Who makes Kling, Seedance, and Veo? The companies behind the AI video models, and why it matters

Kling is made by Kuaishou, the Chinese company behind the short-video apps Kuaishou and Kwai. Seedance is made by ByteDance's Seed lab, the company behind TikTok and CapCut. Veo and its successor, Gemini Omni, are made by Google DeepMind, the AI lab of the company that owns YouTube.

All three makers run a video app with millions of creators on it, and two of them already put their model inside it: Seedance in CapCut, Gemini Omni in YouTube Shorts. This post covers each maker, the other names you will see in model lists, and what the maker tells you before you pick a model. Every fact carries its source and date.

Kling is the video model of Kuaishou, a Chinese short-video company. Kuaishou runs the Kuaishou app in China and Kwai everywhere else, and Kling AI is its video-model unit. Latin America is the center of Kuaishou's business outside China (China Daily, March 2025). Kwai has about 60 million monthly users in Brazil (Caixin, December 2025), and it launched in Mexico in April 2021 (Expansión).

• Kling 3.0 makes 3 to 15 second clips with sound, up to 6 shots in one generation, and lip-sync in five languages, Spanish among them (Kling, July 2026).
• Kling 4.0 (October 15, 2026) makes up to 30 seconds in one pass, takes up to 15 reference images and videos and 10 keyframes, and adds stereo sound and HDR (Kling; Bernama, October 19, 2026).
• What Kling says it is best at: keeping the same character across shots, physics-based motion, and readable text in image-to-video (Kling).

Seedance is made by Seed, ByteDance's AI lab. ByteDance owns TikTok, Douyin (TikTok in China), CapCut, and Dreamina, its AI creation app, and it sells Seedance to developers through BytePlus.

• Seedance 2.0 (February 2026) makes up to 15 seconds with sound from text, images, audio, and video references. ByteDance put it in CapCut for paid users first in Brazil, Mexico, and five countries in Southeast Asia (press reports).
• Seedance 2.5 (July 31, 2026) makes up to 30 seconds in one run and takes up to 30 reference images, 10 videos, and 10 audio clips (TechNode, July 31, 2026).
• Seedance 2.0 Fast and 2.0 Mini are the cheaper tiers, at 480p and 720p (BytePlus).
• OmniHuman is ByteDance too: a model that turns one portrait and one audio track into a talking video.
• Where it ranks: Seedance 2.5 was #1 on Artificial Analysis's image-to-video board when we read it on November 14, 2026.

Veo is made by Google DeepMind, Google's AI lab. Veo 3.1 launched on October 15, 2025, with sound, up to three reference images, and first and last frame control, in clips of 4, 6, or 8 seconds (Google). On May 19, 2026, at Google I/O, Google announced Gemini Omni, a new family of video models that takes text, images, audio, and video as input (Google; TechCrunch).

Google's own deprecation page lists the Veo 3.1 models in the Gemini API as shut down on October 22, 2026, and names Gemini Omni 1.2 Flash as the replacement. Veo 3.1 is still listed on Google Cloud. So if you build on Google's API, Google's video model is now Omni.

Hook Studio runs Gemini Omni 1.2 Flash since November 14, 2026. Veo 3.1 cannot start from a still through our video provider, so Hook Studio does not offer it. For Kling against Google's model on the same shots, see Kling vs Veo.

• Gemini Omni 1.2 Flash (November 5, 2026) makes 3 to 15 second clips, vertical or horizontal, from text, an image, or references, and edits a clip when you ask in plain words. It costs $0.07 a second at 720p on Google's API (Google, November 5, 2026).
• Its home turf: Omni Flash was free to use in YouTube Shorts and the YouTube Create app at launch (Google, May 2026). Omni 1.2 Flash is in Flow, Google Vids, and Google Ads (Google, November 5, 2026).
• Where it ranks: Gemini Omni Flash 1.2 was #5 on Artificial Analysis's text-to-video board on November 14, 2026. Veo 3.1 was #25.

Kling, Seedance, and Google's models are the names creators compare most, but they are not alone. These are the other makers you will meet in model lists and price tables:

• Alibaba: Wan. Wan 3.0 came out of beta on August 24, 2026. It makes up to 30 seconds at up to 1080p, and its weights are closed (The Next Web). It was #1 on Artificial Analysis's text-to-video board on November 14, 2026. Alibaba's Wan Animate, released as open weights in September 2025, puts your character into an existing video.
• MiniMax: Hailuo. MiniMax runs the Hailuo AI app. Its H3 model came out on July 31, 2026, with clips up to 15 seconds, 2K, and stereo sound. MiniMax says it built H3 for following instructions and getting text and brand names right, for ads and online stores (MiniMax).
• Shengshu: Vidu. Shengshu, a spin-off of Tsinghua University in Beijing, makes the Vidu models. A preview of Vidu Q4 came out on October 7, 2026 (China Tech News).
• xAI: Grok Imagine. Elon Musk's AI company released Grok Imagine Video 1.5 in June 2026, and claims sharper realism, better physics, and faster generations (GIGAZINE, June 18, 2026).
• Runway: Gen-4.5 and Aleph. Runway, a US company, announced Gen-4.5 on December 1, 2025, and lists its own limits on that page: effects that come before their cause, objects that vanish, and actions that succeed too often. Aleph 2.0 (May 21, 2026) edits footage you already have, up to 30 seconds at 1080p (Runway).
• Higgsfield: Genjutsu. Higgsfield is an AI video app that offers many makers' models. Its own Genjutsu (August 31, 2026) rebuilds part of a clip you already have: Motion Transfer puts a new character on the motion, and Object Swap replaces one person or product and keeps the rest of the shot (Higgsfield). Hook Studio's recasts run on Genjutsu.

For what a recast looks like in practice, see how to recreate a viral video with AI.

OpenAI announced on March 24, 2026, that it would discontinue Sora. The Sora app closed on April 26, 2026, and the Sora API on September 24, 2026 (OpenAI help center; The Decoder, March 28, 2026). OpenAI gave no reason. Press reports pointed to the cost of running it and falling use. If a list of the best AI video generators still ranks Sora, it is out of date.

• The maker decides where you can use it. Seedance shows up in CapCut and Dreamina, Gemini Omni in YouTube Shorts, Flow, and Google Vids, Kling in the Kling AI app. Tools like Hook Studio run several makers' models side by side: Kling 3, Kling 4, four Seedance models, and, since November 14, 2026, Gemini Omni 1.2 Flash from Google DeepMind.
• The maker decides when it goes away. Google retired its Veo 3.1 API models about a year after launch, and its own page lists Gemini Omni 1.1 Flash, released on August 27, 2026, for shutdown as early as December 3, 2026. OpenAI closed Sora. Keep your stills and prompts, not your loyalty to one model.
• The rankings move every few months. On November 30, 2025, Runway Gen-4.5 was #1 on Artificial Analysis's text-to-video board (Runway). On November 14, 2026, Wan 3.0 was #1. A blind vote on short clips says which clip people liked, not which model follows your direction or costs less.
• Language support is the maker's call. Kling 3.0 lip-syncs in five languages, and Spanish is one of them (Kling). If you post in Spanish, check that before you pay for a talking clip.

So the maker tells you where a model lives and how long it may last. It does not tell you which one wins your shot. For that, run the same still through two or three models and compare. Our guide to making AI videos shows the steps on a real session, from the photo to the finished reel.