Research Scientist, Real-Time DiT Video Models

Responsibilities

• Design, develop, and optimize diffusion-based AI models for high-fidelity video synthesis with attention to temporal consistency, realism, and style control.
• Research and implement state-of-the-art algorithms for controllable video generation and real-time inference.
• Work with large-scale video datasets covering human motion, gestures, facial expressions, and scene context; lead dataset curation and preprocessing pipelines.
• Prototype and evaluate novel diffusion architectures (including DiT-style and transformer hybrids) for video tasks.
• Define and implement robust validation strategies and custom evaluation metrics comparing synthetic vs. real gameplay and interactions.
• Collaborate with product, engineering, and animation teams to integrate models into interactive real-time systems deployed from our UAE office.
• Stay current with top-tier literature (CVPR, NeurIPS, ICCV, ICML, SIGGRAPH) and translate research into production-roadmap milestones.

Requirements

• PhD
• Video diffusion
• Research publications
• Real-time ML
• Python
• PyTorch
• Neural architectures
• Dataset handling
• Evaluation metrics
• Model optimization

Preferred Qualifications

• Track record of publications at CVPR, NeurIPS, ICCV, SIGGRAPH or ICML.
• Experience with DiT-style models, video transformers, or autoregressive video methods.
• Background in human motion and facial expression modeling.
• Experience deploying ML models for low-latency, real-time applications.
• Strong software engineering practices: reproducible experiments, CI for models, and scalable training pipelines.
• Prior experience working in cross-disciplinary teams (animation, graphics, audio, product).

Benefits

• Annual leave
• Medical insurance
• Work visa sponsorship
• Annual ticket (flights)
• End-of-service gratuity

About the Company

Valka is a spin-off from the Realms Group (parent company of Oddin.gg) building an interactive human-digital platform that enables co-created, real-time generative content. Our mission is to move content from passive consumption to active participation: virtual characters that respond dynamically to voice, text, gesture and context. The role is based in the UAE (Dubai) and will collaborate closely with engineering and product teams to bring research prototypes into production for gaming, entertainment, education and beyond. We encourage applications from motivated ML researchers eager to innovate at the intersection of computer vision, generative models, and real-time systems.