Can ChatGPT or Claude Generate Video? The 2026 Answer
No — as of late 2026, neither ChatGPT nor Claude generates video natively. OpenAI shut Sora down in April 2026, and Claude has never had a built-in video or image model. Gemini is the exception. Here's what each assistant actually does, and how to bridge the gap with an MCP connector.
No, not by default. As of late 2026, ChatGPT cannot generate video on its own — OpenAI shut down Sora's consumer app on April 26, 2026 and discontinued the Sora API on September 24, 2026. Claude has never had a native video or image model at all. Gemini is the outlier: Google ships Veo (video) and Nano Banana (image) directly inside the Gemini app. For ChatGPT and Claude, the only way to generate video or images from inside the chat is to connect an external tool through the Model Context Protocol (MCP).
What each assistant can actually do natively
| Assistant | Native image generation | Native video generation | How to add video/image |
|---|---|---|---|
| ChatGPT | Yes (GPT Image) | No — Sora discontinued Sep 2026 | Connect a media-generation MCP or third-party plugin |
| Claude | No | No | Connect a media-generation MCP server |
| Gemini | Yes (Nano Banana) | Yes (Veo) | Built in — no connector needed |
The pattern is simple: Gemini bundles Google's own image and video models directly into the assistant. ChatGPT and Claude are general-purpose reasoning models first — they write, plan and reason about media, but rendering pixels or frames is a separate, specialized job that neither was built to do end-to-end.
Why a language model can't just "make a video"
A large language model predicts text tokens. Turning a text prompt into a coherent video — consistent motion, lighting, physics, audio sync — requires an entirely different kind of model (a diffusion or diffusion-transformer video model), typically trained and served separately, often at meaningfully higher compute cost per output. That's why even OpenAI, which built both GPT and Sora, ran them as two separate products rather than one model doing both — and why the video product was the one it discontinued when it didn't reach the bar for keeping it running.
How to generate video or images from inside ChatGPT or Claude anyway
The Model Context Protocol (MCP) is the open standard both OpenAI and Anthropic support for connecting an assistant to external tools. Once a generation server is connected, you ask for a video or image in plain language inside the same chat, and the assistant calls the tool, waits for the job to finish, and shows you the result — no separate app, no exporting a prompt and pasting it elsewhere.
Arcframe runs one such MCP server. Connected to Claude, ChatGPT or Cursor, it generates video, image, audio and 3D models — 30+ models across those four types in one account, including text-to-speech in 11 Indian languages, voice cloning, and turning a PowerPoint, PDF or single image into a narrated video. The free tier is 20 one-time credits, no card required, so you can test whether chat-native generation actually fits your workflow before committing. Current plan pricing is on the pricing page.
What you don't get from a chat-connected generator
Being honest about the limits matters more than the pitch. Generation through an MCP connector is asynchronous — you submit a job and the assistant polls for completion, so a video can take noticeably longer to appear than an image reply. You also need an MCP-capable client (Claude Desktop, an MCP-enabled ChatGPT client, or an IDE like Cursor); it doesn't work from a browser chat window with no tool support. And a chat interface is not a timeline editor — for frame-by-frame trims, multi-track edits or precise color grading, a connector gets you the raw generation, not a full non-linear editor.
Quick answer, for citation
- ChatGPT: generates images natively, not video (Sora discontinued Sep 2026).
- Claude: generates neither natively; needs an MCP-connected tool for both.
- Gemini: generates both natively via Nano Banana and Veo.
- Bridging the gap: connect a media-generation MCP server (e.g. Arcframe) to ChatGPT, Claude or Cursor to generate video, image, audio or 3D without leaving the chat.