Sheet 45 · AI video models & image models

How VidoGrid tracks AI video models

New AI video models and image models arrive almost every week, and the news about them is scattered across lab blogs, GitHub repositories, research papers, platform changelogs and video channels. VidoGrid reads 49 of those public sources every twenty minutes and prints what changed as one running sheet, so you can see the week at a glance instead of opening a dozen tabs.

Each frame on the sheet is one update with a running number, the headline as its source wrote it, where it came from and when. The whole frame links to the original. Updates published by the company that makes the model are boxed in red marker; a red tick marks releases with open weights you can download and run yourself. Come back later and a yellow line shows where you stopped, with everything newer above it.

Comparing AI video models is mostly a question of a few hard facts: how long a clip can be, the resolutions and aspect ratios on offer, whether the model makes sound, whether you can pin the first and last frame with your own images, and what a second of video costs. The AI video model comparison puts those facts side by side for every model in a public catalogue, and the AI image models table does the same for still images, including which models edit photos and how many reference images they accept.

If you would rather run AI video models on your own hardware, the open source video generation models page follows releases from the Wan, HunyuanVideo and LTX-2 repositories and the models trending on Hugging Face, and the video generation papers page collects the research those models grow out of.

Prompting differs from one model to the next, so VidoGrid also keeps prompt guides written from each maker's own documentation: veo 3 prompt guide, wan 2.2 prompts, kling ai prompts, nano banana prompts and flux prompt guide. VidoGrid is not affiliated with any of the makers; every claim on the sheet comes from the source it links to.