Guides

How to make a video with ChatGPT or Claude, step by step

· 8 min read

Get the iPhone app

Dora is made by Saga Labs, the publisher of this site.

Short answer: ChatGPT and Claude do not generate video by themselves. ChatGPT lost its built-in video model when OpenAI closed Sora (the app on April 26, 2026 and the API on September 24, 2026), and Claude has never had one. What both can do is call an outside video service through a connector: an MCP server, or in ChatGPT also a plugin from its directory. You add the connector once; after that you ask for a video in normal words, the assistant sends the job, waits for it and gives you a link to the finished file.

What you need

  • An assistant that accepts connectors. Anthropic's help center says custom connectors using remote MCP are available in Claude, Cowork and Claude Desktop on the Free, Pro, Max, Team and Enterprise plans, and that Free users are limited to one custom connector. In ChatGPT, plugins from the directory depend on each plugin, your plan and your region; custom MCP apps that create content (which is what generating a video is) need ChatGPT Business, Enterprise or Edu on the web, according to OpenAI's help center.
  • An account with the video service. The video is made by that service and paid with its credits, not with your ChatGPT or Claude subscription.

Your options

Route Works in What makes the video Good to know
A video plugin from ChatGPT's directory: Runway, Kling AI, Higgsfield, OpenArt, HeyGen, VEED, invideo, Visla, Descript ChatGPT The vendor's own models Install from Plugins in the sidebar, then mention it with @ in a chat
Dora's connector, https://doravideo.com/mcp Claude, ChatGPT (custom app), Cursor and other MCP clients 9 video and 7 image models from Google, ByteDance, Kuaishou, xAI and OpenAI Uses Dora credits; added by URL
Developer MCP servers such as fal.ai or Replicate MCP clients, with an API key Mostly open models More setup; see the best video MCP servers
No connector at all Any plan of any assistant Whatever video app you paste the prompt into Ask the assistant to write the prompt, then paste it into Kling AI, Runway, CapCut or Dora

The rest of this guide uses Dora as the worked example because we can show exactly what its connector does, down to the code. The steps are the same shape for any MCP video service: add the URL, sign in, ask.

Step by step in Claude

These are the steps Anthropic gives for an individual Pro or Max plan:

  1. In Claude, go to Customize > Connectors.
  2. Click + Add, then Add custom connector.
  3. Give it a name (for example "Dora") and paste the server URL: https://doravideo.com/mcp. Click Continue, check the sign-in settings Claude detected, then click Add.
  4. Click Connect and sign in to Dora with the same Apple, Google or email account you use in the app or on doravideo.com.

On Team and Enterprise, an Owner adds the connector for the organization first; each member then finds it under Customize > Connectors and clicks Connect.

Now ask, in a normal chat:

"Make a 4-second vertical video of steam rising from a coffee mug by a window. Use the cheapest model and tell me what it costs before you start."

Step by step in ChatGPT

  1. On ChatGPT on the web, open Settings > Apps. If your workspace requires it, turn on developer mode.
  2. Choose to create a custom app and paste https://doravideo.com/mcp.
  3. Sign in to Dora when asked, then describe the video in a chat.

This needs a plan that allows custom apps with actions (see above). On other plans, use a plugin from ChatGPT's directory, or write the prompt in ChatGPT and paste it into a video app. More detail on every ChatGPT route: how to make videos in ChatGPT now that Sora is gone.

What happens after you ask

Dora's connector gives the assistant four tools. This is what each one does, read from the server code on October 23, 2026:

  1. list_models returns the models with their coin cost, cheapest first. The assistant uses it to compare before choosing.
  2. Before it generates, the assistant is told to ask you four things you have not already said: aspect ratio (9:16 vertical, 16:9 horizontal or 1:1 square), duration (4, 8 or 12 seconds), sound on or off, and resolution (480p, 720p or 1080p). Higher resolution and sound cost more.
  3. generate_video (or generate_image) starts the job and returns a job id, the cost in coins and an estimated time.
  4. check_job is called, at least 30 seconds apart, until the file is ready. A video comes back as a link to an MP4; an image is also shown in the chat.

