
The Innovation Behind Stadium-Scale Avatars
How IGA turned hockey players into branded 3D avatars using generative AI
During Montreal Canadiens home games, the Bell Centre lights up with a larger-than-life guessing game. On the arena’s massive 8K jumbotron, a 3D avatar slowly takes shape: big eyes, bold features, familiar vibe. Kids lean in, racing to name the player before the final reveal.
These avatars aren’t random. They’re built in IGA’s signature animated style, with soft shapes, warm lighting, and that playful, cartoon-like look pulled straight from the grocery chain’s ad campaigns. IGA, a longtime Canadiens partner, commissioned ten custom portraits that blend real player features with the brand’s visual DNA.
The activation required solving a technical problem: creating jumbotron-quality portraits that work at 60-foot width while preserving both player likeness and brand identity. Traditional 3D modelling would have taken weeks per player. Instead, Sid Lee built a custom AI pipeline, training a diffusion model on existing brand imagery to generate avatars that look authentically "IGA" at stadium scale.

Training a Model on IGA's Visual Signature
The team trained a diffusion model using IGA's comprehensive visual archive, including Christmas campaigns, in-store displays, and advertising creatives. The model learned IGA's specific shader behaviours, lighting conditions, material qualities, and proportions, instead of mimicking surface appearances, and understood how the brand's cartoon warmth is actually constructed.
A custom rigging system standardized camera angles and facial expressions while preserving individual player characteristics. Cole Caufield needed to feel like Cole Caufield, not just another 3D face or mistaken for someone else. Since these avatars had to hold up on a 60-foot 8K screen, the team developed a multi-stage upscaling pipeline that protected tiny details like hair strands and fabric texture without introducing artifacts.

The Artisan's Hand
The AI got the avatars most of the way there. Sid Lee’s artists took them across the finish line. Every portrait went through rounds of refinement: proportions adjusted, lighting balanced, materials polished until everything hit broadcast-quality standards.
Each athlete required unique generation parameters, developed through consultation with the players themselves, to preserve what makes each player distinctive and ensure their individual character came through authentically. Recognition accuracy mattered more than technical perfection—fans needed to immediately identify their favourites in stylized form.

Beyond the Jumbotron
Ten fully realized avatars were delivered, each one capturing a real player’s identity inside IGA’s animated world. Building them with traditional 3D tools would have taken weeks per character, with a full production team behind it. For a game-day activation, that just doesn’t scale.
But this tech isn’t just a one-off. IGA’s entire brand platform revolves around its 3D characters, used everywhere from TV spots to in-store displays to billboards. With this AI pipeline, that universe just got bigger.
Now, the brand can bring real people into its world. From the employees, to local legends, community faces, even ambassadors. It’s a new tool in the creative kit. One that keeps IGA’s style intact while opening up new ways to tell human stories, at any scale.