Anime Inbetween Cleanup QA Ops 2025 — Designing operations that balance AI assistance and human review

Among sequential shots, inbetween cleanup is the task most likely to explode in cost. When line drift or paint slips surface during rush checks, the photography and finishing schedules get compressed end-to-end. With AI-assisted inbetween tools becoming standard, QA teams have to build Ops that combine human visual inspection with automated verification so every rush improves both quality and lead time.

TL;DR

• Generate cleanup_manifest.json per rush and spell out the risk score and priority per shot.
• Register three presets—line cleanup, paint fill supplement, output formatting—in Batch Optimizer Plus to slash rework.
• Run diff checks in Image Trust Score Simulator so metadata integrity is verified alongside visuals, giving reviewers a single source of truth.
• Log review decisions in Audit Inspector and escalate severe issues through cleanup_incident.md.
• Track Δpixel, lineGap, fillLeakRate, and reviewTime as core QA metrics, then sync weekly with Anime Color-Managed Pipeline 2025 to keep color quality aligned.
• Before each rush release, reuse the postmortem template from AI Retouch SLO 2025 and apply improvements within 24 hours.

1. Rush management grounded in risk

1.1 Shot classification

Class	Criteria	Primary risks	Response
S tier	Action / complex camera work / multi-layer line art	Line drift, paint gaps	AI assist + dual human review
A tier	Facial close-up / high screen occupancy	Micro line offsets	AI assist followed by one QA pass
B tier	Background-driven, locked camera	Paint leakage	Automated check only, spot review when flagged
C tier	Limited animation	Mass-processing slips	Batch inspection only

In cleanup_manifest.json, record the class, owner, lead-time target, and required tools for every shot. Keep it under Git to capture rush history diffs that double as audit evidence.

1.2 Shot scoring

• Line complexity (number of Bézier control points)
• Inter-frame difference (Δpixel)
• Fill closure ratio (0–1)
• Character exposure ratio

Scale each attribute to 0–100: averages of 70+ map to S/A, 40–69 to B, and 39 or below to C. The thresholds keep judging consistent across teams.

2. Automated cleanup and batch handling

2.1 Designing batch presets

• Line shaping: median_filter=1, edge_enhance=0.6
• Paint gap fill: inpaint_threshold=0.15, alpha_safe=true
• Output formatting: export_format=PNG, metadata_copy=true

Register the trio as presets inside Batch Optimizer Plus and reference them directly from cleanup_manifest.json. Store logs under logs/cleanup/*.json and have Slack fire alerts on anomalies.

2.2 Diff checks and metadata audit

• Tag diff layers (for example line, fill, noise) so reviewers can skim what changed.
• Use Image Trust Score Simulator to confirm C2PA metadata and ICC info survive the pipeline.
• Only flip status to approved within cleanup_manifest.json after QA signs off.

3. Making human review efficient

3.1 Two-stage review

Step	Owner	Goal	Exit criteria
Primary review	QA artist	Inspect AI-assisted line corrections	Diff heatmap deviation < 5%
Secondary review	Lead artist	Check alignment with directorial intent	Comments match direction notes

Log comments straight into Audit Inspector and tag them so subsequent rushes can be filtered by issue type.

3.2 Time tracking and bottleneck analysis

• Capture review start/end timestamps in Firestore or a Notion database.
• If any shot takes over five minutes, push a Slack alert so the rush manager can assign backup.
• Visualize review time in Looker Studio as a heatmap and share it with photography and finishing teams.

4. Incident response and continuous improvement

4.1 Defining incidents

• Line drift surfaced after the rush shipped
• Paint gaps that become obvious after CMYK conversion or P3 delivery
• AI inbetween tools misbehaving and causing drastic frame jumps

For each, create a severity=high record in Audit Inspector and append a timeline to cleanup_incident.md.

4.2 Postmortems

• Produce root-cause analysis and permanent fixes within 24 hours.
• Update cleanup_playbook.md with the countermeasures and brief them in the weekly QA Ops sync.
• Track remediation as CLEANUP-* Jira tickets and re-measure metrics once closed.

5. Dashboards and studio rollout

5.1 Visualizing the KPIs

• Monitor average Δpixel, fill leak rate, and P95 review time in Grafana.
• Log incident count and time-to-recover alongside AI Retouch SLO 2025 so leadership decisions are grounded.
• Aggregate cleanup_manifest.json status in Looker Studio to compare rush throughput by studio.

5.2 Sharing across production lines

• Link QA Ops procedures to background and compositing teams to avoid workflow gaps.
• Run a 90-minute onboarding session for new hires and partners, handing out cleanup-checklist.md on the spot.
• Package the quality gains into pitch decks to support next season's production budgeting.

Summary

Raising inbetween cleanup quality takes more than adopting a single tool. Risk scoring, automation presets, review logging, and incident management have to work together. Start refining the cleanup_manifest.json template today and apply QA Ops at the rush level—hitting both schedule and quality suddenly becomes realistic.