Can ChatGPT Make Videos? (2026 Guide)

Type "make me a video about our Q3 results" into ChatGPT and you will get something back. It will be a script, or a shot list, or a storyboard described in words. What you will not get is a video file.

That gap is the whole question, and the answer has moved twice in 2026. OpenAI's video model, Sora, has been through a standalone app, an integration into a paid ChatGPT surface, and a developer API — and two of those three have been shut down this year. A lot of the advice still circulating was written when all three were live.

So here is the current picture, built from OpenAI's own documentation rather than from the aggregator posts that keep recycling last year's answer.

The short version, if you are here for one line: no, not in the chat, and OpenAI's video capability is smaller today than it was a year ago.

Can ChatGPT make videos?

Not in the chat window, no. ChatGPT generates text and images in a conversation. It does not render video there, and asking it to will produce a description of a video rather than a video.

OpenAI's video generation has always lived on a separate surface powered by the Sora model, not inside the chat thread. That distinction matters more than it sounds, because it means "can ChatGPT make videos" and "does OpenAI have a video model you can use" are two different questions with two different answers.

The second question is where 2026 gets complicated. OpenAI retired the standalone Sora app this year, and on 24 September 2026 it shut down the Sora models in its developer API entirely. OpenAI's own model documentation now states it plainly: "The Sora 2 models and Videos API were shut down on September 24, 2026 and are no longer available." The page marks the models Deprecated and Legacy, and OpenAI notes there is no one-to-one replacement API.

That is an unusually clean fact in a space full of rumour, and it is the single most load-bearing thing in this article: if you are a developer, the endpoint you are looking for does not exist any more.

Your content plan deserves better than guesswork. Get a free SEO audit and see which pages are actually earning attention.

What remains is the consumer surface. OpenAI's help centre still documents video generation for paid ChatGPT plans, and the Sora web address still resolves to a page describing text and image to video generation. But with the app retired and the API gone within six months of each other, this is a capability in retreat rather than one to build a workflow on. If you are planning around it, check the current state yourself before committing.

What actually happened to Sora in 2026

The timeline is worth knowing, because it explains why answers to this question disagree so violently depending on when they were written.

The app

Sora launched as a separate consumer app, and for a while it behaved like a social network — a scrolling feed of generated clips, closer to TikTok than to a production tool. That version did not last.

On 24 March 2026, OpenAI announced it was winding the app down. Its statement was short: "We're saying goodbye to the Sora app. To everyone who created with Sora, shared it, and built community around it: thank you." No reason was given at the time. Reporting put the operating cost at roughly a million dollars a day against a user base that had fallen from around a million to fewer than half that.

The underlying model did not die with the app. Coverage at the time of the announcement noted that the Sora 2 model remained available behind the paid ChatGPT tiers — the app was the thing being retired, not the capability.

The distinction between a model and a product is the thing to hold onto here. Retiring a consumer app says something about engagement and running costs; it does not necessarily say anything about whether the model behind it still works. In this case it kept working, on a different surface, for people on paid plans.

It is also a reminder of how fast the framing on this question changed. An article written in February 2026 would have told you to download the Sora app. By May that advice was dead, and by October the developer half of the answer was dead too.

The API

The developer story ended more definitively, and on a published schedule. OpenAI's deprecation notice records that "On March 24th, 2026, we notified developers using the Videos API and Sora 2 video generation model aliases and snapshots of their deprecation and removal from the API on September 24, 2026."

That removal covered the whole family: `sora-2`, `sora-2-pro` and the dated snapshots behind them. The recommended replacement column in OpenAI's deprecation table is empty. There is no migration path, because there is no successor product.

Both announcements landed on the same day in March, which tells you this was one decision about a product line rather than two separate operational calls.

Credit where it is due on process: six months of notice with a published date and a named list of affected model aliases is how a deprecation should be run. The problem for anyone affected is not the communication, it is that there was nothing at the other end of it.

What OpenAI still documents for paid plans

OpenAI's consumer help centre article on generating videos with Sora remains published, and it is the most authoritative statement available on who can still do this.

On eligibility it says: "Sora is fully available to ChatGPT Plus, Team, and Pro users." And on who cannot: "ChatGPT Free, Enterprise, and Edu accounts are not eligible for Sora access." That exclusion list is worth a second read, because it is the opposite of what people expect — the enterprise tier has less access here, not more.

If you are on an Enterprise or Edu plan and were expecting video, that is the line to take to whoever administers your workspace. It is a plan-level exclusion rather than an admin toggle, so no amount of settings archaeology will turn it on.

On output, the documentation describes generating "videos up to 20 seconds long while maintaining visual quality and adherence to your prompt," with controls to "change the aspect ratio, resolution, duration, and the number of variations."

Two editing features are documented. The storyboard, which "enables you to create a video by selecting the individual frames, by timestamp," and Remix, described as "describe changes and create new videos based on your generation." Those are the difference between a slot machine and a tool — being able to adjust one element without regenerating from scratch is most of what makes iterative work possible.

The aspect ratio control is the one that matters most for marketing use, because it decides whether a clip is usable vertically for short-form social or horizontally for a landing page. Generating in the wrong ratio and cropping afterwards loses the composition the model built.

Stop publishing into the dark. Request your free audit and find out what your site is ranking for right now.

A caveat on all of the above: help centre articles outlive the products they describe, and OpenAI has retired two Sora surfaces this year without updating every page that mentions them. Treat these quotes as what OpenAI documents, and verify availability in your own account before you plan a deliverable around it.

