Pixel-style field portrait of Bai Ziye operating a cinema rig, used as the public identity image for film-trained AI production control.

Pixel-style portrait of Bai Ziye operating a cinema rig in a warm yellow field.

Beijing Film Academy direction x AI-native execution

Film-trained AI production control.

I translate director judgment into source selection, AI video shot packages, agent handoff, retake gates, artifact review, and lesson writeback for non-engineer operators.

我把导演判断、影像生产、AI coding agent 和产物验收接成一条可审查的控制链。

Start from the failure

Three copyable routes from real breakdowns.

Three public method cards showing AI video continuity locks, DirectorShotIR, and Project Clinic routing.

Public method library

Copyable control surfaces for real failures.

Five continuity locks

Character, space, prop, action, and emotion locks for AI video retake decisions.

DirectorShotIR

Performance, blocking, camera, transition, asset roles, review rule, retake target.

Project Clinic

Intent, source, capability route, first safe action, review checklist, writeback.

Video admission ladder

Raw artifact, static evidence, playback, relation checks, owner seal.

Case tracks

Synthetic examples that show structure, not private outcomes.

The public cases explain structure, control decisions, and review boundaries. They do not publish private scripts, raw media, provider state, customer material, or final acceptance truth.

Copy-ready starting points

Start with a route, then copy a blank tool.

GitHub proof appendix

Supporting repos, not the headline.

Public discovery loop

Star the library if it saves a false start.

The useful part is the public control language: AI video continuity, DirectorShotIR, Project Clinic, workflow rescue, and video admission.

Star as bookmark