# AI Video and Speech Models

> Veo makes video with sound, and Whisper turns speech into text. Sora shows how a closed model can be switched off for good.

**Course:** [AI Models and Providers](https://learn.benibix.com/learn/ai-models-providers)  
**Time:** 10 min

**After this lesson you can:**

- Name the makers of Veo and Whisper
- Say what happened to Sora, and when
- Explain why a speech transcript needs checking

## Which models make video and turn speech into text

Video and speech have their own model families. Google's Veo makes short video clips from words, and since Veo 3, in May 2025, it makes the sound too. OpenAI's Whisper turns speech into text in many languages, and it is free to download under the MIT license. OpenAI's Sora video app closed in April 2026.

Veo is made by Google DeepMind. You describe a scene and get a short clip, with voices, music and noise to match. Veo 3.1 came out in October 2025. You can use it in the Gemini app or through an API from Google. A longer film is made by joining many short clips.

Sora was OpenAI's video model. OpenAI first showed it in 2024, and Sora 2 added sound and a social app of its own. Then OpenAI shut it down. The app and website closed on 26 April 2026, and the API on 24 September 2026. Users were asked to save their videos before the data was deleted.

Whisper listens to speech and writes down the words. It can also turn speech in other languages into English text. It comes in six sizes, from tiny, with 39 million parameters, to large, with about 1.5 billion. Its code and weights are free under MIT, so it runs on your own laptop.

These models sit inside tools you already use. A meeting app that writes notes often uses a speech model like Whisper. A video tool may call Veo behind its own buttons. You may never see the model's name, only what it makes.

Sora is a reminder that a model can end. A closed model lives on its maker's servers. When the maker stops it, every app built on it stops too. An open model like Whisper stays on your disk for as long as you keep the file.

People trust a speech model's text as a perfect record. Whisper can mishear names, and in long silences it sometimes writes words nobody said. For a contract, a court case or a note from a doctor, a person checks the text against the sound.

![Four rows, each reading input, model, output. Words go into Stable Diffusion or FLUX and a picture comes out. Words go into Veo and a video with sound comes out. Speech goes into Whisper and written text comes out. A picture and some captions go into CLIP and the best match comes out.](https://learn.benibix.com/diagrams/ai2-media-models.svg)

### Example: Jack subtitles his podcast

Jack's podcast runs 45 minutes. Whisper, on his own laptop, writes the whole transcript in about 5 minutes.

He reads it against the sound and finds 6 misheard names and one line nobody said. He fixes all 7 before he posts it.

**Words in this lesson:** [Veo](https://learn.benibix.com/term/ai/veo), [Sora](https://learn.benibix.com/term/ai/sora), [Whisper](https://learn.benibix.com/term/ai/whisper)

**Question:** Why did apps built on Sora stop working in 2026?

**Answer:** OpenAI shut the closed model down, and nothing could run it. Sora was closed, so it only ran on OpenAI's servers. Those were shut in September 2026, and every app built on it stopped.

---

Source: https://learn.benibix.com/learn/ai-models-providers/ai-video-and-speech-models
Author: Behnia Zarei (a developer in Vienna, Austria, who builds and teaches data, AI and business intelligence at BENIBIX)
Publisher: BENIBIX Learn (https://learn.benibix.com/)
Last updated: 2026-11-18
Licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Quote freely with attribution.
