Type a Scene, Get a Movie: How AI Video Generation Went Mainstream, and Why OpenAI Just Walked Away From It
Type "golden retriever puppy opens a birthday present in slow motion, warm morning light" into a text box, wait about a minute, and out comes an eight-second clip that looks like it was shot on a real camera - lighting, motion blur, believable physics, no puppy required. That used to be the novelty. In 2026, it's just what a text-to-video tool does by default, and enough people are using one that it's changed what "making a video" means for someone with no camera, no editing software, and no budget.
It's also, as of this week, a market with a cautionary tale sitting right in the middle of it: on September 24, 2026, OpenAI shuts off the Sora API for good, six months after pulling the plug on the consumer app that made "AI video" a household phrase in the first place.
How it actually works
Tools like OpenAI's Sora, Google's Veo (accessed through an app called Flow), Runway, and Kling all work the same way a chat assistant does, just pointed at moving pictures instead of text. You describe a scene - subject, camera movement, lighting, mood, sometimes dialogue - and the model generates a short clip in one pass: no storyboard, no crew, no editing timeline. Some tools also take a photo or an existing clip as a starting point instead of a blank prompt, which is how most of the "put yourself in this video" features work.
The scale here is easy to undersell until you see the numbers next to each other. Across all AI video platforms, monthly active users passed 124 million in January 2026. Google's Flow alone generated over 275 million videos in roughly five months, and one 2026 usage tracker put Veo's share of all AI video generations at 96.4%, with Sora down at about 2%. That gap matters for what happened next.
What people are actually making with it
The use cases split into a few clear buckets, and most of them look nothing like a Hollywood demo reel.
Occasion videos that used to require a videographer. Wedding invitations, birthday announcements, save-the-dates, graduation montages - AI tools now generate polished, on-theme video invitations from a text prompt or a handful of photos, with pricing under $500 in most cases versus an average of $2,800 for traditional wedding videography in 2025. For a lot of families, that's the difference between having a professional-looking video at all and not.
Small businesses making their own marketing spots. A jingle, a product demo, a 15-second ad for a local business's social feed - work that used to mean hiring a videographer or buying stock footage now often starts with a generated clip. The adoption curve backs this up: 49% of marketers say they've incorporated AI video into their regular workflow, up from just 18% in 2023, and 78% of marketing teams now use AI-generated video in at least one campaign per quarter.
Creators sketching ideas before committing to a real shoot. Similar to how musicians use AI generation as a fast first draft, some creators use text-to-video to previsualize a shot - camera angle, pacing, mood - before booking a location or actor, treating the AI clip as a mood board that happens to move.
Putting yourself, or a friend, into the video. This is the feature that made Sora go viral last fall: a "cameo" system that scans your face, runs a liveness check, and then lets you (or, with your consent, people you approve) appear inside AI-generated scenes. It's also the feature that ended up sinking the product, which is the next part of this story.
Why OpenAI walked away from its own breakout hit
Sora's consumer app was a genuine hit - fast growth, wall-to-wall social media coverage, real cultural moment. It was also, per OpenAI's own account, burning around $1 million a day in compute, with the company's Sora lead reportedly saying in November 2025 that its GPUs "are melting." Video generation is dramatically more expensive per output than text or images, and usage was declining anyway - Sora peaked at roughly 1 million monthly active users and had fallen under 500,000 by the time OpenAI announced it was done.
Cost wasn't the only problem. The cameo system that made Sora fun to use was also trivial to abuse: deepfake-detection firm Reality Defender said it bypassed Sora's anti-impersonation safeguards within 24 hours of launch. Users generated AI videos of deceased public figures, including Michael Jackson and Martin Luther King Jr., doing things they never did, which drew an outcry from actors' unions and family estates and forced OpenAI into reactive content crackdowns. Advocacy groups and researchers raised the broader alarm too: a tool this good at putting a real person's face into any scene you can type is also a tool for non-consensual and defamatory content, at a scale no moderation team can fully keep up with.
OpenAI announced the shutdown on March 24, 2026, discontinued the consumer web and app on April 26, and the developer-facing API goes dark on September 24 - this month. Meanwhile Google's Veo, without a standalone social feed built around impersonating real people, kept growing and now dominates the category. The lesson isn't "AI video is dying" - usage across the category is up, not down. It's that the specific combination of cheap compute, a viral consumer app, and a face-cloning feature turned out to be a business OpenAI didn't want to keep running, even at the top of its popularity.
What to know before you post one
- Check whether the tool you're using needs a face scan or consent flow, and what happens to that data. Any tool that lets you put a real face into a generated scene should ask for that face's consent before generating - if it doesn't, that's a sign the platform hasn't taken the impersonation risk seriously.
- A "safeguard" against impersonation is not the same as an unbreakable one. Sora's own anti-impersonation system was reportedly bypassed within a day of launch. Don't assume a platform's guardrails will catch a misuse case just because the platform says it has them.
- If you're generating something with a real, identifiable person who isn't you, get their actual consent first - not just the app's checkbox. This applies double to anyone who's deceased or can't consent themselves; several platforms have had to build reactive policies here only after public backlash, which means the rules you're relying on may not exist yet for your specific case.
- Disclose AI-generated video the same way you'd disclose AI-generated music or text. If you post it anywhere public, say so. Our Ethical AI Usage lesson has language you can adapt, and the same disclosure logic we wrote about for AI-generated music and AI slop generally applies directly here.
- For anything commercial - a wedding video you're selling, an ad for a client - read the specific platform's commercial-use terms. They vary by tool and by pricing tier, and "the free tier let me export it" isn't the same as "I'm cleared to sell it."
Key takeaways
- AI video generation crossed into genuine mainstream use in 2026 - over 124 million monthly users across platforms, real cost savings on things like wedding and event videos, and adoption climbing steadily in small-business marketing.
- OpenAI's Sora is the cautionary half of this story, not the whole thing. Its consumer app is shutting down (API goes dark September 24, 2026) over compute cost and impersonation risk, even as the broader AI video category keeps growing under Google's Veo.
- The "put a real face in a generated video" feature is the highest-risk part of this technology - safeguards against impersonation have already been shown to fail, so consent and disclosure aren't optional extras, they're the whole ballgame.
- Most everyday use is low-stakes and genuinely useful - invitations, small-business marketing, and creative previsualization, not attempts to deceive anyone.
If you're building general AI literacy before going deeper on any one tool, AI 101 is the place to start. For the wider toolset beyond video generation specifically, see Audio & Video AI Applications. And before you generate or share anything involving another person's likeness, Ethical AI Usage is worth reading first, not after.
This post was developed with AI assistance and is regularly reviewed for accuracy.