Thumbnail image

Playing With Dall E 2

Wed, Oct 26, 2022 2-minute read

Hello again dear reader!

I was a little bit concerned about not writing so much this 2022, but the pandemy has relaxed and now the real world is taking my whole energy.

So instead of giving you my words, this time I’ll give you a more visual content.

To give a little bit of context Dall-E and Dall-E 2 are machine learning models developed by OpenAI. Those are state-of-the-art machine artificial intelligence models, that basically means that some nerdy guys took a lot of data, used some crazy math on it and then resulted on a couple of models cappable of generating images based on a provided description.

Ok, let’s step back a little. Usually machine learning models are used in classification and regression problems, however all those techniques and also some others, are used with big corpora text to give the computers the skill (train them) to “understand” what certain text means.

Once you give a computer that skill, you can now use it to generate images, also based on a given training set.

All this means that although the computer is “generating” this images, all this variations and new imagery, come from the initial training data or in this case, imagery set.

Having said that, let’s go with the images I’ve generated using Dall-E 2, there are some creepy ones and I’ve made a curatory work as well.:

leonardo da vinci painting the mona lisa realistic painting


a mermaid playing an harp on a full moon night, watercolor surrealist


the cyborg unicorn goes to the mechanical vet in a cyberpunk dystopia


an elephant riding a bicycle watercolor


a baby with a skull and a knife in his hands cyberpunk digital art


a teddy bear painting himself surralism


the impresionista devil dancing with a goat in a lake of fire with a big chain and a motorcycle


Buddha’s dream of having a son with a nebula and aurora borealis


I hope you have enjoyed my selection, there are other images that I’ve found interesting, but I think this ones are the most interesting ones. Also I’ve found that the examples provided by OpenAI are those that lead to better results, for example adding painting styles to the description, or adding a specific color palette is a good way to get better results.