03 of 04

Machines that move

Video

When the model adds time, what starts to drift?

Video turns image generation into a temporal coherence problem. Identity, physics, camera language, and background continuity become visible by failing across frames.

A.

What we make visible

01

Frames

Video is a sequence where each frame has to relate to what came before.

02

Drift

Small changes accumulate when a system lacks a stable anchor.

03

Coherence

The same subject, setting, and physics have to persist over time.

04

Failure evidence

Hands, faces, objects, backgrounds, and camera position reveal what the model is modeling.

Fig. 03B — Drift stripsame subject, unstable time
B.

Choose a pathway

Use

Run a video test

Generate or compare A/B/C motion prompts and document what broke across time.

Observe / Critique

Label failures

Analyze pre-generated clips or still sequences for drift, physics breaks, and camera jumps.

Teach / Design

Adapt Temporal Telephone

Use drawing, whiteboards, or still frames to teach coherence without direct video generation.

Build / Code

Make drift inspectable

Prototype a frame viewer, overlay, or annotation tool for temporal failure modes.

Critical / No-AI

Ask what should count

Critique video as evidence, consent object, classroom artifact, or synthetic record.

C.

Tools for this session

Go deeper

explore on your own · studio / async
D.

Run of show · 60–90 min

0-10

Welcome and framing

Name the problem: video claims persistence across frames.

10-25

Drawing telephone

Run a fast human chain to make drift and anchoring tangible.

25-60

Temporal Telephone

Compare previous-only and anchored drawing runs, then play back the sequence.

60-80

Video Failure Gallery → report

Scrub curated failures in the Video Failure Gallery Viewer, label each failure mode, then document categories and one evidence-based claim on the Video Test Report.

80-90

Project bridge

Choose a final project pathway and name what evidence you will bring.

E.

Discussion prompts

What changed first: identity, proportion, background, or style?
Did the anchor prevent drift, or only slow it?
Which failure mode matters most in your teaching or creative context?
What would you need to trust generated video as evidence?

Low-AI / No-AI pathway

Participants can complete the session with Temporal Telephone, curated clips, or an unplugged drawing sequence. Direct video generation is optional. Open the full No-AI pathway.