Sections 6 parts
TL;DRToo long, didn't readExpand summary
External character QA works when every review gate answers a production decision. Separate visual judgment, technical compliance, import proof, feedback severity, and final package evidence so approval moves the asset forward instead of reopening every question at once.
- Review rule
Each round should name the stage being judged, the owner of the decision, and the condition that closes the gate.
- Main risk
Open-ended taste feedback can hide blockers, contradict earlier approvals, and turn final delivery into a moving target.
- Final proof
Acceptance should evaluate the delivered package: import behavior, dependencies, source files, textures, skeleton, notes, and known exceptions.
External character reviews fail when they become open-ended taste sessions. QA should answer a narrower question: can the asset move safely into the next production stage?
That does not make visual judgement less important. It makes the review sequence clearer. Silhouette should be challenged before retopology. Topology should be challenged before baking. Import behavior should be challenged before final acceptance.
For producers, outsourcing managers, and art leads, the protocol turns feedback into decisions. It also protects the artist from contradictory notes that should have been resolved by a gate owner.
Review goal
QA should unlock the next gate
Each review round should know which production stage it is judging.
Technical blockers should be separated from quality improvements and preferences.
A final approval should evaluate the package, not only the image.
01. Review the Right Thing at the Right Gate
Each milestone should have its own acceptance criteria.
Data table
QA review gates by stage
Each gate should focus on one stage decision so feedback can close that milestone.| Gate | Review focus | Common mistake |
|---|---|---|
| Blockout | silhouette, proportion, scale, major construction | debating surface detail too early |
| Sculpt or primary model | anatomy, forms, costume logic, design fidelity | reopening concept decisions without a change request |
| Retopo and UV | deformation topology, UV layout, material IDs | approving beauty shots without wireframe or UV proof |
| Texture and materials | PBR response, map list, roughness, wear, color, masks | giving vague notes without naming the material property |
| Rig or deformation | bind pose, skin weights, morph targets, cloth, accessories | checking only a neutral pose |
| Final handoff | clean import, dependencies, naming, scale, source files | accepting a render instead of an importable package |
A review that mixes all gates at once creates churn. A review that names the gate creates progress.
Gate decision
What each review should decide
| When | Choose | Avoid |
|---|---|---|
| Blockout review Shape gate | Approve silhouette, proportion, scale, and major construction before surface detail starts. | Debating pores, fabric wear, or final material treatment. |
| Retopo and UV review Technical gate | Check deformation zones, UV layout, material IDs, and topology before baking or texturing locks in. | Approving beauty shots without wireframe and UV proof. |
| Final handoff review Acceptance gate | Validate import, dependencies, source package, textures, skeleton, and known exceptions. | Accepting a render as proof that the package is production-ready. |
02. Technical QA Checklist
Before discussing polish, check production blockers.
- Scale: asset uses the expected unit system and matches a reference object.
- Origin and pivots: character root, prop pivots, sockets, and attachments are intentional.
- Scene cleanup: no stray cameras, lights, hidden meshes, construction leftovers, or unresolved helpers in the delivery.
- Naming: mesh, material, texture, skeleton, and animation names match the agreed convention.
- Topology: deformation zones have usable edge flow; no unintended non-manifold data, loose vertices, or duplicate faces.
- UVs: layout, padding, mirroring, UDIMs, and material IDs match the spec.
- Textures: all maps are present, named, packed correctly, and imported with correct color space.
- Skeleton: hierarchy, root, bind pose, and bone names match the target pipeline.
- Skinning: shoulders, elbows, knees, neck, hands, face, cloth, and accessories deform under representative motion.
- LOD behavior: lower LODs preserve silhouette and do not introduce material or skeleton problems.
- Import test: the final export opens in a clean target scene without missing dependencies.
Technical QA should be binary where possible. Either the file imports, or it does not. Either the texture path resolves, or it does not.
03. Visual QA Checklist
After blockers are cleared, evaluate visual alignment against the approved target.
- silhouette reads from the intended camera distance
- proportions match the locked concept or approved iteration
- anatomy supports the pose and character type
- costume construction makes physical and design sense
- material families are distinguishable under neutral lighting
- roughness, metalness, subsurface, opacity, and emissive response match the intended surface
- wear, dirt, decals, damage, and patterning support the narrative without hiding form
- face, hands, hair, and signature details meet the required fidelity level
- the asset still reads when viewed in the target engine or an equivalent preview setup
Avoid notes like “make it pop” or “the vibe is off.” Name the visual property: silhouette, value grouping, edge sharpness, material response, proportion, contrast, or anatomical landmark.
04. Feedback Priority Ladder
Use severity labels so the artist knows what must be solved first.
- P0 - Blocker: prevents import, animation, review, or handoff.
- P1 - Required correction: violates approved concept, spec, or milestone acceptance criteria.
- P2 - Quality improvement: improves fidelity, polish, readability, or surface response.
- P3 - Optional exploration: subjective alternative that should not block the current gate.
Each note should include:
- screenshot or frame reference
- location on the asset
- expected change
- reason the change matters
- severity
- decision owner
If a note changes a previously approved requirement, label it as a change request. Do not hide scope changes inside normal QA.
05. Acceptance Packet for Final Review
The final review should evaluate the package, not only the artwork.
Request:
- final export files
- source scene files when included in scope
- texture maps and source textures when included in scope
- export preset or documented settings
- import screenshots from the target engine or validation scene
- wireframe, UV layout, and material ID proof when needed
- deformation proof for rigged assets
- LOD preview when LODs are in scope
- technical notes with known exceptions
This packet protects both sides. The client can validate the delivery, and the artist can show that the work meets the agreed requirements.
06. Review Anti-Patterns
- Render-only approval: a beauty render can hide import, scale, UV, skeleton, and dependency problems.
- Unowned feedback: notes arrive from many people but no one owns the final decision.
- Late concept drift: approved proportions or design features are reopened after topology or textures are locked.
- Vague adjectives: words like “premium,” “cooler,” or “more polished” are not enough without visual or technical direction.
- No severity: every comment feels equally urgent, so blockers compete with preferences.
- No change control: new scope is treated as a normal revision instead of a schedule and budget decision.
The best QA loop is not the longest one. It is the one that turns each review into a clear production decision.
If a note cannot say which gate it blocks, who owns it, and what evidence would close it, it is not ready to send as production feedback.
Related production links
Connect QA decisions to spec and delivery evidence
External sources used to ground the production guidance in this article.
Review support
Turn character review into a clear acceptance gate.
Set review ownership, technical checks, and handoff criteria so external delivery is judged by production readiness, not only taste.