I’ve been experimenting with auto-generated and AI-generated video.
This section is just my explorations of the current state of the technology, as I begin thinking about how I might use this in my workflow.
Everything below is nothing more than my experiments so far, and my thoughts about what works well, and what doesn't.
As everyone is aware, we are just at the beginning of this technology. Advances are happening very quickly. But this is what I’ve discovered so far. And my opinion about the results is that they are … mixed.
An experiment in AI-generated avatars, using an AI-generated voice, and a version with my own voice.
Broadly speaking, I've created two types of videos:
Type 1:
AUTO-GENERATED (PROMPT-DRIVEN)
There are two versions:
Prompt-Example 1: Prompted AI to create a video based on a topic
Prompt-Example 2: Took results from example 1 but supplied much of the content with better content than the AI came up with in the original.
Type 2:
AI-GENERATED (MY SUPPLIED TEXT)
There are two versions:
AI-Generated Example 1: AI generated text/VO, images, videos, etc.
AI-Generated Example 2: I provided the script, and AI created VO, videos, images, etc.
I prompted an AI presentation creation tool with a title, and had it create a presentation based on that.
I chose a template from the available options, and made very minor modifications to the resulting deck.
This example is the second version I created, after the first version (the example to the right) where I used an AI-generated voice. As with most things AI - the results weren't great, so I thought my own voice would be better.
This videos uses:
In his video was the first attempt, which is 100% AI:
First, here are two examples of auto-generated video, based on supplied text. I let the video generator “decide” how to divide up the text.
The system then:
Note also, the sound on these is not good, as these are only screen recordings, taken to capture these first attempts, these are not downloaded video files.)
Example one of prompt-based video. I prompted AI to generate a video, based on a simple topic.
Results:
I found a few of the visuals to be somewhat appropriate, but most seemed to be completely random, and had nothing to do with the supplied text. (TBH, some were absolutely bizarre.)
I also found the narration to be unusable, repeating the same cadence over and over.
For the second one, I began with the first version, but then I:
Results:
The results are “better”, but I’m not sure that there’s a big advantage at this point to using this workflow, since much more work would still be needed to get the results I would be happy with. At this point I feel creating something like this on my own would be faster to get the same or better results.
The two videos below show one example using using AI to generate everything - voiceover text, voiceover voice, video, images, transition, music, etc., and a different version where I swapped out my own text instead of the tool using the AI-generated text.
For these, I supplied fairly simple/basic prompts, and let the system create everything from them.
I found the results to be mixed, but much better than the non-AI assisted video generator examples above. Specifically, I find that the video / visuals vary widely - some seem pretty good, some are so-so, some are bad (or even completely unrelated to the content) and sometimes just weird (one sample includes video that is displayed sideways!) There is also a lot of repetition of visuals.
That being said, the voiceover is, in a word, amazing. The AI-generated voiceover text was very impressive for a first pass, and the AI-generated voiceover on both of them – the AI-generated text, and the voiceover of the text I supplied – sounds to me almost indistinguishable from voiceover actors.
NOTE: VIDEOS IN THIS SECTION ARE NOT FINALS DELIVERABLES –
THESE ARE EXPERIMENTS – ASSETS AND VIDEO ARE WATERMARKED
For this, I supplied the following prompts:
Results:
I find the results to be mixed, but some of it is good. Some of the visuals are good, but as I mentioned above, sometimes bad, wrong, or just weird. (This first one includes the clip that is displayed sideways). There is also repeated use of the same clips.
The AI-generated voiceover text is quite good, and the delivery is excellent.
For this second AI-generated video, I supplied the copy. The system then applied similar rules as it did in the first example, creating visuals, transitions, etc. but using my text instead of AI-generated text.
Results:
I find the results again to be mixed – some good, some not, etc. and some not really appropriate to the text, and even more repeating clips than the others.
The AI-generated voiceover of the supplied text is excellent.
EVTOL deck
I have already created an EVTOL PowerPoint deck, so I wanted to see what this AI tool would come up with, prompting it to create something similar to what I had already created.