As a founder, I do a bit of everything, but I still stay hands on with the design of the product. This mini case study is on how we built the core feature of TRASH, our instant editing (patent pending).
AI Video Editor
The challenge & opportunity:
Make creating a short video as easy as taking a Polaroid.
Design, product (working closely with my co-founder Dr. Geneviève Patterson on science and tech)
Curiosity & hypothesis
As a creative person, I’d noticed how much easier it is to edit than start from nothing (the blank canvas problem), so I was curious to try to make a rough cut first for users.
I also set a bar for us that if we could make video creation easy as taking a Polaroid (similar to what Edwin Land did for photography), that had a chance at cultural and scientific impact.
Exploration & Plan
The timeline: Early on in the exploration process I threw out the idea of starting with a timeline, getting feedback it was too technical and intimidating for amateurs as a starting point.
The camera roll: Since there are already great cameras on the market (and people have habits around shooting), I focused on the camera roll, letting us interoperate with other cameras. Also, I suspected the camera roll could serve as inspiration, and inspiration is critical to get a user into the motivational state to create. The camera roll seemed like a quick way to test the idea. I was inspired by a UXR interview I did with a film editor who said he always starts his projects by looking at his ‘coverage’ first.
One button: I wanted it to work with one button. This is where a lot of work went into designing the algorithms (a year of R&D).
Tapping one button to create a rough cut eventually became the “ah-ha” moment in usability testing and the key moment we attempt to deliver on with our funnel metrics. However, it took us a long time to get there. Early versions of the app were plagued with slow analysis of the video clips, resulting in unhappy users waiting for videos that sometimes never delivered. Another problem was people putting in one clip vs. 20, it was difficult to tune the algorithm for these different cases to ensure delight the first time someone tries it. This feature required a ton of optimization.
Adding controls to let users edit their rough cut came after and even these are all collaborations with the AI too. A big focus was keeping the timeline editing simple, we went with a drag and drop approach like re-ordering a playlist.