Moving the light inside a Vermeer
This is another way I tried to bring a painting to life. A painting has its light painted in, so it never changes. I used an AI depth model to guess the shape of the face, and from that shape I worked out which way every part of it faces. That means a light can sit wherever your mouse is.

Try it. Press play, then move your mouse or finger over the painting, and the light will follow it.
The painting is Johannes Vermeer's Girl with a Pearl Earring, from about 1665. It's in the public domain.
What I learned
My first version used a video to light the painting, and the bright spot jumped around with whatever was happening in the video. It felt random, so now the light only goes where you put it.
I also learned that less light looks better. At full strength the whole painting washed out, but a dim light that fades quickly looks like a lamp held close to her face.
What it showed me
This showed me that I can add a new light to a painting and put it wherever I want. The light and shadows that Vermeer painted are still in the picture, because the new light is added on top of the painting's own colours. From a guess at the shape of the face I could work out which way every part of it faces, and that's what lets the light sit wherever you put it.
What stayed true is that the light looks best when it goes where you put it and when it stays dim. The question I still have is whether a scene in a video could move the light without it feeling random.
Next question
Could the light come from a scene in a video, like a fire or a screen, and move with it?
More experiments
How much can I control what a world model invents?
I asked a world model not to invent a bridge, and it added one anyway. When I painted the bridge out of its 360-degree view, the bridge stayed gone, but my rough fill showed up as a smudge in 3D.
I gave an AI agent one rule for a world model
I gave an AI agent seven tools and one written rule, which was no bridges outside the painting's frame. It removed three invented bridges and changed 4.99% of the 360-degree picture's pixels, but a thin black line still shows in 3D where one bridge stood.
I posed Lola like a teapot
I built a small app where you drag Lola's elbows, wrists and hands into a new pose, then press Regenerate. An image model redraws her in that pose, and the recording shows her posed like a teapot.



