It's no longer necessary to follow Elon Musk's Twitter or subscribe to Two Minute Papers on YouTube to witness how machine learning is changing the world. In 2019, "Deep Fakes" are a mainstream concern, neural style transfer and image colorization are free consumer products, and a serious presidential candidate is even running on the platform of concern about automation.
How might these emerging technologies affect the animation industry in the next decade? If I knew for certain, I'd probably be too busy launching a software startup to blog about it. Be that as it may, here are my 7 predictions for how the animation industry will evolve on account of machine learning advances:
1. Generating Textures From Photographs Become the Norm
It becomes increasingly clear every year that manually building texture maps is a dying art. The technology to generate high-quality textures from just one or two photographs is already here. I am quite confident that the surfacing workflow at animation/effects studios will shift on account of these techniques. In the coming years, increasingly robust tools will make it even easier to iterate upon photo-scanned textures and the quality will continue to improve.
2. "No-Suit" MoCap Approaches Quality of Traditional Solutions
Modern, high-quality motion capture can be prohibitively expensive for smaller studios. In the next decade, I predict that machine learning will elevate lower-budget markerless solutions into a legitimate alternative, especially for small and medium-sized studios. In the long term, "no-suit" MoCap may eventually replace standard methods in all but the most particular of use-cases.
3. Rigging By Hand Becomes a Rare Chore
For many 3D artists, rigging is a difficult and tedious job required before the creative work can continue. Auto-rigging tools in software such as Autodesk Maya have already massively simplified the process of character rigging, and machine learning techniques have the potential to continue this trend. While I don't expect skilled rigging artists to be entirely replaced in the next decade, they will become more specialized and rely increasingly on machine learning assistance.
4. Automated Speech Animation Software Outperforms Human Animators
My last two predictions hold major ramifications for the future of speech animation. Manually animating a facial rig to realistically form words is a tedious task animators spend many man-hours working on. Researchers today are working on machine learning systems to automate speech animation with (and without!!) MoCap data. The technology isn't quite there yet, but in the next decade I predict it'll match human animators in quality, and far surpass them in speed.
5. Industry-Grade Style Transfer Software on the Big Screen
Style transfer is one of the more well-known techniques in the field of machine learning. Once only possible with still images, it is now improving rapidly within the realm of video as well. I predict that a major animation or film studio (perhaps in partnership with a tech giant) will release a film mostly or entirely processed with cutting-edge style transfer software. Just look how good the results already were in 2016! As a result of this technology, "Loving Vincent" may be the first and last film of its kind.
6. Neural Network Based Rendering Engines Begin to Take Shape
Physically based rendering engines work by running millions of computations every second to create photo realistic graphics. With machine learning, it may be possible to produce graphics of similar quality with much less processing power. Building a new breed of rendering engine from the ground up is a monumental task, but I predict great progress will be achieved in the next decade. Researchers at NVIDIA are already taking paradigm-shifting first steps towards this goal. It will be fascinating to see how machine learning affects the already accelerating adoption of real-time rendering.
7. Animation Career Opportunities Become More Creative, Less Monotonous
Many of the changes machine learning could bring to the animation industry effectively replace human labor with software. In the next decade, I expect artists will spend fewer man-hours creating textures, rigging characters, cleaning up MoCap data, and animating speech. Despite this, the animation industry will continue to grow. Artists will spend more of their time working creatively, and less time tediously managing things by hand. In the long term, animators may have to worry about losing jobs to automation, but for the next 10 years at least, I am optimistic.