AI Roundup October 2022: Google's Imagen Video, Stability AI's $101M, and Jasper Hits Unicorn

TL;DR

October 2022 was the month generative AI stopped being a Discord curiosity and started looking like an industry. Google answered Meta's Make-A-Video with two text-to-video models in the same week, Stability AI banked $101M at a billion-dollar valuation while Stable Diffusion was still warm, Jasper turned GPT-3 wrapping into a $1.5B unicorn, and Shutterstock cut a deal with OpenAI instead of suing it. The diffusion gold rush had officially gone mainstream.

Google Drops Imagen Video and Phenaki in the Same Week

Meta showed off Make-A-Video at the end of September, and Google was clearly not in the mood to be second. On October 5, Google unveiled Imagen Video, a cascade of seven diffusion models that generates 1280x768 clips at 24 frames per second from a text prompt. A day later, a separate Google team surfaced Phenaki, which trades resolution for length and coherence: feed it a long, multi-sentence prompt and it stitches together video running past two minutes.

The split personality is the interesting part. Imagen Video chases fidelity, Phenaki chases narrative duration, and neither shipped to the public. Google explicitly held both back over training-data and safety concerns, which tells you the legal and PR temperature around generative media was already rising. For builders, the takeaway was less "go run this tonight" and more "text-to-video is now a real research front, and the open-weight version is coming whether the labs like it or not."

Stability AI Raises $101M and Becomes a Billion-Dollar Company

On October 17, Stability AI announced a $101 million round led by Coatue, Lightspeed, and O'Shaughnessy Ventures, reportedly at a $1 billion post-money valuation. This is roughly two months after Stable Diffusion's open release, which is a stunning pace even by 2022 standards.

The thesis Stability sold was simple and a little subversive: open-weight models, downloaded and forked by hundreds of thousands of developers, are a distribution moat that closed APIs cannot easily replicate. If you run a local image pipeline on your own GPU, this round is part of why that ecosystem exploded. The flip side, which the next couple of years would make painfully clear, is that a research-grant vibe and a venture-backed P&L are not the same animal. But in October 2022, open models had real money behind them for the first time, and that mattered.

Jasper Turns AI Copy Into a $1.5B Unicorn

A day after Stability, on October 18, Jasper announced a $125M Series A led by Insight Partners at a $1.5 billion valuation. The company said it had crossed 70,000 customers and roughly $45M in revenue, built largely on top of OpenAI's GPT-3.

This was the raise that crystallized the central anxiety of the application layer: how durable is a business that is, at its core, a very good UI on someone else's model? Jasper's answer was speed, workflow, and a browser extension that dropped its writing into any text box on the web. The skeptics had a point that would get louder, but credit where due: Jasper proved buyers would pay real money for AI writing well before ChatGPT made it a household reflex. If you are an indie hacker shipping a thin wrapper, this is the month that both validated and put a clock on your model.

Shutterstock Partners With OpenAI Instead of Fighting It

On October 25, Shutterstock said it would integrate DALL-E 2 directly into its platform, letting users generate stock images from prompts, while OpenAI gets to keep training on Shutterstock's licensed library. Shutterstock also said it would ban AI images created elsewhere and set up a contributor fund to pay artists whose work helped train the models.

This was the rare grown-up moment in a month full of "move fast" energy. While Getty was heading toward an outright ban, Shutterstock chose licensing and revenue share. It was an early, messy attempt to answer the question that would dominate AI for years: who gets paid when a model trained on human work spits out something new?

• Generate: users could make synthetic images natively inside Shutterstock.
• Train: OpenAI got a clean, licensed data source.
• Compensate: a contributor fund aimed to route royalties back to artists.

Key Takeaways

• Text-to-video became a real race: Google matching Meta within a week signaled that video, not just stills, was the next frontier the big labs would fight over.
• Open weights got funded: Stability's $101M proved investors would bet on open distribution, fueling the local-AI ecosystem that homelabbers still live in.
• The wrapper economy went unicorn: Jasper's $1.5B valuation validated AI applications while quietly raising the question of model dependency every builder now has to answer.
• Licensing beat litigation, for one company: Shutterstock's OpenAI deal was an early template for paying creators instead of scraping and hoping.
• Capital outran caution: the funding ran hot while the unreleased Google models hinted that safety and legal questions were already lagging the tech.