A few months ago, my wife Valentina was pitching our product Storique in the Hungarian version of Shark Tank. It was an amazing experience and I want to share a bit of the backstory with you here.
How it all started
Two years ago, when our first wedding anniversary was coming up, I started brainstorming ideas for a special and unique gift - one that would be good enough for this special occasion. At the time, Stable Diffusion was just starting to gain popularity and I saw adventurous tinkerers generate all kinds of (mostly anime) images.
I started to experiment with the technology, first trying to run it locally on my own laptop. But because I had to wait like twenty minutes for a single image to generate, I quickly gave up on this and instead rented a Google Cloud VM with a strong GPU for a few bucks. It was interesting to learn about all the parameters that influenced the outcome of such an image, like the prompt, the number of steps, the seed, the CFG (classifier free guidance) value, the various schedulers, etc. After that, I started to learn about LoRA (low rank adaptation) and learned how to use custom checkpoints. On Civit AI I found a model that I liked and started to play around with it.
After that, I learned how to fine-tune my own LoRA models using images of both of us. I had to properly prepare the training data set, define unique tokens for the two of us, label all the images, and then start the training process. It was tricky to figure out all the correct hyper parameters, but I eventually got it to work. And that was very exciting, because now I could generate images where the characters actually resembled us!
I then came up with a rough outline for a book that I wanted to write, detailing our life together. Especially when I had very specific ideas for images, it was increasingly difficult to generate them. For Stable Diffusion 1.5, a prompt can only be 77 tokens long, and the prompt adherence was not very good either. So I had to learn about additional technologies like ControlNet and inpainting.
In the end, I was able to generate all the images that I needed, wrote all the text, and put together a cover and the content in InDesign. Then I got it printed through a local print shop. The whole process took me easily 80 hours and cost a few hundred dollars, whereas the printing itself cost me $100. To be honest, the end result was not very good. The style of the images was not consistent at all, we were not always recognizable, and the overall image quality was somehow lacking. But it was a fun project and ended up being a very unique gift for Valentina. She was very happy and got really emotional when she opened it.
After we looked at the book a few times together, we thought to ourselves: this could be so cool if it was 100x easier, 100x better, and 100x cheaper. And so the idea for Storique was born!
Building Storique
At around the same time, Stable Diffusion XL was released and we were really excited to try it out. It was a big step forward in many ways, but most importantly it was able to generate bigger images and had a much better prompt adherence. Soon enough, we started training our own style model which defined this characteristic watercolor illustration style that we're still using today. The characters contained just enough details so that they were distinctively recognizable.
We also started to build a web application around it, where users could define the characters they want to include in their story, and upload a few images of them which we use to train their own LoRA models.
At first, users had to write all the text themselves but could generate images to go along with it. Then we added the functionality to let AI write pages for them, which already made it so much easier than what I had to do manually. By now, we have a full-stack system where users only have to define their characters, upload a few photos of them, and then pick a unique story which was generated just for them - or write their own abstract, of course. After that, the whole book is written and illustrated by our agentic AI workflow, and the users can directly see the first draft of their book to which they can still make changes.
We then partnered with various print shops in different countries, integrated with their APIs, and started to ship the books to our customers. It took a while to get the print setup right, but the results are amazing now!
Shark Tank
Both Valentina and I are great fans of Shark Tank and entrepreneurs ourselves, so one day we had the crazy idea to submit Storique to be part of the Hungarian version of Shark Tank. We had to prepare a pitch deck, a business plan, and a few other documents, and then we were ready to go.
After we submitted the application, we didn't hear anything for a while at first, but then we got a call and made it to the first interview! From there, a lot of other interviews followed, but finally, we were selected to be part of the show🎉
For the show itself, we prepared personalized books for all the sharks. Valentina wrote a blog post about how she created them, so make sure to check it out!
When the episode aired on March 23rd, we threw a small viewing party with our friends. It was a lot of fun watching the show together and seeing the reactions to it in realtime. We got amazing feedback and many new customers! And - spoiler alert - Valentina managed to secure funding from one of the most prolific investors in Hungary in the show.
You can check out the episode here.