Hyperstack and SiliconFlow roundup

last edited

This Hyperstack and SiliconFlow roundup tracks open-weight video generation models as of late 2026. Vendor lists name many families, but primary model cards give the checkable details. The page extends the open source video landscape and the image generation snapshot.

Wan family

Wan 2.2 uses a 2-expert mixture-of-experts design for video diffusion. Its code and model cards on GitHub and Hugging Face carry a permissive open license. It remains the newest open-weight base model in the family. Later releases are API-only, with no open weights on the official organization.

A music-to-dance model arrived in July 2026. A character-animation release followed in August 2026. Background on the family is in Wan 2.2 and FramePack.

Newer open releases

Lightricks shipped LTX-2.5 in August 2026. The March 2026 release is now the previous generation.

MiniMax H3 was announced in late July 2026. Weights followed in early August. It is an open audio-video model.

Kandinsky 6.0 Video published open weights in October 2026 under the MIT license. It ships in two sizes, 29B and 3B.

Open video models pair with leading image generation models, which can supply source frames for animation.

Older options and adapters

Hyperstack's September 2026 list still names earlier generations, including cinematic, speed-focused and short-clip models. Those entries are not the current flagships.

Motion adapters add animation to Stable Diffusion workflows without a separate video model. Existing checkpoints and controls stay in place. ComfyUI Core covers node integration.

Related pages: AI art aesthetics and visual trends and open-source generative art techniques and workflows.

Sources