Introducing Clip.fm

Short-form content is ruling the internet and the major platforms have already caught on. YouTube has shorts, Instagram has reels, Facebook has reels, Snapchat has stories, Twitter has suggested videos, and of course there’s TikTok which needs no introduction.

Problem

In order for creators who solely create long-form content (i.e. podcasts, shows, lectures) to stay relevant in the clip era, they will be forced to either (a) clip their own content or (b) hire an editor to do it for them.

(a) is very time demanding, and we’ve found that some creators spend up to five or six hours editing clips per one hour of content. Creators with momentum get bogged down with the workload and some lose motivation to create content.

(b) can be very expensive and difficult to find the right editor for your content. Telling a human editor to clip “the valuable parts” might mean different things to them than it does to you. You will have to invest time and money to train them.

So for the creators who have traction but can’t yet hire a full-time editor, clipping content is a real pain in the ass.

Clip.fm

Clip.fm is an AI tool that generates ready-to-post clips from videos or podcasts with just a YouTube URL. We use a combination of open-source transcription and text completion models to find the timestamps of the most interesting parts of a piece of content and then use these timestamps to output new clips in customizable screen orientations.

V0

We were able to build a working Python script in four days that takes in a YouTube URL and outputs AI generated clips to your local filesystem. So far about 6/10 clips we generate are accurately picking up the start/end timestamps of interesting discussion topics or stories that fall between 10-60 seconds in length (perfect for most platforms) and are ready to post with no further edits needed.

We showed some of these early clips to a podcaster in our network with 10k followers and he was so excited he offered us money on the spot to make clips for three of his podcasts. He said he would normally pay $500-1000 for this amount of work. I didn’t accept his money and told him to consider it a free trial while we work out the bugs.

To test how our clips perform online, we began posting them on YouTube, TikTok, and Instagram. In the first week of posting, we did:

TikTok: 375k views on 23 posted clips
YouTube: 62k views on 20 posted clips
Instagram: 45k views on 16 posted clips

These channels have been driving eyeballs to our landing page and we’ve had several people discover and join our waiting list without us promoting it.

For the next few weeks, our plan is to keep the process manual (we run the script locally and upload the clips to a shared Google Drive) so we can focus on tuning the model and further validate this is something people are excited to pay for. In addition, we’ve hired a college student to handle making clips for our social platforms so that we can focus on engineering and working with early customers.

Looking Ahead

For the next 1-3 weeks, our focus is continuing to work with existing creators and connecting with more who are willing to pay us for clips despite the process being so manual.

We’d like to roll-out a UI for this within the next 4-6 weeks where creators will be able to (a) signup/pay (b) input a YouTube URL, (c) select which clips they want to export, (d) make final tweaks to the duration of the clip using a crop tool, and (e) export. If we use Tesla as a metaphor, the long-term goal for this is to be fully self-driving, but for now we will keep a steering wheel and a few other manual overrides in place in case the AI doesn’t get it perfect.

In addition, growing the social channels with our own clips will continue to serve as validation and organically drive eyeballs to our site. We believe this will be our biggest growth channel.