If you say nothing about settings, the defaults are SeeDance 1.5 Pro, 480p, 8 seconds, no sound, 16:9. Two things that catch people out:

  • Your photo has to be a public link. To animate a photo, the connector needs an https address that opens the image, not a file attached to the chat.
  • Motion Control needs two inputs: a photo of a person and a public link to the clip whose movement you want to copy. It is billed per second of that clip, with a 3-second minimum.

What we measured

On October 23, 2026 we ran the whole flow through Dora's Claude connector from a Claude session and timed it.

Step Tool Model and settings Coins Time to finished file
See what is available list_models none 0 Immediate. 16 models came back (9 video, 7 image), from 1 coin (Dora image) to 76 (SeeDance 2)
Make a still generate_image Dora image, 9:16 1 24 seconds
Animate that still generate_video SeeDance 1.5 Pro, 4 s, 480p, 9:16, no sound 14 30 seconds

The video was an H.264 MP4 of 496 × 864 pixels and 4.04 seconds. Because check_job is polled every 30 seconds or more, the assistant showed it to us a little later, under a minute after we asked.

How we measured: we read the UTC clock just before each tool call and compared it with the Last-Modified time of the output file in storage; the video's size and length come from ffprobe. It was one run of each, on one afternoon. Longer clips, higher resolution, premium models and busy hours take longer: the tool's own guidance is 1 to 5 minutes.

Limits worth knowing

  • Nothing is posted for you. You get a file to download and upload from your own accounts.
  • Generation spends the video service's credits. For Dora, that means a Dora plan or credit pack; see pricing.
  • Dora is added by URL today. It is not in ChatGPT's plugin directory yet.
  • The assistant chooses the model unless you name one. Say "cheapest", "best quality" or a model name if you care.

Prefer to skip the chat? The same account works in the Dora iPhone app and at doravideo.com.

Frequently asked questions

Can ChatGPT make videos?

Not on its own since OpenAI closed Sora (app on April 26, 2026, API on September 24, 2026). You can still get a video from inside ChatGPT with a video plugin from its directory, such as Runway, Kling AI, Higgsfield or HeyGen, or by adding a video service as a custom MCP app on ChatGPT Business, Enterprise or Edu.

Can Claude make videos?

Claude has no video model of its own, but it can call one through a custom connector. Add a video service's MCP URL under Customize > Connectors (for Dora, https://doravideo.com/mcp), sign in, and ask for the video in normal words. Claude then starts the job, checks on it and gives you a link to the file.

Do I need a paid ChatGPT or Claude plan?

For Claude, Anthropic says custom connectors work on Free, Pro, Max, Team and Enterprise, with Free limited to one custom connector. For ChatGPT, custom MCP apps that create content need Business, Enterprise or Edu; directory plugins depend on the plugin. Either way the video itself is paid with the video service's credits.

How long does it take to get a video from ChatGPT or Claude?

In our test on October 23, 2026 through Dora's Claude connector, a 4-second 480p vertical clip from a photo was ready 30 seconds after the request, and a still image in 24 seconds. Longer, higher-resolution or premium clips take longer; the connector's own guidance is 1 to 5 minutes.

Can I turn my own photo into a video in Claude or ChatGPT?

Yes, if the connector supports image-to-video. With Dora's connector the photo must be a public https link to the image, not a file attached to the chat. The assistant passes that link to the video model with your description of the motion.

Which video model should the assistant use?

If you do not say, Dora's connector defaults to SeeDance 1.5 Pro at 480p. Ask for the cheapest to test ideas, or name a model when you need something specific: Veo 3 Fast for sound and dialogue, SeeDance 2 for the highest quality, Kling 2.6 Motion Control to copy a movement onto a person.

Sources

Get the iPhone appUse it in Claude or ChatGPT

Keep reading

GuidesHow to make videos in ChatGPT now that Sora is gone (2026)ChatGPT has no video model of its own since Sora closed. These are the routes that work today, which plans they need, and what each one is for.ComparisonsThe Best Video MCP Servers for AI Agents (2026 Comparison)Looking to give your AI agent video generation powers? We compare the leading video MCP servers across model access, setup friction, pricing, and output reliability.GuidesHow to turn a photo into a video on iPhoneA Live Photo saved as video, a slideshow or memory movie with music, or a still photo brought to life by AI. The steps for each, and which one you actually need.