OpenAI's Sora launched in late 2025 with more attention than any AI video tool before it, and it was gone from the App Store by April 2026. That timeline alone answers a question a lot of people have been typing into Google: is AI video generation good enough yet? The honest answer is that it depends entirely on which tool you mean, because the one everyone talked about is not the one actually winning.
What actually happened to Sora
Sora's app peaked at roughly 3.3 million downloads in its first month, according to TechCrunch's reporting on the shutdown. By February 2026, that number had fallen to about 1.1 million. Over its six-month life, the app generated around $2.1 million in in-app purchases, a small fraction of what it likely cost OpenAI to run given how computationally expensive video generation is. OpenAI notified developers of the shutdown on March 24, 2026, closed the consumer app on April 26, and is winding down the Sora 2 API entirely by September 24, 2026.
The company never gave a detailed public explanation, but the pattern is familiar: a viral launch, a steep drop-off in daily use, and a cost structure that could not be justified by the engagement. OpenAI has been shifting its resources toward coding tools and enterprise products, and a video app that could not sustain its own downloads was not going to survive that shift.
Is AI video generation good enough yet, or was Sora just the wrong product?
Sora's downfall was not really about video quality. It was about what people did with it. The app's "cameos" feature let users insert real people's likenesses into AI-generated scenes, and it was quickly used to create disrespectful deepfakes of public figures, including footage involving Martin Luther King Jr. and Robin Williams, prompting their families to request removal. Users also generated video using copyrighted characters like Mario and Pikachu without permission. Disney had reportedly invested close to a billion dollars and negotiated a licensing arrangement with OpenAI, but that deal fell apart when the app shut down.
None of that answers whether the underlying technology is good enough. It mostly shows that a viral novelty app built around putting real people's faces into AI video was always going to run into trouble, separate from whatever the model could actually do.
The tools that are quietly winning instead
While Sora dominated headlines, two other players were building real, boring, revenue-generating businesses. Kuaishou's Kling AI passed 100 million users across 224 countries and regions, and its revenue exceeded 850 million yuan in the second quarter of 2026 alone, up more than 200 percent year over year, according to Kuaishou's own reporting. First-half 2026 Kling revenue topped 1.5 billion yuan.
Google took a different route. Instead of a standalone viral app, it folded Veo directly into Google Ads, letting advertisers turn a handful of product images into ready-to-run YouTube video ads. Google reported that advertisers generated nearly 70 million Gemini-powered creative assets in the fourth quarter alone, a threefold increase from the prior year. Apparel brand Rhone is one of the named advertisers using Veo and Google's Nano Banana image tool this way, turning existing product photos into video ads instead of commissioning new shoots.
Neither Kling nor Veo needed a viral consumer app to succeed. They got embedded into workflows people were already running: social content creation and paid advertising.
Where the technology genuinely still falls short
Setting aside the business drama, the technical limitations are real and consistent across every major model in 2026, including Veo 3.1, Kling 3.0, and Runway. Close-up shots of people talking or emoting still tend to look artificial. Text on signs, labels, and documents inside a generated video is frequently garbled. Physics involving water, cloth, and complex object interactions often looks slightly wrong even when the overall shot is impressive. Most usable clips still run 5 to 10 seconds, which is why almost everyone using these tools professionally works in short cuts rather than long continuous takes.
That is the practical version of "good enough": these tools are reliable for b-roll, social ads, product demos, and short marketing clips, but nobody is using them yet for a scene that requires a convincing close-up performance or a long unbroken shot.
What this means if you are actually trying to use AI video
- For short marketing or social clips, Kling and Veo are the tools with the track record and the pricing to back it up, with Kling charging around $0.10 per second of generated video.
- For anything requiring a recognizable real person's likeness, treat it as a legal and ethical risk, not a shortcut, regardless of which tool you use.
- For long-form or dialogue-heavy work, AI video is still a supplement to traditional production, not a replacement for it.
The takeaway
The company that generated the most headlines about AI video was not the one that built a sustainable product. Sora proved that novelty and controversy can move fast without moving anywhere useful, while Kling and Veo grew by quietly attaching themselves to work people were already paying to do. If you are wondering whether AI video generation is good enough yet, the more useful question is good enough for what. For short, disposable marketing content, the answer is already yes. For anything that needs a human face doing something specific and convincing, or a story longer than ten seconds, the honest answer is still not quite.
