Wan 2.2 14B Image to Video

The rest are this month's best-liked on Civitai, shown from there; each belongs to the person who made it. Open one for its maker.

About

Trained on data labelled for lighting, composition, contrast and color tone, to give control over cinematic style. Built to keep video stable, avoid unrealistic camera moves and handle stylized scenes. The video keeps the shape of the original picture, whether it is tall or wide.

How it compares

In LM Arena's public vote, it places third of the four open models rated for making video from a picture.

Good to know

One part of the model plans the overall layout, and a second part refines the details.

It was trained on 65.6% more images and 83.2% more videos than the earlier Wan2.1.

The makers released the Wan models and their code openly so others can build on them.

Sources: Wan-AI/Wan2.2-I2V-A14B model card, Wan: Open and Advanced Large-Scale Video Generative Models, Outmake's speed estimates, LM Arena and Hugging Face. Every line comes from one of these.

Run It Anywhere

Wan 2.2 14B Image to Video is an open model. For ComfyUI, with the ComfyUI-GGUF nodes installed, put one of the GGUF files in ComfyUI/models/unet.

• ComfyUIWindows, Linux, MacFiles