FOR-A America continues to expand its product portfolio in new directions, moves that are also giving the company more stickiness when it comes to a North American market that historically has not been as well-versed in the company’s strengths comparedwith other Japan-based manufacturers. Ernie Leon, SVP/head ofsales and strategic growth, FOR-A America,explores the buzz around the IMPULSE live-production and software-defined platform; how its joint project with Nippon Television, viztrick AiDi, is expanding into new sports; and where hardware still fits as the industry leans further into software-defined production.
What generated the most buzz for FOR-A at IBC this year?
Certainly, the flexibility of our IMPULSE platform, which keeps evolving. It’s a software-node production suite so you can build a control room however you want and then essentially copy and paste that structure, which is great for local news and station groups. Beyond broadcast, we’re seeing it used for things like podcasts, where, one day, you have one guest and the next you’ve got 10 people miked up. Depending on how many cameras you have, you can scale the flexibility of the solution without being limited by your M/Es. That’s coming directly out of our innovation center, where we’re moving from hardware to software-defined solutions.
And on the Go Vertical AiDi side, which automatically crops video for the 9:16 social-media format?
AiDi keeps evolving into more sports opportunities and structures. We’ve gone from basketball to soccer, now basketball and golf. We got a request just yesterday from Dome Productions as they’re doing lacrosse work in Canada — and hockey is becoming a real interest, too. We’ve also had a request for IndyCar racing. In every case, it’s about making sure we have the right modeling in the AI structure to follow the correct car, or cars, correctly.
What is driving the need to turn around 9:16 content quickly?
In live sports, when someone scores a goal or a touchdown, you want that reaction to hit social media as close to the live broadcast as possible — especially for a large broadcaster. It matters less for a lower-end, non-live post; you can create a file and add effects after the fact. But, if you’re live, latency has to be very, very low. Every second counts on social; it’s about who can get it out first.
Walk us through what’s actually involved in training a new sport into the model.
It starts with getting data from the broadcaster — whoever owns the content — and training the model on it. That’s video: we take a file and run it through the system.
We can also take something like a PDF with faces and train the model on facial recognition. If you’re tennis player A, the system recognizes your face and automatically inserts your name into the graphic. No typing required, it’s already pre-created. We’re continuing to evolve the AI product to be more flexible and modular for organizations that want to implement these solutions with fewer resources. That’s what’s driving a lot of this.
How long does it typically take to train the model for a new sport?
It depends on the sport and its complexity. Hockey, for example, is more complex: we need a lot of data to nail down puck speed and tracking. We can do it, but it takes training. With basketball, on the other hand, we have a lot of content to draw on. We’ve trained on U.S. and Japanese basketball, and we’ve done the same with baseball, including something like 30 years of Japanese baseball games.
Those sports follow a fairly common structure. With pitcher pitches, batter swings, wide shot to follow the ball, baseball is probably our most perfected sport, with basketball close behind. Golf gets more complicated. And there isn’t a lot of hockey content in Japan, so we’d need to work with U.S. content providers to train that properly. Broadly, the more content we have, the better. But, within a month or two, we can typically train a good chunk of the model to handle the major things you want to track.
Does it keep getting smarter as more events run through the system?
Exactly — that’s from an initial standpoint. The more data we have, the more perfection you’ll get in that tracking.
Is AiDi part of the industry conversation around DMF and MXL or more of a standalone approach?
Today, it’s very linear, very SDI-based. We are having conversations around DMF and MXL, trying to be in the right place at the right time within those forums — though that’s more managed at the headquarters level. But, today, the AI platform specifically is linear: install it on-prem, SDI in, SDI out. For Impulse, we’re having a lot more conversations around software-defined products, making sure we’re building solutions that conform to the MXL and DMF standards. MXL is a big focus in our product structures.
Zooming out, do you see a future where dedicated hardware always has a place, or is everything eventually moving to software?
If you look back to COVID and the Amazon era — I was at Amazon at the time — everybody was moving everything to the cloud. Every production, everything. But today there’s a different consideration. Cloud will be there forever, but it’s not everything; that has been proven.
I think it’s the same from a software standpoint. There’s always going to be some on-prem component and a lot of hybrid applications, where the majority moves to the cloud or some non–on-prem structure, largely to reduce cost and figure out total cost of ownership. As things move to the cloud and to software-defined resources, human capital — a big chunk of the financial ecosystem — gets reduced, too. On-prem isn’t going away; you can’t put everything into software, not if you want speed and minimized latency. I think we’ll see more of a balance between the two going forward.
What do you see as For-A’s advantage in the marketplace today?
My approach, especially working with FOR-A, is that we can’t do everything for everybody. Honestly, that’s true for Sony, Grass Valley, and Ross, too. You try, but it’s impossible.
Where this works for us is that we have a certain product structure, but I need the customer to tell me what they need and push me. We can adapt and provide products that meet that demand, because I want that business. What also makes FOR-A a bit different, and something we’ve heard from a few customers, is that we’re more of a boutique-style firm. We’re easier to work with and easier to adapt for than a larger vendor, which comes with a lot more red tape and a lot more challenges moving fast.
Do you think FOR-A gets to that point: AI agents building out custom features on demand?
In many ways, I think we will, for certain applications. If a customer asks, “Can you do X?” the answer increasingly becomes, “Yeah, sure. Give me a week.” That’s where software-defined solutions make sense, because you can move faster and reconfigure.
The exception is porting something across operating systems entirely: Linux to Windows to Mac. You design for one, then have to ask whether it’s worth building an entirely new kernel for Mac if your base is Windows and you’ll sell only two units. That’s still a real challenge. But from an AI standpoint, you can move faster than you used to. Ten years ago, porting to a new OS might have just been a flat no. Today, what used to take a year might take five months instead. That’s a big improvement, even if it’s not instant.
