Mateusz Bykowski
CG Supervisor · Realistic CGI & Generative AI · Synthetic Data for Computer Vision · Industrial CG
- Location
- Kraków, Lesser Poland Voivodeship, Poland
- Experience
- 10 years
- Work preference
- Remote · On-site · Hybrid
- Availability
- Open in 1–2 months
About
CG Supervisor with 10+ years in 3D production and an MFA in fine arts. I care most about images that come from a clear, scalable pipeline. At Sky Engine AI, we build realistic simulations of the world, rendered as images and sensor data that train AI systems. For three years I've set the look for that data-accurate CGI and held it across hundreds of assets and millions of procedurally generated variants, leading a team of up to 19 artists. I define references, lighting, camera and materials before production starts, including for proofs of concept, hold the bar through the whole pipeline and supervise generative AI with artists in control. >Due to an NDA with my current employer, I can provide a demo reel on demand.<
Professional Experience
CG Supervisor
SKY ENGINE AI
• Supervision across the CG pipeline. One of two CG Supervisors reporting to the Head of CG. Asset supervision on every project; full supervision on some (assets, layout, shots, simulation, shading, compositing), specific asset groups on others (heads, hair, props, vehicles). 7 commercial and 3 R&D projects delivered on schedule, many of them extended. • Realistic assets at scale. Hundreds of hard-surface assets (vehicles, furniture, props, full buildings, a manufacturing area with robotic arms), mostly on procedural shaders with exposed parameters and variants. Topology, UVs and optimisation built to the rules of a proprietary engine, so one asset could feed thousands of variations. • Look development. Set references, lighting, camera and material targets before production, including for proofs of concept. Built two full skin shader systems in Substance Designer; lit and composed marketing key shots. • Digital humans, re-engineered. Replaced hand sculpting with an AI-assisted head pipeline I co-designed and supervised: 2D prep, 3D generation, then procedural topology fixes in Houdini. It produced about 60% of 420+ heads and cut time per head by about 80%. AI never reached the final image. • Generative AI with artists in control. Led R&D and validation of ComfyUI workflows (texture generation, denoising) and rolled them out for repetitive texture and geometry stages. Reviewed AI output for hallucinated detail, anatomy errors, colour shifts and inconsistent quality before anything reached a dataset. • Specs, feedback and tooling. Defined asset specs, naming, review and publishing rules. Feedback based on the root cause of an error and a tested fix, so recurring issues were solved once, team-wide. Regular workflow audits; commissioned automation for texture and geometry stages. One year of hiring and onboarding. • Ground truth as the quality bar. Every asset passed validation against pixel-perfect semantic masks, keypoints and accurate scale. Output was analysed frame by frame, so a flaw film or games would hide counted as a defect in the dataset. • Working with data science and ML. Translated client and data science requirements into asset and scene specs for artists and QC. Built diverse human data (adults and children across ethnicities, sexes and developmental stages), with anatomy checked against scientific publications to reduce bias in facial recognition models. • Working with rendering developers. Matched output to client sensor specs, from high-end telemetric cameras to low-quality CCTV, inside the in-house rendering engine. • Environments from real data. Forest and farmland terrain for UAV clients built from GIS data, set-dressed for low-altitude drone cameras with scatter-ready LODs, in a small R&D team working directly with data science. • Scoping and estimates. Daily man-day estimates, proof-of-concept scoping with the Head of CG and production.
3D Animator
SKY ENGINE AI
• Driver and occupant monitoring. Character and vehicle animation for automotive safety scenarios (Euro NCAP) and animal animation for AI monitoring systems. • Ground truth in motion. Keypoints, UV-based semantic segmentation and eye gaze vectors embedded in the output, where accuracy had to be absolute. Defined rigging and asset requirements for the animation team. • Stepped up. When the CG Lead left mid-project, I took over as supervisor and coordinator. We delivered on time, the project was extended, and I was promoted to CG Supervisor.
Digital Technologies Specialist
NED-Project
• Reality capture. ~35 tanker and container vessel projects built from laser scan and point cloud data, integrated with BIM geometry, for engineering clients with zero tolerance for inaccuracy. • Digital twins. Led a 2-person R&D team delivering 6 VR digital twin projects in Unity, from raw geometry through optimisation, materials and real-time lighting. • Visualisation. Eight years of 3D visualisations, animations and technical 2D graphics for naval engineering.