The term AI is pretty general
I'm not sure if this is entirely on topic, but I've been thinking about it a lot.
The term "AI" as it's been commonly used in the past few years is really just a marketing or branding word that doesn't have any real meaning.
Technically "Artificial Intelligence" means a machine that can perform tasks that usually require a human mind, but as soon as such a machine is created, those tasks no longer require a human mind, so the machines are no longer AI.
Before the invention of the calculator, all mathematical operations required a human mind. "Computer" used to be a job title that a person would hold. By this definition, a relay could be considered a form of AI.
So the term AI should only really be used for hypothetical future inventions, or to describe an area of study centred on making machines perform human tasks which they can't yet perform. But more recently the term AI has been used to describe pretty much any kind of algorithm with no real pattern to what programs are "AI" and what programs aren't.
I think this is a big part of what's lead to the irrational hatred of "AI". The term makes it sound like it's some big new invention when it's really just lots of small developments in increasingly complex algorithms that aren't intrinsically any different from the algorithms that find a route on Google Maps.
The other big problems people have with AI is specifically with generative AI, firstly in the case of image generation people say it isn't real art, which maybe it isn't, but who cares? The doodles I do on a napkin aren't real art but that doesn't mean I shouldn't be allowed to doodle. People complain that is just absorbs and regurgitates other people's work, but isn't that how everyone learns to create art? The actual "creative" part is in putting things together in new ways, which can still be done using AI as a tool (it's what everyone here uses AI for).
For voice generation people complain that it could be used for nefarious purposes, and by now we've all seen the adverts on YouTube where someone uses an AI voice to pretend to be Richard Hammond or some other celebrity advertising an investment scam, but this argument is kind of like saying "Cars can be used for drunk driving, so you couldn't use cars." the fact that a tool can be used immorally doesn't make the tool immoral. If it did, your own free will would be the most evil thing you own.
The other complaint people have about voice generation is that it's inherently immoral to use someone's voice without their express permission. But if that were true, simply doing an impression of a celebrity, or remixing their words in a YTP or similar work, would be immoral. You could say that you have to make it clear that it's not their real voice, but I think that depends on context. Does it really matter that Jason Mamoa and Amber Herd never said the couple of lines about going to Sicily that I put in my Snyderverse edit? I'd like to see them try to sue me over that.
And the final biggest complaint that people have against AI is that it's taking away people's jobs, which it is. But what technology doesn't take away people's jobs?
Thousands of 1-Hour-Photo employees lost their jobs because you (and everyone else) prefer the convenience of digital cameras.
Is using Chroma Key tools immoral because you didn't hire someone who trained all their life to be a rotoscope artist?
I don't want to come across as heartless. It's always a sad thing when people lose their jobs, but it's part of the progress of technology. The only way to prevent people's jobs being replaced is to ban the invention of new technologies altogether, and the Luddites already lost that battle 200 years ago.
There is actually one other complaint against AI image generation that comes specifically from artists.
It goes along the lines of "I trained for years to create this stuff, and now other people can do it easily! Why should I bother!"
I didn't include this in the list because it's not really an argument against AI, it's just people being petty. There's no more or less reason to learn to be an artist now then there was ten years ago. The fact that other people can create something has no bearing on whether (or how) you should want to create things. Even if you are one of the top 5% of artists in the entire world, there are still 400 million people who can create art as good or better than yours.
Nothing has changed on that front.
//rant over
Back to the original point of the thread. I can't wait for Elevenlabs to improve to the point where you can get reliably good takes on every generation.
I did some experiments with Retrieval-based Voise Conversion which uses a voice sample as a base and creates a copy with the generated voice. I tried to create a version of Bad Lip Reading's Twilight video using the actual voices of the actors, the idea being that I could use my own voice to create the lines exactly how I want them to sound, but I found that this method takes even longer than Elevenlabs and you need a very big sample of the generated voice for it to sound good. I'm sure in a few years Eleven labs will have RVC capability, or at least some more options to control emotion, timbre and rhythm.
A great tool I don't see talked about enough is Ebsynth. It's designed for forcing an art style onto a video (Making a video look like an animated watercolour, for example), but I've been using it mainly for special effects. I've found that the results really depend on how much effort you're willing to put into trouble shooting and refining the shots. I'll make some comparisons of things I've done with it and post them here.
As for pipe dream AIs, I've seen some promising development on tools for animating still images. Those could be vey useful for lip-synching AI generated dialogue if they become good enough.
