cineinfini-audit 0.4.10.1

CineInfini: Adaptive Multi-Stage Video Quality Audit Pipeline

🎬 CineInfini — Video Quality Auditor

The only no-reference, modular, per-shot, multi-dimensional video auditor that runs in under 1 GB of weights and exposes every gate to the user.

CineInfini scores AI-generated and natural videos across 19 native metrics and produces output compatible with VBench, VideoScore, DOVER, and FAST-VQA — so you submit the same audit to any leaderboard.

---

TL;DR

pip install cineinfini-audit
cineinfini bootstrap                                       # Tier-1 weights (~835 MB)
cineinfini audit my_video.mp4 --config cfg/profiles/postproduction.yaml
cineinfini score    output/                                # → 5 axes + composite
cineinfini export-vbench output/                           # → VBench JSON

---

What you get from one audit run

A single cineinfini audit produces all of these output formats simultaneously:

Format	What	Compatible with
data.json	Native gates + per-shot ACCEPT/REVIEW/REJECT	—
dashboard.html / .pdf / .md	Human-readable reports	—
*.vbench.json	16-dim VBench-leaderboard format	Vchitect/VBench
5-axis VideoScore + composite	Visual / Temporal / Dynamic / Alignment / Factual	TIGER-Lab/VideoScore
dover_aesthetic, dover_technical	DOVER scores	DOVER (when wired)
fastvqa_score	FAST-VQA score	FAST-VQA (when wired)

Performance profiles

cineinfini audit video.mp4 --config cfg/profiles/realtime.yaml         # < 2s/min CPU
cineinfini audit video.mp4 --config cfg/profiles/ultralight.yaml       # < 5s total
cineinfini audit video.mp4 --config cfg/profiles/postproduction.yaml   # default for serious users
cineinfini audit video.mp4 --config cfg/profiles/academic.yaml         # all 21 modules
cineinfini audit video.mp4 --config cfg/profiles/low_memory.yaml       # < 4 GB VRAM

See docs/USER_MANUAL.md for the full reference and docs/benchmarking/PROFILES.md for performance.

How CineInfini compares

Capability	VMAF	DOVER	VideoScore	VBench	EvalCrafter	CineInfini
No-reference	❌	✅	✅	✅	✅	✅
Per-shot diagnostic	❌	❌	❌	⚙️	⚙️	✅
Disk footprint	5 MB	200 MB	16 GB	10+ GB	30+ GB	835 MB
Plug-in extensibility	❌	❌	❌	❌	❌	✅
YAML-driven config	❌	⚙️	❌	❌	❌	✅
Causal/physics modules	❌	❌	⚙️	VBench-2.0 only	⚙️	✅

Full comparison: docs/benchmarking/COMPARISON.md.

---

Installation

# Minimal — pure-CV modules only
pip install cineinfini-audit

# With ML modules (CLIP, ArcFace, DINOv2)
cineinfini bootstrap                            # ~835 MB models

# Full — including DOVER, FAST-VQA cross-benchmark wrappers
pip install torch torchvision dover-vqa fast-vqa
cineinfini bootstrap --include-optional         # +395 MB

Detailed guide: docs/INSTALLATION.md.

---

Documentation index

Doc	What
docs/USER_MANUAL.md	Complete user manual — every CLI command, module, profile, output
docs/QUICKSTART.md	5-minute getting started
docs/INSTALLATION.md	Install + competitor dependencies
docs/STATUS.md	Exhaustive status: done, not done, what to install
docs/MISSING_ASSETS.md	Tier 1/2/3 asset inventory + training recipes
docs/benchmarking/COMPARISON.md	Head-to-head vs 12 industrial/academic tools
docs/benchmarking/COMPETITORS.md	One-liner install for each competitor
docs/benchmarking/PROFILES.md	Per-profile performance characteristics
notebooks/	Runnable Jupyter notebooks

---

Citation

@software{cineinfini,
  author  = {Benbrahim, Salah-Eddine},
  title   = {CineInfini: Video Quality Auditor},
  year    = {2026},
  url     = {https://github.com/CineInfini/Video_Quality_Auditor},
  version = {0.4.10.1}
}

License

MIT. See LICENSE.