At the beginning, the reaction of the people was quite varied. Some felt aversion, everyone was astonished, and for the more sarcastic ones, simply seemed something more to make fun of it. The first totally autonomous and functional droid had been created. Everyone would had cried if they had known what the future was keeping for them.
Al principio, la reacción de la gente fue bastante variada. A algunos les causó cierta aversión, a todos les asombró, y a los más sarcásticos, simplemente les pareció algo más de lo que hacer burla. Había sido creado el primer droide totalmente autónomo y funcional. Todos habrían llorado si hubieran sabido lo que el futuro les deparaba.
AI Art Experiences
Here we go with a new round of my weirdo droids!
So as I was telling yesterday, get decent drawings seems more difficult than get nice paintings. My main problem is to get suitable backgrounds for the main subject, something that was happening almost by itself when I was creating paintings. But I think that have got some improvement in the composition and the backgrounds of the images.
After ran many test, I had some kind of epiphany and figured why I was getting these quite messy images. One has to consider which kind of data was used to train the AI. If we think about the paintings, a huge part of them fulfills the whole canvas, creating cohesive artworks with its main subject, background and so. But the drawings.. there are tons of them with only a main subject, without background, or minimalistic ones... it's a different scenario than the paintings. The AI has a hard time to figure where to place the main elements and what to do with the backgrounds. So what I have done in this round of droids, have been just specify in the prompt that I wanted the drawing on a cyan background. For the AI, it seems that to know what to do with the background, helped it to place the elements with more good taste as well, as they were not "floating" in an uncertain background anymore. And yes, I know, the backgrounds are not flat cyan as I have requested... because the AI does what it pleases. But I influenced it enough to at least, get more decent results than before.
Changing the subject, I mention in the footer of my posts, that I also use Photoshop for these images. If you wonder what I'm doing to them, to be honest I just fix some lack of contrast and saturation that happens when you run a lot of iterations for an image. It seems that with the more iterations and detailing, the more contrast and saturation lose. I like to show the images as they are given by the AI, so there are not Photoshop tricks, just fix the lose.
And that's all for today, I'll be back with more stuff about my AI art experiments in the next days.
¡Allá vamos con una nueva tanda de mis raros droides!
Como estaba diciendo ayer, consigue unos dibujos decentes parece más difícil que conseguir unas pinturas que se vean bien. El problema principal es conseguir unos fondos adecuados para el sujeto principal, algo que pasaba por si sólo cuando estaba creando pinturas. Pero creo que he conseguido una mejora en la composición de las imágenes y sus fondos.
Después de bastantes pruebas, tuve una especie de epifanía y me di cuenta de porqué mis imágenes eran tan confusas. Hay que considerar que tipo de datos se han usado para entrenar a la IA. Si nos paramos un momento a pensar en las pinturas, la mayoría de ellas llenan todo el lienzo, creando arte muy cohesivo, con su sujeto principal, fondo y lo que sea que el artista puso ahi. Pero en el caso de los dibujos... hay montones de ellos que no tienen ni fondo, o fondos minimalistas... es un caso muy distinto al de las pinturas. Así que IA lo pasa mal para saber donde emplazar los elementos y que hacer con los fondos. Así pues, lo que he hecho para esta ronda de droides, ha sido especificar en el prompt, que quiero que haga el dibujo sobre un fondo de color cyan. Y parece que para la IA, saber que hay que hacer con el fondo la ha ayudado a repartir los elementos del dibujo con mejor gusto compositivo, ya que ellos ya no están "flotando" en un fondo incierto. Y sí, lo sé, los fondos no son cyan como le pedí... esta IA hace lo que le da la gana. Pero al menos he logrado influenciarla lo bastante como para conseguir resultados mejores que los anteriores.
Cambiando de tema, en el pie de mis publicaciones nombro que uso Photoshop para mis imágenes. Si alguien se pregunta que les hago, la verdad es que sólo les arreglo el contraste y la saturación que sucede cuando la imagen ha necesitado muchas iteraciones. Parece que cuantas mas iteraciones y detallado, se produce más perdida de contraste y color. A mí me gusta mostrar las imágenes tal cual salen de la IA, así que no uso trucos de Photoshop, sólo corrijo lo que he comentado.
