When Half-Life 2 launched in 2004, the character animation team at Valve had built something called the Flex system. Rather than animating faces with bones alone, they created dozens of individual muscle targets that could blend together. Alyx Vance could furrow one brow independently of the other. Her lip corners could pull asymmetrically. No major game had done this at that fidelity for a real-time character.
For freelance character artists, this release quietly shifted client expectations. Studios began asking for facial blend shape libraries as deliverables, not just rigged meshes. A basic facial rig before 2004 might include 12 to 20 shapes. Post-Half-Life 2, requests for 40 to 60 shapes became common on mid-tier projects within a few years.
The technical process involved sculpting each shape as a morph target in the modelling software, then exporting them as a stack that the engine blended procedurally during dialogue. Valve combined this with lip-sync data extracted from audio files automatically. As a freelancer, you typically deliver the shapes and the rig; the studio handles the audio pipeline.
If a client asks for expressive characters and does not specify blend shape counts, ask them directly. The difference between 20 shapes and 60 shapes is significant in both time and price. Half-Life 2 is the clearest historical anchor for that conversation because most art directors know exactly what it achieved.