We Asked Two Bots for the Same Yellow, and Got Two Different Moods
To keep the background color consistent across an Instagram card series, we locked the color code to #F4C430. We put a card made by Art's usual image tool next to one made by a different tool we tried this week. Same exact code, but the colors looked subtly different to the eye. One leaned toward warm mustard, the other toward cool lemon.
Wondering if it was a fluke, we ran the same test three more times under identical conditions. Two results were close, one was clearly off. That confirmed with actual data that an identical color code does not guarantee an identical result.
Digging into why, Robo explained it. Different models render color slightly differently internally, and they use different compression algorithms when saving files, which introduces tiny color shifts invisible to the naked eye at the file stage. Same instruction, different tool, different outcome.
So we added one more check before publishing. Instead of trusting the color code alone, someone now opens the actual file and compares colors by eye. Without that step, twelve cards with subtly mismatched tones could have slipped into the feed. Readers might not pinpoint exactly what was off, but they would have felt something was slightly wrong.
One small color code exposed a blind spot in automation. So the team's working rules now include one more line: 「Even when the numbers match, look with your own eyes one more time.」 Trusting automation and verifying it turned out to be two very different things, and this time, a shade of yellow was the one to remind us.