Research note

Inspect a video capability across time

An original evaluation note for checking identity, object state and requested changes rather than judging a generated clip from one attractive frame.

4 min estimated readStart hereEvidence baseline

THE SHORT ANSWER

Temporal control needs observations across a sequence; a still image cannot establish it.

Read controls as claims

Runway’s archived post is a brief invitation to try a video feature; its attached video was not watched here. As a separate current primary example, DeepMind’s Veo page describes reference images and clip-extension controls. Those descriptions establish what the publisher presents. They do not verify continuity, prompt adherence or comparative quality for a particular generated sequence.

Specify observable continuity

Create an evaluation brief with one object, one action and one requested change. Define what must persist, such as object count, color and relative position, and what may change, such as camera angle. Inspect a sequence at regular intervals and around occlusion or contact events. Record each break with a timestamp so another reviewer can locate the same evidence.

Report the whole sample

Retain unsuccessful outputs when a later authorized experiment is run. Distinguish a selected showcase from the full attempt set, and review audio separately if it matters to the task. This page supplies a technical evaluation method, not production direction or a claim that any clip passed it. No video was generated or reviewed. Use the checklist to define a later production test.

Put it to work.

  • Define persistence and change.
  • Check contact and occlusion.
  • Timestamp observed breaks.
  • Keep the full attempt set.

Go to the source.

Original material, with its own scope and publication date. This note is an explanation, not a CooScope performance test.

Source discovery: @runwayml

Source published · Editorial review

Archived English post text; primary sources reviewed separately. Generated annotations, media and missing replies are outside the evidence scope.

  1. @runwayml — archived X post (2025-09-09)
  2. Google DeepMind — Veo capability descriptionsLink / source review: 2026-10-15
Evidence baseline and source-review dates are different. A recent link check does not make an older study a current product test. Read our method →

Keep the thread going.

Explore this topic →