Some thoughts on Metadata

There are many kinds of metadata, and previously I’ve noted six that are important to production. There’s a whole second category of metadata that’s more related to asset management, but they share the common goal of being able to find a specific shot, as quickly as possible. In this post I’m talking about what would be Added Metadata of the six.

For scripted material, this is relatively easy, particularly with tools like QR Slate, Movie Slate, and the like now available. For scripted material you have known assets: Scene, shot, take, character, etc., as well as one or two free form fields like Comment. It is with un-scripted material that we run into problems not solvable with those solutions (at least in their current forms).

For reality and documentary production, and the many corporate and educational parallels, there is no advance script. No shot breakdown, because it’s unknown what might happen. Typically, this information is added inside the NLE during the logging phase. It’s this problem that Adobe are addressing with Prelude, and that Randy Ubillos addressed for Final Cut Pro before his producer’s logging tool became iMovie ’08! It’s the problem I’m trying to address while working on the Solar Odyssey show.

Now, I set out to shoot a show about the first part of Ra’s journey around the Great Loop waterway – an up to five month adventure. Turns out I shot the “how the boat was finished, very much delayed, and its first journeys” instead. Lots of good material, but no clear story, particularly early on.

Logging this material was killing me. So, we started working on a metadata entry tool, and I kept logging stuff the old fashioned way – after ingest into Final Cut Pro X. Some days I used the metadata logging tools, and some I did not. There were days before the first iteration of the tool, days where the action happened remote to where it could be logged, and days where I had to shoot and couldn’t devote any time to logging as we went.

That is the goal: to get basic log notes by tagging what’s happening on camera in real time, and translate that into useful metadata inside the NLE. Working both ways  in parallel, forced me to think about logging more consistently (like the machine would) and what logging I wanted. That’s where I discovered that I can’t have my perfect tool just yet.

Tracking and logging Person – who’s on camera; What are they doing; About what, Where. Person, action, noun, location. Clip names are built from Person(s)+Action(s)+Noun. And that’s perfectly fine for b-roll, or material that has obvious action. It’s also a huge benefit to get that within seconds of ingesting, transferring time from the edit bay to the production day. The assumption being that someone has time to log on set – to an iPad app rather than on paper as it probably is being done now. Note too that we’re not logging the specifics of a shot, just what’s happening during the time frame. This is applied to media shot in that time frame.

What I found is that I still need to do a lot of fairly manual logging, even with this organizational structure  in place, to get detail from the dialog. Those long takes where someone is talking, that help drive story and character, are still the logging bottleneck until some transcription technology automates the process. (I’d love file access to the Dictation function coming in OS X Mountain Lion, but I don’t think they’ll be open to developers this time round.)[Update: for those who are really observant! As of the start of Solar Odyssey, we are using Dictate for a “comment” keyword. The words dictated become the Note for that Keyword.]

The use of these snippets, that tend to become selects, leads to the question of how best to integrate the information. I’m working in Final Cut Pro X and the first, obvious response, would be “Keyword Collections”. Except each quote snippet would be a unique Keyword Collection and that would be clumsy.  I quickly moved to using Favorites for the job. Placing the transcript or summary of what was said, into the Favorite name. Not ideal, but the text is searchable. Then I was reminded that Favorites (particularly not names) do not translate to the XML, so even if we captured this information in the field, or generated it programmatically, there would be no way to sent it to Final Cut Pro X. Oops!

My current strategy is an evolution. I use relatively few Keyword Collections – mostly to gather the main themes of the day, or by character, and other information. Now, we categorize this source added metadata into People, Places, Actions, Keywords internally, and that’s important metadata that we don’t want to lose. Unfortunately Final Cut Pro X doesn’t support this type of key-value pair, so we used prefixes like p: k: a: to the keywords. This also makes it easy to drag these into an optional People, Places or Actions  folder (as we cannot do that via XML yet).

To those Keyword Collections I added some generics: Action, Quote, Problem. I apply these as Keyword Ranges, but fill in specifics into the associated notes field (which is supported in the XML, so I could export to a different type of database for asset management, if I wanted). The Notes field can apparently be any length and is great for transcript, semi-transcript or summary details.

An advantage of this approach: it allows me to use a Favorite to highlight, perhaps, a particular Quote. Action is used to break up a long clip that has many different identifiable actions happening. Again the detail goes into the Notes field. Each entry has its own Notes option, so it’s very versatile. (I now have the decision to make as to whether or not I manually swap over my Favorites to Quotes in the earlier logging!)  We’re still working on how best to add these notes from the field, when there’s generally not sufficient time for typing.

In a perfect world, I’d be able to mark certain time periods as “transcribe for me” but I don’t see that coming for at least another year or two. Even when it does, I’d probably still want to listen through large sections to get tone, but if I can get to that stage faster, then my goal is achieved.

As to why entering metadata isn’t easier? Along with most worthwhile things in life, it comes with some work attached. That doesn’t mean we can’t work to make it easier.