Why it is so hard to build a AI Video Editor?

[asaayushii]

Now Sora is in town but still I can't find any platform that converts raw clips to edited,engaging clip with B-Rolls,transitions and subtitles in single click and no manual intervention. Should I think about building one?? IS IT WORTH IT? What say?

[PaulHoule]

So far you chose to fail in the exact same way that 99% of A.I. dreamers choose to fail. How is that: bad smell number one is "...single click and no manual intervention"

(1) If you hired somebody to make a video of course you would talk to them about the video you want you want made. At some point in the creation process you might look at the video and give some feedback and your video producer could adjust accordingly.

Since there are many decisions that have to made to make a video (say 20 in a one minute video), even if the system makes the right decision (for you) 90% of the time it is inevitable it will make a terrible decision if the length of the video increases.

(2) The ability to use feedback would be a secret weapon against people who choose to make their product impervious to feedback.

(3) If you get that feedback over the long term the system can use that as a training/test set and will help the system make the right decision more often.

Training data: the way out is the way through.

[asaayushii]

when AI can be a good product manager, it can also be a good video editor..the prompt enginnering and fine tuning plays a very imp role here.

To be more precise...It will make video editor and content creator task easy..not a replacement for them.

[PaulHoule]

I think it can be absolutely great but you have to be able to say "I want a different typeface for that title" and get it. If you can do that by changing the prompt, fine. If your goal though is to have the thing process 10,000 videos and post them to YouTube automatically I hope you get banned.

I have more than one "A.I. agent" that expand the range of what I can do by an order of magnitude or so but people frequently dismiss them out of hand because: (1) the agents propose actions that I approve and (2) I collect data on how well it works. If you want to use fine tuning, you need that data.

[asaayushii]

Not at all, no bulk video editing..that will miss out the real beauty of video editing. Goal is to make content creator/video editor task easy. GPT of video editor + Freedom to make changes if needed

[asaayushii]

Yeah data plays key role here..and the fine tuning ..need to see how it can be doable

[gmreads]

Klipme claims to do this exactly. I've been on the lookout for such tools, to generate reels/vlogs from my huge raw content. Happy to bounce ideas and contributing to build one.

[asaayushii]

I checked Klipme..it can generate shorts with little bit editing that only we need to do. I was looking for something where that editing(B-Rolls/transitions/subtiles/BG Sound) also get added by AI .

[gmreads]

The company behind it has more robust cloud tools too, meant for film studios. If you're serious about building this, please reach out using the email in my profile.

[asaayushii]

Do share your channel. will love to check it out

[codemusings]

> Should I think about building one??

Definitely. Seems like you have a good grasp on the problem space.

[asaayushii]

Thanks! Will soon start with the enginnering after user validation.