Sora and Runway are used for short clips: animating a product photo, building an intro, making a video for social media. Open video models do the same inside the company — scripts, frames and client footage never reach a third-party service, and the number of attempts is limited only by your hardware, which matters for video because a good take rarely comes first. Video is heavier than images though: smaller models run on a gaming GPU, larger ones expect server cards, and each clip takes noticeable time to render. Clip length, face stability and motion physics are weaker in open models than in mature cloud services, so a complex narrative piece is still out of reach. The realistic use is short clips and animated stills, with editing and sound added on top.
Sber's Russian family of image and video generation models. Understands Russian-language prompts and Russian cultural context well; released under MIT.
Text-to-video and image-to-video; the small version runs on a gaming GPU. After 2.2 only applied models are open: editing (VACE), audio-driven talking characters (S2V), dancing to music (Dancer).
A fast video model; with LTX-2 it generates video with sound and speech in one go. Camera and pose control, lightweight versions available.
Tencent's video model, one of the first open ones on par with closed services. Version 1.5 is lighter (8.3B) and runs on consumer GPUs.
A 2–5B video model that runs on a single gaming GPU. A popular base for research and add-ons.
Wan and Kandinsky are Apache 2.0 and MIT, while LTX-2 and HunyuanVideo carry their own community licenses with restrictions (HunyuanVideo's does not apply in the EU, the UK and South Korea), and CogVideoX 5B allows commercial use after registration. Have the terms of your chosen version checked by a lawyer.
An open model can run on your own server: data stays in-house, there is no per-request fee, and the model can be fine-tuned on your documents.
