    int         flexcontrollerui_index;

    float       vertAnimFixedPointScale; // ??
    int         unused2;
    
    /**
     * Offset for additional header information.
     * May be zero if not present, or also 408 if it immediately 
     * follows this studiohdr_t
     */
    // studiohdr2_t
    int         studiohdr2index;
    
    int         unused3; // ??
    
    /**
     * As of this writing, the header is 408 bytes long in total
     */
};

Model flags

Bit Name Dev comment Description
0x00000001 STUDIOHDR_FLAGS_AUTOGENERATED_HITBOX This flag is set if no hitbox information was specified Set by StudioMDL if no $hbox QC commands were used
0x00000002 STUDIOHDR_FLAGS_USES_ENV_CUBEMAP NOTE: This flag is set at loadtime, not mdl build time so that we don't have to rebuild models when we change materials. Set by game engine if VMTs used for the model have $envmap env_cubemap.
0x00000004 STUDIOHDR_FLAGS_FORCE_OPAQUE Use this when there are translucent parts to the model but we're not going to sort it Set by StudioMDL if the $opaque QC command is used
0x00000008 STUDIOHDR_FLAGS_TRANSLUCENT_TWOPASS Use this when we want to render the opaque parts during the opaque pass and the translucent parts during the translucent pass Set by StudioMDL if the $mostlyopaque QC command is used
0x00000010 STUDIOHDR_FLAGS_STATIC_PROP This is set any time the .qc files has $staticprop in it

Means there's no bones and no transforms

Set by StudioMDL if the $staticprop QC command is used. Required for prop_static and prop_physics.
0x00000020 STUDIOHDR_FLAGS_USES_FB_TEXTURE NOTE: This flag is set at loadtime, not mdl build time so that we don't have to rebuild models when we change materials.
Confirm:Does "FB" mean fullbright, or framebuffer?
0x00000040 STUDIOHDR_FLAGS_HASSHADOWLOD This flag is set by studiomdl.exe if a separate "$shadowlod" entry was present for the .mdl (the shadow lod is the last entry in the lod list if present) Set by StudioMDL if the $shadowlod QC command is used.
0x00000080 STUDIOHDR_FLAGS_USES_BUMPMAPPING NOTE: This flag is set at loadtime, not mdl build time so that we don't have to rebuild models when we change materials. Set by game engine if $bumpmap or $normalmap are present in any of the models's VMTs.
0x00000100 STUDIOHDR_FLAGS_USE_SHADOWLOD_MATERIALS NOTE: This flag is set when we should use the actual materials on the shadow LOD instead of overriding them with the default one (necessary for translucent shadows) Set by StudioMDL if the $shadowlod command has the use_shadowlod_materials parameter
0x00000200 STUDIOHDR_FLAGS_OBSOLETE NOTE: This flag is set when we should use the actual materials on the shadow LOD instead of overriding them with the default one (necessary for translucent shadows) Set by StudioMDL if the $obsolete QC command is used.
0x00000400 STUDIOHDR_FLAGS_UNUSED N/A
0x00000800 STUDIOHDR_FLAGS_NO_FORCED_FADE NOTE: This flag is set at mdl build time Set by StudioMDL if the $noforcedfade QC command is used.
0x00001000 STUDIOHDR_FLAGS_FORCE_PHONEME_CROSSFADE NOTE: The npc will lengthen the viseme check to always include two phonemes Set by StudioMDL if the $forcephonemecrossfade QC command is used.
0x00002000 STUDIOHDR_FLAGS_CONSTANT_DIRECTIONAL_LIGHT_DOT This flag is set when the .qc has $constantdirectionallight in it

If set, we use constantdirectionallightdot to calculate light intensity rather than the normal directional dot product

only valid if STUDIOHDR_FLAGS_STATIC_PROP is also set

Set by StudioMDL if the $constantdirectionallight QC command is used.
0x00004000 STUDIOHDR_FLAGS_FLEXES_CONVERTED Flag to mark delta flexes as already converted from disk format to memory format Used by engine at runtime
0x00008000 STUDIOHDR_FLAGS_BUILT_IN_PREVIEW_MODE Indicates the studiomdl was built in preview mode Model has not been optimized into tristrips, and DMX quads have not been triangulated.

Set by running StudioMDL with -preview parameter.

0x00010000 STUDIOHDR_FLAGS_AMBIENT_BOOST Ambient boost (runtime flag) Contrary to the dev comment, this can be set by StudioMDL using the $ambientboost QC command.
0x00020000 STUDIOHDR_FLAGS_DO_NOT_CAST_SHADOWS Don't cast shadows from this model (useful on first-person models) Set by StudioMDL if the $donotcastshadows QC command is used.
0x00040000 STUDIOHDR_FLAGS_CAST_TEXTURE_SHADOWS alpha textures should cast shadows in vrad on this model (ONLY prop_static!) Set by StudioMDL if the $casttextureshadows QC command is used.
0x00080000 N/A N/A (undefined)
0x00100000 N/A N/A (undefined)
0x00200000 STUDIOHDR_FLAGS_VERT_ANIM_FIXED_POINT_SCALE flagged on load to indicate no animation events on this model

Secondary header

This header section is optional, and is found via the studiohdr2index value.

struct studiohdr2_t
{
        // ??
        int    srcbonetransform_count;
        int    srcbonetransform_index;

        int    illumpositionattachmentindex;

        float  flMaxEyeDeflection;    //  If set to 0, then equivalent to cos(30)

        // mstudiolinearbone_t
        int    linearbone_index;

        int    unknown[64];
};

Texture data

struct mstudiotexture_t
{
        // Number of bytes past the beginning of this structure
        // where the first character of the texture name can be found.
        int    name_offset; // Offset for null-terminated string
        int    flags;

        int    used;        // Padding?
        int    unused;      // Padding.

        int    material;        // Placeholder for IMaterial
        int    client_material; // Placeholder for void*
	
        int    unused2[10]; // Final padding
        // Struct is 64 bytes long
};

Skin replacement tables

Each "skin" that a model has (as seen in the Model Viewer and choose-able for prop entities) is actually referred to as a skin "family" in the MDL code. (Additional skins may be created with the $texturegroup compile option.) For the purposes of this section, a "skin" an area where a single material (aka texture) may be applied, while a skin-family defines a list of materials to use on skin zones 0,1,2, etc.

The skin-family section of the MDL is a sequence of short (2-byte) values, which should be broken up into a table. To illustrate how it works, we will consider a model and then examine how it would be represented on the byte-level. For this example, imagine a crate-model with three skin-families labeled (fam0,fam1,fam2) two skins-zones (mainbody,trimming) and three textures (lightwood,darkwood,metal).

Skin table
mainbody trimming
fam0 lightwood metal
fam1 darkwood metal
fam2 metal metal

Let's assume that the various textures are given ID values in the MDL like so:

Texture IDs
ID Name
0 lightwood
1 darkwood
2 metal
Note.pngNote:This ordering frequently matches the "VMTs Loaded" display of HLMV

In the MDL data, this table relationship is broken down into a stream of bytes, with two-byte texture IDs replacing the literal names. Thus the final series of bytes that corresponds to these skin-family relationships would (as little-endian short values) be:

00 00   02 00
01 00   02 00
02 00   02 00

The total number of bytes is numskinfamilies*numskins*2.

Note.pngNote:MDLs may often have a texture-replacement table which is larger than necessary, with additional columns which are never used.
Todo: Discover how to accurately detect which columns are meaningful in too-large replacement tables.

See also