Whenever a new version of an AI tool comes out, I try to test it against others and see how responsive (or useful) they may be. But certain instructions can still elude common sense with Generative AI – often when you give it even simple commands and directions, it has a mind of it’s own like an over-confident toddler headed toward a busy street.

I am often finding myself arguing with the AI chatbots like OpenAI’s ChatGPT  and Google Gemini Nano-Banana, as they will quite reularly not only botch-up the instructions but they clearly do not understand them no matter how clear and direct, simplistic and accurate your prompt directions are. In most cases it will reply that it’s followed the instructions in the text repsonses but in reality it has performed NOTHING to the image! So it brings little pleasure in arguing with it and playing nice after several frustrating fails… I usually just move on to another tool that will provide me what I really need.

I mean, look how far the tools have rapidly evolved since I first started reporting on AI imaging and Video Tools back in 2023! (Remember the HANDS?!!!) 😛

We’ve come so far in just a couple of years, as predicted we would, but now people are taking Generative AI tools more seriously and producing a lot more content than ever before. Having these tools at your disposal helps you be creative and develop new ideas in very little time. But to rely on them to do an entire video production is only going to disappoint you, and your audience.

Now that AI has been flooding our social and marketing channels over the past couple years, often people are looking for cues that give away to the fact they’re actually watching an AI video, so even the public is getting more keen to the phenomenon of the Uncanny Alley.

Yet many folks have posted amazing productions online that make it look like AI is doing all this awesome imaging, but in reality, it’s no different than any other form of content creation. For every one good result, there are most likely dozens of failures behind that nobody sees. Most of the resulting content has been edited and filtered and massaged until it looks ready for prime time.

What could go wrong?

The unexpected, of course!

Here, a simple product placement test using HeyGen’s Avatar IV with product placement. It seems one of their default models has too many limbs and it’s using ALL OF THEM! 😛

For this scene, I tried every tool just to get the camera to pan/arc around the center of this group – but ended up with extra people and so much spinning!

Finally, some home lifestyle attempts to add people moving about to static home images… and check out the shoes on the guy on the patio with the fireplace. I DID NOT ASK FOR THOSE! 😛

Scene 2, Take 247…

A couple projects that I’ve shared here in my previous reviews (all disclaimers still stand on these videos as outline in each article).

From my article AI Tools: Video & Animation Come to Midjourney, I had to try several attempts at each step to get the characters to respond to my prompts. Here’s the final rendered video:

Despite how the resulting video(s) appear, there were several FAILS along the way at each stage. Here are just a couple prompt fails that didn’t make the cut.

From my recent article AI Tools: HeyGen Avatar IV Gets Real, I tried every AI Tool I could get my hands on to create the various motions and movements this AI model was to make based on the AI-generated script.

Here’s the final edited video (WARNING: PG-13):

When I tried to instruct her to just “pull the trigger on the drill” to make it spin, I got warnings in some AI tools that the language was prohibited so having to creatively reword the prompts to try to get a result I needed. You can see that it just wasn’t going to happen…

Several other tools and attempts were made and I finally got enough to work with in the end, but some of these are just too hilarious – right up to the end! 😛

But just remember this…
THIS IS THE WORST THE TECHNOLOGY WILL EVER BE!