AI Generated Videos

We all hear a lot of buzz around AI and its lollapalooza. Since I am from the software field, I have some sense of the preiminent imperfections in AI and I understand why it’s not gonna take over the world anytime soon. So, I decided to take a deep dive into seemingly famous tools that are being used widely for content creation on YouTube. I reason that YT is a fad and if AI can give a commendable challenge to human generated content on YT then it indeed has a great future.

First of all, here are the videos I generated completed out of AI tools. Have a look at it and decide yourself on the state of the art quality of AI:

How to be happier

How to live a great life!

Now, to develop these videos, I used the following tools and spent less than 15 mins from start to finish.

1. Thinking of a topic: I thought of a random topic on which I could get sufficient content on the web. This part could be further automated by creating a list of generating suggestions.
2. Search text articles on the web: This will form the base of our content and provide raw taxonomy. Such as points, breakup, action items, etc are obtained.
3. Text Summarization: AI is powerful enough to summarize text into bullet points. I used rytr.me which ensures that my content will be plagiarism free and ensures novel content.
4. Stock Image Search: I used free/stock images from various websites to generate a video. Few such websites are everypixel.com, pexels.com and shutterstock.com.
5. Video Generation: By now, we have enough text and images. All we need is a tool to add stock audio and club it all together into a clip. In my previous demo videos, I used to record screens and/or narrate when needed but tools such as invideo.io do it for free.
6. Watermark or private label removal: Several times the generated video includes a photo with a watermark or unwanted personal details. A watermark remover tool is super handy in those cases. I used it to remove the InVideo’s watermark.

All the steps above could also be combined via a script so that the whole process takes even less time and zero manual invervention. Now before we go gaga about automatically generating videos and putting them on youtube to earn ads money, the biggest question to ask yourself is: Are these AI's content nearly as neat as any human-generated content?

I will let you judge that by yourself. But asked about my opinion, then in the context of the time spent (~15 mins), I think it’s insanely awesome. Maybe the production quality is low. But the expected outcomes from these videos might be to quickly show a demo video than to generate production quality content. If measured on metrics of obtained views, engagement, or any other parameter, it might flop as well. Honestly, these videos had a wow feeling for me at the start but it quickly faded away and there was little to no good feeling about generating/watching them later. Hence I wrote a blog to ensure that I generate something of value from using these tools rather than just creating videos.

For someone re-attempting what I have done, here are some frictions for which I will advise to have patience with:

1. Text summarization causes variable-length content which is a problem if one is generating a fixed-length video or no of words in a speech.
2. Content tends to be meaningless at times because of using web searched stock photos and videos.
3. Memes are very difficult to incorporate unless searches are very smart.
4. Downloading any AI tool’s source from GitHub and testing on our source images is painfully prohibitive.
5. No free lunches while using AI tools. These tools are paid/not available if used via APIs and not so productive to use without automation.