There is no built-in ChatGPT video generator in late 2026. OpenAI shut down the Sora app in April 2026 and the Sora API in September, and the promised Sora integration in ChatGPT never shipped. What people call a "ChatGPT video generator" today is one of the third-party video apps you can connect inside ChatGPT: Runway for short generated footage, HeyGen and Synthesia for an AI presenter, invideo for a narrated video built from stock and generated clips, and Descript for editing a recording by chatting. ChatGPT writes the brief and the script; the app renders the video on its own servers, using your account and credits with that company. Pick the app by the kind of picture you need, not by the name in the search result.

What changed, and why the old advice is out of date

Most articles written in 2025 tell you to "use Sora in ChatGPT". That route is closed. OpenAI announced on 24 March 2026 that it was discontinuing Sora, turned off the app and web experience on 26 April, and ended the API on 24 September. The integration into the ChatGPT chat window, announced in early 2025, was never released.

So ChatGPT on its own now produces text, images, spoken replies in voice mode, and files from its code tool. It does not produce a narrated, edited video. Everything that does is a separate company plugged into the chat. Can ChatGPT make a video? goes through exactly what ChatGPT can output by itself, line by line. This page is about the apps.

The four kinds of video app inside ChatGPT

The video apps in ChatGPT's directory look interchangeable from their descriptions. They are not. Each one makes a different kind of picture, and the kind of picture decides whether the video is any use for your job.

There are smaller apps in the directory as well, and the list changes month to month. Use the table as a sorting tool: when you open a new one, ask which row it belongs in.

Generated footage: Runway

Runway's app lets you ask for a clip in the chat and get generated video back, and edit or restyle clips you already have. The output can be striking. It is also footage, which means it is a moving picture of something, with no guarantee that any text, chart or label in it is correct. Generated footage invents details. Ask for "a diagram of the water cycle" and you will get something water-coloured and cyclical with labels that may be nonsense.

Use it when the shot is atmosphere: a city at dawn behind a narrator, a slow push into a server room. Do not use it as the explanation itself.

AI presenter: HeyGen and Synthesia

Both connect to your account with that company and turn a script into a video of an avatar speaking it, often with captions and simple backgrounds. This is the closest thing to "type a script, get a finished video" inside ChatGPT, and for a short message from a person (a welcome from the head of the department, a product update) it works.

The weakness is structural. The picture is a person talking, so anything the viewer needs to see, the steps of a process, the parts of a system, a before and after, has to be added separately or does not appear at all. If your topic is an explanation rather than a message, a Synthesia alternative without an AI avatar covers what to use instead.

Brief to stock edit: invideo

You mention invideo in the conversation, describe what you want, and it returns a draft video with a script, scenes, a voiceover, music and visuals pulled from stock libraries or generated. You then refine it in invideo itself. For a 60-second promo or a "five facts about" short, this is fast.

The limit is that the visuals are chosen to match the topic of each sentence, not to show its claim. A sentence about interest rates gets a shot of coins and a stock-market screen. That is fine for mood and poor for teaching, because the viewer sees money, not how a rate change moves a bond price. The best AI video generator for educational content goes into this difference in more detail.

Edit by chatting: Descript

Descript's app is mostly an editor you can direct in plain language: remove filler words, pull clips from a long recording, add captions, translate. It can also start a draft video from a prompt, which then lands in your Descript account for editing. It is the right choice when you already have a recording and the problem is the editing.

A test you can run in ten minutes

Before you commit to any of them, give each the same brief and judge the result against the job. Here is a brief that exposes the differences quickly:

Make a 60-second video explaining why the Moon has phases. The viewer should understand that half the Moon is always lit by the Sun, and that what changes is how much of the lit half we see from Earth.

What tends to come back:

• Footage: a beautiful Moon in the sky, changing shape. It shows that phases happen, not why.
• Presenter: a person explaining it accurately, if your script is accurate, with no diagram of Sun, Earth and Moon.
• Stock edit: night skies, telescopes, maybe an astronaut. Pretty, and the "half is always lit" idea never appears on screen.
• A drawn explainer (outside ChatGPT, see below): the Sun on the left, Earth in the middle, the Moon drawn at four positions with its sunlit half shaded, and the view from Earth drawn beside each one.

Only the last one makes the explanation visible. Which one you need depends on whether the video is meant to explain something or to set a mood.

When to leave ChatGPT for the video step

For explanations, lessons, revision videos and training, the useful split is: ChatGPT for the thinking and the first draft of the script, then a tool that draws the explanation from that script. Drawn tools are not in the table above because they work from a pasted script rather than from a chat command, and that is the point: the script you checked is the thing the video is built from.

The workflow, including what to strip from a ChatGPT script before you paste it, is in how to turn a ChatGPT script into an animated video. The short version is below.

Make it

1. 1

   Decide what the picture must show

   Write one line: the thing a viewer must see to understand. If it is a mood, a footage app is fine. If it is a mechanism, you need drawn scenes.

2. 2

   Draft the script in ChatGPT

   Ask for narration only, two to four sentences per scene, with one short line saying what is drawn. No timestamps, no music cues.

3. 3

   Check every fact yourself

   Names, dates, numbers and the direction of every arrow. ChatGPT is fluent and sometimes wrong, and a video makes the error look official.

4. 4

   Paste it into openCanviz

   Choose Keep my wording if the script is final, set the target length, and pick a style: whiteboard for a process, poster or painted for a story.

5. 5

   Fix scenes one at a time

   Watch with the sound off. Any scene that does not show its own sentence gets redrawn or rewritten in the editor.

6. 6

   Replace the voice if you need to

   Keep the generated voice, or record your own narration over the same scenes if a teacher or client asks for it.

Common questions

Is Sora still available anywhere? No. OpenAI discontinued the Sora app and website in April 2026 and the Sora API in September 2026. Third-party sites that still use "Sora" in their names are not OpenAI products.

Do I need a paid ChatGPT plan to use these apps? It depends on the app and your ChatGPT plan, and both have changed often. In practice the bigger cost is the video company's own account, because the rendering happens there and uses that company's credits. Check both before you plan a project around one.

Which ChatGPT video app is best for school? For a video that has to explain something to a teacher, none of the four kinds is ideal on its own, because none draws the explanation from your script. Write the script with ChatGPT's help, check it, and make the video in a tool that draws scenes from it.

Can ChatGPT itself turn its images into a video? Its code tool can sometimes stitch images into a short silent clip or animated GIF if you ask, but it has no voiceover, no timing to narration and no editing. It is a curiosity, not a production route.