Changelog

All notable changes to modelBridge.ai are documented here.

v0.13 — Category Gating & E2E Certification (March 2026)

Three-tier category gating — ALLOWED (11 categories), WARNING (reserved), BLOCKED (13 categories). Five independent gates prevent unsupported models at every entry point
Fail-safe unknown category handling — new categories from fal.ai are automatically blocked until explicitly allowlisted
End-to-end verification for all 11 supported categories — full pipeline from search to cost log
Audio pipeline — dedicated audio track import with 3-tier track selection (targeted empty > any empty > auto-create)

Defensive filter in saveCustomModels() preventing blocked-category models from persisting

101-check category gating audit — zero failures
277 total tests across 6 automated suites (94), 6 manual suites (82), and category audit (101)

Active Generations Panel — persistent panel tracking all background generations with status dot, model name, elapsed timer, step label, and expand chevron per row
5-step progress tracker — Sent → Queued → Generating → Downloading → Importing with real-time updates and fal.ai queue position
Input details in expanded view — prompt, duration, resolution, aspect ratio, endpoint displayed per generation
Auto-dismiss on navigation — navigating to a completed model’s card auto-removes the background row
Failure handling — failed rows stay visible with red state and “See error” link, never auto-dismissed
Snackbar notifications — floating success/failure toasts with quick-action buttons

Four-layer pricing pipeline — curated supplements for exact pricing, official API for base rates, family heuristics for related models, honest “unavailable” fallback
Three cost badge states — Estimated (live parameter updates), From (base rate with disclaimer), Unavailable (no data)
Context-aware timeline import — one button with automatic routing: source-based models replace in-place, text-based insert at playhead, audio inserts on audio tracks
Contextual import instructions — each preview card shows what the import will do before you click
Audio preview player — play TTS and sound effect results directly in the preview card before importing
Selection-driven Dual Mode results — clickable preview cards with active-state border, import per slot
Same-source-clip handling in Dual Mode — second slot auto-switches from “replace” to “insert at playhead” after the first imports

Preview buttons simplified from 4 to 3: “Save and import to timeline”, “Save to project bin”, “Discard”

Three-layer error handling — Prevent (schema-driven preflight) → Learn (self-improving constraint cache) → Translate (human-readable messages with WHAT / WHY / WHAT TO DO)
Self-learning validation — full Learn → Read → Enforce chain for 6 constraint types plus generic fallback parser
Dual persistence for learned constraints — localStorage + disk file survives cache clears and updates
Adjacent clip detection — real-time adjacency detection for interpolation models
Direct two-clip timeline import — generated video replaces both source stills spanning combined duration
Server resilience — 90-second heartbeat, automatic crash recovery, pre-flight health checks, exponential backoff
Schema-driven file parameter detection — zero hardcoded model names, priority cascade from schema structural signals
Image dimension fallback — reads from file via browser Image API when Premiere Pro returns 0x0 metadata

Dual-frame selection card — dual-column UI for models accepting start + end frames with Required/Optional badges per slot
Unified media card design — all media-requiring models share the same card component
Drag-and-drop from Finder — drop files directly onto media cards, single/dual-slot modes, mixed mode, pinned files
Mask editor with zoom/pan — adjustable brush size and opacity, eraser mode, zoom and pan
Source Monitor preview with In/Out point selection before committing results
11-stage generation progress with background recovery polling up to 30 minutes
Prompt optimization — one-click AI rewrite of prompts
“New” filter chip and NEW badges — models added within 14 days highlighted in search
Commercial license filter — filter chip + badge sourced from API metadata

Intelligent clip ordering — dual-frame models respect selection order with info notice when order differs from timeline position

Real-time cost tracking per project and client with CSV export
Multi-currency support — USD, EUR, SEK with pre-generation estimates
Error translation system — raw API errors translated to plain-language recovery guidance
HTML report generator — self-contained offline reports with Chart.js, 4 KPIs, cost-by-model and cost-by-deliverable charts, compliance badge
External AI costs via CSV import for third-party platforms (Midjourney, RunwayML, etc.)
Deliverable tagging per generation for cost breakdown by output

Dynamic form generation from OpenAPI schemas — 500+ models, zero hardcoded interfaces
Two-layer input validation preventing wasted spend
Continuous media selection monitoring with contextual feedback