+The photo galleries I have at the moment, like the Panic Cell example above,
+are made by using an external script to parse XML gallery descriptions (lists
+of image filenames, with metadata such as titles), and using this to write
+IkiWiki markup into a directory which is then used as an underlay. This is a
+hack, but it works. The use of XML is left over from a previous attempt at
+solving the same problem using Django.
+
+Perhaps a better approach would be to have a setupfile option that names a
+particular underlay directory (meeting the objective of not having large
+photos under source code control) and generates a source page for each file
+in that directory during the refresh hook. The source pages could be in the
+underlay until they are edited (e.g. tagged), at which point they would be
+copied into the source-code-controlled version in the usual way.
+
+The synthetic source pages can be very simple, using the same trick as my
+[[plugins/comments]] plugin (a dedicated [[directive|ikiwiki/directives]]
+encapsulating everything the plugin needs). If the plugin automatically
+gathers information like file size, pixel size, date etc. from the images, then
+only the human-edited information and a filename reference need to be present
+in the source page; with some clever lookup rules based on the filename of
+the source page, not even the photo's filename is necessarily needed.
+
+ \[[!meta title="..."]]
+ \[[!meta date="..."]]
+ \[[!meta copyright="..."]]
+ \[[!tag ...]]
+
+ \[[!galleryimageviewer p1010001.jpg]]
+
+However, this would mean that editing tags and other metadata would require
+editing pages individually. Rather than trying to "fix" that, perhaps it would
+be better to have a special CGI interface for bulk tagging/metadata editing.
+This could even be combined with a bulk upload form (a reasonable number of
+file upload controls - maybe 20 - with metadata alongside each).
+
+Uploading multiple images is necessarily awkward due to restrictions placed on
+file upload controls by browsers for security reasons - sites like Facebook
+allow whole directories to be uploaded at the same time, but they achieve this
+by using a signed Java applet with privileged access to the user's filesystem.
+
+I've found that it's often useful to be able to force the creation time of
+photos (my camera's battery isn't very reliable, and it frequently decides that
+the date is 0000-00-00 00:00:00), so treating the \[[!meta date]] of the source
+page and the creation date of the photo as synonymous would be useful.
+
+### Images are the viewer's source - special filename extension
+
+Making the image be the source page (and generate HTML itself) would be
+possible, but I wouldn't want to generate a HTML viewer for every `.jpg` on a
+site, so either the images would have to have a special extension (awkward for
+uploads from Windows users) or the plugin would have to be able to change
+whether HTML was generated in some way (not currently possible).
+
+### Images are the viewer's source - alter `ispage()`
+
+It might be possible to hack up `ispage()` so some, but not all, images are
+considered to "be a page":
+
+* srcdir/not-a-photo.jpg → destdir/not-a-photo.jpg
+* srcdir/gallery/photo.jpg → destdir/gallery/photo/index.html
+
+Perhaps one way to do this would be for the photos to appear in a particular
+underlay directory, which would also fulfil the objective of having photos not
+be version-controlled:
+
+* srcdir/not-a-photo.jpg → destdir/not-a-photo.jpg
+* underlay/gallery/photo.jpg → destdir/gallery/photo/index.html
+
+## Proof-of-concept implementation of "viewers' source page is the gallery"