AI video: the best models today

What you'll see in this lesson

1. AI video in October 2026

   Six families lead in the cloud and there are four open models for your computer; almost all now generate image and sound together.

2. What happened to Sora

   OpenAI shut down Sora in 2026: first the app, in April, and then the API, which went dark in September. It no longer works in any workflow.

3. Google: Veo 3.1 or Gemini Omni Flash

   Veo 3.1 is the quality specialist; Gemini Omni Flash, introduced in May 2026, generates and edits video through conversation.

4. Kling 3.0: several shots in a single generation

   Kling 3.0, from Kuaishou, creates sequences of up to 15 s with several consecutive shots and the same character in all of them.

5. MiniMax H3: 2K, fast and with open weights

   MiniMax H3, also called Hailuo 03, generates 4 to 15 s in 2K with stereo audio and can also run locally.

6. Seedance 2.5 for long shots, Runway for editing

   Seedance 2.5 generates up to 30 s in a row with audio; Runway excels at transforming video you've already recorded.

7. On your computer: Wan 2.2, LTX-2.5 and HunyuanVideo

   Open models run free on your graphics card with ComfyUI; for video you'll want 24 GB of video memory or more.

8. Which model to use for each job

   Choose by shot type, not by fame: each model wins at something different and the price per second varies a lot.

9. Summary

   There's a model for every shot type: learn their strengths and plan the video shot by shot.