“Wan AI model” usually refers to Alibaba’s Wan family of video-generation models, but it does not identify one version or one API. Wan 2.1 and Wan 2.2 are distinct releases; the model itself is also separate from a provider that hosts it. Choose by the task you need—such as text-to-video, image-to-video, or video editing—then confirm that your chosen API provider supports that exact version and workflow. Alibaba’s Wan repositories document model capabilities; a hosted API’s endpoint, parameters, and price are provider-specific. (github.com)
What is the Wan AI model?
Wan is a family of video-generation models associated with Alibaba. The official Wan 2.1 repository describes tasks including text-to-video, image-to-video, video editing, text-to-image, and video-to-audio. The Wan 2.2 repository describes it as an upgrade and highlights a 5-billion-parameter model that supports text-to-video and image-to-video at 720p and 24 frames per second. These are model-repository descriptions—not a promise that every hosted API exposes every task or setting. (github.com)
This distinction matters in production: a model release may have downloadable weights and inference code, while a platform API may expose only selected versions, input types, output settings, or controls. Check the actual provider’s documentation before planning an integration.
Which Wan version should you choose?
Start with the output and workflow requirements, rather than an unverified claim that one version is universally “best.” The official Wan 2.1 and Wan 2.2 repositories describe different capabilities; a separate hosted listing may describe additional versions, but that does not establish that they are available from Alibaba directly or through your intended API.
| If you need… | What the available documentation supports | Selection guidance |
|---|---|---|
| Text-to-video or image-to-video | Both the Wan 2.1 repository and the Wan 2.2 repository describe these tasks. | Compare the exact version and parameters exposed by your intended host; repository support alone does not confirm an API route. (github.com) |
| Editing or other video tasks | The Wan 2.1 repository lists video editing and video-to-audio among its tasks. | Verify that the hosted version you plan to call exposes the specific workflow you need. (github.com) |
| 720p output at 24 fps | The Wan 2.2 repository describes its 5B model as supporting text-to-video and image-to-video at 720p, 24 fps. | Treat this as a model-repository capability; confirm resolution, frame rate, and limits on the selected service. (github.com) |
| First- or last-frame guidance, or particular aspect ratios | A Layer listing for Wan 2.1 describes reference images, first- and last-frame inputs, and square, portrait, and landscape formats. | These are Layer-specific product details, not a universal Wan specification. Ask the provider for the exact accepted values and output limits. |
| Native audio or longer clips | Layer’s Wan 2.1 page describes Wan 3.0 as supporting native audio and clips up to 30 seconds. | This is a third-party listing, not independently verified here. Confirm version availability and specifications with the provider before choosing on that basis. |
The Wan 2.1 official repository also labels its repository license Apache 2.0. If you plan to self-host or use model files commercially, review the applicable license for the specific code and weights you download; a third-party API has its own service and usage terms. (github.com)
How to verify a Wan production API
Before integrating, confirm each item against the documentation for the provider you will actually use:
- Exact model and version: Confirm the catalog name and model identifier. Do not assume “Wan” means the same version across providers.
- Supported task and inputs: Check text-to-video, image-to-video, frame guidance, editing, and any reference inputs you require.
- Output constraints: Verify resolution, aspect ratio, duration, frame rate, audio support, file formats, and maximum input sizes.
- Request and authentication: Confirm the endpoint, credential type, request fields, and how input files are supplied.
- Job completion: Determine whether requests return a result immediately or an inference/job ID to poll; check response fields, errors, timeouts, and polling guidance.
- Operational fit: Ask about rate limits, concurrency, account or region access, data handling, and any stated availability commitments.
- Cost: Confirm the billing unit and current price for that route. If the provider uses credits, ask how credits convert to money and what output settings affect usage.
Example: a hosted Wan 2.1 route is not a universal Wan API
Layer’s Wan 2.1 listing shows a provider-specific flow: submit a prompt with bearer personal-access-token authentication, receive an inference ID, and poll a GET endpoint until completion. The listing says completed responses include asset URLs. It also lists a prompt-only cURL example and additional parameters such as aspect ratio, quality, duration, and first- or last-frame guidance. This is an illustration of Layer’s interface only; it is not an Alibaba or TokenForge endpoint, and the available listing does not provide a literal response payload or polling interval.
For that reason, do not copy a model ID, endpoint, or request schema from one host into another. Use the selected provider’s current reference for the runnable implementation.
TokenForge Cloud availability and pricing
The Token Forge Cloud homepage currently describes production APIs for Qwen, DeepSeek, GLM, and Seedance. The research available for this article did not confirm a Wan listing or a Wan API specification there. That is not proof that Wan can never be offered; it means Wan availability, model ID, endpoint, authentication, limits, and price should be treated as unverified until TokenForge Cloud’s current documentation or service team confirms them. (tokenforgecloud.com)
A third-party Layer listing gives Wan 2.1’s price as 4 Creative Units (CUs) per megapixel-second. That is a Layer-specific rate, not a currency amount and not a TokenForge Cloud or Alibaba API price. The available listing does not provide a CU-to-currency conversion or enough billing detail to calculate a monetary cost per clip. Request the applicable billing unit and price from the provider whose route you plan to use.
Practical takeaway
For general video-generation orientation, Wan 2.1’s repository lists multiple tasks, while Wan 2.2’s repository describes 720p, 24-fps text-to-video and image-to-video support for its 5B model. Use those details to narrow your evaluation—not as proof of a hosted API’s exact behavior. Before production, test the chosen version against representative prompts and inputs, and verify its model ID, request/response flow, operational limits, and price with the provider. Wan support on TokenForge Cloud is not confirmed by the documentation found for this guide.
Sources
Wan-Video/Wan2.1 official repository