Made on
Tilda
New software pipelines use standard RGB cameras or depth sensors (like Apple's TrueDepth) to track a performer's face.
represent a paradigm shift. Instead of moving vertices, a neural network learns to approximate the effects of a full volumetric physics simulation. This offers the best of both worlds: the expressive control of blendshapes with the anatomical realism of a simulation. The results automatically avoid volume loss and self-intersections while adding subtle details like wrinkles and volumetric elasticity, all while running in real time on consumer CPUs. This approach only requires a single neutral face mesh as input for its minimal setup. morph target animation new
Enter . This technique uses Compute Shaders to move the entire animation pipeline—animation sampling, blending, inverse kinematics (IK), and even ragdoll physics—onto the GPU. The CPU is freed to handle only the most minimal control parameters each frame. This paradigm shift not only unlocks unprecedented scale but also simplifies the rendering pipeline, enabling massive real-time experiences and simulations. New software pipelines use standard RGB cameras or
Historically, creating hundreds of distinct facial blend shapes (such as the standard Apple ARKit set of 52 shapes) required dozens of hours of meticulous sculpting. New AI-driven tools can analyze a single base mesh scan and automatically generate a complete library of anatomically correct morph targets in seconds. This offers the best of both worlds: the
| Feature | Skeletal Animation (Bone Chain) | Morph Target Animation | | :--- | :--- | :--- | | | Very Low (stores only joint rotations) | High (stores vertex positions for every frame) | | Resolution | Limited by bone count (can cause "candy wrapper" pinching) | Infinite resolution (every vertex moves independently) | | Complexity | Hard to achieve complex volume preservation | Excellent for volume preservation & squashing | | Collisions | Requires runtime calculations (expensive) | Pre-baked collisions "free" at runtime (lookup) | | Best For | Interactive physics, long duration loops | Unique, complex movements, stylized "squash & stretch" |
While skeletal animation tells you how the structure moves, morph targets tell you how the surface changes.