While working on the convolutional art generator, I felt it did not have good enough compositions. Before I had the idea to procedurally generate shapes with the drunken walk method, I instead downloaded a small sampling of low-resolution images, that the generator then could use as input. The idea was for it to, er, elaborate on the images themselves.
One of the images came from a search after the term "Portrait" and was the following:
And though generally, I dislike the feature, I have a few interesting elaborations on the Portrayed, that I will put together here.
One of the more interesting versions, the piece titled Blue has very few and very pale blues compared to the original. The only colour that goes directly through all the layers is that of her lips. Her eyes seem to be disappearing, but somehow, she only looks serene.
Lively Figs is one of the older pieces with the Portrayed, and one of the ones I had printed out. The picture, though the colours are different, stays true to the original with every line but the most important ones around her cheeks and mouth. In a way, she looks .jpged.
The newage piece, Warheads Bobsleds is the one that most looks like a deliberate photo-shop piece by someone quite technically skilled but not exactly aesthetically inspired. It feels overdone though all that has been done is by random chance of a chaotic algorithm. If someone wants to be a musician and need an album cover, feel free.
Our final piece, Peroxides Nutrient has chosen to go for a harmonic colour-palette, but it does not really experiment with the additional lights and darks. Instead, the Portrayed becomes an icon, condensed to her most distinguishing features.
Evaluation
Saying that I only put in the pre-made motives because I had not figured out how to procedurally generate shapes is a bit of a cop-out. Truth is that I admire photo-collages, aesthetically speaking, and I was hoping to be able to do that on top of the normal generator. Some of these pieces are close to reaching that which I was hoping for, but I can only assume it is some sort of luck and something to do with the high contrast of the original.
With the Convolutional art generator, there is only feast or famine, rarely anything resembling a balance. A lot of the pieces which technically use the Portrayed somewhere in the algorithm show no resemblence at all, and others of the other motifs turn out as clean reproductions without adding anything new:
Okay, so it is kind of interesting that the image falls into rainbows in the top corner, but it is hardly transformational, which it would have to be for me to be okay with stealing these images.
To conclude, if I were to redo the Convolutional art generator so that I might have a bit more control, a bit more balance between extremes, I might still implement the same feature again. Until then, I am happy to have left it as a very rare occurence, like a sort of easter-egg, because sometimes, like the case of the Portrayed, the results might be worth the more derivative failures.
One of the images came from a search after the term "Portrait" and was the following:
Blue
One of the more interesting versions, the piece titled Blue has very few and very pale blues compared to the original. The only colour that goes directly through all the layers is that of her lips. Her eyes seem to be disappearing, but somehow, she only looks serene.
Lively Figs
Lively Figs is one of the older pieces with the Portrayed, and one of the ones I had printed out. The picture, though the colours are different, stays true to the original with every line but the most important ones around her cheeks and mouth. In a way, she looks .jpged.
Warheads Bobsleds
The newage piece, Warheads Bobsleds is the one that most looks like a deliberate photo-shop piece by someone quite technically skilled but not exactly aesthetically inspired. It feels overdone though all that has been done is by random chance of a chaotic algorithm. If someone wants to be a musician and need an album cover, feel free.
Happy Climber
With rings of gold in her hair, and the blue light turning into a waterfall, Happy Climber takes an abstract approach but remains quite loyal to modern beauty standards. It also displays the approach taken quite clearly: One dimension of the picture determines the "transparency" of the original image at every given point. There is a clear distinguishing line down the middle of her face here, for instance.
Peroxides Nutrient
Evaluation
Saying that I only put in the pre-made motives because I had not figured out how to procedurally generate shapes is a bit of a cop-out. Truth is that I admire photo-collages, aesthetically speaking, and I was hoping to be able to do that on top of the normal generator. Some of these pieces are close to reaching that which I was hoping for, but I can only assume it is some sort of luck and something to do with the high contrast of the original.
With the Convolutional art generator, there is only feast or famine, rarely anything resembling a balance. A lot of the pieces which technically use the Portrayed somewhere in the algorithm show no resemblence at all, and others of the other motifs turn out as clean reproductions without adding anything new:
Healthy Levee
Okay, so it is kind of interesting that the image falls into rainbows in the top corner, but it is hardly transformational, which it would have to be for me to be okay with stealing these images.
To conclude, if I were to redo the Convolutional art generator so that I might have a bit more control, a bit more balance between extremes, I might still implement the same feature again. Until then, I am happy to have left it as a very rare occurence, like a sort of easter-egg, because sometimes, like the case of the Portrayed, the results might be worth the more derivative failures.