Independent visual research / Public editionReviewed 04 Sep 2026
Motion fieldnotesOpenAI film study
About this study

Presenter demo / 02:15 / Published 2026-04-21

Instruction Following with ChatGPT Images 2.0

Interview coverage alternates with extreme prompt crops and output examples that make concrete instruction details visible.

Watch official film ↗3 screen examplesAll 100 films
Representative frame from Instruction Following with ChatGPT Images 2.0
OpenAI · representative sample 01:28.000 · Source moment ↗

Useful to borrow

Put the specific constraint on screen before showing the result.

Do not infer

One clock or object arrangement is not a general accuracy benchmark.

8 shot groups.

Read across from the observed content to its treatment and possible adaptation. Enlarge the frame for detail.

Approximate study intervals; horizontal scrolling reveals all columns on small screens.

Moment / sourceReference frameWhat is on screenFraming / editUseful adaptation
Shot group 01
≈ 00:00–00:26
People / environmentWatch from here ↗
Sample 00:16.000
Interview opening

Wide and close presenter angles include a production-slate moment.

Human coverage with topic overlay.Establish the speaker before detailed examples.
Shot group 02
≈ 00:26–00:50
App / browser screenWatch from here ↗
Sample 00:36.000
Detailed prompt

A filmed laptop becomes tight crops of typed instructions.

Extreme text emphasis.Crop only the constraint the narration discusses.
Shot group 03
≈ 00:50–01:18
App / browser screenWatch from here ↗
Sample 01:12.000
Example gallery

Presenter coverage surrounds a gallery of image examples.

Browser overview between human shots.Use the gallery to show range, not to count every image as a workflow.
Shot group 04
≈ 01:18–01:30
Isolated outputWatch from here ↗
Sample 01:28.000
Clock example

A short prompt insert leads to an isolated green clock.

Input-to-clean-output cut.Hold an inspectable detail.
Shot group 05
≈ 01:30–01:54
App / browser screenWatch from here ↗
Sample 01:48.000
Spatial instruction

Another laptop insert exposes a list of object-placement constraints.

Wide input then text detail.Make the test conditions visible.
Shot group 06
≈ 01:54–02:06
Isolated outputWatch from here ↗
Sample 02:00.000
Object and pattern results

A blurred loading tile resolves into object imagery and a cat pattern.

Output-only detail sequence.Do not infer that every listed constraint was met.
Shot group 07
≈ 02:06–02:12
People / environmentWatch from here ↗
Sample 02:08.000
Presenter close

The interview returns.

Matched close angle.Keep commentary separate from evidence.
Shot group 08
≈ 02:12–02:15
Title / identityWatch from here ↗
Sample 02:12.000
End mark

White identity card.

Centered wordmark.Close simply.

Reference frame

Enlarged reference frame

Watch this moment on the official channel ↗