There is a war between artists and
Artificial Intelligence people. AI is primarily useful today to those people who want to avoid the hassle of doing something impressive, while still maintaining an image of impressiveness. Artists, on the other hand, who's whole being is in grinding themselves into true impressiveness are not satisfied with AI being used to replace their labor with cheap, algorithmic knock-offs. One such machine learning algorithm, though, had found its way into millions of artist's work-flows, which they don't seem to care much about. And I'm talking about the Intel's
Open Image Denoise found inside of
Blender.
The denoiser's job is very simple. Blender's Cycles rendering engine produces an image that is very noisy. To make it less noisy you can increase the time spent to render the image. But not always the time is available. So a denoiser is therefor used to smooth out the noise, so it looks like there is none. I used the denoiser inside of Blender for my
films, if you want an example of what this looks like.
Contrary to the generative AI that artists are so against, the denoiser is not generating anything by itself. You still have to put all the work into the modeling and texturing and lighting and animation of the scene, and then even render it out, before the denoiser can be applied to it as a final touch, to simply smooth out the noise. And yet somehow it still feels weird to use it as an artist.
In
Free / Libre Software, like Blender, knowing how a thing works is a key to it being Free. It is one of the four essential freedoms. Though if you look at AI from a certain perspective, it is impossible to know how it does what it does.
You can argue that "the software" in this case, is whatever algorithm is used to decipher the AI data. And you would be not wrong. It is similar to, say, a
VLC Media Player playing a video. The video is not a part of the program, but without it, the experience will not work.
Yet with AI it seems like the training data is so important to its operation that without it the decipher program is pretty much useless. Making the Blender Denoiser kind of, not Free, in one sense of the word.
Of course, these kinds of questions were brought up to the people responsible for the definition of "Free Software", and their answer was not that pessimistic. They believe that since further training of the data is possible, it is therefor possible to change the program, and therefor the freedom number 1 is preserved. And knowing about what it does, well, it does analytical analyses of the data. And looking at the decipher software and the trainer software should be enough. Others point out that looking at the input data is also important.
The Denoiser inside of Blender is not causing the planet to heat up. It's not running on a massive cluster of servers, like ChatGPT. It is built into the software and it is running on the computer of the artist. The Denoiser inside of Blender is not violating user's privacy, since it doesn't need to send anything out. And it is not trained on private data of people. To train a denoiser you need to give it two images, one with noise, one without. And it is possible to do algorithmically, generating those images on the fly. The Denoiser inside of Blender is not generating any art. It does not rub anybody of work. There was no such a thing a denoising artist before. There was just more time given to a computer to render out a picture.
The only problem I can see with the Denoiser in Blender is that it is not absolutely clear how it works. The
deciphering software is Free Software. But it still uses a bunch of highly human-unreadable data to do the actual job. Does that make it evil? Or is it that this particular AI algorithm is exempt from being evil, by the rest of its virtues?
On
Moria's Race web-page thingie, I said thus:
Made By Humans: Moria's Race was not AI-Generated but rather written, designed, modeled and animated by real humans. There was one machine learning algorithm used in the production of the movie and it was Intelยฎ Open Image Denoise. It takes a noisy picture and outputs a picture without noise. It doesn't generate anything on its own.
Did I lie? Or is using this denoiser, not considered as using AI?
Happy Hacking!!!