AI-generated 3D geometry needs retopology before it can move

The picture

Two comparison documents, both retrieved 16 September 2026, describe the same structural limit in current text-to-3D generators: the mesh a model outputs is usually not built the way an animator needs it built. A trade comparison of eight AI 3D generators states that 'AI models often have non-manifold geometry, inconsistent wall thicknesses, and impossible undercuts' - a finishing-pipeline problem - and reports 2026-era tools increasingly adding 'finishing pipelines (retopology, UV, conversion)' rather than a default. A vendor's own head-to-head, Meshy's comparison against Tencent's Hunyuan3D, states Hunyuan3D's open releases produce 'static meshes only' with 'no built-in rigging or motion library', while Meshy's own product offers 'Auto-rigging' for 'Humanoid and quadruped' characters bundled with '600+ motion presets.'

What the documents show

The trade comparison distinguishes output usable as delivered from output needing a manual stage. It reports Tripo AI generating 'clean topology ready for game engines' and separately that 'Meshy AI's generated meshes require cleanup for manufacturing' - a claim about manufacturing, not animation, and the two should not be merged. The Meshy-versus-Hunyuan3D page is Meshy's own marketing comparison, so claims about its own auto-rigging are the vendor's own figures; claims about Hunyuan3D describe a rival's release and should be checked against Hunyuan3D's own documentation before repeating as fact. Together, both draw the same line: a static prop's mesh can often ship as generated, but a character meant to deform at a joint needs genuine auto-rigging or a human retopology pass.

What it is allowed to decide

A generated static prop mesh - a chair, a crate - can plausibly support a class 2 creative-approval decision once confirmed watertight and free of non-manifold geometry; neither document is a peer-reviewed or standards source, so this is a comparison-level finding, not a certified one. A generated character mesh cannot be assumed animation-ready without stated auto-rigging or manual retopology, because Physical authority - whether the mesh deforms correctly at a joint - is exactly what raw topology does not establish by default.

The disclosure label

A disclosure label for a generated 3D asset, dated 16 September 2026, would read: this is an AI-generated mesh from a named text-to-3D tool, decision class 2 at most (creative approval) unless topology has been separately checked, holding no default Physical authority for animation until rigging or manual retopology is confirmed, and no independent Provenance authority over training data unless the vendor's licence terms are separately checked. A pipeline supervisor validating incoming assets is the plausible party asserting this label, since only someone opening the mesh can confirm its topology.

• Is this asset static set-dressing, or does it need to deform at a joint - and has that been checked?
• Does the source tool offer stated auto-rigging, or does readiness depend on a separate manual pass?
• Which claims are the vendor's own, and which are a third party's untested assertion about a rival's product?

Treated as a line rather than a ranking, this is genuinely useful: it tells a supervisor what to check before approving a generated asset, and keeps a fast-moving comparison from being mistaken for a verified specification.