Wan Video β€” hosted API access

This repository is not the model weights β€” it documents running Wan Video through a hosted API on apimodels.app, a multi-model gateway where one key and one OpenAI-compatible endpoint reach 170+ image, video, audio and language models.

  • Original model & weights: https://huggingface.co/Wan-AI (the open Wan2.1 / Wan2.2 releases)
  • Authors: the Wan team (Alibaba Tongyi)
  • License: Apache-2.0 (the open Wan2.1 / Wan2.2 weights)

Independent hosted-API integration. Not affiliated with or endorsed by the authors; does not redistribute their weights. All credit for the model is theirs.

Which version the API serves

apimodels.app serves Wan 3.0 as wan-3.0-video, a newer version of the Wan family offered as a hosted model. We found no Wan 3.0 weights in the Wan-AI organization on Hugging Face (checked November 2026); the weights link above points to the open Wan2.1 / Wan2.2 releases.

What it does

One model name covers four ways to drive it:

  • Text to video β€” a prompt alone.
  • First frame / first + last frame β€” first_frame_url (+ last_frame_url).
  • Omni reference β€” up to 10 reference images, 5 reference video clips (15 s total) and 5 audio clips (15 s total).
  • Document or web page to video β€” file_url (pptx, docx, xlsx, pdf, txt, md and more) or link_url.

Frame mode and reference mode cannot be mixed in one request. Output is 480P, 720P or 1080P, 2 to 30 seconds as one continuous shot at 30 fps, with an audio track. Two speed tiers share the same capabilities: mode: "standard" (default) and mode: "prime" (faster; on the same 480P 5-second job we measured 58 seconds on prime versus 356 seconds on standard).

Wan Video β€” hosted API access sample

A sample produced through the apimodels.app API.

Use it via API

On apimodels.app the model is served as wan-3.0-video. Get a key at apimodels.app, then:

# text-to-video (async): create, then poll
curl -X POST https://api.apimodels.app/v1/video/generations \
  -H "Authorization: Bearer YOUR_API_KEY" -H "Content-Type: application/json" \
  -d '{ "model": "wan-3.0-video", "prompt": "aerial shot over ocean waves at sunset", "resolution": "720p", "aspect_ratio": "16:9", "duration": 5 }'
# -> { "data": { "taskId": "..." } }
curl "https://api.apimodels.app/v1/video/generations?task_id=TASK_ID" \
  -H "Authorization: Bearer YOUR_API_KEY"   # state -> completed, resultUrls[]

Defaults: resolution 720P when omitted, duration 5 (2–30, or -1 to let the model choose), ratio / aspect_ratio adaptive (16:9, 4:3, 1:1, 3:4, 9:16 also accepted), audio true. Optional: mode, prompt_extend, seed, watermark (default false), callback_url.

Pricing

Billed per second:

Resolution standard prime
480P $0.045 / second $0.068 / second
720P $0.09 / second $0.14 / second
1080P $0.18 / second $0.28 / second

A 5-second 720P clip on standard is $0.45. Without a reference video you pay for output seconds only; reference images, audio, documents and web pages cost nothing extra. With a reference video (video edit, extend, character swap) you pay for input video seconds plus output seconds, and the two together must stay within 30 seconds. With duration: -1 the 30-second worst case is held and the difference is refunded once the real length is known. Turning the audio track off does not change the price.

Failed jobs are not charged. Result files are kept for 30 days; download anything you want to keep.

Links

More browser demos, a dataset and API guides: apimodels.app API demos collection

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support

Collection including stimi/wan-video-api