Azure
Платформа и инфраструктура облачных вычислений для создания и развертывания приложений и служб, а также для и управления ими через глобальную сеть центров обработки данных, управляемых корпорацией Майкрософт.
MAI-Image-2.5-Flash: Tennis Racket Handling Issue
![](data:image/svg+xml, %3Csvg xmlns='http://www.w3.org/2000/svg' height='64' class='font-weight-bold' style='font: 600 30.11764705882353px "SegoeUI", Arial' width='64'%3E%3Ccircle fill='hsl(134.4, 31%, 23%)' cx='32' cy='32' r='32' /%3E%3Ctext x='50%25' y='55%25' dominant-baseline='middle' text-anchor='middle' fill='%23FFF' %3EV%3C/text%3E%3C/svg%3E)
Vanya
0
Баллы репутации
Bug report
This happens in Bing Image Creator. I haven’t checked it in Playground.
I’m currently testing my prompt-generation bot across different topics, and I decided to try tennis. Surprisingly, I noticed an interesting issue with MAI-Image-2.5-Flash.
DALL-E often struggled to correctly hold weapons, but MAI has no such problem. However, it sometimes struggles with other objects. A clear example is a tennis racket: in some variants, the model doesn’t understand how to hold it with both hands, placing both hands at the same point instead of one above the other. In other cases, it creates a third arm.
The prompt included two hints related to the tennis racket: “holding tennis racket” and “ready for forehand”.
Azure
Войдите в систему, чтобы ответить