Today I saw a Techcrunch announcement for a new Y Combinator company - Shred Video. It's a video editing app that intends to be absolutely painless and practically instant. The basic workflow is as follows:
- Add video files.
- Add music file.
- Profit from a fully-edited video, with the action shots trimmed and cut perfectly to the music.
I'm an action sports enthusiast who likes to take his GoPro on fun outings. I also enjoy and appreciate movie editing, when I'm trying to put together anything remotely professional. But lately, this has rarely been the case, and I've usually just wanted to throw some clips together of the day to share with friends. Basically, I'm right in the target market for this app, so I thought I'd try it out.
I recently went cliff jumping with my brother and cousin. Nothing too special, but I wanted to put together an edit to be able to share and look back on the day. Usually I open iMovie and do the video editing that way, but even when I'm being sloppy it can be tedious. To try out Shred Video, I used the cliff jumping clips and an Arcade Fire song, pressed 'Shred', and got my video:
That's the result I got with no effort involved. Their algorithm follows a pattern (which they talk about on their site), where the video starts with some happy face shots, moves on to the tricks, and finishes with the bangers at the end of a 'beat'. You can see that the above video kind of follows that pattern - some smiley shots in the beginning, and then the 'action' for the rest of the video.
Except for one problem - it totally missed the action. All of the actual cliff jumping shots are either of right before or right after the actual jump - there are no shots of a whole jump itself! This is a bummer, because it's pretty much the main selling point of the software.
The thing about that, though, is that it's ok with me. I honestly don't expect us to solve the computer vision problem of determining exactly when a trick begins and ends. Frankly, I feel that due to the massively unstructured nature of action sports, it is practically impossible to create an algorithm that could detect this and work for every sport. Take, for example, skydiving. It's a pretty broad sport and people enjoy it in different ways. Some people would like to get shots of them exiting the plane - maybe the like to show their wingsuit inflating or some flips out of the plane. Others will want to show the freefall - maybe they're free flying, or maybe they're trying to fly in formation. People also like to include a shot of the canopy deploying. Other people only care about showing themselves swooping while landing.
Shred Video anticipates this and allows you to manually trim your shots to indicate where the action is. I went ahead and did this with every shot, as well as hitting the star next to every clip. I clicked 'Shred' and saw my video - much better! However, the clips were cut just a bit shorter than I had trimmed them, and there weren't any 'lifestyle' shots. I actually prefer that the video didn't start with the 'happy faces' - it feels cheesy and just not the how I'd do it. But I still want a few lifestyle clips. I did some tweaking and came up with a 'formula' that I liked:
- Star every video, and trim it to be just a bit longer than the trick itself.
- Click the '+' to duplicate each clip, and un-star it.
- Click 'Shred'
- A pretty good edit!
Here's my final result:
I love it! All the tricks are cut as I'd want them, and a lot of them are well-timed to the music, with a few splashes happening pleasantly on the beat. The best part is that they ended the video perfectly - I don't think I would've done it any differently. The song was winding down, and they finished off with a perfect 2-angle shot of my brother doing a gainer! I don't know if the 2-angle finish was on purpose, but it was really really well, and if they can re-create that consistently I'll be very impressed. However, on later attempts to follow my formula with tweaks, I ended up with a totally different video with a very boring ending (the same ending as the first video). This leads me to some of the feedback I have:
- If I star and trim all of my clips right to the tricks, the software should still look around the clip for lifestyle shots to include. My workaround for this is to duplicate every clip.
- It would be nice to be able to mark the 'banger' of an edit - the very best trick to end on. Even more advanced would be to rank the tricks according to some scale, so the algorithm could mix around some shots according to skill as well as the flow of the clip.
- It always opens file dialogues to the place you last were, but you usually open the file picker for a different reason than the previous time. You usually pick the clips, pick the song, then save the movie in a certain folder. It would be nice if the app remembered my default location for these actions.
- Keep working on the 'trick detector' - it won't be perfect but it also didn't seem totally horrible, as it did detect a part of the trick out of ~30-second shots.
Overall I am not worried about the 'trick detecting' functionality, mostly because of what I said before about how I don't think it could be perfect. The software is still great because it's designed for action sports and still mix together your clips with action sports in mind - other software doesn't do that (that I've seen). I ended up paying the $30 for unlimited saves, and I plan on using this app as my go-to editor for edits like this. For an edit that I care more about (anything less 'home video'), though, I will probably still manually edit it, unless Shred Video provides more manual controls style="margin: 0px auto; display: block;" for the video.