来自聊天中的AI视觉的[grid]

[sam (Sam Saffron)]

我开始研究这个问题，AI 中的修复其实很简单，问题是我们还需要在聊天中进行修复 @zogstrip / @j.jaffeux

具体来说，一旦我们移除 [grid]，情况就会变得奇怪（tm）。

对于提供给聊天的类似这样的块：

![Create an image in a humorous illustration style, featuring a cartoon monkey with exaggerated features. The monkey is wearing a vibrant-colored Hawaiian shirt and sunglasses. The spendthrift character has a wide grin and is holding a banana as if it's a microphone. The illustration features bright, playful colors similar to those found in children's books. Set against a background depicting a tropical beach with palm trees and a blazing sunset.](upload://6mq3rDixX44cdKzsXb2efl0GfwG.jpeg) ![Generate an image of a photo-realistic humorous monkey dressed in a business suit, seated at an office desk looking serious. The desk is strewn with banana peels, and the monkey is holding a banana as if it were a telephone to its ear. The office environment should depict a sense of chaos, with papers fluttering around, creating an amusing contrast.](upload://yLUEHwfiDZFeNeNgJmyi7YzD88q.jpeg)

聊天渲染为：

这在几个方面都有点问题。

1. 文字太多
2. 它只抓取了 markdown 中第一张图片的标题，这会造成混淆
3. 点击链接时，它不会使用灯箱，这很不方便

我们可以通过添加前缀文本来避免这些错误，得到类似这样的结果：

但这感觉有点像一种 hack。

…

通过移除 grid 来修复会使情况变得更糟

---

更糟糕的是，我们需要确保在这种情况下有适当的上传引用，这样我们就不会被孤立的图片困住。