Alibaba: Wan 3.0

Alibaba: Wan 3.0 is a video generation model from Alibaba. It belongs to the video category and is identified on api14 as alibaba/wan-3.0. The model is built for turning text and images into video, and it sits alongside Alibaba's other generative work rather than being a general purpose text model.

Wan 3.0 accepts text and image inputs and returns video as its only output. It supports text-to-video, image-to-video, and reference-guided video generation. Output can be produced at 480p, 720p, or 1080p, in aspect ratios of 16:9, 4:3, 1:1, 3:4, and 9:16. Durations range from 2 to 30 seconds. The model can also generate audio alongside the video. For image-driven work, a first_frame image can be supplied. No context length is listed for this model, so plan around the video parameters rather than a token window.

Those capabilities point to a fairly specific set of uses. Text-to-video suits short clips generated from a written description, such as a scene, product shot, or concept preview. Image-to-video with a first_frame is useful when you already have a still and want it to move, which fits storyboard animation, ad variations, or bringing a single frame to life. Reference-guided generation helps when the output needs to stay close to a supplied visual reference. The 2 to 30 second range covers short social formats and brief narrative beats, and the square, vertical, and landscape ratios mean the same model can serve feed, story, and widescreen placements. Audio generation means you do not have to add a separate sound step for clips that need it.

You call Wan 3.0 through api14's OpenAI-compatible API using the model id alibaba/wan-3.0. The request accepts max_tokens, presence_penalty, response_format, seed, temperature, and top_p. Seed is worth setting when you want reproducible output across runs, and temperature and top_p give you control over variation. Because the output is video, response_format is the parameter to check when you need a specific return shape. The video settings themselves, such as resolution, aspect ratio, duration, and the first_frame image, are part of the request you build for this model.

There is no context length listed, and maxOutput is reported as 0, so treat this as a generation endpoint rather than a long-context reasoning model. If your task is text, code, or analysis, a language model will fit better. Wan 3.0 is the right choice when the deliverable is a short video clip built from text, an image, or a reference.