Director: Stephanie Soho
Prod. Company: BWGTBLD
Agency: Media Monks
Client: BMW
The brief was genuinely exciting from the first read. Lil Miquela — one of the world's most recognised virtual influencers — experiences a series of real-world scenarios: watching children play on a beach, losing herself in a club, navigating everyday life alongside the new BMW iX2. Director Stefanie Soho's treatment was clear and ambitious, and we said yes immediately.
What followed was one of the most technically and creatively demanding projects we've taken on. Thirty-one face replacement shots we had to build from the ground up. A top-secret car that didn't officially exist yet. And a protagonist we were contractually obligated not to change.
The problem with making her too real
Working with an existing IP like Lil Miquela means you're a custodian, not a creator. DapperLabs provided animation caches so we could match her expressions down to the pixel. Every decision had to satisfy BMW, Stefanie, Media.Monks, and Lil Miquela's management simultaneously. We couldn't alter her proportions to better match the actress, for instance, even when that would have made our lives considerably easier.
The deeper challenge, though, was calibration. The entire film is rooted in authenticity — real moments, real emotion, real life. Miquela needed to feel present in that world without dissolving into it. Too real, and the tension that makes her interesting disappears. The comparison process — actress side by side with AI output, overlaid with reference imagery of Miquela — was meticulous and ongoing throughout the project.
Constructing LilMiquela
Our face replacement method was built on top of the open-source DFL (Deep Face Lab) framework, but pushed well beyond its standard capabilities. The most significant departure was resolution: while the convention sits at 768 pixels by 768 pixels, we had to create a UHD image — a decision that demanded far more compute but was non-negotiable for the quality we needed.
We generated over 20,000 images of Miquela at 4K×4K resolution, capturing her face across a wide range of lighting scenarios, and ran more than 3,000,000 training iterations over two weeks of continuous cloud rendering on AWS. When the model showed a preference for certain lighting conditions, we generated more images to recalibrate it. The model needed to be versatile — the footage spanned wildly different environments — and getting it there required constant testing and fine-tuning.
The compositing process was where the output came alive. Highlights, eye reflections, wetness, skin texture — all were refined in comp, including in extreme close-up shots where AI replacement typically struggles most visibly.
The car that didn't exist
The iX2 was still under wraps during production, which meant every city shot — day and night — was filmed with a camouflaged stand-in. Our job was to re-skin it in post, which meant building a true digital clone: BMW's paint has genuinely complex optical behaviour, and the light clusters have intricate internal geometry that refracts light in distinctive ways. We brought in one of the industry's leading CG automotive artists and preserved as much of the real car as possible — wheels, certain windows, structural elements — to keep the replacement invisible.
A global team, one vision
The project ran on our distributed model — 28 artists across Canada, the US, Brazil, the Seychelles, Spain, France, Armenia, India, the UK, and more. Each was selected specifically for their role, not pulled from a general roster. Weekly live reviews with BMW, Stefanie, and Media.Monks kept everyone aligned across time zones and creative departments.
It's the kind of project that tests every system you have. Ours held.
What we think this means
We think BMW's Make It Real does put the bar very high as a genuine creative use of a set of tools to enable the director to do just that, direct.
We're grateful to BMW AG, Stefanie, and the entire Media.Monks team for the trust they placed in us to go somewhere new. And to the 28 artists who made it happen — thank you.