Testing the AI ​​that creates Amazing Images.

Thanks to advances in AI and neural networks, there are currently several programs or programming languages ​​that, from a given sequence of data, generate a given response, which in this case we are talking about generating images from text .

This tool that I am talking about is VQGAN + CLIP, they are two neural networks that are linked, the first one learns from patterns and generates characteristic elements of the input that we give it, this is called Codebook, this is important data because we can say how many elements the neural network must learn, in short, the network learns to decompose patterns of many similar images that will allow it to recompose a new image.

The second acts as a regulator of the results to see if it fits the input that we give in the form of text to the system, and allows us to generate impressive images.

The link if you want to try it, is in Google COLAB and is completely free, that's good because it allows you to experiment and play with additional models.

The instructions once the page is opened is as follows, it is to execute the subroutines sequentially one by one,

without skipping any subroutine that depends on each other, if any subroutine does not load properly it will throw an error and will not give results.

In this subroutine you can choose the model, there are two main ones that indicate the amount of parameters that the AI ​​will take into account when creating the image; we have other models but they are very heavy.

Además puedes ir leyendo las instrucciones que se encuentra en algunos apartados.

When we get to the subroutine of "Tools for execution:" we can modify several parameters, the main parameter is the "Text", here you must enter the concept or idea in keywords or phrases that you want the image to be created, it has to be in English; You can also enter other additional ideas separated with the symbol "|" (vertical bar) that will provide unique characteristics and allows you to separate ideas or concepts.

You can modify the width and height of the image, these values ​​are the number of pixels. You can also change the model, the image intervals it shows, by default it is 50, which means that every 50 interactions it shows us the results.

As for the initial_image and target_image parameters, I'm not sure what it does specifically, in another post I will tell you what they are for.

As for the default "seed" parameter is -1, which means a random value, the program starts from a seed that we do not know and will create the image; If you leave it at -1 and the other parameters do not vary, it will give you different results each time you run the program, if you put a specific number on it, it will give you the same image.

Finally, "The execution is done" and we wait for each result according to the interactions that we have previously given.

Initially it will give an initial image, which is the seed that the program randomly chose, but little by little the images are showing results.

This is the second image after 50 interactions, it is not what I expected but let's wait for the results of the AI ​​to see how it interprets the pattern phrase that I gave it.

This is the third image, it is already taking on more color and definition, it looks like a painting to me, I did not expect it.

This is the fourth, it seems to me that if you are on the right track, since you can see the crystalline structures of the gems and their typical shapes, I definitely like this AI.

It is important to clarify that this image that is being created is unique, since I do not even know the "seed" with which it was created. I use the program, there is a very difficult way to determine it but it can be done.

This is the Fifth image, since the pattern or the main idea of ​​the image is defined, it will not change, and the AI ​​will define each element that makes it up and give it greater definition, and the images that are generated differ very little between them, we will see the result.

You can also create a video of the images created in sequence, in my case I did not give this option, but you can do it.