A compilation of some limitations I found with Google Lens (mostly Manga-related)

[Mordraug666]

Tested on version 0.7.2, later versions might have fixed this already, thank you if so!:)

Hey there! First of all, thank you so much for this incredible program, it's truly a game-changer! However, with Manga in particular, there are still some issues that I'd like to share just in case (it's probably well known, but hey...). Google Lens in particular (since Paddle isn't very usable anyway). I'll share some pictures that show it on some random stuff/panels I tested/was notified about by somebody else.

With Manga (vertical text), it tends to flip the text of the bubble or part of the text of the bubble from the typical right-to-left direction into a left-to-right direction which can be a bit annoying/confusing (especially when only part of it changes like this). The scanning accuracy of the Kanji and kana is fine though otherwise. The styling can be different from the text too (especially the space between characters, exclamation points, dots etc. don't always match the original style, though that's a smaller issue mostly).

Then, I found that if a bubble has these weird big dots next to the characters, it will completely destroy the scanner and will produce a mess by giving some massive characters and weirdness, sometimes it keeps part of the proper bubble text on the sides, but it's a mess nonetheless. This also happens with some more stylised wacky text that has layers, like in Persona 5. And yep, sometimes the output gets too massive (like it can go outside the screen).

And lastly, not sure if there can be anything done with this, but when there's furigana, it can be a bit confusing in the output window, sometimes it works fine, but sometimes the furigana is misplaced. It'd be better if furigana was ignored outright (like how mostly Mokuro has it).

Anyhow, not sure if anything can be done with it, but I heard the 0.8 version will bring improvements in this, so I'm pumped!
Anyway, this is arguably the only real limitation with Yomininja I found - Manga (especially with furigana and weird dots next to lines) and vertical text (and wacky/stylised text that has layers of small and big font). Horizontal text with Google Lens works really well. And obviously, possibly the fact that it's not suited for stuff you can't pause, like cutscenes (but I heard there's now an autoOCR with an option to copy to a text-hooker page or something? That would solve this limitation in terms of not being able to reliably scan/look-up video stuff that can't be paused.

PS: Though I'm impressed that Lens was almost able to recognise this low-res, small blob on the top panel without even zooming in on it. It got the beginning Kanji and end right, just fumbled the middle (and missing おしらせ). Impressive.

[Mordraug666]

Hm, this is sus, ngl:D Careful anyone, right after I uploaded this too. Don't click or download stuff from randoms. Uploaded from Russia yesterday, hm... DEFINITELY not sus. The profile I reference in the picture doesn't exist anymore, heh. Now a new account appeared doing the same, which seems to have been hacked from someone who downloaded it since I found this ICDM-UESTC/MMRA#3, where the exact same comment is, and the poor fellow responded showing he downloaded it and likely ran it... yeah, I don't believe this comment/link. Careful, again!

EDIT: The profile and the thread is now gone, at least GitHub is fast about it;D Let's see if more will come.

And now my stuff is closed, great, why does Github have it that you close stuff upon commenting... egh. Okay, reopened:D