Eisspin, the Oh So Terrible / Wormcatcher / The Salt Bag / Alpha channel in machine learning / ORIONO / Cyclic

Eisspin, the Oh So Terrible: Based on the novel "Der Schrecksenmeister" by Walter Moers, the team combined state-of-the-art technology like previs and virtual production with traditional narrative methods from live-action film in an extremely complex production process lasting several years. The project was created as a bachelor's and master's degree film at the Film UniversityBabelsberg KONRAD WOLF. The approximately 4-year development process was characterized by the intensive collaboration between the departments of the film university with the VFX, acting, C-Tech and digital media culture courses as well as the great supportof fans of the novel from the Zamonien community as part of a successful crowdfunding.

Filmuniversität Babelsberg KONRAD WOLF: Sebastian Esposito  (VFX Compositing Artist), Neele K. Siemann (VFX Compositing Artist)

Wormcatcher and How to make footsteps sound scary:  Blood? No. Guts? No. Feet? YES! - Wormcatcher is an abstract narrative horror game about the disintegration of a relationship, in which walking through the remnants of the relationship becomes an uncomfortable experience. In abscence of gory imagery, Wormcatcher leans on sound design and the breaking of kinesonic synchresis to create horror. 

University of Applied Sciences Upper Austria – School of Informatics, Communications and Media:  Lilith-Isa Samer  (Audio Lead & Writer),  Anton Schneeberger (Concept Lead & 3D Artist), Simone Feldbacher (Gameplay Programming Lead)

Made out of 1597 images of carefully positioned salt, The Salt Bag is a stop-motion animated film telling the eerie story about why a close friend always carries a small bag of salt with her. Combined with a spooky sound atmosphere, the film uses the characteristics of salt to create an array of visual sequences to make the story come alive. The film itself has been completely animated and edited by one person in the span of 11 days.

Hochschule Mainz — University of Applied Sciences:  Thomas Range (director/animator)

Alpha channel in machine learning: The importance of the alpha channel in machine learning: Generating ground truth and using the result for compositing.

Part 1: With the introduction of machine learning tools within VFX, it is important to understand how to get the best results from them. Inevitably this is down to the quality of the data that the network has been trained on. Keying & rotoscoping are functions that every compositor is aware of and would ideally like to reduce the amount of manual labour involved. The question is "what is a good key", or even more apt is "what is the perfect key". Given the simple compositing equation C=A+B(1-a), it is easy to see how we have a single equation with 2 or 3 unknowns, which means it is not possible to get a single perfect answer. This is certainly true when filming, however if the structure of the problem is altered it is possible to generate a set of simultaneous equations to generate the perfect ground truth. We have been investigating this as a process for keying tulle. This netted fabric has a translucent property, highlighting when its alpha matte doesn’t have the correct falloff. Excitingly, the possibilities do not end there, with many fine details, such as hair, being effectively keyed by going through the following process.

Part 2:  The project topic is an investigation of neural style transfer methods and techniques to implement them into the compositing pipeline. My research focuses on ways to transfer an artistic style to an image sequence to preserve the most characteristic qualities of the style, including, correct representation of artist gestures, used tools and mediums, but also takes into consideration important aspects of compositing such as working with alpha mattes, control over separate layers and temporal coherence. The artist whose style I picked for the project was XIX-century painter Auguste Renoir.

In the research process, I tested various existing methods of neural style transfer from CNNs and GANs to Diffusion models and compared them in terms of time efficiency accessibility and ease-of-use and all already mentioned aspects. Moreover, I trained project-specific LORA models with the aim of achieving desired results. The research goal is to choose the best combination of techniques to meet the set requirements and perform a style transfer over the composited sequence of images to produce a shot showing a “brought-to-life” Renoir painting, the breaking of 4th wal.

The investigation was supported by various academic resources covering the science behind machine learning and different neural networks but also insights and experience of industry professionals and computer scientists who used machine learning-based style transfer in their work.

Arts University Bournemouth: Oktawia Sokalska (Researcher, Compositor), Megan Thomas (Researcher, Compositor)

Oriono: Despite having no arms or legs, Armin Nielsen is going to be the first human being to put a foot on the planet, farthest away from earth: Uraban 5.  But no one knows what expects him there. Oriono is a 3D-animated short film with stunning textures, 2D-animated VFX Animation and a thrilling story.

Technische Hochschule Ostwestfalen-Lippe: Max Laux  (Director), Jan Brakhage (Technical Director, Director of Animation), Ali Halawi (Art Director, Director of Animation)

In a mesmerizing spectacle of creation and renewal, Cyclic invites viewers on a spellbinding journey through diverse landscapes as an abstract sphere undergoes a cyclic odyssey. This short film immerses the viewer in an impressive world in which a sphere rolls through unique environments.

Hochschule der Medien: Philipp Kamm (Art & Technical Director), Célime Mörbt  (3D Generalist & Compositing Artist)