OpusClip - Experience

[Mustak3]

Hey everyone! I’ve been thinking about starting content creation on YouTube, TikTok, and Instagram by posting clips from podcasts like "The Joe Rogan Experience," "Modern Wisdom," "The Iced Coffee Hour," "Full Send Podcast," and others.

The clipping process will be probably very time-consuming since I'll need to watch the full episodes and manually select the clips. I was thinking of using an AI tool called "OpusClip."

Has anyone here used it? How are the views and monetization with this app?

 

[Opulentos]

Opus is good, used it for shorts. It finds clips for you and cuts it up for you and adds subtitles

 

[hadeyjumor]

I used Opus. and my channel got monetized from YT. It's a good tool. you can use it.

 

[Spicy Chromosomes]

The app is good, the idea is a bit saturated

 

[Ryan_2]

Opus is actually a very nice tool

 

[Mustak3]

so i decided to create my own app using ai. Curretly finishing it and i might publish it for free for everybody to use!
Created the interface and now working on generating good captions for posting. I made it run on my pc(local network) using my gpu and creating. Creating 5 shorts from a video of 30minutes in like 6min, running in a problem where the captions are pretty much a transcript of the video.

open to any suggestions!

 

[Gants-P]

so i decided to create my own app using ai. Curretly finishing it and i might publish it for free for everybody to use!
Created the interface and now working on generating good captions for posting. I made it run on my pc(local network) using my gpu and creating. Creating 5 shorts from a video of 30minutes in like 6min, running in a problem where the captions are pretty much a transcript of the video.

open to any suggestions!

Use lavllama, it has vision and can extract the scenes into captions.

It's wonky but there's different flavours with the text encoder.

You can use in your CSV file with regax to catch fulltops, or even srt, then you use ffmpg to cut it, so when a full stop appears in text it's a new clip


It's all easy said with out providing the script.

 

[Mustak3]

Use lavllama, it has vision and can extract the scenes into captions.

It's wonky but there's different flavours with the text encoder.

You can use in your CSV file with regax to catch fulltops, or even srt, then you use ffmpg to cut it, so when a full stop appears in text it's a new clip


It's all easy said with out providing the script.

I am currently using Gradio . app as a machine learning software but it's a bit slow.

 

[Gants-P]

I am currently using Gradio . app as a machine learning software but it's a bit slow. View attachment 432758View attachment 432759

Yes gradio has overhead, it's easier in command line, but people here would prefer it in gradio.

But you need to learn ffmpg it can stitch 50 videos in seconds, cut it, and based on the text in the file it can add a overlay and cut right when there's a full stop, so your guaranteed a cut without any edits, you can even pull videos down from YouTube and have ffmpg turn them into PNG, then wash them then rebuild them then repeat 4 or 5 folders then stitch them all together.

Then you can add a voice over to read the entire stitched video, and once done you cut it like a sausage with either the person's voice, or the srt then you can do what ever you want with it.

So that's a simplification but it's automated run and just monitor it in command line, and if you want to edit it, sure ad it in capcut.


I still recommend you keep going with your project you've already dedicated enough time.


But you'll need: https://github.com/haotian-liu/LLaVA

To complete the pipeline this will help

While these projects are dated they give you a idea their purpose on vision and text labeling:

https://github.com/camenduru/LLaVA-colab

Picture the vision as ai with a voice looking and assessing the video, it does it in text, so your getting text, and a layer of voice for free through vision recognition, now I don't know the overhead when using gradio but it could be 5% 11%

 

[Mustak3]

I am using Ffmpeg because for automated short‐form clipping MoviePy + FFmpeg is enough, but i am using it just indirectly via MoviePy’s Python interface.
This is how the videos get generated but still working on the title and caption to be better because now they are a "transcript" style

 

[Gants-P]

I am using Ffmpeg because for automated short‐form clipping MoviePy + FFmpeg is enough, but i am using it just indirectly via MoviePy’s Python interface.
This is how the videos get generated but still working on the title and caption to be better because now they are a "transcript" style

View attachment 432762

If you want to make it unique like, I see, there two people talking, one with no hair, the other with glasses, and you can tweak the llama model to narrate it, well something to consider,

Well if you wish to bake the text to the video you'll need to be aware of fon't ratio scaling, and colour bleeding.
And aspect ratio

You may want to add style fonts in,

 

[Mustak3]

Because I am not that skilled into this sector and have not worked a lot in it. I am using ChatGPT o3-mini-high to program the whole thing with overlooking it myself because I have a background in this. I made it look a bit more tight and a easier to understand layout removed the unnecessary things.
Any recommendations on which AI I can use to get a better coding assistance?