Making Intelligent Use of AI

Lee Griffin, VP Business Development, EditShare

There is no doubt that, with ChatGPT making waves with immediate writing capabilities and stable diffusion enabling creatives to create incredible imagery based on simple prompts, artificial intelligence is the hot topic of the moment. Making sense of it all is quite a challenge, especially for us in the media and entertainment industry.

The talk at the moment is about the “creative” capabilities of AI. But that raises a lot of ethical questions for content producers. Who is in charge of the work? Where are the boundaries between fact and AI generation? What are the intellectual property implications? When AI can create completely convincing images from a brief text description, what is truth, and when does it matter?

These ethical dilemmas can make media businesses a bit uneasy, but let’s not forget that AI can be a lifesaver in some areas. It’s like this: people are great at the fun, creative stuff, while computers excel at the dull, repetitive tasks. AI is just another example of this age-old truth.

It’s all about the metadata

Asset management is the way we find stuff in our content stores. We describe the content using metadata, and the more detailed the metadata, the easier it is to find what we are looking for. Indeed, with really good metadata – and maybe some AI help in the search – we may find content that will take our productions off in a new, unexpected and delightful direction.

At EditShare, we’re all about helping you keep track of your content. Whether you’re producing blockbuster movies or TikTok videos, our FLOW asset management software bundled with our storage systems is the way to go. One fundamental truth with all asset management platforms is the better the metadata the easier it is to find the raw material to tell your story.

The real timesaver: Intelligent software

Completing good, detailed metadata is a time-consuming business and sometimes, it is not an especially rewarding task – which can lead to human fatigue and, in turn,  errors. Spending hours or even days in a darkened room manually tagging hours of video can be tough, but it is needed if you want to be sure you can later find the people, places and items you need; to sort the best takes from the OK; to find new and informative B-roll to tell the story.

But computers are good at repetitive tasks, so this is an ideal application for AI. If the software is intelligent enough, it can reliably tag objects, people, locations, and even emotions. It can process audio, transcribe the speech and lock the script to timecode, so that users can quickly jump to the part of the clip they need. Processing audio and video together helps fine tune the script tracking and understanding of emotions.

This all sounds great in theory – a real timesaver at the point of ingest, and throughout the extended life of the content. But can it be done in practice?

FLOW AI for Superhuman Vision

At EditShare the team studied the market, and saw that a start-up based in Berlin had all the right vision coupled with an incredibly powerful technology stack. The company is called Mobius, and their AI platform is called Superhuman Vision, which is pretty much what we wanted to achieve.

Facial recognition

So we have integrated it into our FLOW asset and workflow management system. We call the add-on FLOW AI. It adds yet another layer to our already highly automated environment. FLOW AI can process content for keywords in a scene, automatically tag people based on facial recognition and improve the post environment with intuitive scene and  detection algorithms. 

One of the operational considerations in AI is that it has to be trained to know what you know. FLOW AI is supplied with a lot of basic knowledge already in its database: as soon as you plug it in it can recognize more than 5000 objects, emotions, actions and even 10,000 famous people. And you can teach it more – you can add people to the database, for instance, and it will index them every time they appear in your archive.

Detailed image analysis

FLOW AI also includes intelligent support when searching for content. User requests can be very specific or quite vague, and the retrieval system will suggest relevant clips beyond the usual, so that stories become more interesting.

Obvious applications are systems with a high throughput of content, like newsrooms and sports broadcasters. Say a golfer hits a hole in one: you can quickly find all the other times this golfer has done it; or it has been achieved on this hole; or even disastrous triple bogeys at the same place.

It is a powerful tool for other users, too. Say you are producing a multi-part drama serial, and you need some cutaways to manage the pace of the story. By framing your request by mood, weather, time of day or people in shot, you can help the editor by offering clips that will match and flow seamlessly. For post houses versioning content for different markets, it can help by identifying the points which may need conformance editing, like nudity or swearing.

The partnership with Mobius is already proving very worthwhile. The algorithms in their software not only save time but offer more accurate, richer results, and as users get into the capabilities of FLOW AI, the benefits continue to grow. Object detection is just one string to our AI capabilities, with other integrations including speech to text and translation, but that’s something for a future blog!

It is said that AI is a powerful tool when used to support human decision-making. That is why we see FLOW AI as a great enhancement for our storytelling platform.

Want to find out more? click here to book a demo, or a chat with your local EditShare team member.