Very interesting new paper from @DeepMind: https://t.co/DTls99ToGV https://t.co/RD8rkPvXe6

Finally, #InvokeAI 2.3 (well, the RC) is out with support for #stablediffusion 2.x models & for the new diffusers format.

More importantly, @lincolndstein & team have published one of the clearest release notes the #AI community has ever produced 🙂

https://t.co/d8CBvOFips

Let me guess: cloud service providers report a slower cloud growth (that will persist over time). Something I’ve predicted a LOOOONG time ago.

Let me try again:

Azure growth decline will be offset by the massive consumption of OpenAI services + internal model training.

Let me guess: cloud service providers report a slower cloud growth (that will persist over time). Something I’ve been tracking for a LOOOOOONG ago.

Let me try again:

Azure growth decline will be offset by the massive consumption of OpenAI services + internal model training.

While that is fair to the #AI company as they allocate computing resources to generate your voice as you do your tests, the whole thing might discourage many potential users because you don’t really know how much you’ll end up spending after all rewrites.

It’s much more useful to have a voice designing system that gives you the legal ownership of a unique voice you have created.

@elevenlabsio’s voices are already very good. I’m looking forward to trying their Voice Design system when it comes out.

3. There is no guarantee whatsoever that the #AI voice synthesis corp will keep available the voice you have selected for your project. And no way to export/reuse that voice elsewhere.

So if you are building a long-term project, your #1 fear is that your voice will disappear.

There are 3 problems with #AI voice synthesis offerings today:

1. Depending on the voice you use, you might want to rephrase certain sentences. Sometimes, you have to rewrite a whole paragraph. Multiple times.
The way credit systems are implemented forces you to pay for tests.

2. In most cases, there is no way to “craft” a voice. Voice cloning is fun but extremely dangerous from a legal standpoint and unless you have a paid voice actor and you know how to deal with copyright law, you should stay away from that option.

This is to say that there’s a long way for #AI voice synthesis to become a consistently viable tool for media production and the issues are not about voice quality (which is making incredible progress – I hope to show you something about this later this year).