Prague '89

Prague ‘89 is a research sprint conducted by the Carl Laemmle Institut in collaboration with Filmakademie Alumnus Adrian Meyer, now working as AI & VFX Supervisor. Its primary purpose was to seek further insights into the capabilities of generative image and video AI technology (GenAI) for artist-controllable film creation processes.

 

The narrative context is based on the events in the fall of 1989 at the West-German embassy in Prague, where foreign minister Hans-Dietrich Genscher confirms the granted travel permit for hundreds of East German refugees. This event was originally captured under poor lighting conditions and in low resolution.

 

Prague ’89 explores the extent to which GenAI can be used to reimagine a historically significant, visually well-defined moment — with controllable visual language, an approximation of the authentic architecture, a recognizable historical figure, and a reconstructed atmosphere. The moment is consciously approached as a reenactment, not as a representation of a singular "historical truth".

 

The idea to re-stage this historical moment had its origin in the R&D Department many years ago in the context of the Digital Actor in Film. Its intent is to re-create this moment in a more contemporary visual appearance with a clear message: To keep believing in peaceful change in times of growing global crisis.

 


The research sprint surveys the capabilities of GenAI to stage a specific setting under established film creation processes – rather than a “prompt and go” approach. Adrian Meyer first started with a rough Animatic. He then moved on to create a vertical slice for one key shot on the balcony. This also required spatial accuracy with the architectural reference still existing. The findings, pipeline and challenges of this vertical slice were presented at FMX2026 during the Carl Laemmle Institut’s introduction and the project was completed afterwards. In terms of complexity, the sprint pointed out how many iterations and manual adjustments were required to achieve the desired output; hundreds of generations were made. But it also proved, that very believable and controllable results are achievable with enough skill and effort. The workflow was documented and builds the basis for a verifiable creative contribution as compared to a fully machine-driven prompt result.


Additionally, the research sprint involved expanding the capabilities of our render farm (RoyalRender) to also distribute and coordinate ComfyUI workflows in our network including workstations and dedicated GPU clusters. This significantly contributed to our effort on sovereign solutions as ComfyUI data can now be processed locally. The next step will be scaling this infrastructure to a regional cloud provider, ensuring more compute on demand in secured environments.