Show HN: I built an AI video filter to watch R-rated shows with my family (clipfilter.com)

I wanted to watch Peep Show (really funny British comedy series ever, free to watch on youtube, season 1 here: https://www.youtube.com/watch?v=wXX4XJzsqQ8&list=PLAbD_h_-yg...) with my family, and lots of other shows/movies that are just too explicit for their taste, both in terms of language as well as visual elements (violence, sex, etc). Couldn’t find a clean version or any good software that could clean up the content for me, so I built it for personal use, and now have thrown it up on ClipFilter.com in case anyone else would find it useful.

How it works: Two options - Target rating: choose the rating level you want: G, PG, PG-13, R — each one has different presets for what kind of language and visual elements to filter out - Custom: design your settings from scratch for each of these elements: violence,

Right now only one option to remove visually inappropriate elements: 1. Skip

But realized some scenes that are critical for plot were rendered incomprehensible with all the skipping, so also working on adding: 2. Blur

Underlying technology, simply put - for language, speech is converted to text; when a word in the blocklist is found it identifies the timestamps at beginning and end of word and mutes audio during - for visual elements, really poor experience with a content moderation apis, especially OpenAI's free endpoint.Instead, frames are fed to vision LLM along with requirements for what is inappropriate and what is not based on user input; frames that are found noncompliant are flagged and either skipped.

Because I am using OpenAI APIs I can’t make it totally free but all signups get 5 free minutes of content filtering on signup, and there's no verification on signup yet so if you're enjoying and run out of credits feel free to make a new signup.

Always looking for feedback on both the functionality/output quality as well as pricing.

Don’t expect there to be a ton of demand for this use case anyway but was a fun project and might be some similar use cases out there so eager to hear any and all thoughts!