Y esto es todo por hoy, volveré con más de mis experimentos a lo largo de los siguientes días.
AI Artworks
All the content of this post is from my own.
Todo el contenido de este post es de mi autoría.
Images generated with VQGAN+CLIP and edited with Photoshop.
Imágenes generadas VQGAN+CLIP y editadas con Photoshop.
100% AI free writing.
Textos 100% libres de IA.
!discovery 30
!PIZZA
@lallo! I sent you a slice of $PIZZA on behalf of @jotakrevs.
Learn more about $PIZZA Token at hive.pizza (2/10)
Nice...I was hungry..lol
!PIZZA
@jotakrevs! I sent you a slice of $PIZZA on behalf of @lallo.
Learn more about $PIZZA Token at hive.pizza (1/10)
This post was shared and voted inside the discord by the curators team of discovery-it
Join our community! hive-193212
Discovery-it is also a Witness, vote for us here
Delegate to us for passive income. Check our 80% fee-back Program
These did turn out really well. Which data set library did you use for this one, if i may ask? I am just now running a session using the wikiart for the first time. From the initial image it looks to be taking a more artistic approach already. Great notion to figure out the difference in the background was the main issue between drawing and painting from an AI perspective. It made perfect sense.
Keep up the great work bro!
I have been using the default one, imagenet. I remember tried wikiart at the beginning, didn't notice much difference with imagenet, but well, I was very noob, probably didn't appreciated the details.
In the other hand I have tried the only datasets that can be used to make NFT's and they just suck to make artistry stuff lol... what a crap.
I guess blunts are magic, at least time to time lol
Regards and thanks for your kind comment! Oh and your !PIZZA, don't forget it!
Always love the Pizza! Thanks bro. I have been using the Flickr dataset because it can be used for commercial purposes. Originally neither wikiart data set was working but now the one with the bigger number seems to load and it's ok. I think wikiart should be as well since the art is all creative commons zero attribution license?! Either way... keep up the great experimentations! hahah
Did you managed to get anything that looks like a decent painting/drawing with this dataset? I only got crap lol
About wikiart dataset, it seems it has a mix of stuff under Public Domain and copyrighted without permission to use for commercial purposes. So as all the comes packaged in the dataset, there is not permission.
From Wikiart TOS:
We respect the intellectual property rights of others. It is our policy to respond to any claim that Content posted on the Service infringes the copyright or other intellectual property infringement ("Infringement") of any person. WikiArt presents both public domain and copyright protected artworks. The latter are showcased in accordance with fair use principle: as historically significant artworks as used for informational and educational purposes as readily available on the internet as low resolution copies unsuitable for commercial useAnd yeah thanks, I'll keep goingwith my crazy experiments :D
Thanks for clarifying on the WikiArt usage details. I have had some interesting results... but they are more fractal or repeating chaotic around the iamge than oriented in a cohesive way. Kind of like a creature in the movie The Thing. Since i wasn't going for a drawing or painting look... i don't have much to guess how it would do if prompted to properly. I will let you know after some experimenting. Cheers!
Well, in my experience the free to use datasets start with a random mixture that seems... a photography of little stones in the sand... then it start to transform it... the AI tries to do a painting as I have asking for... but the results are just atrocious lolol
Idk, I'll keep trying stuff. And I'm interested in the results of your experimenting as well!
Regards!
@castleberry! I sent you a slice of $PIZZA on behalf of @jotakrevs.
Learn more about $PIZZA Token at hive.pizza (4/10)
Excellent results here!
Thank you, I'm glad you liked them!
!PIZZA
@nikv! I sent you a slice of $PIZZA on behalf of @jotakrevs.
Learn more about $PIZZA Token at hive.pizza (3/10)
Congratulations @jotakrevs! You have completed the following achievement on the Hive blockchain and have been rewarded with new badge(s) :
Your next target is to reach 800 comments.
You can view your badges on your board and compare yourself to others in the Ranking
If you no longer want to receive notifications, reply to this comment with the word
STOPTo support your work, I also upvoted your post!
Check out the last post from @hivebuzz: