AI Video Generation
The Rip-Off Everyone Pretends Isn’t Happening
I use a lot of AI lately — and have spent literally a couple thousand dollars this year — to create content for my YouTube channel, or images to use with my posts (like this one).
Frankly, it’s so exhausting and frustrating that I sometimes wish AI had never been invented. Now that it’s here, you almost have to use it.
But G-d!
After a particularly exhausting morning, I recently had a back-and-forth with Google’s Gemini about their Veo 3 video model.
My Own Failed Experiments
As I noted, before I even asked Gemini about Veo 3, I’d already been down this road myself. I’ve burned through credits on Enhancor trying to animate jail doors, on MidJourney trying to get characters to face the right way, on Runway when I just wanted a smooth sequence without the software hallucinating cars, limbs, or random objects that didn’t belong.
What should be a simple request — “start with this image, end with that one” — turns into days of wasted runs. Most of the outputs are either unusable or just plain bizarre. I’ve had characters swap clothing mid-scene, props vanish, or new objects suddenly sprout in the background. The credits don’t care. Every misfire still costs.
So when I put the question to Gemini about Veo 3, it wasn’t out of idle curiosity. It came from lived frustration: hours of work, hundreds of credits, and more abandoned projects than finished ones.
Veo 3 on Paper
On paper, Veo 3 is supposed to be “cinematic,” “professional,” and even able to spit out synchronized audio. Gemini gave me the corporate-sounding rundown: Pro plan users get around 3 videos per day, Ultra plan users maybe 5. All managed by a credit system. You can use a “start image,” but you can’t guide the end of the video. Maximum length? Eight seconds. That’s it.
Compare that to MidJourney, which now lets you feed it both a start and end image to create deliberate transitions. Veo 3 doesn’t do that. Google’s line is that Veo is “different, not less sophisticated.” Maybe. But in practice, the limitations matter more than the branding.
When It Works
Don’t get me wrong: when these tools actually deliver, they’re often brilliant. Working with Veo 3, I’ve managed only two great videos so far — Luis’s Arrest and Miguel’s Arrest. (You can watch them below in this same section.)
I told Veo to reuse the same arrest scene, but swap in Miguel for Luis, and it did exactly that. I’m going to — chop them up a little — to create some jump cuts in an upcoming Spanish video on my YouTube channel.
The two sequences line up beautifully: the officer checks the cuffs, then guides each arrestee toward the cruiser. And it worked this way because after the starting image for first video was created, I simply told the AI to generate that same image and we went through a series of iterative changes.
The conversation went like this:
Gemini:
The image you provided is a painting of a man being arrested by a police officer. The man is wearing a white shirt and is being handcuffed. The scene is dark and appears to be taking place at night with the lights of a police car in the background.
Me:
Can you keep everything as it is including the pose, but change the man to be a scruffy-looking down-on-his-luck-appearing young man wearing a red t-shirt and blue shorts?
Gemini generated an image that matched my request.
As my post is about the disparity between people who can afford bail and people who cannot, I was trying to get a scruffy, poor-appearing character I’d named “Miguel Poblete” to contrast with “Luis Rico” and so…
Me:
Make the arrestee look less attractive but not quite ugly.
And I got what I wanted: the character wasn’t actually ugly, but definitely looked a little “rougher” in the face. To add to that…
Me:
Give the arrestee about three-to-four days of beard growth.
And that led to a nearly identical image, but with Miguel in place of Luis. I used those two images to create the only two videos I was allowed to make today. (One my third image I was going to make, I forgot to upload the reference image. When I realized my mistake, I cancelled the job, thinking I could try it again. That’s when I learned that even a cancelled job counts towards your “three videos” for the day.)
Luis Rico is arrested in this first video and Miguel Poblete in the second.
The only difference? Luis — dressed well — gets nudged slightly right in the frame. Miguel — the low-income Hispanic kid — drifts left. Otherwise, the shots mirror each other almost frame for frame. For once, the machine followed directions.
And that’s what makes the rest of this so maddening. Because after the rare moment of brilliance comes the reality — the part where the system starts to feel like a scam.
The Credit Scam
Here’s where the whole thing stinks. Google (and every other player) hides behind “credits.” Gemini explained it like this: computation is expensive, GPU farms cost money, training costs money, and so on. Credits are how they “bundle value” and cover those costs.
But here’s the truth: the majority of users aren’t well-funded film studios. We’re creators scraping to get something watchable. And with these systems, one usable 8-second clip can easily cost 10, 20, even 100 times the sticker price because you burn credits on failures. The companies call that “iteration.” I call it a money pit.
When you spend days or weeks running prompts, throwing good money after bad, only to abandon the project because nothing works? I wouldn’t call that a “creative process.” I’d liken it to a casino where the house always wins.
Branding on Your Dime
As if the credit system weren’t enough, there’s the insult of watermarks. Veo, for example, stamps “Veo” in the lower right corner of every clip — free advertising for Google, courtesy of you. You already paid for the credits, you already paid with your time, and now you’re forced to either crop out part of your video or slap your own logo over theirs if you don’t want to look like a walking billboard.
And Google isn’t alone. Many of the other AI video tools pull the same stunt. Some do give you an option to download without a watermark — but usually only at a higher subscription tier, which means you’re effectively paying twice: once for the tool itself, and again for the privilege of not promoting the company in your own work.
It’s a business model that says, in effect: “We’ll charge you to create, then we’ll make you advertise for us.” Then they tell you it’s because you’re getting such a great value from them.
If that isn’t gaslighting the creator, I don’t know what is.
Ask Not For Whom the Bell Tolls: It Tolls For Thee
These systems don’t just drain your wallet or brand your work. They take a toll in time. Every generation takes minutes — and that clock never stops ticking. By the tenth or twentieth failed attempt, you’ve lost days. And for what? An eight-second clip that still isn’t usable.
Worse, platforms like Veo 3 throttle how often you can even try. Credits run out, cooldowns kick in, and you find yourself staring at the clock, waiting for permission to burn more hours on a machine that already wasted the first round.
The bell tolls for every creator caught in this loop. You don’t just pay with money or by ceding advertising space. The bell tolls with days of your life, rung out one failed iteration at a time.
Gaslighting Users
What makes this worse is the gaslighting. Companies (and their AI spokesbots) talk as if you’re paying for a service that reliably produces content. In reality, you’re buying lottery tickets. If you happen to get something usable, congratulations. If not, tough luck. Try again. And again. And again. The entire risk is pushed onto the user, while Big Tech keeps cashing subscription fees.
It’s not just frustrating — it erodes trust. These companies want to position themselves as indispensable creative partners, but when the cost-to-value ratio is this skewed, they’re undermining the very market they claim they’re serving.
The Bigger Problem: AI Pollution
Even if you get past the money issue, there’s the environmental and informational pollution these models create.
Energy & water drain: Training and running these systems burns obscene amounts of electricity and water. One failed clip after another isn’t just a financial sinkhole — it’s a planetary one.
Garbage in, garbage out: The internet is now flooded with AI sludge. That sludge becomes the training data for the next generation of models, which just means more garbage.
Search destroyed: Google’s own search engine — once the crown jewel — is collapsing under this weight. Between SEO spam and AI-generated filler, it’s barely usable. And their “AI Overviews”? Half hallucinations, half nonsense.
So you pay through the nose for tools that rarely work, while they wreck both the environment and the internet ecosystem that made them possible.
The Disconnect
The companies see each 8-second video as a unit of value worth charging for. Creators don’t. For us, that one usable video is the end result of a slog through dozens of failed generations. The cost of the failures is the real price, and it’s astronomical.
Until these companies acknowledge that reality, they’ll keep bleeding early adopters dry while pretending they’re democratizing creativity. Right now, the only thing democratized is disappointment.
That’s where we are: a marketplace that feels like a scam without technically being one. The tech works sometimes, just enough to keep people hooked. But the system is broken. And unless it changes, it won’t just drain our wallets — it’ll bury the internet itself under a mountain of its own trash.
Exit Wound
Video isn’t the only place AI stumbles. As some of the images I’ve posted here show, the rot runs deeper than just motion — the mistakes bleed into stills, too.
Video just happens to burn through more credits, faster.
And more than once I’ve caught myself muttering: “A picture may be worth a thousand words, but it sure as hell isn’t worth that many credits.”
Great information! Thanks Rick - excellent read as always.