Wildberries

I built the motion design system behind Wildberries advertising from scratch. It runs across every type of campaign, from seasonal sales like Black Friday and the summer sale to always-on formats that stay on screen all year. The same rules apply to the core brand and to its services, including WB Travel, WB Taxi, WB Resale, WB Books and WB Bank, and they hold in all eight countries where Wildberries operates: Russia, Belarus, Kazakhstan, Uzbekistan, Kyrgyzstan, Armenia, Georgia and Tajikistan.

Most of this motion runs out of home. The group has its own advertising network: screens in more than 150 Russian cities, in the Moscow, St Petersburg and Kazan metros, on trains and at stations, in 45 airports and in about 10,000 Wildberries pickup points. Its combined reach is over 88 million people a month. Every one of those screens has its own size and ratio, and the system is built so a campaign can move between them without being redesigned.

In two years I worked on more than fifty ad campaigns. Each one looked different: some were 3D videos, some were 2D, and for many I was involved from the idea. My part was the master animation. From there the formats were either produced automatically or resized by a team of motion designers working from my masters.

The work I enjoy most is building a custom procedural setup around a campaign's identity, where one controller drives the whole animation. Some of these setups are shown below.

Over two years the work kept changing. Most of our 2D and 3D production moved toward AI generation, because it can be automated, it's faster and it gives more output for the same effort. A large part of my job became building generation pipelines for different tasks.

One of them was built for Rive Gauche, the perfume and cosmetics chain within the Wildberries group. The input is a single link to a product photo. The pipeline generates three views of the product, builds a 3D model from them and hands it to a prepared Blender scene, where a script imports the model, applies a neutral material and renders a motion sequence. A video model then uses that sequence as its motion reference and puts the real product back on top.

Done in classic 3D, every video would mean modelling the product, unwrapping UVs, setting up light and animating it. With the pipeline it takes two clicks and under five minutes to generate a 5–10 second video. The result looks like a 3D product shot, stays stable from run to run and can be tuned.