In Search of Inspo

So, as I was trying to crawl out from the depressing state last week, I was trying to recenter myself to the project and look out for existing works that critques on Big Data and raising awareness.

Smile-To

An installation on seeing as machine, the slow interaction with the mirror makes the audiences to experience what it feels like to have your emotion read by the computer and what is it like to feign our emotion to a non-human entity. The compounded expression in the manipulated smile questions what it means to have a relationship with fully conscious artificial intelligent.



Data Labor

Another studio that did a installation with the theme of "Datafication", which closely relates to my research topic as well. They speculated the future where humans are made to generate data as a new form of earning for their income.

Thereafter, I found myself hitting the wall with data as currency since it’s already an existing thing and there’s even a marketplace for companies to purchase large amount of data. Off course, there is so types of data, I could dwell in into one of it and talk about the issue but I felt like it’s not really necessary OR maybe I should think deeper of what to do with such datas? However, moving forward, I will to be looking into what exactly data is and how it works. Perhaps working with computer vision.

Replicating Diffusion Methods

So, as I was researching on the ways to present data and how data potentially causes an issue, the process of AI synthesizing images to generate a new image intrigued me. So through the process of diffusion where noise is introduced to an image and from the diffused image, the process will then be reversed.By generating the image from the noise as steps, it will will iterate many times from the noise and thus generating a "new" image.

Exercise 2

Replication of Diffusion Process

I wanted to test the concept that noises are what determines the generation of image for AI image generators. I will first generate a noise image from an image and then reverse engineer with a existing Image to Image AI generator. The left image is the original image and the right noise image was generated using ChatGPT with the prompt to generate a noise image from it.

Processing the Noise Image

So using the generated noise image, I used DEZGO AI for it's image to image generator with prompts. So the consistent prompt that I had was " Woman in pain" and having the alteration strength changed at a gradual increase of 10% for each output.

Observation

Ultimately, they look different and the problem might be of that the prompts are slightly different and but what I interpret the pictures generated from the pure noise is that it started out with a rather small figure at 45% of strength to alter the noise image, subsequently at 70%, it showed the weirdest pose and the head dislocated. However, there is a somewhat uncanny resemblance in between the last 2 images, in terms of the framing, color, expression and background.

A Different Noise Map

I wanted to see if I could denoise any other random generated noise map with the same prompt of “woman in pain”, BUT at 80% onwards, its starting to generate outputs that are NSFW, therefore I will not be including the images.

Conclusion

As a conclusion to this exercise, I believe there are is still some gap in how I tested the reverse diffusion and if it's the correct way. However, for now the results will remain as something that I interpret the process to be.