Landscapes of Ladakh, India: AI vs Real
Recently I have been to Ladakh, India for a relatively long trip over June-July. I shot hundreds of photographs of the incredible landscapes, so I have plenty of personal images of the landscape of Ladakh. Also I visited the spot recently, so I have the memory of the those incredible landscapes burned freshly at the back of my mind.
 |  |
|---|---|
| Bing AI Image Generator | Looking down/south from Thiksey Monastery, Ladakh |
PS. This is on the side, but I must write this for 
|Image1|Image2|
|-|-|
|caption1|caption2|
Alright, this post is already completely disjointed. My apologies, please bear with me. I have a lot in my mind and trying put this down quickly as kids are trying to make a lava lamp!! Yeah, don't ask! Their Mom is travelling on a business trip!
Back to the images, and I promise I will try and stay on topic.
Intent of this post is to discuss the relatively recent proliferation of AI generatied images based on text prompt. I am writing this for new Splinterlands players who are trying their hands in blogging at hive.
I am not going to get into the moral debate of whether it is good or bad at hive. I will say my opinion on them. In my view the technology is briliant. However, I hate to be a digital artist today, or maybe even a photographer.
Personally I use Bing Image Creator (DALL-E3 Model). I find a simple supportive image for my text content very useful and readily available without having to worry about copyright restrictions of available free images from the internet. I am sure you have noticed, that I do use them in my blog.
I prefer not to use them, where the main content is art or photography. Especially when an user is presenting an AI generated image as manually created art. I think such activity is fradulent. With that said, please look at the two images of Ladakh, India side by side. They are both landscapes.
I have used the following prompt for the image to the left
Ladakh India Landscape Green settlement town in the middle lower right surrounded by barren area
I was trying to get as close to my own image taken from the rooftop of Thiksey Monastry, about 30 min to the south of Leh, Ladakh by car. Mind you, landscapes (photorealistic) are rather difficult to generate by AI. Most of the current publicly available text prompt based AI image generators uses a technique called Diffusion Model. Although the math behind is complicated, but the general idea is simple to understand. Diffusion models work by pairing text description with actual images in a massive database. Typically the model starts with random noise, and matches pixel-by-pixel with the given word prompt based on the database of known images. The model is not copy-pasting random parts of different image objects in a database to generate a new image, but it is subtly shaping random noise to create an image to match a target prompt by trial and error.
If you look at my result, and mind you, I didn't really spend too much time on this; it is probably obvious that which one is AI image, it is rather good. However, I must say, if I am using that landscape as a background of a slide deck regarding, I don't know, say Hydrogeology of Ladakh, the result would be perfectly fine and soothing. Often we use an image to improve the monotony of text, like this writeup right now! I think that AI image will do a good job.
This image is generated by a prompt:
Ladakh India Landscape photorealistic
Basically four words, and you have a passable background for your presentation, or a supporting image for your blog, and you don't have to worry about copyright. No wonder, they are everywhere. Just please, do not go overboard with them, and use it as a tool that it is not meant to be used.
