«GES-2”—a former power plant from 1907 on the Bolotnaya Embankment, repurposed as the V-A-C Foundation House of Culture. The building, with its two blue smokestacks, has become a canvas for video mapping.
Loading...
The premise is based on the story of a spirit-like girl who lived within the station for a long time and finally revealed herself when the building found new life.
Loading...
Loading...
To bring the idea to life, we integrated illustrative imagery with outline illustrations that carry the identity of GES-2 and the V-A-C foundation.
Loading...
- Character Development
The character concept and visual characteristics were formulated using Claude. Based on these, we generated a character sheet in ChatGPT Images 2.0 (gpt-image-2), detailing facial features, proportions, hairstyle, color palette, and rendering style. This sheet served as the primary reference for all subsequent generations, ensuring character consistency.2. Prompt Engineering
The prompts were constructed and iteratively refined using ChatGPT and Claude. Each prompt independently specified composition, stylistics, pose, facial expression, and movement characteristics; following each iteration, the wording was clarified based on the results.3. Image Generation
Key frames and styleframes were generated in Nano Banana 2 (Gemini 3.1 Flash Image) and ChatGPT Images 2.0 using image-to-image mode: the original character image combined with a text description of changes to pose, angle, or detail. The environment for rendering the 3D facade model was generated in ChatGPT Images 2.0 using a photographic reference of the location, without altering the building model itself in the frame.4. Video Generation
Videos were generated in Kling 3.0 (Video 3.0 Omni, Motion Control) and Seedance 2.0 using image-to-video and first/last frame modes. The input included a key frame, a style reference, and a description of the motion. The stop-motion aesthetic was defined in the prompt through a low frame rate and the absence of smooth interpolation.5. Generation from Hand-Drawn Animatic
For certain scenes, the storyboards were hand-drawn frame-by-frame and compiled into an animatic. In Seedance 2.0, the animatic was uploaded as a reference video alongside a style reference image. File roles were assigned in the prompt using @-mentions: the video provided composition, timing, and motion trajectory, while the image provided the visual style. This ensured the motion remained original while the rendering matched the approved aesthetic.6. Post-Production
The generated clips were assembled in After Effects. 2D outline animation and effects were added on top, and the material was then adapted to the geometry of the facade for video mapping.
