Repository navigation
studiomdl_v2 — Release Notes
A cross-platform (Linux, Windows) 64-bit port of Valve's Source Filmmaker studiomdl, plus dmxconvert.
New features
.mppcompile (DMX model description): pass an.mppinstead of a QC; Valve'smdlcompileloaders are restored.
Binary and keyvalues2 both load. The format is documented inMPP_FORMAT.md.- Flex optimization (
-flexopt auto|on|off,-flexopt_min_static N, QC$flexoptimize): moves faces no flex touches
into separate static meshes, keeping models under the engine's 65,535-vertex flex cache limit. Without it, flexes move
vertices of other meshes. Prints a budget check and before/after statistics. - Weight fitting (
-fitweights, experimental): reduces each vertex to the maximum number of bone influences with a
least-squares fit over sample poses, instead of dropping the smallest weights. - UTF-8 paths everywhere: non-ASCII (e.g. Cyrillic) QC, source and output paths work on Windows and Linux.
- Rig constraints in DMX: point, orient, aim, parent and twist constraints, including targets under
DmeRig/DmeRigHandle. - No external DLLs: a single static executable, with no Steam or game binaries required.
Raised limits
- Flex controllers: 128 (
MAXSTUDIOFLEXCTRL) - Materials/skins: 128 (
MAXSTUDIOSKINS, was 32) - Flexed vertices per model: 65,534 (
MAXSTUDIOFLEXVERTS)
Fixes vs SFM's studiomdl.exe
- Model splitting (vertex clamp): no crash for materials other than index 0; faces keep their material; flex
controllers and rules are no longer duplicated per piece; flex keys that move nothing in a piece are culled.
Pieces are named<model>_clamped<N>. - Hitboxes: the last hitbox is no longer written into every slot; auto-generated hitboxes have no garbage orientation.
- VTA
flexfilewas ignored; DMX animations overflowed the heap (missing root bone slot);-ttexture replacement
read out of bounds;$collisionprecisionhad no effect; DMX$eyeliddidn't find its model. .mppfixes:mdlPathprefix handling,castTextureShadows,noForcedFadespelling, missingmotionControlno longer crashes.- Output is deterministic (no runtime pointers or uninitialized bytes in the files). The MDL checksum is correct on 64-bit.
Performance
- The VTX vertex-cache optimization uses meshoptimizer instead of NvTriStrip: a large model now compiles in ~1.3 s instead of ~6.5 s,
and vertex cache efficiency is better.
Under the hood
- Physics (
$collisionmodel/$collisionjoints) via a slim built-invphys_lite. - Messages use
std::format; one structured global state; std containers instead of fixed C arrays; scoped enums for studio flags. - Removed: threading, Xbox/PS3 code, Perforce, material system, rendering, the MDL/data cache, the Dme undo system,
and SFM-only Dme classes (animation sets, film clips, tracks, cameras, game models, and others). DMX files containing
removed types still load.
Known issues
- Old-style VTA flexes are not carried into split pieces.