'Good' Deep Fakes 👪
This Week in MediaTech: Friday 5th March, 2021
One thought for the week 💭
Deep learning is democratising content creation
[3-minute read]
A series of deepfake videos of Tom Cruise has been going viral on TikTok over the past week. The videos are so realistic that millions of users are having a hard time believing that this is not actually him.
Deepfakes are so-named because they use deep learning to create fake content. This can take the form of an image, video, text or audio.
The quality and fidelity of deepfake content has improved markedly over the past few years, and it is fast becoming indistinguishable from traditional media.
The potential for misuse is obvious. Perhaps the most troubling outcome would be the total collapse of trust in digital content.
However, there is a brighter side to deepfakes. As with many instances of new technology, big things start out looking like toys. The same holds true with the technology underlying deepfakes. This seemingly pointless and sinister technology is ushering in the next major evolution of media creation.
Deepfakes ⊆ Synthetic media
Deepfakes are a subset of synthetic media - which is a term used to describe digital content that is generated or modified by AI. Due to the evolution of Generative Adversarial Networks (GANs) over the past few years, the capabilities of synthetic media have grown significantly. It is currently being used to generate content in applications which include:
Converting structured data into written reports (e.g. AX Semantics), producing creative texts (e.g. AI Dungeon) or creating interactive conversations (e.g. Replika).
Synthesising speech for virtual assistants, computer games, podcasts, or audiobooks (e.g. Resemble, Sonantic, Descript).
Generating music (e.g. Popgun.ai, AI Music,Loudly, Jukedeck).
Generating stock images (e.g. Rosebud AI, Icons8).
Generating ‘virtual product placements’ (e.g Ryff).
The evolution of Media [→ Media 3.0]
Media technology has seen two tectonic shifts to date. Thanks to advances in synthetic media, we are on the brink of the third.
This is because synthetic media will streamline or completely automate the creation of content. This will drastically reduce the cost, and therefore the barriers to entry of content creation.
‘Old media’ →mass distribution for a select few creators, through TV, radio and print. Enabling technology: broadcast technology.
‘New media’ →democratised distribution for everyone through social media and Web 2.0. Enabling technology: the internet.
‘Synthetic media’ → democratised creation for everyone. Enabling technology: deep learning.
The state of play
Where are we in the transition to Media 3.0? That depends on the content type in question.
Images - 75% there: Over the past decade, there has been a major reduction in the barriers to entry for image creation. Smartphone cameras, computational photography and tools like Illustrator, Photoshop and Canva have democratised the creation of high-quality digital images. The marginal cost of producing images is approaching zero, which has led to an abundance of content across the internet. There is still room for further automation. For example, OpenAI showcased the DALL·E model a couple of months ago which can produce images from any text prompt.
Music - 30% there: Music DAWs like Ableton and Logic democratised access to music production, albeit to a lesser extent than imagery. High-quality music can now be produced by anyone, from anywhere, without the need for prohibitively expensive studio equipment. However, a significant level of skill and knowledge is still required to create something that sounds even remotely professional. New tools are emerging - like Endless - which continue to reduce the friction in music creation. However, there is still plenty of room for deep learning to streamline processes and reduce complexity.
Video - 10% there: Professional video creation has not made the same progress as images and sound over the past decade. A huge delta in quality still exists between Hollywood studios and YouTube creators.
The reason for this is that unlike image and sound, editing and generating high-quality video requires specialist cameras, actors, studios and VFX. Weeks are required to capture and edit a few minutes of video content. The process is highly laborious, expensive and complex.
However, synthetic media is slowly disrupting several key aspects of the video creation process. For example:
Post-editing and correction of videos and audio can be done using Overdub by Descript.
Photo-realistic landscapes can be generated using GauGAN by Nvidia.
3D motion animations can be created without the need for hardware or expensive suits using RADiCAL or Sway.
VFX can be generated on any mobile using Beamm.
Videos can be automatically dubbed in multiple languages using Papercup.
Each of these processes uses deep learning to automate a traditionally manual and expensive task.
These solutions are not currently suited for high-end use cases. However, as per Clay Christensen’s take on innovation theory - the most disruptive technologies enter at the bottom of the market and continue to move up-market as they (inevitably) grow in sophistication. Eventually, the emergent technologies completely displace incumbents.
That is the future of content creation that I believe we are about to enter. This will start at the bottom of the market. Deep learning will continue to make traditionally ‘Hollywood-grade’ processes accessible to anyone. Over the coming years, these impacts will compound. Eventually, companies and individuals will be able to produce 10x the content that they do today, with 10x the quality at a fraction of the cost. Production times will accelerate due to a shrinking gap between idea and content creation.
Totally new forms of communication and storytelling will be born as a result.
News from this week 🗞
Gaming 🎮
Epic Games is acquiring video game studio Mediatonic, the maker of the smash hit "Fall Guys," for an undisclosed sum. The deal marks one of Epic’s higher-profile acquisitions of late, following its 2019 purchases of social video app Houseparty and Rocket League developer Psyonix. Link
Esports betting platform Rivalry raises $20m. Rivalry launched officially in February 2018 with a core focus on esports wagering but has since expanded into traditional sports offerings and proprietary casino games. Notably, the betting platform owns all of its intellectual property, forgoing white label solutions with in-house software development, which has helped provide a more intimate and customised user experience. Link
Simplebet raises $15m. Simplebet is a three-year-old developer of sports betting software that allows users to wager on quick odds, like which player will make the next catch in football. The company has now raised $50 million altogether. Link
Theorycraft Games raises $37.5m. Founded by developers from studios like Blizzard, Bungie, Riot, and Valve, Theorycraft Games will be using the influx of funds to recruit talent and develop its first project, a community-driven PvP game available on multiple platforms. Link
Audio 🎧
Meeting transcription service Otter.ai raises $50M. Over the past year or so, voice transcription startup Otter.ai doubled down on the future of remote work by integrating its product with meeting apps like Zoom and Google Meet. Otter has transcribed over 100 million meetings with more than 3 billion minutes and has seen an 8x increase in revenues during 2020. Link
Sounder raises $4m. Sounder is an end-to-end podcast management and monetization platform for creators. Its monetization solutions have already yielded meaningful impact for creators, with some partners seeing an average of 35% increase in streams, an 80% average increase in bid requests for programmatic campaigns, and ad rates more than 25% above the industry average. Link
Kings of Leon is releasing an album as an NFT, with tokens that unlock special perks, including limited-edition vinyl, front-row seats to future concerts, and audiovisual art. Link
Square acquires the majority of Tidal, Jay-Z’s streaming service for $297 million. Link
Video 📹
Live-streaming platform Maestro raises $15m. Those investing in Maestro include Sony Music Entertainment and Chinese tech giant NetEase. In addition to live-streaming tech, Maestro offers its music clients monetisation tools that enable them to sell tickets and merch, set up pay-per-view and subscription events, and to interact/give “shout outs” to fans who tip on the platform. Maestro is capable of hosting live, virtual, and augmented reality concerts, or a mixture of the three. Link
Celebrity video platform memmo raises $10M. memmo is a direct competitor of Cameo. The company is currently operating localised marketplaces for Sweden, Germany, Finland, Norway, the United Kingdom, Spain, Italy and Canada. Link
Around, a video collaboration startup has raised $10 million. Around helps companies communicate collaboratively over audio and video without the formal meetings of today’s remote world. Link
Interactive live event platform Looped raises $7.7M. The startup powers music concerts, comedy and other entertainment. It has now raised $8.8 million in total. Link
Other 🤷♂️
Newsela, the ‘replacement for textbooks’, raises $100M and becomes a unicorn. Newsela is a SaaS platform for K-12 instructional material It defines its content as “material that isn’t purpose-built for education, [but] purpose-built for being interesting and informative.” If Newsela is doing its job right, the content can replace textbooks within a classroom altogether, while helping teachers give fresh, personalized material. Link
Sorare, the NFT digital collectible football platform announced that it has received €40 million in Series A financing, led by Benchmark, with additional funding from Accel and new business angels. The angels include Reddit co-founder Alexis Ohanian, Vaynermedia CEO Gary Vaynerchuk, football players Antoine Griezmann and Rio Ferdinand. The financing brings Sorare’s total funding to €48 million. The company, which is already profitable, has experienced an explosive 52% month-on-month growth over the past 12 months, going from €50K worth of cards traded on the platform in January 2020 to €3.5 million in January 2021 across 120 countries. Link
Google moves further away from tracking. While Google had already announced it would be phasing out support for third-party cookies in Chrome, it went further today by declaring that “once third-party cookies are phased out, we will not build alternate identifiers to track individuals as they browse across the web, nor will we use them in our products.” In fact, Google’s David Temkin argued in a blog post that attempts to build alternative approaches to ad-tracking will not “meet rising consumer expectations for privacy, nor will they stand up to rapidly evolving regulatory restrictions, and therefore aren’t a sustainable long term investment.” Instead, he pointed to Google technologies like its interest-based Federated Learning of Cohorts. Link
Interesting data from this week 📈
Shopify generated $120bn of GMV in 2020… that is a lot of D2C activity.
Source: Ben Evans