Concept

As AI/ML progresses, there will be massive disruption and efficiencies created within creative industries.

Longer Description

Ever since the advent of Generative Adversarial Networks from Ian Goodfellow et al we have believed that a wide variety of businesses that control the middle and application layers could be created to enable more efficient, scalable, and impressive creativity.

We’ve invested against this thesis multiple times since 2017 and have seen it come to fruition increasingly throughout 2022 with models like Dall-E, Stable Diffusion, and more reaching the mainstream.

Investments include:

• Shadows
  • We incubated Shadows, a ML-first animation studio. The high level is seeing progress of GANs and unsupervised learning largely in japanese academic communities focused around anime datasets was eye opening (long thread here) and the shifting media landscape valuing IP and animation specifically was really compelling. Dylan (ceo/co-founder) was leading a group in new media at CAA so knew the space incredibly well, Corey (CCO, co-founder) runs a large animation studio called Shadow Machine (responsible for BoJack Horseman, among others), and so our three lenses of industry experience and technical experience on my end made for a compelling vision. Ultimately the thesis of building in-house tools to scale animation played out well in short-term, though in the longer-term the company focused less on native ML-tooling and more on unique 3D pipelines.
• RunwayML
  • We led Runway’s seed round (blog post here) and have been large supporters ever since. Our thesis was novel product primitives enabled by AI/ML would enable an Adobe competitor to emerge.
  • The learnings from shadows showed that there was a large oppotunity in ML-enabled creativity, but also in the tooling. Shadows was taking the approach of being fully vertically integrated, while we also felt that a non-technical platform for utilizing machine learning was needed across multiple industries. The Runway team had built the most beautiful and fully featured version of that we had seen.
• Stealth File Explorer Company
  • The concept of using machine learning to enable creators also extends to tackling the scale of problems. We had done prior work in understanding graph theory and ML-enabled data clustering across largely images and audio.

Other Thoughts

• We generally believe value will accrue at the middleware and product level, with fine tuning of models perhaps being a core short to mid-term moat. We are less interested in backing pureplay “ML Model” companies.
• Our investment in Runway precludes us from a variety of creative tooling investments, however we believe there are areas surrounding creativity they will not touch moving forward.

Comparable Companies

Related Reading

• We’ve written at length about this in our deep dive into animation and spoke about this thesis on Invest Like The Best as well as another podcast on ML and Creativity here.

Older Research Repo: