Thumbnail Ai is a simple Ai tool that is able to give a rating to a Youtube Thumbnail and predict how well it will perform. It's helpful to help you maximize clicks and get maximum exposure.
Hey PH ππΌ!
I'm Yassine an indie hacker from Montreal π¨π¦.
Today I'm launching my second project of 2023 already and it's called ThumbnailAi.
It's a simple tool that allows you to get a rating on your Youtube Thumbnails and help you make them better.
I've trained the AI using TensorFlow and have been posting some updates on Twitter @ybouane.
For the frontend, it's built in low code using Sktch.io a powerful project I launched last year that allows web developers to build frontend faster than ever before in nocode or lowcode!
I hope you'll find this tool useful. As I mentioned on the website, take the rating and the tips with a grain of salt π§, Ai is a blackbox and it will definitely give you weird results especially if you try a weird picture.
Nevertheless, on the validation dataset, it got a 91% accuracy which is great!
I'm excited to hear the feedback from the community and if you want to stay up to date for my future projects, I do frequent updates on Twitter: @ybouane
how do you discern the clickability of a thumbnail against the concept and title? in other words how is 91% accuracy weighted for the thumbnail rather than just the concept of the video itself?
@matthiasiam You're definitely right that title play a big role in the clickability of a video. The idea though is that Thumbnail-Ai tries to identify if the thumbnail is eye-catchy or not. It does make many assumptions though. A pink thumbnail with an interesting title could be very clickable but rate badly using the tool so as mentioned you have got to take the rating with a grain of salt!
How's everything, people? Great job on getting Thumbnail AI on Product Hunt! You have done an outstanding job. I'm going to test it. I've only got one inquiry. How does the Thumbnail AI algorithm work?
@denis_galka Hey Denis, I built it using Keras/TensorFlow. It's a simple Neural Network that I trained using a dataset of Youtube video thumbnails that I built.
Going to enjoy testing this out. When you say it has 91% accuracy what kind of testing did you do for that? That's so cool.
I'm assuming you uploaded a series of videos and compared their performance to the predicted performance?
Nevertheless, congrats on the launch! I'm going to enjoy using this in the future.
@carter_barnett Hey Carter,
It's the validation accuracy that TensorFlow reports when training the model.
I tweeted about it here:
https://twitter.com/ybouane/stat...
It's important to remember that if your dataset is 80% class A and 20% class B and that the AI classifies everything as class A, it will have an 80% accuracy which sounds great but isn't. I'n my case, the dataset was approximately 50/50 Good thumbnails/Bad thumbnails.
SyncFlow β Notion-Webflow Sync