Wan2.1-FLF2V-14B-720P

Wan2.1-FLF2V-14B-720P, released on 17 April 2025 under Apache 2.0, is the first-last-frame member of the Wan 2.1 family and the most controllable of them. The other image-to-video models are given a starting picture and left to decide where the shot goes. This one is given both ends - the frame it must open on and the frame it must land on - and has to construct a plausible path between them at 1280x720. That is the form of the problem film and animation work actually poses. A storyboard fixes where a shot begins and ends; the labour is the movement in between. Constraining both ends also removes the failure that makes open image-to-video hard to use in sequence, namely that each clip drifts somewhere the next clip cannot pick up from. The model shares the generation's backbone - a 40-layer diffusion transformer of width 5120 with 40 heads, the umt5-xxl text encoder capped at 512 tokens, the CLIP vision encoder and the Wan autoencoder - and its weight files count 16.4 billion parameters against the 14 billion in its name, the surplus again being the conditioning path. Only 720p is supported; there is no standard-definition build. The caveat is stated by Alibaba itself and deserves to be repeated rather than buried. The first-last-frame model was trained primarily on Chinese text-video pairs, and the maker recommends writing prompts in Chinese for better results. Every other model in the family is presented as bilingual without qualification; this is the one place where the documentation admits that the two languages are not equally served. Usage reflects its narrower job: around 1,800 downloads a month, a fraction of the plain image-to-video builds, against 228 bookmarks - the interest is disproportionate to the traffic. As with the 720p image-to-video sibling, the LICENSE.txt file the model card links to is not present in this repository.

▸News

Alibaba sells its open video models as bilingual — on one of them, its own card recommends writing in Chinese

9/4/2026 · Wan-AI model card, Hugging Face

▸Videos

No videos yet.

▸Reviews