Prompting Claude for Physics-Based 2.5D Motion Videos
A detailed breakdown of Apil's workflow for generating physics-based 2.5D motion videos with Claude, from bloom lighting and camera tracking to high frame rate
When using AI models to create motion graphics code, achieving convincing depth requires strict physics constraints and disciplined staging rules. Creator Apil (@apilpirman) shared an eight-step prompting guide for Claude to build physics-driven 2.5D motion videos, combining dynamic camera tracking, bloom lighting, and frame rate conversion.
Physics Dynamics and Optical Staging
Apil's method begins with a 2.5D space where every asset moves strictly according to physics. Requiring physical simulation eliminates rigid linear easing, giving objects authentic inertia and weight. Additionally, any element that emits light must include a bloom light effect to create atmospheric glow.
The author also prescribes emission effects to produce lingering trails. Apil specifically used a bitmap effect upon an image's appearance to leave a distinctive visual trace. Crucially, each element must emerge through a progressive process rather than appearing abruptly.
Camera Tracking and Downsampling Techniques
Staging extends to scene-wide cinematography. Apil mandates dynamic camera tracking movement that follows the subject, amplifying spatial depth. Audio design is equally important: sound effects (SFX) must be carefully selected so the audio matches the visual tone without sounding cheap.
For motion fidelity, Apil employs a two-tier frame rate technique. Claude is instructed to generate the sequence at 120fps first to capture fine-grained physical movements, which is subsequently converted down to 30fps. This downsampling retains intricate trajectory data and produces significantly smoother motion.
Output Constraints and Token Exhaustion
Despite its effectiveness, this prompting workflow encounters clear operational constraints. Apil noted that the resulting video was not entirely perfect because Claude ran out of tokens during the generation process.
Calculating physics trajectories, bloom shaders, and high-frame-rate code arrays consumes significant output volume. Users attempting this pipeline should modularize complex scenes or break prompts into incremental passes to avoid token cutoffs.