Procedural Animation vs Hand-Keyed: When to Use Each

The point about secondary motion is where I think most people underestimate hand-keyed work. Procedural secondary is great for predictable physics responses — hair, loose cloth, tail sway — but the moment you need secondary motion to tell a story (a character hesitating, their coat settling as they pause to think), procedural systems tend to produce physically correct but emotionally inert results.

I've started treating procedural as a first pass even for cases where I'll hand-key the final. Running a quick rigid body sim on a cape gives me a physically plausible baseline to key on top of, rather than starting from a T-pose. Cuts my blocking time significantly. The two approaches aren't really competing — procedural is often the best reference material for good hand-keyed work.

The framing I've landed on after a few projects: procedural handles the physics of the world, hand-keyed handles the intention of the character. A foot planting correctly on a slope is physics — procedural wins. A character reacting with shock to seeing a ghost is intention — hand-keyed wins.

The area I'd push back on slightly is locomotion blending. I used to hand-key walk/run cycles and blend between them, but switching to motion matching (we used a stripped-down version inspired by the GDC talks on the Ubisoft/Unity approach) removed an entire class of foot-sliding bugs I'd been fighting for months. It's technically procedural but it's sourcing from hand-animated clips, so it's more of a hybrid than a replacement.

For anyone working in Godot, the new AnimationTree with blend spaces handles a lot of this without needing to roll your own — worth exploring before building custom blending logic.

The framing of procedural vs. hand-keyed as a binary always bothers me a little — in practice the most interesting results come from using them as layers. Motion matching is the clearest example of this middle path: you hand-key (or mocap) a library of clips, and a runtime system selects and blends between them based on trajectory and pose matching. The animator controls quality and feel; the system handles responsiveness and transitions.

For smaller teams that can't afford full motion matching infrastructure, a simpler version of this is just being intentional about which layer owns what: procedural IK and secondary motion on top of hand-keyed base poses. The animator focuses on the performance; code handles the environmental adaptation. Kept that way, neither discipline is fighting the other.

Where I'd push back on the "procedural for environment adaptation" rule: foot planting IK can actually destroy an animator's intentional weight and timing if the constraints are too aggressive. Sometimes the character sliding slightly is less jarring than IK wrenching the pose. Tuning the blend weights matters a lot.

Procedural IK for climbing is the one that humbled me. Conceptually simple: raycast, place hand, solve chain. But getting it to feel good took embarrassingly long. The IK math was fine, the problem was all the blending logic around it. How fast does the hand move to the new target? What happens when the character is mid-leap and the surface angle changes? Does the spine compensate for reach? Every edge case revealed another edge case.

Ended up being a hybrid anyway, hand-keyed reference animations driving the blend weights, procedural IK doing the final placement. Neither approach alone got there.

100% agree. The best results I've seen are hybrid — like The Last of Us Part II where hand-keyed combat animations blend with procedural IK and physics responses.

The key insight is that procedural should handle context adaptation while hand-keyed handles artistic intent.

One hybrid approach I don't see mentioned enough: using procedural animation as a corrective layer on top of hand-keyed clips rather than as a replacement. The base walk/run/idle are fully hand-keyed for artistic control, but a lightweight IK pass runs on top every frame to handle foot placement, and a spring-bone system handles secondary motion on hair/clothing.

The key insight from my project was that animators stopped fighting the procedural systems once they understood they were only responsible for the intent of the motion — the procedural layer handles environmental adaptation. That division of responsibility actually sped up the animation pipeline because cycles could be authored on flat ground without worrying about terrain edge cases.

Where I'd push back slightly on the original post: physics-based secondary motion can actually be worse than hand-keyed for characters with strong personality. A bouncy cartoon character's hair should overshoot and settle in a very specific way that communicates character — physics gives you physically accurate, not expressively accurate.