What changed in AI video
in October 2026

What were the main AI video releases in October 2026?

Five releases matter to creators this month. Two are video models, two work on sound, and one moves a whole toolkit into ChatGPT.

What is new in Kling 4.0?

Kuaishou says Kling 4.0 officially launched on 15 October. The headline is length: up to 30 seconds in a single generation. Veo 3.1 clips stop at 8 seconds and Kling 3.0 at 15.

The second change is references. Omni Reference takes up to 15 at once, counting images, videos, voice profiles and elements together: up to 10 images, 5 videos and 7 subjects, in Kuaishou’s figures. Kuaishou names “three-view character sheets, grid images, storyboards and wireframes” among the images it accepts, which matters if you keep the same character across shots.

Kling’s own user guide adds the working details:

• Length is set from 3 to 30 seconds in 1-second steps, 10 by default.
• Prompts can run to 8,000 tokens.
• Audio input is voice only; music and effects are generated from the picture.
• Kling 4.0 Flash, the faster version, stops at 20 seconds and 720p, and “does not currently support start-and-last frames or multiple keyframes”. It went to a limited early-access group on 28 September (Kling).

One feature is not live yet. Kuaishou says video extension “will soon support multiple extensions of up to two minutes”.

Did AI video sound get better?

Two new tools work on the sound of a finished video rather than on the picture.

ElevenLabs launched Instant Soundtracks on 15 October. You upload a video, the model scores it with original music, then finds moments that need an effect and places each one on the timeline. Through the API it can return music and effects as separate files, so you can mix them yourself.

Mirelo released SFX 2.0 a day earlier. It is aimed at the music bed a video model adds when nobody asked for one: it removes it, keeps the dialogue and builds effects and ambience around it. Before you publish anything made with either tool, check the licence: our guide to music and sound for AI video covers ElevenLabs’ plans; for Mirelo, read its own terms.

Is Gemini Omni new?

No. Google launched Gemini Omni Flash on 19 May 2026 for Google AI Plus, Pro and Ultra subscribers, in the Gemini app and Google Flow, and said it was “rolling out at no cost to users on YouTube Shorts and YouTube Create App”. Google’s pitch is editing: you change a video by describing the change in conversation.

Flow’s credit table prices Omni Flash at 12 credits for an 8-second clip at 720p, 6 at 360p, and 40 to edit a video. Free accounts get 50 Flow credits a day.

What about MiniMax H3?

MiniMax launched H3 on 31 July 2026: clips of up to 15 seconds at 2K, with stereo sound generated alongside the picture. The model is also on Hugging Face, where MiniMax calls it an “initial open-source release”.

What else changed?

• Sora is fully gone. OpenAI closed the Sora API on 24 September, five months after the app and website closed on 26 April. Our Sora alternatives page compares the replacements.
• Seedance 2.5’s prompt guide was updated on 16 October. ByteDance confirms clips of up to 30 seconds and up to 50 reference files per request.
• Pika in ChatGPT adds 250 bonus credits to your Pika account after your first generation through the plugin, says Pika.

What should you try first?

• Long takes: test Kling 4.0 or Seedance 2.5 on a 30-second shot before you stitch four short clips together.
• Silent or messy clips: run one through Instant Soundtracks or Mirelo and compare it with the model’s own audio.
• Cheap tests: FLUX 3 Max’s launch price ends on 7 November; the list price on fal is $0.17 a second at 720p.