A few questions: Is this compatible with existng solutions like Mastodon? And why just images? Why not videos/documents/music or even events? How long will the images be available? Who is paying for storage? How does the technology compare to IPFS?
Pixelfed is fairly new, and probably should get it's minimum viable product working before worrying about supporting more formats. And there's a lot of ActivityPub-based software out there, it makes sense to focus on a specific experience people are looking for.
You host Pixelfed where you want to, or use someone else's Pixelfed server. As long as either you or someone else is paying to host it, it's there.
ActivityPub style federation isn't like IPFS. Data is largely still contained on the server it's posted on, but you can communicate with other servers so that people can be notified, interact, and comment on things posted on other servers. But it's a much more traditional server model than IPFS. When a Mastodon or Pixelfed server goes offline, it's content is gone.