ARD Olympics 2026

2026ARDMotion Design / Broadcast Branding / Art Direction Support

A sporting dream has to feel extraordinary and still be recognisable as sport. For ARD’s Milano Cortina 2026 campaign, UnitedSenses returned to the cloud athlete: a figure that can move between human effort and an imagined landscape. AI image and video generation offered a new way to develop that idea across trailers, idents and social formats, with the sporting gesture holding the visual world together.

The brief — a cloud-athlete idea returns

UnitedSenses’ 2010 ARD campaign for the Vancouver Winter Olympics established the cloud athlete as a recurring visual idea: sporting figures taking shape in the sky, woven into a journey through winter landscapes. The brief for Milano Cortina 2026 returned to that idea and proposed a new expression of it through AI image and video generation.

Under the working theme “Der große Traum” — “The Big Dream” — clouds offered a way to connect anticipation, athletic ambition and the host landscape. A sporting figure could form, move and dissolve into another discipline or a mountain range. The Dolomites and Milan’s cathedral anchored the proposed dream journey in the setting of the Games, giving its transformations a shared destination.

The ambition extended beyond a single image film. The concept proposed a shared visual language for event teasers, openers, advertising breaks, studio backgrounds and social formats. Recognisable sporting gestures and a coherent cloud aesthetic would allow individual disciplines to stand on their own while belonging to the same campaign.

Original campaign — Vancouver 2010

AI Cloud Athletes

AI Fans

“Built as a system. Powered by AI. Still very human at heart.

Campaign approach and quoted statements: UnitedSenses — ARD Olympics 2026.

Making of — learning the shape of a sport

A cloud athlete needs more than a convincing silhouette. The posture, balance and rhythm have to read as the sport itself: a low stance through an alpine turn, the extension of a ski jumper, or the characteristic movement of another winter discipline.

Custom LoRAs were trained for different models and continuously refined to bring that sporting specificity into the cloud characters. The process combined model-specific training with repeated generation, comparison and selection. Across thousands of generations, suitable images and animation sequences emerged gradually, with realistic sporting movement as the guiding criterion.

Direct comparison with a sport reference kept the cloud treatment accountable to the movement underneath it. Pose and silhouette had to remain legible as the generated material changed. The ComfyUI comparisons document that image-development work; animation required its own iterative generation and selection. A convincing still was a starting point, rather than proof that the figure would move correctly.

ComfyUI Workflow

Bringing the reference, pose and depth guides, generated variations and comparisons into one ComfyUI workflow made it possible to judge the same sporting action across successive interpretations. The ski-jumping study documents that feedback loop: retain the mechanics of the gesture while developing the cloud character around it.

Qwen LoRA — finding the sporting gesture

The Qwen LoRA studies test how much cloud volume and variation a sporting gesture can carry before it loses clarity. Keeping the reference and pose guide alongside each result makes the criterion concrete: the lean of a slalom turn, a biathlete’s stance and a cross-country skier’s balance should remain distinct, even within the same visual material.

Credits

Client

ARD

Motion Design · Broadcast Branding · Art Direction Support

Matthias Lein

Agency

UnitedSenses