When translating a website, Kagi Translate URL-encodes any space that appears in the srcset of an image. This is not correct, since srcset uses spaces between image candidates, and between an image candidate and its condition descriptor. These spaces need to be preserved.
The result is that any image that uses an srcset will fail to load in Kagi Translate.
For example, try to load https://habr.com/ru/companies/timeweb/articles/776040/ in Kagi Translate. All images will be missing. Now look at the first image in the original site. It's defined with:
<img src="https://habrastorage.org/r/w1560/webt/2h/l_/2y/2hl_2ykodvjs5zkodjmxix4ehke.jpeg" sizes="(max-width: 780px) 100vw, 50vw" srcset="https://habrastorage.org/r/w780/webt/2h/l_/2y/2hl_2ykodvjs5zkodjmxix4ehke.jpeg 780w,
https://habrastorage.org/r/w1560/webt/2h/l_/2y/2hl_2ykodvjs5zkodjmxix4ehke.jpeg 781w" loading="lazy" decode="async">
But Kagi Translate has transformed it into:
<img src="https://habrastorage-org-adblock.kagiproxy.com/r/w1560/webt/2h/l_/2y/2hl_2ykodvjs5zkodjmxix4ehke.jpeg" sizes="(max-width: 780px) 100vw, 50vw" srcset="https://habrastorage-org-adblock.kagiproxy.com/r/w780/webt/2h/l_/2y/2hl_2ykodvjs5zkodjmxix4ehke.jpeg%20780w,&" loading="lazy" decode="async">
So the browser will try to load https://habrastorage-org-adblock.kagiproxy.com/r/w780/webt/2h/l_/2y/2hl_2ykodvjs5zkodjmxix4ehke.jpeg%20780w,&, which of course doesn't exist.
Kagi Translate should not URL-encode spaces in the srcset attribute of images.