Purpose#
V2 Generation Pipelines govern AI-assisted creation of cosmetic variants, stage/decal candidates, and commentary lipsync assets before Bellona cook. Generated assets are content-build artifacts only and cannot influence rollback simulation.
Packages#
- Cosmetic generation plan:
@v2/isis-cosmetic-pipeline. - Commentary lipsync:
@v2/isis-commentary-lipsync-pipeline. - Base generation chain:
@isis/3d-generation,@isis/ai-texturing, and@isis/universal-rigging. - Launch-blocking originality and content gates:
@themis/music-shield,@themis/visual-shield,@themis/text-shield,@themis/video-shield, and@themis/design-shield. - Runtime model metadata and Model Card hosting:
@nous/safety. - AI-system-of-record and conformity audit export:
@themis/accountability. - V2 adapter:
@v2/eu-ai-act-surface.
EU AI Act Classification#
Risk class: Limited risk.
Disclosure text: AI-assisted V2 assets are labeled in creator and provenance surfaces before they can ship as cooked content.
Inputs#
- Approved V2 cosmetic, stage, decal, and lipsync generation requests.
- Source asset URI, source format, SHA-256 content hash, and tag metadata.
- Fighter skeleton and slot metadata for rigged cosmetics.
- Shield, anomaly, provenance, and quality-gate attestations.
Player And Creator Controls#
Creator tools can disable generated-asset suggestions. Player-facing generated cosmetics expose provenance and report controls where the asset is inspectable.
If generated metadata, shield results, or provenance are unavailable, V2 uses the manual-authored asset path and dead-letters the generated candidate before Bellona cook.
Safety And Evaluation#
- Rollback isolation: generation is content-build-only and has zero gameplay RPCs.
- Originality gate: every Themis shield result must be present and passed before cook.
- Provenance retention: source hash and AI generation metadata remain attached to the Bellona cook request and the cooked artifact.
- Audit export:
@themis/accountabilityretains the AI-system-of-record with regulator-ready evidence links.