Finding the India Connect
India is a vast country with thousands of years of history and cultural heritage. Which also means every region has a different art form, historical and traditional paintings and styles of buildings, sculptures, and so on.
Unfortunately, in the very early months as I began exploring the world of AI Imaging, I found very few images that represented the "look and feel" of India. There were images that had an India connect, but they were mostly fan art images of Bollywood celebrities. Some others represented images of Gods and Goddesses. But very often, the images suffered from a similar set of limitations- the image quality was not upto the mark, or the limbs were too many or too few.
You might ask, "What is wrong with the below image, after all, it shows a scene from Rural India, and it seems alright"?
I believe there is nothing wrong with the picture, except it is too Diensy-ish or Pixar-ish, or too much like the children's animated stories that one finds on YouTube. There is nothing unique about the above style. When we take the history and tradition into context, generating above kind of images is a problem in my (probably ill-informed) opinion.
Many creators simply began to mention the style of artists such as Greg Rutowski and Artgem, or the images on Artstation in their prompts. Which ultimately meant that many of the better quality images had a similar look and feel. This made me take up a project of my own- to generate as many India themed images as possible.
My aim behind this project was twofold:
a. The images I would generate could maybe inspire others to create their own interpretations, which would ultimately result in a reasonable sized library of artwork. The images need not be historical, mythological or about celebrities. They could be images from everyday life, or just some creative work about life in India. Below is one such example- depicting a man walking down the dark alleys of my hometown, Bengaluru, in 2045. The second image shows a couple from rural India, painting in the style of artist Paul Gaughin.
b. My second aim was to generate artwork for our podcasts. I have mentioned this aspect previously in this book. The stories we published on our podcasts- Baalgatha, Veergatha, Devgatha, and Fairytales of India- are kids and family friendly, and have a very deep India connect. If I was going to invest time in learning images, then I needed to understand how the images would fit well with our audio shows.
The first image below depicts an AI generated image that shows a golden palace. While the building style may not be typically found in India, such an image could be used in a podcast for bedtime stories or Fairytales about a King who lived in a Golden Palace.
Speaking of Kings and Queens and Palaces, the below image came out quite well when I wanted a cover art for a story about a wicked Queen.
In contrast, I would have to really figure out where to fit the below image about a Super Villian, because it is quite a generic representation. The image is quite good by itself, by the way, but would not fit for our intended purpose.
To extend the idea further, I created some images for Baalgatha, our podcast for Children's Bedtime Stories. The below image fits very well with a story titled The Blue Jackal, which is a classic tale from the Book I of the Panchatantra Tale Collection. I have compressed the image, so you may observe some imperfections, and the image itself may need some basic editing. But for most part, I think it is great visual representation for our audio story.
#Mumbai City and Monsoons
Mumbai city in the Monsoons is a theme that is near and dear to my heart. Yes, the city gets very heavy rains between June and September every year (and more frequently in October in the recent past). Yes, the roads get flooded, and yes, life becomes a bit difficult. But the city has its own charm during the monsoon months.
The above image is indeed great, but I cannot find 'Mumbai' in it. The below image, however can be more passable.
#Getting into the specifics
AI imaging presents a great opportunity to create images that distinctly convey the essence of stories for our podcasts. This is particularly true when we consider the sights, the colors, the seasons, and the setting of a particular image. For example, we produce a podcast called Veergatha, which honours the brave acts of the young boys and girls who have won the National Bravery Award in India over the years.
Let us take the example of an episode in which we narrate the story of a brave child who saved her friends from drowning. Let us say the hero of our story lived in the hilly regions in North Eastern India, or near the Himalayas in northern India. She performed a brave act during the monsoon season, when the river had flooded a village. The below image shows three girls who are walking near the banks of a river, and this image fits well with the theme of the story.
It takes quite a bit of trial and error before the right image gets generated. By the 'right image' I mean an image that is most suited for your stated end use. But in order to generate such an image, very often the renders might not be as expected, or the right keywords might be missing from the prompt.
If the story had a somewhat different setting, say a boy who saves his classmates form drowning, the image could be something like the one below.
Consistently generating relevant, high quality images is both an art and a science. And a matter of chance of luck!
in order to generate amazing images in a short span of time, there are two approaches: Be prepared spend a lot of time in experimentation, and a lot of frustration later, you will be able to figure out the right combination of words for prompts.
The other approach is to visit sites such as lexa.ai, where you can search from millions of images, simply copy and paste the prompts (or copy+ paste + edit) to start generating quality images.
Third approach is to learn form one of the many excellent tutorials online, particularly on YouTube. In hindsight, I would highly recommend this approach.
#Colours and lighting matter
Needless to say, I followed the first approach, namely, trial and error. I also realized during this process that using terminologies like volumetric lighting, and vibrant colours makes a lot of difference. This is because because bright colors work very well for India, the setting unless you're looking for occurs at nighttime. Hot and dry summer and mild winters, or wet season (monsoons) are other defining characteristics one can play around with.
#Artists form India
There are very few Indian painters or artists whose works are available in the AI image databases right now. Raja Ravi Varma is the most popular or the most common name one can find. Jamini Roy, Venkatappa, Amrita Shergill are some other painters of repute. In much of my early experiments, I was forced to add styles of Western, mostly European painters such as Paul Gaughin or Edvard Munch. The results were quite dramatic, and the quality of my AI renders improved considerably.
There are various styles of painting in India itself. Some of the more well known styles include Pahadi, Pichwai, Kalamkari, Rajsthani, Madhubani, Warli, and Tanjore painting. Over the past few months I have spent quite a bit of time or so and it's been a very short but a very steep and a very exciting learning curve.
On a lighter note,
Below is a random Image I generated using "India" in the description. Since the setting is in 2045, maybe in the next two decades my country would resemble this scene. Who knows?