Delta Protocol: Live media reinvented

By treating live video as a sequence of meaningful scenes, Delta Protocol aims to reduce transmission bandwidth and compute, while maintaining broadcast-grade fidelity, virtually eliminating content theft and significantly enhancing personalisation.  

The Delta Protocol: Live Media Reinvented project has seen an explosive evolution, according to Sandeep Tiku, Chief Technology Officer, DAZN Group.  

“[It is progressing] successfully, and faster than we expected. Six months ago, this was a whiteboard idea. Today, it is a functioning end-to-end proof of concept. We are moving away from focusing on how we compress and ship pixels faster, where gains are becoming increasingly incremental.  

“Delta Protocol asks a different question entirely: what if we shipped meaning instead of pixels? The goal is to disrupt live streaming’s foundations, so that the industry moves beyond pixel transmission to an architecture of pure intelligence. 

“We've proven the hardest part: taking live sport, having AI understand what's happening in the scene, and turning that understanding into instructions sent as a lightweight, brand-new stream of scene intelligence that a device can act on in real time. This is now running code with multiple partner technologies plugged into the pipeline.” 

The experience is now being polished for the demo this weekend at IBC2026, which will be based around live sport for a host of reasons, as Tiku explains: “For 30 years, live sport has been delivered through brute force. Every pixel is encoded, shipped, decoded, and then is instantly everywhere. It's one of the great engineering achievements of our industry, but it's also breaking under its own weight due to pressure from huge infrastructure, soaring CDN costs, latency expectations, formats we can never fully standardise, and piracy we can’t fully stop. 

“And personalisation? Today it's an illusion. Graphics are bolted onto a stream, forced to look the same for everyone, and propped up by server-side stitching and MCR overhead. If the stream doesn't understand the customer, context, event, and bring it all together, we can’t achieve true hyper-personalisation.”  

The project has set about structuring an AI-native transmission model by defining a live event as a set of scene instructions, which has a wide range of benefits, according to Tiku: “We are going to demonstrate to the industry a proof of concept for an ecosystem designed to solve the industry's core challenges of enhancing the consumer experience with hyper-personalisation, collapsing the cost and complexity of delivery, and most importantly, building a model where piracy is structurally impossible.”  

The Delta Protocol as an overall solution 

Getting down to the nuts and bolts, Delta Protocol is powered by AI models operating in real-time. Tiku explains the heart of the project: “We have developed a real-time AI scene-understanding layer for Delta Protocol. It ingests the live broadcast and creates compact, structured descriptions of what is happening with player details, weather conditions, crowd sentiment, and any meaningful changes in the scene, all encoded as delta instructions at a fraction of the bandwidth. The potential is near-limitless. As the AI models learn from more content, the range and precision of what they can detect, interpret and describe continues to improve. 

“These delta instructions are precisely synchronised to the video frame and delivered to the device in real time. That is what makes the approach genuinely AI-native, rather than AI-adjacent, as the intelligence is live.” 

“The client receives scene instructions and adapts the experience locally, based on the device's capabilities and the viewer's preferences. Contextual overlays, personalised moments and picture enhancement can all be changed by updating the delta instructions, not by repeatedly re-encoding and distributing new video variants - that is the shift from AI around the pixels to AI inside the delivery pipeline.” 

Technical challenges across the value chain 

The technical challenges are considerable and are being borne by a team of technical partners including V-Nova, which is being evaluated as an alternative feed source for AI analysis, NeoViews and Google, which provide a richer understanding of what's happening in any given scene, and finally Eluv.io, which is used to keep the video in sync with the intelligence track. 

Inevitably, for such a visionary project, there have been a considerable number of challenges, not least the question of how to define a change within a scene. Tiku is enthusiastic: “It has been the most fascinating problem in the whole project. Video compression already knows how to spot pixels changing. We had to teach the system something harder: which changes matter. A player jogging is noise. A goal is signal. A crowd erupting is context. A rainstorm changes what the device needs to render. 

