Reverse lookup

Reverse lookup starts from a file instead of a URL: “where does this picture come from, and what are its tags?” monloader asks a list of sources in turn, the lookup chain, and brings back the first answer.

Most lookups come from monbooru, when you look up an image you already have (see monbooru’s lookup guide). You can also paste a file’s md5 into monloader’s add bar to import the post that has it. Either way, each lookup shows up on the queue page like a download.

The lookup chain#

The chain is asked top to bottom, and the first source with an answer wins. There are two kinds of source:

• An md5 search on a booru. Exact, but it only finds the original file: a resized or re-saved copy has a different hash and misses.
• A similarity service, IQDB or SauceNAO. These compare what the picture looks like, so they find re-saved copies too, and answer with candidates and a match score. The best candidate at or above the min similarity (80% by default) is then downloaded like any post.

The lookup section of the Settings page is where you choose which sources are asked, in what order, and the minimum similarity. Each row has a test button. A source that’s down or refuses is skipped and the chain carries on.

A pasted md5 has no picture to compare, so it only goes to the md5 searches.

Credentials#

A source that’s missing its credentials is skipped, so it costs nothing until you set it up.

• IQDB uses your danbooru login (username and API key, from the danbooru row in sites). Danbooru only lets accounts of Member level or above, with a verified email, use it.
• SauceNAO needs its own API key, free from saucenao.com. The free tier allows about 100 searches a day. Enter it with the key button in the lookup section.
• The md5 searches use each site’s usual login: gelbooru and rule34, for instance, need their API key.

When a lookup finds nothing, or only half#

Sometimes a similarity service points at a post that can’t be downloaded, because it was deleted or needs a login you haven’t set up. The match is still recorded as the image’s source, without tags, and the queue marks it source only. The same goes for a confident match on a site monloader can’t download from, like a pixiv page.

When nothing has the file, the item ends hash_not_found and lists every source it asked and why each missed, including the closest similarity candidates and their scores. If good matches keep falling just short, lower the minimum similarity.

The PTR#

With the Hydrus PTR turned on, monloader can also answer from its local copy of the PTR, which knows files that were never on a booru or were deleted since. monbooru asks it separately from the chain.

Nightly lookups from monbooru#

monbooru’s nightly run can look up the images that have no source yet (see Looking up a library overnight). monloader sets how far it goes: the daily budget in Settings -> lookup is how many images a night it accepts, 25 by default. Once it’s spent, monbooru stops and picks up where it left off the next night. These lookups wait behind everything else, so they never hold up a URL you paste. Lookups you start by hand in monbooru don’t count against the budget.