This is a draft of a patch for something similar to issue #18 -- but it was not implemented exactly according to the requirements described in that issue.

Instead of a single string containing both the positive and negative clauses, I think it would be cleaner if the additive and subtractive phrases used separate command line parameters, like:

rclip zebra --minus="black and white" --plus="red and blue"

More details are mentioned in a comment under #18 .

If you think this is a good direction, I could clean it up more (add examples to docs; and remove a no-longer-used method) and re-submit it.

Still a work in progress for #20 . Not yet ready for merging; but creating the pull request early for comments on the patches.

A couple open issues/questions:

• What user agent string (if any) should rclip use when it makes http(s) requests? I think it should have a user agent, because some websites like Wikimedia Commons are pretty picky and block requests without one ( https://meta.wikimedia.org/wiki/User-Agent_policy ). I just put in a temporary placeholder for now, mostly so wikimedia.org doesn't complain when I test with things like: ./bin/rclip.sh -f https://upload.wikimedia.org/wikipedia/commons/c/c1/Variegated_golden_frog_%28Mantella_baroni%29_Ranomafana.jpg

• So far this has only been minimally tested. I've been using this:
  ./bin/rclip.sh ./tmp/d2/img0015.jpg looking for a score of 100% from the identical picture, and this ./bin/rclip.sh -t 999 "" --sub "" --sub ./tmp/d2/img0015.jpg | tail looking at a score of -100% to make sure that adding and subtracting is working; but I haven't really tried many combinations yet. I haven't tried on Windows at all yet.

• This commit undid the optimization from the previous pull request of calculating many text vectors in a single statement. Not sure it's important. If so, I suppose it could use itertools.groupby to group the user's inputs into text vs image requests; and at least process all the text ones in a single group.

Sometimes it's more convenient to search for "a picture similar to the one I have in mind", instead of trying to describe it with words. Especially considering the extremely short phrases CLIP allows.

I'd like to be able to run:

 rclip https://i.redd.it/a7jf5rlghsz51.jpg

to return pictures of similar buildings like these.

It would also be nice to be able to search for pictures similar to a local file, like rclip /tmp/turtle.jpg and/or rclip file:///tmp/turtle.jpg.

This branch has a working work-in-progress for both of those features; but probably isn't ready for a pull request yet because it has a python3.9 dependency (str.removeprefix is a 3.9 feature). If you're interested, I'll remove that dependency and open a pull request.

E.g., like rsync has the --exclude=PATTERN that can be repeated...

I see this in main.py:

EXCLUDE_DIRS = ['@eaDir', 'node_modules', '.git']
EXCLUDE_DIR_REGEX = re.compile(r'^.+\/(' + '|'.join(re.escape(dir) for dir in EXCLUDE_DIRS) + r')(\/.+)?$')

Do you have any interest in having additional exclude directories specified as command line arguments? Or perhaps an e.g. .rclipskip, .rclipexclude, .rclipignore file that contains directory patterns?

A good example of what I mean is http://image-search.0ape.com/search?q=naked%20-person.

Thanks, @ramayer for the rclip-server and the inspiration!

Requirements:

• the current interface rclip "query" should continue to work
• it should correctly support following examples: rclip "multi word query -word1 -word2" rclip "query -'multi word negative'" rclip 'query -"multi word negative"' (i.e., it should work with any type of quotes)

rclip "multi word query -word2 some other words" should error with the suggestion to use quotes around the negative since it's unclear how to parse it

Switch to the largest OpenAI's CLIP model. This change brings quality improvements for a price of a huge performance penalty. The difference on my small test set is 16.34s vs 192.72s (+1079%) on image ingestion and 4.49s vs 7.41s (+65%) on image search.

The quality benefits are to be evaluated before merging this change.