“Get the threshold wrong and you either drown the system in noise or starve the device of what it needs. Tuning that balance using live football, not lab data, has taught us more in months than years of theory could. We're still iterating on this. It's the kind of problem where the answer keeps getting sharper the more you work with real content at real speed.” 

Reconstructing contextually relevant content at the edge of the network is a core requirement for Delta Protocol, and not straightforward. Tiku: “We treat this as one system, not a tug-of-war between cloud and device. The principle is to understand the scene where it's cheapest to understand it, transport only what matters, and let the device do what devices do best: render an experience tailored to the viewer based on device capabilities. 

“Freshness beats completeness. A slightly simpler instruction now beats a perfect one late. And critically, everything is timed against the live video, so the device always knows exactly when to act. Today, that rendering is smart and deterministic. Tomorrow, as on-device AI matures, we can make it progressively more generative. We've deliberately built the protocol so that evolution doesn't require reinventing it.” 

More widely, AI and the pace of industry iteration have also created challenges for this Accelerator, which Tiku breaks out into two linked points: “First, [the challenge] is building all of this to be fit for live broadcast. Offline AI can take its time. Live sport gives you milliseconds. Every model and every decision has to earn its place in the latency budget. 

“Second, building for a moving target. AI, devices, and rendering are improving every month. We had to design something disciplined enough to work today, and open enough not to be obsolete next year. Reaching that balance between shipping now and not painting ourselves into a corner has been the real engineering challenge.” 

IBC2026: The Delta Protocol – Live Demonstration 

The IBC2026 show demonstration will centre around a football match to provide a strong proof-of-concept, according to Tiku.  

“For the IBC demonstration, we have built a reference end-to-end implementation: ingesting a live feed, generating Delta instructions, transporting them in sync with the video, and rendering them frame-accurate through a client-side overlay SDK. A live command centre visualises the protocol in action. Two distinct customer devices deliver the experience that is composed just for them. 

“The long-term vision is transformative; the video path becomes entirely optional. The client device reconstructs the viewing experience locally using neural networks, consuming only the semantic track. That unlocks a future of unprecedented personalisation, efficiency, and intelligence at the edge. This goes beyond an incremental improvement to streaming. It's a new generation of sports broadcasting that is intelligent, hyper-personalised, and built for the AI age.” 

The wider industry context: “Exactly what innovation needs to thrive.” 

Tiku is bullish about the value of the Delta Protocol project to the wider industry, both from a practical point of view and as a waymarker. “The fact is that AI in broadcasting doesn't have to just sit around the stream, making highlights and metadata. It can become part of the stream and its architecture. The moment scene understanding becomes real-time and timed to the video, a completely different set of possibilities opens up, including true personalisation, live augmentation, and device-native enhancement. These are all things pixel delivery was never designed to do. 

“The second learning is that we don't need to invent new physics to get there. The building blocks, computer vision, GPUs, modern transport, capable devices, already exist. The opportunity is connecting them intelligently. The question is no longer whether AI can watch video, but what happens when watching becomes part of delivery.” 

Tiku also sees considerable value for the technology and media industries in driving innovation at speed:   

“Broadcasting is at an inflection point. The economics of traditional delivery are under pressure. Audiences expect personalisation that linear models can't provide. And the technology to do something genuinely different, not just marginally better but categorically different, finally exists.  

“The IBC Accelerators programme forces you out of theory and into a working system, alongside partners we wouldn't otherwise be building with at this pace. That collision of capability is exactly what innovation needs to thrive.” 

“We joined to set the pace and define what comes next. For DAZN, this has been a powerful vehicle to pressure-test ideas central to our long-term strategy, in the open, with some of the best minds in our industry. The learnings flow straight back into our product thinking, and the Delta Protocol demo is the first working proof of where this industry is heading.” 

The Delta Protocol: Live Media Reinvented Accelerators Project will be presented in the Accelerator Zone at IBC2026, 11-14 September at the RAI, Amsterdam. 

 

Champions:  

DAZN 

EBU 

IET

RAI 

 

SMPTE 

SBA (Saudi Broadcast Authority) 

 

Participants:  

Eluv.io 

Google Cloud 

NeoViews 

V-Nova