If you were building on the Videos API

This is the group with a real problem, and the guidance is unusually blunt because OpenAI's own is.

There is no drop-in replacement. The deprecation table lists none, and OpenAI's model page states there is no one-to-one replacement API. Any code calling the Videos endpoint has been returning errors since 24 September 2026.

If that describes your application, the failure mode is worth checking rather than assuming. A video call that fails inside a queue worker can go unnoticed for weeks while the user-facing symptom is simply that nothing ever finishes.

Practical steps, in order:

1. Audit your codebase for calls to the Videos API and for any of the retired model aliases, including dated snapshots that may be pinned in config rather than in code.
2. Check whether video was load-bearing or decorative in your product. A generated B-roll clip in an onboarding flow is a different problem from a video generation feature customers pay for.
3. If it was load-bearing, move to a vendor whose video generation is its core business rather than a side product — that is the structural lesson here.
4. Build the integration behind an abstraction this time, so the next deprecation is a config change rather than a rewrite.

The broader read for anyone planning AI features: OpenAI shipped a video app, a video API and a paid-tier video surface, then retired two of the three inside six months with no replacement. That is a reasonable thing for a company to do and a terrible thing to have bet a roadmap on. Depth of commitment matters more than capability when you are choosing what to build against.

The same test applies to any AI feature you are considering buying or building on: is this the vendor's main business, or is it an experiment they can afford to close? The answer is usually visible in how they talk about it.

What to use instead

If what you need is video, the honest answer is that OpenAI is not currently the obvious place to get it, and the market has several vendors whose entire business is video generation. Those are the ones with an incentive to keep the lights on.

Rather than name a winner that will be stale in a quarter, evaluate on the things that do not change: whether the vendor's video product is its main product, whether it publishes a deprecation policy, whether the API has versioned snapshots you can pin, and whether output licensing is clear enough for commercial use. The Sora episode is a reminder that the last two matter as much as output quality.

One more criterion that people skip: whether you can export the raw output rather than only sharing from the vendor's platform. A clip you cannot download is a clip you lose when the product closes, which is precisely the scenario this article is about.

For many business cases, though, generated video is the wrong tool. A screen recording with a voiceover converts better than a synthetic clip for software demos. A charted animation beats a generated scene for data storytelling. Generated video is strongest for mood and texture — backgrounds, transitions, atmosphere — and weakest for anything that has to be precisely, factually right.

That maps neatly onto a practical rule. If the video's job is to make a point, record or animate it. If its job is to set a mood behind a point made elsewhere, generation is a reasonable shortcut.

Your traffic should not depend on which tool is fashionable. Get a free SEO audit and build on what actually works.

Limitations worth knowing

Even where video generation is available to you, the constraints are real.

Length. Twenty seconds, per OpenAI's documentation, is a shot and not a sequence. Anything longer means generating pieces and editing them together, which reintroduces the editing workflow the tool was supposed to replace.

Consistency across clips. Characters, lighting and style drift between generations. This is why the storyboard and Remix features matter — regenerating from a fresh prompt rarely lands in the same visual world twice.

Text in frame. Generated video handles lettering badly, which rules out anything where a logo, a price or a label has to be legible and correct.

Factual content. A generated clip has no relationship to truth. It is suitable for illustration, not for depicting anything that happened.

Rights and disclosure. Commercial use terms vary by vendor and tier, and expectations about labelling synthetic media are tightening. Check both before anything generated goes into paid distribution.

Review time. Generated clips need watching end to end before they ship, because the failure modes are visual and arrive without warning: a hand with the wrong number of fingers, a reflection that does not match, a sign with nonsense on it. Budget the review, not just the generation.

Video versus images in ChatGPT

It is worth separating these, because the two capabilities have had very different years.

Image generation inside ChatGPT is a first-class, in-conversation feature. You ask, it draws, and the result arrives in the thread. It is available across plans with usage differences, it is integrated into the chat rather than parked on a separate site, and OpenAI has not retired any part of it.

Video has been the opposite: a separate surface, a separate app, a separate API, and two retirements in six months. If your mental model is "ChatGPT does images and video," correct it to "ChatGPT does images, and OpenAI has a video model elsewhere that is being wound back."

For most marketing work, that is less limiting than it sounds. Static visuals carry blog posts, social cards and documentation perfectly well, and they are far easier to get factually right.

And if video is genuinely the format your audience wants, the bottleneck is almost never the rendering. It is having something worth saying and knowing which topics are worth saying it about — which is a research problem, not a generation problem.

Find the gaps before your competitors do. Claim your free audit and see where the opportunity actually is.

Final Thoughts

ChatGPT does not make videos in the chat window, and it never has. OpenAI's video capability lived in Sora, on its own surface — and in 2026 OpenAI retired the Sora app in March and shut down the Sora models and Videos API on 24 September, stating plainly that they "are no longer available" with no one-to-one replacement.

What is left is the paid consumer surface that OpenAI's help centre still documents for Plus, Team and Pro, with twenty-second clips, storyboard editing and Remix. That may be all you need. Just verify it in your own account first, because the documentation has outlasted two products this year already.

If you were really asking what ChatGPT can generate for you today, the answer is images, and that part works well in the thread. Our guide to whether ChatGPT can generate images covers what the image model is good at, where it still fails, and how many you get on each plan.