The story of Comixify.ai

Read-o-Meter 5 minutes

Find out more about our story!

by Maciej Pęśko

It has been 2.5 years since our five-member team has known each other and frankly speaking it has been a really crazy and intensive time. We decided to share our story with you to inspire everyone to work on making dreams come true. 

How It all started

It was February 2018, we were working in different companies and we barely knew each other. What we had in common was a strong computer vision background and passion to create something great. At that time, three of us, namely Paweł, Adam and I were still Warsaw University of Technology students looking for their thesis topics. We knew that we wanted to do something special and non-trivial concerning Deep Learning and Computer Vision, hence our supervisor choice was Tomasz Trzciński. One of the greatest Computer Vision and Machine Learning specialists at the University. There we met and rapidly came up with the very first idea. Adam is a great fan of comics and he proposed an extraordinary idea, “Let’s create an end-to-end algorithm to transform videos into comics”. We all agreed on it, split the problem on smaller tasks, redistributed work and started our research.

Fig. 1. comixify.ai web page look in December 2018 (Left) and September 2020 (Right)

It took over 8 months to develop a working algorithm with a prototype application demonstrating it. We called it comixify. When the work was done we wanted to share it worldwide to give everyone an opportunity to use it. To broadcast information about comixify we demonstrated technology behind it at the PyData Warsaw 2018, which is a large conference concerning state-of-the-art Data Science and Machine Learning stuff. Full presentation can be watched here. We also wrote a white paper that we published on arxiv and then in Fundamenta Informaticae. At that time no one thought about commercialization, until December 2018…

Everyone loves it!

What happened then was unbelievable for us. Everything started with this tweet by David Ha, highly followed Google Brain scientist. Within a month our platform became extra popular. We had around 25 000 unique users, 50 000 views and around 35 000 comics generated within a first month. The hype was huge and we were really shocked that there are articles about us on websites like medium.com or engadget.com and many others wanted to perform interviews with us. These days our website was just a prototype and looked horrible in comparison to the current version which can be seen in Fig. 1. The traffic was so big that we struggled with website performance and we had to immediately upscale it as the algorithm itself was quite slow and unoptimized and the website was unable to handle all the users at once. This impulse motivated us to engage in this project full time. In February 2019 we established company comixify.ai, but at this point we did not know much how to develop a start-up. Only Tomek had some entrepreneur experience and now we can say that we had a lot of luck. Since we wanted to go globally we started to search for global start-up accelerators that would match our company profile. We found one that particularly interested us, Betaworks Synthetic Camp. Betaworks is a VC fund and start-up accelerator founded in New York City. The camp was intended for companies that employ generative AI to create or manipulate images, videos or audio in an extraordinary and innovative way. What a perfect match! We were really excited to fulfill our application to participate in the program, we strongly believed we could get there. After three online meetings with several people from Betaworks, including their CEO John Borthwick, we were finally accepted! Among hundreds of start-ups all over the world, only 7 was accepted to the program. We were really proud and super excited to be a part of the Synthetic Camp. What is more, Betaworks invested quite a big amount of money in our company and we knew that it is a huge, crucial step for comixify, the first polish company accepted to Betaworks accelerator.

Fig. 2. comixify.ai team in Betaworks

Heading to NY

The program started on April 1st 2019, but it was not April Fools’ Day, it was real. Positive decision was given to us in the middle of March, so we had around 2 weeks to get a US visa, book apartments and flights. Those were two mad weeks, but we made it and on April 1st we were in New York City, scared as hell but simultaneously very excited. At that time we expanded the team by one new person, Bolek became our CMO & CPO. We searched for someone with big product, marketing and sales skills to help us with talks with potential investors and clarify our business model. To this day the 5-member crew has not changed. Synthetic Camp was groundbreaking for us, we met a lot of new people and we learned how to build and develop a company. It turned out that our main idea concerning transforming videos into comics is really hard to fit to any business model, because it does not solve any real issue. We were taught that a company should start not with a cool idea but with an actual problem of some target group. 

Fig. 3. VFX companies spent around $300-400k a year on manual rotoscoping. What if it might be done cheaper and faster by AI?

Vision upside-down

The time spent in NY was very valuable for us. During those three months we struggled with what problem can be solved by our cool-tech, how to monetize it and what vision should we establish for our company. We have plenty of ideas and there were days that we were 100% convinced about one idea only to drop it the other day due to some wrong assumptions we made. We talked to many people who were guiding us to explore different fields. Two people I remember particularly well, Danika – the Betaworks Camp Manager and Yasser a Betaworks Studios Member. Their huge power was the ability to ask valuable questions and challenge all our ideas. Thanks to that we were always pushed in the right direction to confront our current plans with reality and not to stop on any of our initial thoughts.

Finally! We’ve got it!

In the end of the Synthetic Camp we finally possessed necessary skills and we were ready to make a very hard decision. Comics are not the way we should go, the present and the future are videos. We had already had several conversations with people from the Visual Effects Industry behind us and we finally found the way to go. At that time we also acquired a key advisor, Brian, who is CEO of ZeroVFX. With his help we discovered that the VFX industry has a huge potential to use state-of-the-art Deep Learning algorithms but for the time being this potential is wasted and a lot of tasks are done manually. For instance Rotoscoping that is extracting selected objects such as people or vehicles from their background in the video. This technique is used in almost any post-production project and it is still being done manually and absorbs plenty of time and produces unnecessary costs. We decided that our strong AI background in the Computer Vision field can be truly helpful in this matter. We started with rotoscoping as a first VFX problem to solve and we set our long-term vision to reinvent the way the VFX industry works by using unlimited AI potential. Soon later we discovered other areas where Machine Learning methods can be used as an estimation of video Super Slow Motion or intelligent video upscaling. 

Fig. 4. Comixify VFX Platform screenshot.

VFX

For the last year we have been developing our proprietary AI based algorithm to handle VFX Industry problems. During that time we collected a lot of useful data from VFX studios that we cooperate with and we got familiar with an uncountable number of Machine Learning methods to address existing problems. We implemented, tested and improved selected ones and we still do not slow down the pace by working on new ones. In August 2020 we released the beta version of Comixify VFX demo platform where everyone can try out our VFX tools. To get access just visit our website at vfx.comixify.ai and apply for a free demo account. Soon afterwards, you will get email with access to all currently available modules which are: Style Transfer, Rotoscoping and Super Slow Motion along with full update to all the future tools that we are about to add soon. 

I hope you enjoyed reading about the way we have gone so far. Obviously, this whole time was full of many other situations, plot twists and interesting people that I have not mentioned so that this post would not become a book. Stay tuned and wait for our next steps!

Leave a Reply

Your email address will not be published. Required fields are marked *