Film × AI · Calibration Instrument

What AI Cannot Do
in Filmmaking

A living calibration tool for practitioners working with AI. Eighteen limitations, six clusters, and a 4Ds practice studio.

calibration, not critique

SCROLL

The Index

Eighteen limitations. Six clusters.

0

Limitations mapped

0

Structurally impossible

—

Your avg assessment

Replication ratings are qualitative practitioner assessments, not empirical measurements. Scale: <10% Foundational human capacity · 10–20% Partial approximation possible · 20%+ AI can contribute meaningfully but not replace. Source: M. Nagy, practice-based fieldwork 2024–25.

Step 01 — Delegate STUDIO LIVE

What are you handing to AI?

Delegation is the first act of judgment — not a default. Select every creative task you intend to give to an AI system this session. The more precise you are, the sharper your analysis becomes.

Image generation

Video generation

Script development

Storyboard

Colour script

Sound design

Research support

Pitch document

0 tasks selected

Step 02 — Describe

Describe your creative intent

A prompt is the translation of vision into language a machine can approximate. The gap between what you mean and what it produces is exactly what this instrument maps.

Intent · emotional register · cultural context0 / 1500

Step 03 — Discern

Evaluate before you generate

Discernment is the practitioner's most critical faculty. The analysis below — powered by Claude — names which limitations apply to your task and what you must supply that AI cannot.

UNESCO Ethical Checklist · 2021 Recommendation

0 / 6

Human Oversight. I direct the AI — I do not follow it. Transparency. I will disclose AI's role where relevant. Cultural Integrity. Output does not flatten or appropriate cultural specificity. Subject Dignity. Represented people are held to real-subject standards. Accountability. I carry full responsibility. The system carries none. Proportionality. I use generation with awareness of its cost.

Step 04 — Diligence

Generate. Review. Iterate.

Diligence means never accepting the first output. Generate, judge it against your intent, refine, generate again. The practitioner stays in the loop at every pass.

YOUR PROMPT (FROM STEP 02)

—

Generation model

Field Notes

Your observations from practice

⌘ / Ctrl + Enter to save · stored in this browser, persists across sessions

M. NAGY · LONDON — VOL. 01 ONGOING

4Ds Framework — Dakan & Feller (CC BY-NC-SA 4.0) · UNESCO Ethics of AI (2021)
Grounding: Bhaskar (1975) · Schön (1983) · Archer (1995) · Built for BA (Hons) Digital Film Practice, MetFilm School · EdD, Bournemouth University

What AI Cannot Doin Filmmaking