The introduction of ChatGPT in November 2022 marked a transformative second in content material creation. I’ve been writing about this since 2021 on VC Cafe. As an investor in leisure tech, gaming, and next-gen shopper tech at Remagine Ventures, this subject is of specific curiosity to me. On this submit, I’ll discover among the best-in-class examples of AI content material creation and try and forecast future developments on this quickly evolving subject.
Whereas the seemingly in a single day success of ChatGPT was end result of greater than a decade of iterations in ‘artistic automation’, the progress we’ve witnessed in AI over the past couple of years is poised to endlessly change the panorama of content material creation. McKinsey predicts that Generative AI might add to the financial system between $2.6 trillion and $4.4 trillion yearly. However Generative AI has but to dwell to that hype. As enterprise agency NEA places it, creativity is perhaps Generative AI’s first ‘killer’ use case.
The TMT trade stands on the forefront of this disruption, so it’s not stunning that it’s additionally the trade that has seen the very best adoption price of generative AI know-how (see graph under).
As LLMs turn into extra multi modal, the identical mannequin ought to have the ability to assist with all duties, however in the interim, devoted fashions to a selected sort of output carry out higher. For a extra detailed mannequin comparability, together with open-source fashions, I like to recommend leaderboards like Chatbot Enviornment. This submit is just not meant to be complete, as it could require a separate submit on every modality to do it justice. It’s relatively a sampling on how far, and how briskly this house is shifting, with an vital caveat on enterprise adoption in direction of the tip of the submit.
Textual content technology
Writing
Foundational fashions proceed to dominate this house. Whereas ChatGPT-4 maintains its place as a number one textual content generator, newer fashions like GPT-4 Turbo (known as the ‘strawberry’ mannequin) excel in reasoning duties however could not surpass their predecessors in pure writing capacity. In my expertise, fashions like Grok and Google’s Gemini have a tendency to provide extra robotic-sounding textual content. For essentially the most human-sounding, pure writing, Anthropic’s Claude has constantly impressed me.
Limitations: Mannequin hallucinations stay a major problem in AI textual content technology. As an illustration, when trying to find a ‘quote of the week’ for my e-newsletter and requesting quotes from well-known rappers, the mannequin typically repurposed earlier quotes by merely altering the attributed names to rappers. One other limitation is language variety. Whereas English stays the dominant language for AI textual content technology, new fashions are step by step rising which are skilled on different languages, akin to Hebrew.
AI Analysis
Perplexity deserves particular point out as an AI-enhanced search device. In contrast to conventional search engines like google that present hyperlinks to related data, Perplexity generates direct solutions to queries and cites its sources. Nonetheless, Google is just not far behind, having begun to combine AI-generated solutions (powered by Gemini) into its search outcomes.
Initiatives like NotebookLM (at present in beta) provide highly effective, in-depth analysis capabilities for text-based data. Customers can ask questions, create summaries, generate sensible podcasts for studying functions, and add notes on the fly to develop accessible data.
Rising startups like Unriddle AI (half of the present Y Combinator batch) are additionally making strides on this house, gaining early traction with their progressive approaches to AI-powered analysis.
It seems that AI could be extra artistic than people when arising with analysis concepts!
AI Video
The video area represents maybe essentially the most thrilling frontier in AI content material creation. As the next timeline by A16Z exhibits, the progress in AI video over the previous yr has been important
To present you a way on how far generative AI video has come, check out the video under and a few of these examples.
Textual content to video footage
OpenAI’s Sora was the primary head turner on this class. Kling AI is one other text-to-video AI mannequin developed by Kuaishou, a distinguished Chinese language know-how firm recognized for its short-video platform just like TikTok. Alibaba additionally just lately launched a brand new text-to-video AI mannequin as a part of its broader initiative in AI know-how. To date, the output tends to be quick clips (under 10 seconds) and appears extra like inventory footage.
One other instance from the most recent mannequin, Kling v1.5, which added sensible human emotion (from immediate)
AI video startups Runway and Luma just lately launched APIs, signalling a serious step ahead in accessibility for video technology instruments. In the meantime, Google has built-in its flagship video mannequin, Veo, into YouTube so you’ll be able to create + submit 6-second clips straight on the platform.
Quickly, ordering a Pizza can be a completely automated multimodal AI expertise:
Textual content-to-Video Avatars
We’re used to receiving most data from different folks, and particularly, faces. That’s why hyper-realistic artificial video based mostly on human avatars has turn into so common as a textual content alternative – for training, communication, promoting, gross sales and extra. These digital characters can converse any language and movies could be tailor-made for his or her viewers.
Hour One, a pioneer in AI-powered video creation, has made important strides in producing sensible human presenters for varied functions (disclosure: Remagine Ventures portfolio firm). Their know-how permits companies to create professional-looking movies with digital hosts, dramatically lowering the time and value related to conventional video manufacturing. To see how far AI generated avatars have come, watch how Reid Hoffman created an AI clone of himself within the type of Reid AI, utilizing Hour One.
Different notable examples within the AI video house embrace:
- Heygen/ Synthesia: Enable customers to create AI-powered movies with digital presenters talking in a number of languages.
- D-ID: Specialises in creating speaking head movies from nonetheless photos, enabling the animation of historic figures or the creation of personalised video messages.
- Fliki: Combines text-to-speech and AI-generated visuals to create participating video content material from written scripts.
And the listing goes on. For extra exploration, take a look at this Github repo with an inventory of text-to-video firms.
These developments in AI-generated video are usually not solely democratising video creation but in addition opening up new potentialities for personalised content material at scale. Nonetheless, moral concerns surrounding deepfakes and the potential for misinformation stay vital challenges that the trade should handle.
Quick type movies and video modifying
Munch focuses on mechanically creating short-form video content material from longer movies (disclosure: Remagine Ventures portfolio firm). Their AI-powered device analyzes long-form content material to determine essentially the most participating moments, then generates quick clips optimized for social media platforms. This know-how is especially worthwhile for content material creators and entrepreneurs trying to repurpose current video content material for platforms like TikTok, Instagram Reels, and YouTube Shorts.
Descript allows customers to mechanically transcribe the video and edit the video by modifying the textual content.
Picture technology
Within the realm of AI-generated photos, Midjourney continues to steer the pack with its spectacular high quality and flexibility. Customers can now skip Discord to generate photos, and the standard of the output nonetheless very a lot relies on the standard of the immediate. Nonetheless, competitors on this house has intensified considerably.
Secure Diffusion, an open-source picture technology mannequin commercialised by Stability AI, has gained substantial traction on account of its flexibility and the power for builders to fine-tune it for particular use instances. Stability has very highly effective modifying capabilities and is utilized by builders by way of APIs.
Adobe has additionally entered the fray with Firefly, integrating AI picture technology capabilities straight into its suite of artistic instruments. The photographs are copyrighted, making them protected for industrial use.
I’m a fan of Ideogram, which initially obtained recognition for its capacity to generate textual content in photos.
DALL-E 3, developed by OpenAI, has made important strides in producing photos that extra precisely mirror detailed textual content prompts. Google’s Imagen and Meta’s Make-A-Scene are additionally pushing the boundaries of what’s doable in AI picture technology.
AI Animation
The realm of AI-powered animation has seen exceptional developments, with a number of firms pushing the boundaries of what’s doable in automated animation creation. A16Z mentioned that the subsequent Pixar could possibly be a generative AI firm… I’ve seen generative AI animation firms which have scaled Youtube Channels to over 100K subscribers with AI generated content material. And far of multi billion greenback animation firms like Moonbug’s Cocomelon is data-driven, AI-assisted manufacturing.
- Cascadeur: This AI-assisted animation software program makes use of physics-based algorithms to create sensible character actions. It just lately raised $7.6 million in Sequence A funding led by IOLA Enterprise Capital in 2023.
- Kinetix: A no-code AI-powered 3D animation platform that enables customers to create animations from video. They raised $11 million in Sequence A funding in 2022, led by Adam Ghobarah, founding father of High Harvest Capital.
- Rokoko: Specializing in movement seize know-how, Rokoko’s AI-enhanced instruments make high-quality animation accessible to indie creators. They secured $3 million in seed funding in 2021 from The Danish Development Fund and Vækstfonden.
- Marvel Dynamics: Based by actor Tye Sheridan and VFX skilled Nikola Todorovic, this startup makes use of AI to automate CGI and 3D animation for movie manufacturing. They raised $9 million in Sequence A funding in 2022, backed by Horizons Ventures, Epic Video games, and Samsung Subsequent Ventures.
- Cartoon Animator (previously CrazyTalk Animator): Whereas not a latest startup, this software program by Reallusion has built-in AI to reinforce its 2D animation capabilities, making it simpler for creators to provide high-quality animations shortly.
These developments in AI-assisted animation are democratising the animation course of, permitting creators with restricted technical experience to provide professional-quality animations. This pattern is especially impactful within the fields of leisure, training, and advertising and marketing, the place animated content material can considerably increase engagement and understanding.
3D Content material and AI gaming
The creation of 3D content material is one other space the place AI is making important strides, with functions starting from sport improvement to architectural visualisation, digital twins and digital actuality experiences.
- Playo.ai has introduced the launch of the world’s first basis mannequin particularly designed for sport improvement. Playo’s know-how simplifies the complicated strategy of sport improvement by enabling customers to generate total video games from a single immediate.
- Promethean AI: This AI-powered device assists within the speedy creation of 3D environments for video games and digital worlds. They raised $6 million in a seed spherical in 2021, led by Andreessen Horowitz.
- Situation: Specializing in AI-generated 3D property for sport improvement, Situation raised $6 million in seed funding in 2022, with Play Ventures and Anorak Ventures among the many buyers.
- Luma AI: Whereas talked about earlier for video, Luma AI additionally excels in creating 3D fashions from 2D photos. They raised $20 million in Sequence A funding in 2023, led by Andreessen Horowitz.
- Hypothetic: This startup makes use of AI to generate 3D sport property from textual content descriptions. They raised $3.6 million in seed funding in 2023, with backers together with South Park Commons and NEA.
- InstaLOD: Providing AI-powered 3D optimization and automation options, InstaLOD raised €3.4 million in Sequence A funding in 2022 from Excessive-Tech Gründerfonds and Capnamic Ventures.
- Inworld AI: a platform for creating AI-powered digital characters and interactive experiences for video games, metaverse functions, and customer support.
The impression of AI on 3D content material creation is transformative, considerably lowering the time and experience required to provide complicated 3D fashions and environments. This democratization of 3D content material creation is opening new potentialities in fields akin to:
- Sport Growth: Quicker creation of detailed sport worlds and property.
- Structure and Actual Property: Fast technology of 3D fashions for buildings and interiors.
- E-commerce: Straightforward manufacturing of 3D product fashions for enhanced on-line procuring experiences.
- Digital and Augmented Actuality: Fast improvement of immersive environments and objects.
As AI continues to evolve on this house, we are able to count on to see much more subtle instruments that blur the road between human-created and AI-generated 3D content material. This development will seemingly result in extra immersive and detailed digital experiences throughout varied industries.
A number of massive caveats/ considerations for the way forward for generative AI startups
The way forward for these wonderful artistic tech startups continues to be unsure and I might be remiss to not point out the constraints plaguing these firms. As Deloitte’s Q3 2024 State of GenAI in Enterprise Report reveals, enterprise curiosity in generative AI stays excessive, however precise adoption ranges are nonetheless low. For instance, 75% of respondents have elevated investments in knowledge life cycle administration, a crucial consider enabling large-scale deployments, however the majority of organisations (70%) have moved 30% or fewer of their GenAI experiments into manufacturing.
Causes for restricted Enterprise AI adoption
1. Copyright/ Coaching knowledge – there are a number of lawsuits happening in opposition to firms which are suspected (or confirmed) to make use of scraped coaching knowledge from the Web/ Youtube/ Publishers with out permission. Massive firms (like OpenAI) have deep pockets and may deal with the warmth, however for startups, that may be deadly. To date, trade gamers have adopted two main methods: both negotiating compensation from Massive Language Fashions (LLMs) and foundational fashions for his or her knowledge, or pursuing authorized motion once they can show their content material has been used for coaching functions with out permission. Notable examples embrace Getty Pictures vs Stability AI, in addition to file labels’ lawsuits in opposition to AI music turbines like Suno and Udio.
2. It’s safer to attend and see. As Deloitte and others report, enterprise purchasers are very all in favour of AI, however aren’t in a rush to deploy options. Partly on account of compliance, safety and privateness considerations, and partly because of the situation with copyright talked about above. Till that modifications, a lot of the adoption is coming from SMBs or Prosumers, which might in mixture generate a big enterprise if you happen to’re market chief, however it’s harder when there’s quite a lot of competitors.
3. The massive tech firms are leaning in on generative AI, laborious. Up to now, startups benefitted from the FAANG firms being gradual to undertake new developments, however within the case of generative AI, it’s totally different. Google, Microsoft, Amazon, Nvidia, Meta and others are pouring billions of {dollars} into GenAI – from foundational fashions, to tooling and infrastructure. In addition they see the potential in enterprise adoption and can compete face to face with startups.
4. It’s nonetheless very costly to coach a brand new mannequin. It requires coaching knowledge (both legally obtained or scraped), costly GPUs, and costly expertise (knowledge scientists, ML engineers). That offers an enormous benefit to the big gamers, who have already got many of those assets.
5. Commoditisation and platform threat is actual. The generative AI house is shifting so quick, {that a} new know-how, say textual content to animation, that appears novel immediately, would possibly turn into commoditised tomorrow by one of many giant tech firms or Generative AI scale ups. This makes it very troublesome for buyers to allocate into the house as they like to attend till the mud settles. Each main announcement by OpenAI, Google, Meta and so forth sends aftershocks to startups working in the identical house. Startups can solely discover relative security in niches, the place there’s much less threat that the incumbents will enter.
Conclusion
The speedy developments in AI-powered artistic automation are reshaping the content material creation panorama throughout textual content, video, and pictures. Whereas these applied sciences provide unprecedented alternatives for effectivity and scalability, in addition they elevate vital questions on the way forward for human creativity, copyright, and the authenticity of digital content material.
Competitors and go to market stay a giant situation for Generative AI startups within the artistic house. Within the utility layer, i.e. startups who’ve constructed wrappers round API from different foundational fashions like OpenAI, Anthropic, Stability AI, and so forth, are susceptible to be disrupted by the LLMs themselves, until they’re able to discover a area of interest with relative security. As fashions turn into extra multimodal, it’s safer to count on extra consolidation within the variety of instruments.
For creators, it’s like consuming from a flowery buffet proper now. As startups compete on reaching prosumers who’re prepared to pay a number of {dollars} for entry, amateurs and gifted entry stage creatives are in a position to create content material that might beforehand price far more and require a crew of execs to provide. This could contribute to the flourishing of the creator financial system, if they’re able to monetise their work both by way of the content material platforms (youtube, instagram, tiktok, linkined, X, and so forth) or straight from their viewers.
The approaching years will seemingly see much more integration of AI instruments into current workflows, additional blurring the strains between human-generated and AI-assisted content material. For buyers, content material creators, and know-how fanatics alike, staying knowledgeable about these developments can be key to understanding and shaping the way forward for artistic industries.
I am a former normal associate at google ventures, head of Google for Entrepreneurs in Europe and founding head of Campus London, Google’s first bodily hub for startups.
I am additionally the founding father of Techbikers, a non-profit bringing collectively the startup ecosystem on biking challenges in help of Room to Learn. Since inception in 2012 we have constructed 11 faculties and 50 libraries within the creating world.
