Sunday, January 01, 2023

My Experiments with Dall-E

In November 2022, Dall-E 2 was opened to the public without a waitlist. This is a deep learning AI model that generates digital images from natural language descriptions called ‘prompts’. I signed up just to see what the buzz was about and since then, I have been quite thoroughly enjoying the artificial intelligence experience.

A new user gets 50 credits to play around with, after which you get 15 free credits every month – each prompt or variation uses one point. You can buy credits also. I have not yet explored ‘Outpainting’ which is an editor interface that lets you tweak the images that get thrown up with your prompts.

Alas, I did not use my 50 credits too well. First prompts were extremely mundane – a log cabin in the woods, for instance. Nice but meh.


I played a bit with watercolours and some nostalgia. The effects were pretty.

Photo-real images with my rudimentary prompts were a bit hit or miss.


Then I learnt that with its heavy learning material, the AI interface was aware of thousands of artist styles and aesthetics. A muddled prompt for something in the Mughal Art style threw up a pretty but somewhat confused assortment of elements.


I tried a couple of classical elements together: a koel, a mango tree, and an ornate window in the style of Amar Chitra Katha, and I was really pleased with these images.

Line drawings were satisfactory, illustrations of flowers in the botanical style were near-perfect and even this render of a phoenix was very acceptable.


I wanted to represent the epic road trips my sister and I undertook last year and I was pretty happy with a couple of results in a cartographic rendering.

As you see from the column on the right, I seized the chance to represent this blog’s title in an expressionist output.

I explored what might emerge if I suggested ‘EH Shepard’ or ‘Raja Ravi Varma’ or ‘MC Escher’.


A photo-real request for a Pallas’ Cat (this gorgeous feline was one of the highlights of the trip we took to Ladakh in August 2022) was quite stunning.  


Since many users were bingeing on androids and ghostly apparitions, I didn't go down too much the futuristic route, although a neon digital art image of a mesmerising phone screen delivered to the brief.

As this column says, we may never have to use stock images again, or struggle to express an abstract idea. I have mourned for years that I cannot draw. Now who cares?

No comments: