A.I. For Media Management: We Set One Up On-Prem

The Promise of A.I. in Media Management

The Perifery team was gracious enough to let us setup their Perifery A.I.+ media management solution and we hope to compare to other competitive products in a future video. But for now, we want to give you a snapshot of our first experience setting this up at the office.

Our team will walk you through the setup process and demonstrate how A.I. can integrate into media management systems, enhancing search capabilities without the need for manual metadata entry. This technology promises to make media retrieval faster and more accurate, revolutionizing the workflow for media professionals. We’ll candidly discuss the good, the bad, and the challenges of implementing these new workflow systems.

Why On-Prem Matters?

It’s important to note that Perifery A.I. is an on-premise solution. Why is that important? For some media and entertainment companies, high-profile television and films are not allowed to use cloud services due to security and intellectual property precautions—they are air-gapped from the internet. I’ll compare the on-prem setup versus a cloud-based setup for reference.

Setting Up A.I. Media Management Workflow

What is Perifery A.I.+?

Perifery A.I.+ is an on-prem A.I. platform that can locally run AI metadata processes. The generated metadata can then be moved to your Media Management platform, like Iconik, for your creative team to leverage for faster asset searches.

Essential Components of A.I. Media Management Systems

To set up an A.I. media management workflow with Perifery, you need:

• Media Management Software: We’re using an Iconik server locally at the Key Code office.
• A Server with a Graphics Card: Perifery recommends the NVIDIA A6000ADA, but we used an in-house NVIDIA Grid server with a Tesla V100 card due to availability issues.
• Perifery AI+ Platform: Installed on the same computer as the Iconik gateway, leveraging the graphics card for processing the AI.The system promises to automate the process of labeling metadata for tasks like face recognition, transcription, voice detection, translation, and object identification.

The Process: From Setup to Implementation

Lets start with the basics of how the Perifery system is setup in our iconik asset management system.

Transcribe and Summarize

Perifery can transcribe all of your content and summarize what is happening in a frame of video. This feature is perfect for quickly finding specific dialogue or visual elements in your media.

Object and Face Detection

Perifery can detect objects and faces, helping you find all clips with a specific talent, logo or product shot. For the on-prem version, you need to provide 10-15 still images of the object or face since the system cannot access the internet.

For face detection, we noticed that if you upload photos of the same talent with the same background, the AI can confuse the set for the face. Using diverse talent photos with different backgrounds helps train the AI to understand the difference. We photoshopped the background out, which helped the AI focus solely on the talent’s face.

Auto Translation

Convert any video to another language effortlessly.

Our Experience with the Setup

This brings us back to setting up the system. If A.I. is truly going to reduce our need for metadata taggers what does the setup look like? How do you train an A.I.- especially when its on-prem, locked out from cloud services providers like Google and AWS.

Initial Setup Challenges

Setting up the system involves some manual work, especially for the on-prem version. For objects and faces, you need to provide multiple images for training. Once the images are prepped, the Perifery team needs to manually run scripts to process them.

Manual Intervention

Currently, there isn’t a simple “process AI” button. The Perifery team has to remote into the system to run the necessary scripts. We were hoping for a more user-friendly GUI, but it’s on the roadmap.

Comparing On-Prem and Cloud Solutions

While cloud solutions like Google Photos have vast training data and resources, replicating this experience on a private server is more challenging. Google has 4 trillion assets across 1 billion Google Photo users to train the machine learning on. However, the privacy and control offered by an on-prem solution are significant advantages for large studios and brands.

Cost Considerations

Lets breakdown the basic costs you’ll need to consider when purchasing A.I. driven solution for media management.

• Iconik License: Approximately $49 per month. You will need a MAM to run the system.
• NVIDIA A6000 ADA: Around $5,000-8,000 due to high demand. You will need a high-powered graphic card to process A.I. requests on-prem.
• Server/Computer Costs: Approximately $2,000-4,000. You will need some type of computer or laptop to run the MAM license and Perifery AI.

Perifery AI+ Licensing:

The Perifery AI+ licensing options are either AI+ CX or AI+ FX. The primary difference between the 2 packages is that FX includes a much larger set of additional tools and features specifically for iconik users, including things like our robust iconik migration utility and our automatic scaling AWS transcoder for iconik ingest. The MSRP for CX is $26k / year and for FX is $52k / year. These prices include installation and support.

Conclusion

The potential for A.I.-driven media management is very real. The on-prem offering satisfies major studios and mid-to-large content owners concerned about privacy. Although the setup process is currently laborious and manual, advancements are rapidly improving the technology.

A.I. is iterating faster than any other technology today, and production companies will likely want AI in their media management systems soon.

If you’re interested in evaluating an A.I. solution for your creative systems, please contact us at Key Code Media. Or, if you’re local to our offices, come by, and we can show you how it all works. Contact us on the website for more information.