i2t2i

Mona Lisa

The picture tipped at round 20 and stayed there.

The starting image
start
After 105 rounds
after 105 rounds
Fast: every round, 15 images a second.
Slow: every round for one second, morphing into the next.

What happened

Round 1 already loses the smile: the first description calls it a smile, and no description after it does.

For the next 17 rounds the picture stays an oil painting of a woman with folded hands in front of hills and a hazy landscape. The colours get greener and darker, the face more generic.

Between rounds 18 and 21 the landscape and the oil paint disappear. In their place: a plain studio background, a ribbed knit sweater and a table under her hands. The image is now a modern photograph of a young woman.

From round 22 to 105 nothing more happens to the picture. The descriptions keep rewording the same portrait; only the sweater and the background slowly turn from green and grey to brown.

Every description, round by round, and where the start was lost

How far the words moved

round 0round 104

Each bar is one round: how far its description has moved from the first one. The red bar is round 20, where it tipped.

Words that left, words that arrived

hills
landscape
desaturated
atmospheric
pale
uniform
ribbed
knit
sweater
camera

Point at a column to see that round's picture. A filled cell means the word appears in that round's description. Ochre words belong to the start and fade out; blue words arrive and stay.

Start, tip, after, end

Round 0
start
Round 20
round 20, the tip
Round 25
round 25
Round 105
round 105, the end

Details

date
2026-10-24
rounds
105 of 105
describer
openai gpt-5-mini
describe tokens
235,935 in, 109,503 out
painter
gpt-image-1-mini, quality low, moderation auto
paint tokens
100,740 in, 42,840 out
images
105 at 1024×1536
cost
about $1.46