[Game-Time special article, no reproduction without authorization]
Game-Time report/After ChatGPT became a global hit at the beginning of last year, a large number of star startups attracted widespread attention from technology tycoons, capital circles and gaming circles. Among them, the most noteworthy ones for the game and designer groups are undoubtedly Midjourney and Stable Diffusion,Stable Diffusion has quickly become “the first AIGC tool widely used by gamers around the world” because of its powerful AI drawing function, open source, deployable to local installation, support for plug-ins and secondary development.
Image source: Pixabay
However, after AIGC became popular, the star startups of the year began to have different endings. For example, Stable Diffusion, which is used most by small and medium-sized teams in the industry, recently reported that the CEO ran away.
On March 22, Stability AI issued an announcement stating that “earlier today, Emad Mostaque resigned as CEO of Stability AI and resigned from the company’s board of directors to pursue decentralized AI.” The company did not immediately announce a CEO Instead, the COO and CTO were appointed as interim co-CEOs.
As for the reason for his resignation, Emad explained inexplicably, saying that he could not defeat centralized AI with more centralized AI. “The concentration of power in AI is not good for all of us. I decided to resign and solve this problem at Stability and other places.”
Regarding the AI centralization problem Mostaque mentioned, Chinmay Hegde, associate professor at New York University’s Tandon School of Engineering, said that each company has its own model, and it is difficult to call this situation centralization. However, what is interesting is that he has absolute controlling and voting rights in the company, but he decided to leave when the company encountered a bottleneck in its development, which is very puzzling.
Most of the core team has resigned and the CEO has fled: Stable Diffusion may have no money to burn
Shortly before Mostaque resigned, Stability AI also lost more than half of its key talents. Foreign media said that among the five-person core team that developed the Stable Diffusion model, research team leader Robin Rombach, core research member Andreas Blattmann, who is responsible for the Vincentian graph project, and key Researcher Dominik Lorenz has left. With the “CEO running away”, the future of this once popular AIGC company has added a lot of uncertainty.
Objectively speaking, as a generative AI company, the most direct reason for the drastic changes in Stability AI may be that it encountered problems with capital flow. Although Lightspeed Venture Partners and Coatue Management invested US$100 million and reached a valuation of US$1 billion, the company has not been able to successfully raise funds again for a long time. Although there were rumors that the company had tried to refinance at a valuation of US$4 billion, no follow-up negotiations were reached.
As early as April last year, Bloomberg revealed that the company was facing serious internal problems. Sources said that Stability AI was burning at least US$8 million a month, but it did not have a clear profit direction. Coupled with other expenses such as hardware, it meant that the company The company may not have much financing left.
However, Mostaque did not seem to prioritize revenue growth a year ago. He even expressed ridicule on his social media for the “strange focus on revenue” of generative AI companies, even saying that “the technology is useful, but because of the daily There are new breakthroughs, but they are far from mature.”
Mostaque then cited several examples, including Magic Leap, which had spent billions of dollars before bringing in revenue. In addition, at least more than US$100 billion has been invested in self-driving car research, but there is also no revenue return. However, under continued pressure from investors, Mostaque also changed its attitude. Stability AI subsequently launched a paid product and launched the V3 version in February this year.
In addition to the funding issues, the company also faced legal issues and personal controversy in August regarding Mostaque, who was accused of an “exaggerated history” that included falsifying academic qualifications and spying for Britain.
In addition, the Stable Diffusion model itself comes from Latent Diffusion created by New York AI startups and German university researchers. There is an unclear connection between the two, which leads to doubts about the intellectual property rights of Stable Diffusion itself. This may also be one of the important reasons for the company’s financing difficulties.
Stability is a pity, it is difficult for AI startups to survive without making profits
From a macro perspective, European companies have always had a natural weakness in the technology industry. Take the United States as an example. Giants such as Microsoft, Apple, and Meta have strong financial resources to continue investing in research and development, and have huge enterprise and user markets. Therefore, the two most important parties in the AI war are China and the United States.
On the other hand, European companies, whether it is the Internet, mobile networking, or today’s AI, have always been passive at the capital level. If we go a step further, European companies have also encountered the same disadvantages in the electric vehicle and smartphone markets. The European capital market cannot support Stable Diffusion’s competition with American giants.
In addition to capital, competition for talent from big companies also makes startups unable to resist. In the same week as Stability AI’s “accident”, Inflection AI, another startup that raised $1.5 billion, announced that two of its founders and several other employees had joined Microsoft, the company’s new round of financing. Lead investor.
Thanks to the advantages of open source, localized deployment, and a large number of community developer tools and content support, Stable Diffusion has a higher penetration rate among small and medium-sized teams, and can accelerate the production of art assets for small and medium-sized teams. Therefore, it is undoubtedly regrettable that its developers are in the current predicament.
However, compared with Midjourney, the open source Stable Diffusion itself faces challenges in commercialization, not to mention that its intellectual property rights themselves are not clear. Although Stability AI also launched a fee-based business last year, and its revenue in 2023 is reported to reach US$44.2 million, it is still difficult to maintain the high cost of AI training.
After the high-level exodus, the future of Stability AI is worrying. In fact, foreign media said that the company had sought to sell the company, but could not reach an agreement due to price issues. In addition to the ownership of the model’s property rights, the photography agency Getty Images also launched an infringement lawsuit against the company, claiming that SD Model stole more than 1,200 of its Thousands of copyrighted photos.
Although regrettable, it does not mean that developers will have fewer AI tools at their disposal. For example, during this year’s GDC, Tencent announced the “all-round” AI engine GiiNEX, which can cover the entire cycle of game development. It is foreseeable that with the entry of giants, more and better commercial models will gradually enter the attention of developers.
