Keyboard shortcuts

Press ← or → to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

Animation pack evaluation: Mixamo Locomotion

Technical verdict: Restricted use

Evaluation completeness: partial — all 20 extracted FBX files have current mechanical evidence; vendor mapping, runtime, visual, contact, retarget, license, and cross-pack acceptance remain unavailable.

Confidence: medium

Evaluation date: 2026-09-16

Current evaluator: AnimSmith 0.14.0

Report format: 3

Detailed evidence: Evidence appendix

Technical decision

For a controller-driven (kinematic) prototype, start with an unarmed walk/run speed pair using the selected members. Their source mechanics were measured, but the proposed blend has no accepted loop, foot-contact, transition, target-character, or target-engine result. Use one controller for movement and yaw; decide ownership separately for any additional root-motion-directory state.

First experiment: Import the selected in-place files on the target character, play them individually, then test their blend through speed or direction changes, starts, stops, and interrupted transitions. Review any endpoint warning on other named files before admission: retain an intentional hold after visual review, or request source repair when the intended endpoint differs. The evidence appendix retains the exact source counts, warning totals, skeleton counts, and profile results. The adoption guide compares this limited pilot with broader library and authoring routes.

Capability coverage

This is a full-body humanoid controller hypothesis for visible characters. Third-person gameplay-camera appearance and target-character deformation remain visually untested. Dedicated first-person arms/viewmodel content and close-camera suitability were not established.

Content present

Candidate ingredients in the evaluated filesCurrent use boundary
Walk and run are named and measured. Idle, jump, strafe and turns appear by filename; their controller roles remain unclassified.Check the named states and their transitions in the target game before treating this as a complete controller.

Content gaps and unknowns

Other delivered motion leads, filename only: idle.fbx, jump.fbx, left turn 90.fbx, right turn 90.fbx, and right strafe.fbx. These names occur in the inspected source or issue inventory, but their intended controller roles and gameplay behavior have not been classified or tested. Inspect each source and its boundaries before adding a state; see the retained source evidence. A zero canonical role count does not establish that these motions are absent.

The legacy count of 18 motions has no verified mapping to these 20 files. Classify the other named motions before adding them to a controller.

Evaluation still needed

  • Loop/phase policy, target-engine graph, contact and visual review, target-character retargeting, performance and cross-pack acceptance remain open.

Runtime sets and authored motion

SetController useAdoption decisionExact members
hypothesis/kinematic-speedWalk/run speed blend; controller owns XZ translation, yaw and collisionThe selected walk/run pair has no reported mechanical finding. Prototype with controller-owned movement and project-defined speed thresholds. Six other root-motion-directory states need per-file ownership; idle, left turn 90 and right strafe have source endpoint warnings. No AnimSmith repair was tested.Member timings, coordinates and contracts

The selected files are grouped by name and measurement. AnimSmith measured phase (cycles) as walk 0.4841 and run 0.4529; their measured phase spread is 0.0312 cycles. These phase numbers do not establish matching foot contacts or a usable blend. Loop intent, contact alignment, and gameplay use remain not-evaluated.

Integration recipe

  1. Members/topology: Build a walk/run speed blend from the two linked clips. Set walk and run thresholds from your controller speeds and visible stride matching; stationary roots do not provide those thresholds. Treat other directions and root-motion alternatives as separate, untested sets.
  2. Timing/synchronization: Keep source timing and leave loop settings undecided. Compare normalized phase and visible foot contacts in the target engine before enabling a continuous blend.
  3. State ownership: These selected in-place files supply pose; one kinematic controller supplies XZ translation, yaw, and collision. Root-motion use, where animation supplies travel, requires separate per-file checks.
  4. Composition constraints: Begin with full-body states and transitions. Masks, additive motion, IK, sockets, and layering need separate tests.
  5. Acceptance gate: In the chosen engine, sweep the walk/run speed range in both directions and test rapid speed changes, starts and stops, foot contacts, interrupted transitions, and target-character deformation.

