Minds'Eye: An Image Generation Web Application
Background and Problem Statement
With rapid advancements in technology, more and more individuals are empowered to become creators. Based on my own research and personal experience, one of the biggest problems creators face is finding inspiration for their next project. Most creators would already have some intuition of what they want to do – they just need the extra push to supplement and crystalize that intuition. However, this is usually a long process governed by trial and error.
For example, an artist might possess the habit of scrolling social media sites to find that extra push to start working on their art piece. If he/she manages to chance upon the right image early on in the process, he/she will be able to start he project sooner. However, the problem with such methods lies in the fact that viewed artworks tend to be completed works of art, leaving little room for the imagination to run free. Thus, more often than not, this inspiration finding process will be dragged out, since artists are consuming content that are not stimulating the dormant ideas they already possess.
Furthermore, there is also an increased risk for aspiring artists to feel jaded since they are constantly looking at beautiful pieces of art made by others, with little room to inject their own personality and flair. Thus, I approached this project with the following problem statement:
How might we speed up the inspiration process to help aspiring creators create instead of consume?
Solution (VQGAN+CLIP) and how I made Minds’Eye
After researching and much trail and error, I decided to leverage on a pre-trained image generation model.
Evaluation of Minds’Eye
Strengths
100% Free!!!
Accessible to anyone since it is incredibly easy to use
(One-click)
Fast performance (other solutions in the market requires upwards of 20mins to generate an image)
Limitations
Google colab notebook needs to be running in order for Minds’Eye to stay active
No fixed link
(free version of ngrok)
Only 1 image can be generated at any one time
Improvements
Additional features such as using Google Firebase to add a backend queue system would enable multiple users to click on the button simultaneously and still receive their image (albeit those later in the queue might have to wait more than a minute for the image to be generated).
Final Thoughts
This was an incredibly challenging project since I went head first into it without any prior web development experience, much less integrating a machine learning algorithm with a web development framework. It was a long and tedious process with a relatively steep learning curve - but all the hardwork only made the completion of the project all the more satisfying.
I hope this was also an experience for you, the reader. I hope I have brought you through a journey and have inspired you to start on that project that was always at the back of your mind – or at the very least, entertained you throughout the time that you were here (: