Import Script
AdvancedLocation: /studio/[project]/import
Script Import is a three-stage wizard — Input → Propose/Review scenes → Shot detail — that populates the project from a script. It supports pasted text, file upload (txt/fountain/md client-side; pdf/doc/docx/pptx/hwp/hwpx/odt/rtf via server extraction), whole-folder scanning (media saved as assets, images whose name matches a shot ID auto-assigned as keyframes; docs merged into text), and video-URL ingest. Extraction defaults to keyless deterministic heuristics; enabling AI deep analysis also extracts relations and emotional arcs, falling back to heuristics on failure. After approving scenes it decomposes each into shots, and the Canvas board fills as soon as data is saved.
- 1On Input, set title/genre and paste the script or choose a file (video URL or folder import also available).
- 2Optionally enable 'AI deep analysis' (if a key/CLI exists), then click 'Propose scenes' to extract a scene list.
- 3On Scenes, review the proposals (same scene IDs overwrite existing ones) and click 'Approve scenes' to save.
- 4On Shots, click 'Generate shots' to decompose per scene (skip scenes that already have shots, or regenerate all).
- 5When done, click 'Open canvas' to see results (pack JSON download also available).
Scene/ontology extraction: deterministic heuristics (keyless), or for AI deep analysis BYOK (OpenAI/Anthropic/Google) or local CLI (Codex/Claude Code). Document text extraction & video ingest: server APIs (/api/owntent/extract-text, /ingest-video). Doc-image cleanup: gpt-image via god-tibo-imagen.
In folder import, images whose filename contains a shot ID are auto-assigned as that shot's approved keyframe. Images embedded in documents (e.g. hwpx) are kept as reference assets; 'Clean with AI' re-renders subject-only transparent PNGs via the keyless imagen path. Heuristics work with no key at all, so the AI toggle is optional.