After the driest April on record in Catalonia, rain finally arrived in May. On the 25th, the day with the highest precipitation in Barcelona, we extracted climate data from Meteocat, the Catalan meteorological service, and transformed it into Generative RAIN: an audiovisual experiment that generates stories automatically using artificial intelligence models. Every hour, based on the forecast, the input instructions changed, producing texts, images and sounds that we then selected and organized.
Technical note
The project was made in 2023 for Barcelona’s +RAIN Film Fest, largely with Deforum, alongside the first video models such as Runway’s Gen. The main tool, open source and released in late 2022, animated images with Stable Diffusion frame by frame from a Google Colab notebook. There was no interface. Each shot was programmed by writing parameters, from the seed to camera movements in coordinates, and the result only appeared after hours of computation.
Experimenting with those early models was a huge challenge, but they had an aesthetic potential of their own that we have since seen swallowed by current generators, which tend toward photographic representation. That search was the value of the experiment, a mechanism capable of triggering everything automatically from weather data at a time when videos made from a single sentence did not yet exist.
Direction and Screenplay:
Anna Giralt Gris and Jorge Caballero
Project created for the +RAIN Film Festival.
