SKILL: LTX 2.3 Subject Motion & Performance Direction

Version: 1.0 | Hermes Agent LTX Human Action, Acting & Facial Performance Specialist

DESCRIPTION

Deeply researched prompting doctrine for directing subject motion, human performance, and action sequences in LTX 2.3. While the existing camera movement skill covers HOW the camera moves, this skill covers HOW subjects move — the acting grammar, facial expression triggers, gesture vocabulary, dialogue performance beats, and physics-based motion constraints required to generate believable human (and non-human) motion. Covers micro-expressions, gait patterns, hand gesture reliability, dialogue lip-sync anchoring, and the specific motion types LTX 2.3 handles well versus poorly.

TECHNICAL PARAMETERS

LTX 2.3 Motion Architecture for Subjects:

• Model: 22B DiT with 3D RoPE for spatiotemporal dynamics
• Motion Encoding: GemmaAPITextEncode processes motion descriptions as temporal sequences
• Subject Motion vs. Camera Motion: Must remain in separate prompt clauses — never mix
• Strength: Moderate, natural human motion (walking, talking, gesturing, facial expression)
• Weakness: Complex physics (crowds, chaotic cloth, water splashing, intricate hand manipulations)
• I2V Advantage: Subject identity locks at frame zero; motion deviates less than T2V