WEBVTT

NOTE AI Engineering Visualised: https://understandingdata.com/ai-engineering-visualised/image-editing-masks/

1
00:00:00.142 --> 00:00:12.829
An image edit stacks four planes, the photo, a mask where change is allowed, a reference and the instruction, to swap one white mug and nothing else.

2
00:00:12.829 --> 00:00:24.284
Trace one real edit: with a tight mask around the mug, OpenAI's gpt-image-1.5 worked for 43 seconds and handed back a blue enamel mug.

3
00:00:24.284 --> 00:00:34.615
Diff it against the original: 77 percent of the mug area changed, but so did 9 percent outside the mask, because the model repaints everything.

4
00:00:34.615 --> 00:00:43.501
Widen the mask to 22 percent of the frame, and the mug grows, the laptop shrinks, and a mask check passes half the damage.

5
00:00:43.501 --> 00:00:53.682
With no mask at all, the rest changed 4 and 8 percent, no more than the tight mask's 7 and 9: a mask steers but locks nothing.

6
00:00:53.682 --> 00:01:02.754
A reference pins down the new mug, speckles and rim included, and it drifted least, at 5 and 6 percent, but not zero.

7
00:01:02.754 --> 00:01:13.069
On OpenAI's newest model, gpt-image-2.5, under a tenth of a percent of the rest changed with the tight mask, and as little with no mask.

8
00:01:13.069 --> 00:01:24.121
Widen the mask, and the region comes back solid black, the mug set down outside it, in all six loose runs on gpt-image-2 and 2.5.

9
00:01:24.121 --> 00:01:33.615
So diff every edit, on any model, and when the rest must be exact, paste the masked region back: the only row at zero by construction.