Hacker News new | past | comments | ask | show | jobs | submit login
Show HN: I created a reverse search engine to find the source of anime clips. (cleithral.com)
41 points by Mrvolcano 6 months ago | hide | past | favorite | 22 comments
Hi, my name is Rayn. I am a solo independent entrepreneur. Initially, my only reason for learning was to make tons of money, but after completing the learning process, I was very confused about what my first project on the internet should be. I started browsing the internet for ideas but got bored and wasted time on social media. I noticed that in the comment sections of anime clips, many people were asking for the anime name and episode. That's when I thought, "Let's make an anime reverse search engine," and boom, Cleithral was born. Just upload a video, image, or gif, and it will give you the anime name and episode number.

To be honest, I chose the name Cleithral because I had a failed Shopify shop with that domain name, so I just reused it. That's the story behind my first project.




For reference another site that does this with images is https://trace.moe/


Looks like this site just uses trace.moe, the video link it outputs is at "api.trace.moe"


You are correct, it does use the trace.moe API. I can't afford my own RTX A100 to train a model, so I just use trace.moe's API.


You don't need an A100, you don't even need a GPU.

Trace.moe compares colors and patterns.


I actually used https://trace.moe/ api for this website.


Mmm. I showed it Frieren and it gave me Nanbaka. The scene sure looks a lot like her, but it wasn't her


It's not perfect. It's just a free, simple tool. Also, it will only work if the image you upload is from an anime episode, fan art, or a picture from Google. The uploaded image needs to be from the anime episode. This is actually my first project that i put live on internet.


Thanks for trying it.


It's good! I like anime a lot, and also use it for flashcard training in Anki. So I view an average of 100 short clips/day, ranging from the very popular to the fairly obscure. It got most still frames I tried it out on, and did OK with deliberate fakes (eg stills from ordinary movies, where it found fairly close approximations to the source image). Missed a few submissions completely, if I was picking for very obscure/forgotten anime, but also succeeded with some that I didn't expect.

Minor bug: name and description don't update properly if you upload successive clips/images.


Alright fixed it. For some reason the state of one of my component wasn't updating but it's fixed now


Might work even better as a browser extension.


Thanks for the suggestion. I might have a logic about how the extension should work.


> “ I like anime a lot, and also use it for flashcard training in Anki.”

Sorry, I’m not sure I understand this sentence; what do you mean when you “use it for Flashcard training” ?


Anki is a flashcard system that makes use of spaced repetition to help one learn a topic.

I came across a project/service that used Japanese anime screenshots to teach the language, and incorporated it with Anki cards. Not sure if absoluteunit1 meant that project by what he said.


Ah I see; initially I didn’t understand what you were using Anki for in relation to anime. It’s to learn Japanese; got it


Seems to just use trace.moe ? And begs for donations on every request. Did you really made your own models or are you just using an api and try to make money on their back ?


Yes, this website does use trace.moe. I believe trace.moe prohibits commercial usage of their API, but since I am only asking for donations, I don't think it can be considered commercial use, so I would say I'm good.

Also, I know Python and the Hugging Face transformer API, so I could build my own model. However, I didn't because this is my first project that I put live on the internet, and I don't have an RTX A100 nor can I afford to rent one, so I wanted to keep it simple. Besides, if an API for something I want already exists, why should I build my own?


Guys! I made a search engine! Hey it just uses a Google search API.. Yep!


Wonder what corpus this was trained on or what it sources from.


I use the trace.moe free api for this.


A lot of nyaa.si torrents


I use LLM’s for product and domain names

see if you can get inspiration from those things




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: