Depending on who you ask, generative AI is either massively overhyped or undervalued. Defined as algorithm-driven tech that creates text, art and other forms of media given a prompt, it's captured the attention of major VC backers who've piled hundreds of millions of dollars into firms like Jasper and Stability AI. But generative AI has yet to generate (no pun intended) correspondingly high returns, casting doubt on its near-term profit-making potential.
Emmanuel de Maistre and Hervé Nivon think the problem is the application of the tech rather than the tech itself. While startups such as Stability AI aim to tackle a broad number of use cases with their generative AI, de Maistre and Nivon advocate for a narrower, slightly more focused approach. Their startup -- called Scenario -- lets artists and game developers create their own image generators trained on the specific style of their games.
Scenario launches today, accessible via the web, mobile app or API.
"Using Scenario, game developers -- regardless of the level of technical expertise -- can create dozens or hundreds of custom generators capable of producing entirely new game assets that are perfectly style-consistent with a given style or art direction," de Maistre told TechCrunch in an email interview. "Our solution is the only one available that allows them to train their own AI generator based on a specific art style using first-party training data. So if you’re an independent artist or developer, you can start with a handful of assets in a given style, upload them to Scenario and create a generator specific to those assets."
De Maistre and Nivon co-founded Scenario in 2021 after spending several years in the 3D modeling and data science industries, respectively. Nivon previously was a solutions architect at Amazon Web Services (AWS) working on AI products, while de Maistre sold his other startup, drone analytics firm Redbird, to the since-shuttered Airware. (Nivon was Redbird's CTO.) Prior to AWS, Nivon was at Accenture, leading "innovation transformation" for the company's France division.
AI concepting is amazing! Exploring character possibilities is awesome. These characters were created on https://t.co/gqUKpmzCIU and edited in Photoshop. I'm still working to improve the faces on @Scenario_gg and already loving the results.#GenAI #StableDiffusion #XR #AIart #AI pic.twitter.com/7AdyYV5cL6
— Rodrigo (@rodrigon) January 5, 2023
De Maistre says that he and Nivon were inspired to launch Scenario by generative AI products like OpenAI's DALL-E 2. The raw power of these tools was "obvious," de Maistre believed, but the output was too inconsistent to be useful.
"I knew if we could better direct that power, give users more control and consistency, that would instantly be generative AI's killer application," de Maistre said. "The gaming industry is the best fit for generative AI -- game developers and artists have to continually produce content while time and resources are often limited. That’s why we started Scenario last year. We wanted to provide a solution that lets anyone train their own AI models -- generators -- using their own data so they can generate game assets faster and more efficiently, while keeping consistency and full control over the process."
The game industry indeed presents an opportunity for disruption where it concerns generative AI. Gaming requires a high volume of content -- much of it artwork. Estimates are hard to come by, but one source pegs the cost of creating art assets for a small-scale game at a few dollars to thousands of dollars.
With Scenario, users can upload a set of visuals that define the characters, items, environments or other assets for a given video game or project. Scenario's AI engine then learns and adapts to the visuals' graphic style, generating new assets for games, game prototypes, game marketing materials and more from simple text-based prompts.
In letting developers and artists train their own generators, Scenario hopes to sidestep the major legal challenges emerging around generative AI. Just this week, Getty Images sued Stability AI, the creators of AI art tool Stable Diffusion, for scraping its content allegedly without permission and using it to train art-generating AI systems. Meanwhile, the U.S. Patent and Trademark Office (USPTO) recently moved to revoke copyright protection for an AI-generated comic, saying copyrightable works require clear human authorship.
— Robtheidiot (@Robtheidiot1) January 15, 2023
De Maistre notes that Scenario's terms and conditions require those on the company's platform to only use data that they own -- for example, data they've purchased or have been granted the right to use -- or open source alternatives. Scenario also doesn't claim ownership over customers' generators or images created on the platform, leaving most trademark -- and objectionable content -- decisions in users' hands.
"We advise customers to work with intellectual property (IP) professionals as appropriate to ensure IP and compliance risks are mitigated, especially for commercial projects. We are a design tool and it is the user’s responsibility to ensure compliance with applicable laws and regulations," de Maistre said.
That Scenario's attempting to wash its hands of legal liability won't instill confidence in every customer. But de Maistre claims that 5,000 people have signed up for the platform and that 20,000 more are on a waitlist. Pricing will be usage-based, starting at $20 per month with plans for higher-volume customers to follow.
"Currently, our closest competitors would be generative AI art tools such as Midjourney, DALL-E 2 and Stable Diffusion," de Maistre said. "But as sophisticated as these images are, they are still evolving to fit the controlled use cases required for the gaming industry, and many users still struggle with keeping a high consistency of the outputs ... Our platform has been used to create assets for various types of games, [including] mobile, cards, tabletop role playing, VR and even 3D games."
Suggesting investors are pleased with the early momentum, Scenario recently raised $6 million in seed funding from Play Ventures (who led the round), Anorak Ventures, Founders, Inc., The VR Fund, Oculus co-founder Brendan Iribe, Twitch founder Justin Kan and Hugging Face founders Clem Delangue and Julien Chaumont. That's high praise considering Scenario is but one of several startups in AI-generating game asset space; rivals include Poly, Hotpot and Pixela.ai.
Scenario -- which has a team of eight people -- plans to put the new capital toward bringing on more full-stack engineers, data scientists and product designers, as well as a customer support team. De Maistre believes it's the fastest way to differentiation, and -- with any luck -- setting Scenario well ahead of the generative AI pack.
"We believe that generative AI will be as transformational for game development as Photoshop has been for digital photography, but it cannot get there without the same commitment to consistency and ease of use," de Maistre added in a follow-up email. "We want to open the opportunity this technology brings to the gaming industry: exponentially increased production, dramatically reduced busywork and completely unconstrained creativity from AI-partnered artists."