ASIMOV — Diffusion Video Compression

As part of the ASIMOV platform, led an R&D initiative into diffusion model-based video compression — a novel approach that uses generative AI to reconstruct video frames rather than relying solely on traditional codec techniques.

The result was significant file size reductions while maintaining visual fidelity across the types of high-motion, high-detail content common in live sports broadcasting. This directly addressed one of the core bottlenecks in deploying real-time holographic and streaming content at scale.

The compression pipeline was designed to integrate cleanly with existing delivery infrastructure, allowing IKIN’s partners to benefit from reduced bandwidth costs without changes to their downstream playback systems.

Person & Brand Reliability Through Guided Augmentation — a critical requirement for sports and advertising applications is that the generated or compressed output preserves the integrity of real people and brand assets. ASIMOV addresses this through a metadata-driven augmentation engine that auto-captions source media and uses those captions as grounding prompts during generation. The system explicitly encodes identity-relevant attributes — hair style and color, wardrobe, background, and brand markings — ensuring that augmented or reconstructed frames stay anchored to the source subject rather than drifting toward generic AI output. Motion guidance further locks subject movement across frames, producing results that blend realistically with live footage while remaining fully brand-compliant.