Technical issue register

IDSeverityProblem and impactPrimary ownerCurrent actionFuture AnimSmith potentialEvidence/status
MIX-OWN-001moderateScope: Locomotion_Pack_-_root-motion/idle.fbx; Locomotion_Pack_-_root-motion/jump.fbx; Locomotion_Pack_-_root-motion/left turn 90.fbx; Locomotion_Pack_-_root-motion/left turn.fbx; Locomotion_Pack_-_root-motion/right turn 90.fbx; Locomotion_Pack_-_root-motion/right turn.fbx; reproduce: lint each file with the current archive-level animation-owned XZ declaration; impact: applying that policy can leave these states stationary while gameplay expects displacement. Guidance: not applicable.engine-configDeclare these files controller-owned XZ unless project evidence establishes authored displacement. Accept when per-file declared lint has no ownership error and one target-controller playback applies translation/yaw/collision exactly once.No generic rewrite is justified because the directory label does not establish per-file intent.observed-animsmith; 6 current findings; no runtime acceptance; unresolved.
MIX-TIME-001moderateScope: Locomotion_Pack_-_root-motion/idle.fbx; Locomotion_Pack_-_root-motion/left turn 90.fbx; Locomotion_Pack_-_root-motion/right strafe.fbx; reproduce: empty-baseline duration-sanity; impact: shorter channels clamp-hold at the end and can create a brief frozen joint or transition/loop pop. Guidance.unknownInspect the intended endpoint and slowed target-engine boundary playback first. If the hold is intentional and visually acceptable, document and retain it. Otherwise ask the source author to align the shorter channels to the intended endpoint, rerun duration-sanity, and review boundary playback. Accept only when the intended hold or corrected boundary is visually approved and the remaining mechanical warning is recorded accurately.Deterministic endpoint conformance is plausible only after the intended endpoint and protected channels are declared.observed-animsmith; 3 current source files; visual intent not evaluated; unresolved.
MIX-CONTENT-001noteScope: Locomotion_Pack_-_in-place/X Bot.fbx; Locomotion_Pack_-_root-motion/X Bot.fbx; reproduce: empty-baseline duration-sanity reports Take 001 with no tracks; impact: automatic clip registration can expose an empty state, although these may be reference-character files. Guidance: not applicable.unknownClassify each file as reference-only or intended animation before import. Accept when reference files are excluded from animation registration or the author supplies the intended tracks.No tool should invent missing motion.observed-file and observed-animsmith; semantic purpose unavailable; unresolved.

Engine status

RuntimeEvidence levelTechnical resultRemaining gate
Unity unspecifiednot-evaluatedNo current import, graph, playback, or visual evidenceTest the proposed controller and full-body graph on the target character
Unreal Engine unspecifiednot-evaluatedNo current import, retarget, Blend Space, or playback evidenceTest the exact retarget and controller path
Godot unspecifiednot-evaluatedNo current import, AnimationTree, or playback evidenceTest the exact Skeleton3D and controller path
Bevy unspecifiednot-evaluatedNo current loader, graph, or playback evidenceTest exact animation targets and controller composition

Fit and limitations

Best fit is general unarmed locomotion prototyping where a developer can own movement, define thresholds, and test the exact target character. It is a poor fit for drop-in root-motion, network, motion-matching, layered-animation, or production-ready claims.

Cross-pack compatibility is unknown. Shared Mixamo role resolution and a common 66-bone count on motion-named files do not prove identical hierarchy, rest pose, scale, retargeting, blending, contacts, or style.

Changes between AnimSmith versions

AnimSmith 0.14.0 — The current evaluation reran 996 inspect, measure, and lint commands across the collection. This constituent admitted 20/20 files; baseline and declared-contract findings were revalidated, and the selected controller set was proposed from current files. Its clip roles remain unverified.

AnimSmith 0.10.0 — Retained historical mechanical counts agree with the current rerun; clip mapping and engine acceptance remain open. AnimSmith 0.7.0 evidence remains superseded.

Evidence status

The current tests covered 20 FBX source files individually. The legacy metadata lists 18 motions, but its clip-to-file mapping is unverified for the 20 files. The evidence appendix records the source integrity check, evaluator revision, command results, and digests. See the canonical readiness ladder.

Sources