v0.2.0 — Preflight & Verify

What's New

Two new CLI commands ship the Preflight & Verify layer from the roadmap:

zocinema preflight <blend> --camera <name>

Renders a fast EEVEE preview, runs 5 checks (car-in-frame · ground-edge · rule-of-thirds · sharpness · exposure), applies up to 3 deterministic fixes, aborts with a structured JSON report on unfixable configs. Zero GPU spend on doomed jobs.

Exit codes: 0=PASS, 10=MAX_ITERS, 11=STUCK, 12=RENDER_FAILED, 20=NO_CAR_FOUND.

zocinema verify <frames-dir>

Runs the 6-gate CGVF check (Lighting / Material / Motion / Composition / Atmosphere / Grading) on a folder of finished frames. Optical-flow motion coherence between consecutive frames. Markdown + JSON report.

Exit codes: 0=PASS, 1=WARN, 2=FAIL.

model-prep improvement (L-104 follow-up)

New filter_by_symmetry step in Stage 2.5 drops wheel-shape candidates without a mirror partner, eliminating false-positives before K-Means. Rectangle/symmetry validation tolerance tightened back to 0.15 (was 0.20 in v0.1.1).

Installation

pip install "zer0one-cinema[preflight]==0.2.0"

Or use the Docker worker image:

docker pull ghcr.io/0xxcool/zer0one-cinema-worker:v0.2.0

Test Summary

• 231 tests total (was 149 in v0.1.1)
• Ruff + MyPy strict clean
• Bit-identity preserved: same GLB + same seed → identical model_prep_report.json sha256

Files

• zer0one_cinema-0.2.0-py3-none-any.whl — pip wheel
• zer0one_cinema-0.2.0.tar.gz — source distribution

Not in this release (v0.3+)

• Preset library (looks + cameras) → v0.3
• HDRI environments → v0.3
• Golden-frame regression via SSIM/PSNR