dupeGuru has a great GUI for finding similar looking images (videos are not supported). See How to deduplicate files on Linux with dupeGuru for a tutorial. There's a similar tool for videos called videoduplicatefinder but I did not try it out.

I will now share my experiences with using dupeGuru. I ran it in Picture mode and selected "Match pictures of different dimensions". It was nice to see that it found a few low-quality JPEGs that look identical to higher-quality PNGs. One feature that I really like is the option to "ignore duplicates hardlinking to the same file"; let me explain why: I sometimes download the same image into multiple subfolders of my fan art collection, so I want to keep both images despite them being exact duplicates. This can be accomplished by running rdfind -makehardlinks true . to turn exact duplicates into hardlinks, and running dupeGuru afterwards.

I experimented a bit with with "Filter Hardness". On the first run, I set it to "≥95% match" and the results did not contain any false positives! I later went down to "≥70% match", which reported more duplicates, but also some false positives. If you have many comic doodles in your collection, choose at least 80%. However, if you want to detect crops of other images, you have to choose a lower percentage. On my collection of 3000 images, dupeGuru took about 5 minutes.

dupeGuru has one major downside: The current release version ignores webp images, but you can build it from source (Arch users can install dupeguru-git from the AUR). Another downside is that it automatically chooses the biggest file among all duplicates as the reference file. This is often a good heuristic but fails badly with webp images that have better compression ratio than other formats. You can fix this by going to Actions → Re-Prioritize Results and putting "Dimension: Highest" at the top of the list.