From Novel to Trailer: Complete AIGC Filmmaking Workflow
Given a body of text — a novel, a script, a worldbuilding document — and the goal of turning it into an AI-generated video, what you're missing isn't "a better prompt." It's a complete production pipeline.
This article breaks down the four-layer workflow architecture I currently use: Text Layer, Anchor Layer, Production Layer, and Ledger Layer. This pipeline has been battle-tested across 82 production units for the Silent Era · The Guardian trailer.
Architecture: The Four-Layer Model
Text Layer: Novel → Trailer Script → Shot List (44 shots) → PU Breakdown (82 units)
↓
Anchor Layer: Character Anchors (7) + Scene Anchors + Object Anchors (12)
↓
Production Layer: First-Frame Gen → Preflight Gate → I2V → QA → Composite
↓
Ledger Layer: cost_log + token_log + attempt ledger + decision recordsEvery layer's output must be immediately usable as the next layer's input — no dependency on human memory or context.
The Four Layers
Text Layer
Input: 50-chapter Chinese novel. Output: Trailer script → 44 numbered shots → 82 Production Units. Each PU is an atomic unit of work: one first-frame or one animation segment.
Anchor Layer
The consistency backbone. 7 character anchors, multiple scene anchors, and 12 object anchors. Each anchor is a locked visual reference — once approved, downstream PUs reference it by ID rather than re-describing from scratch.
Production Layer
First-frame generation → Preflight Gate (12-point checklist, pass at 10) → I2V animation → QA review → merge to edit timeline. Candidate caps prevent infinite re-roll loops.
Ledger Layer
Three parallel logs: cost_log (every yuan), token_log (every session), attempt ledger (every PU attempt with outcome classification). The ledger enables pattern recognition across failures.
Key Principles
- Anchors before production — never generate character-containing shots before anchors are locked
- Single-point fixes beat group re-runs — fix one anchor, one problem at a time
- Preflight Gate is non-negotiable — the highest-ROI investment in the pipeline
- Candidate caps prevent sunk-cost spirals — when the cap is hit, change strategy, not prompts
Full Chinese original with complete layer specifications available at 从小说到预告片:AIGC影视化完整工作流