Can ChatGPT make videos?

No, not by itself. OpenAI closed Sora in ChatGPT on April 26, 2026 and shut the Sora API on September 24, 2026, so ChatGPT no longer has a video model of its own. You can still get a video through a ChatGPT plugin such as Runway or invideo, or through Codex, which writes the video as code and renders the MP4 on your computer.

What happened to Sora

Until this spring, the answer was Sora. OpenAI discontinued the Sora web and app experiences on April 26, 2026. Its API deprecation list then shows the Videos API and every sora-2 model shut down on September 24, 2026, with no replacement named.

So any guide that tells you to open Sora from the ChatGPT sidebar is out of date. If you made Sora clips, OpenAI's help page points to an export at sora.chatgpt.com/sunset and warns that the data is deleted once any final export window closes. It also says purchased ChatGPT or Sora credits can still be spent in Codex, which matters for the route below.

What ChatGPT can still do

On its own, ChatGPT now stops at words and stills. It can write the script, plan the shots and draw frames with ChatGPT Images. For moving pictures it hands off to a plugin. Since July 9, 2026, the old app directory lives under Plugins in the sidebar. Three that make video:

• Runway. Install it from Plugins, sign in with a Runway account and mention @Runway in a chat. The finished clip comes back in the conversation. Runway lists models including Gen-4.5, Veo 3.1, Kling 3.0 and Seedance 2.5; which ones you get depends on your Runway plan.
• invideo. An official plugin that turns a prompt into a narrated video with visuals, voiceover and music, and can open and edit projects in the invideo Editor, down to a 9:16 cut for Reels.
• HyperFrames by HeyGen. A plugin built around HeyGen's HTML-to-video framework. In a chat, its hosted connector builds the video and renders it on HeyGen's servers through your HeyGen account; HeyGen labels that hosted path a beta.

Each is a real way to get a clip without leaving the chat. None of them is ChatGPT making video: the rendering happens on the provider's servers, under the provider's account.

Plugin videos bill the plugin, not ChatGPT. Runway generations draw on your Runway credits, and invideo's AI generation uses the credits on your invideo plan. Your ChatGPT subscription covers neither, and whether a plugin appears at all depends on your ChatGPT plan, region and workspace.

What ChatGPT can't do

• Generate a clip with an OpenAI model. OpenAI's deprecation notice names no replacement for Sora, in ChatGPT or in the API.
• Give you a project you own. A plugin returns a clip, or a link into the provider's editor. You don't get source files on your disk that you can reopen, change one line of and render again.
• Render on your machine. Every plugin route runs in someone else's cloud and stops when the credits do.

The last two are the reasons to move the job into a coding agent. If you already have footage and want it cut rather than generated, that's a different question, answered in Can ChatGPT edit videos?

The Codex route: video as code

Codex is OpenAI's coding agent. It comes with ChatGPT plans, Free included, with limits that depend on the plan. The Codex CLI runs in your terminal, where it can edit files and run the tools already installed on your machine. Give it a video framework and it writes the video as code, previews it and renders an MP4 locally.

The result is a project folder, not a sealed clip. Changing a word, a color or a beat means editing one line and rendering again, which is how a reel gets from first draft to posted without starting over.

Three engines do the rendering. HyperFrames, from HeyGen, turns HTML, CSS and animation into MP4 and is Apache 2.0 licensed. Remotion does the same with React; it's free for individuals and for companies of up to three employees, and larger for-profit companies need a company license. ffmpeg handles cutting, joining and encoding, and HyperFrames needs it to render. The trade-offs are in HyperFrames vs. Remotion.

1. Install Codex

   Use the standalone installer, npm or Homebrew. Then run codex in an empty folder and choose Sign in with ChatGPT.

2. Add a video skill

   HyperFrames needs Node.js 22 or newer and FFmpeg; npx hyperframes doctor checks both. Its skills install to ~/.agents/skills, a folder Codex reads in every project. Start a new Codex session afterward.

3. Ask for the video

   Name the skill with $hyperframes, describe the video and ask for the plan first. Codex may ask before it runs a command or uses the network; approve those steps for this folder.

4. Preview, revise, render

   Codex opens a local preview. Say what to change in the same conversation, then ask it to render. npx hyperframes render --output final.mp4 does it by hand.

The install commands

# Codex CLI on macOS or Linux
curl -fsSL https://chatgpt.com/codex/install.sh | sh

# or with a package manager
npm install -g @openai/codex
brew install --cask codex

# HyperFrames skills for Codex (needs Node.js 22+ and FFmpeg)
npx hyperframes skills update
npx hyperframes doctor

Remotion works the same way. Scaffold a project, add its agent skills, start the preview, then open Codex in the folder. Remotion also publishes a Codex plugin in the ChatGPT desktop app; search Plugins for Remotion and type $remotion.

npx create-video --yes --blank my-video
cd my-video
npx remotion skills add
npm run dev

# in a second terminal
cd my-video
codex

frame 000/180№002

ChatGPT vs. Claude vs. Gemini

Ask the three chat apps the same question and you get three different answers. Only one has its own video model in the app today.

Gemini is the pick for a generated, photoreal clip from a prompt. Gemini Omni Flash creates and edits video in the Gemini app, and Google says it replaces Veo there. Editing an uploaded video with it isn't available in the EEA, Switzerland, the UK and some US states.

Claude, like ChatGPT, doesn't generate footage. Anthropic says Claude Motion uses no video generation model; it builds animations as code, which is why you can change one word and keep the rest. The full picture is in Can Claude make videos?

For a reel you can rebuild line by line, the chat brand matters less than the agent and the engine. HyperFrames lists Codex, Claude Code, Cursor, Gemini CLI and Google Antigravity among the agents its skills support, so the same project runs in any of them.