The Wan 2.5 API is Alibaba’s hosted video endpoint for the Wan 2.5 preview models. It makes a 5 or 10 second clip with sound from a text prompt or a first-frame image. Alibaba Cloud Model Studio bills it by the second of finished video: $0.05 at 480p, $0.10 at 720p and $0.15 at 1080p. Wan 2.5 has no open weights, so a hosted API is the only way to run it.
Most pages that rank for this term are host listings, and each one describes its own endpoint. This guide reads the model owner’s pages instead. Every limit and price below comes from Alibaba Cloud Model Studio documentation, read on 4 October 2026.
It is written for a team that makes product video at volume: an agency, an operator who sells clips, or a brand with a few hundred products. DesignerBox publishes this article and sells a different route to the same clips, so read that section with the bias in mind.
Key Takeaways
- The first-party route is Alibaba Cloud Model Studio. The model IDs are wan2.5-t2v-preview for text to video and wan2.5-i2v-preview for image to video.
- The price is per second of finished video. $0.05 at 480p, $0.10 at 720p and $0.15 at 1080p in the Singapore region. A failed request is not billed.
- A clip is 5 or 10 seconds. There is no other length on Wan 2.5.
- Every Wan 2.5 clip has sound. Alibaba’s reference says only the Wan 2.2 and 2.1 models make silent video.
- The video link lasts 24 hours. You must download and store each file yourself.
- Wan 2.5 is not open source. The newest open weights on Alibaba’s Hugging Face account are from the Wan 2.2 family.
- Wan 2.5 is three versions behind. Wan 2.6, Wan 2.7 and Wan 3.0 are all live on the same platform.
What is the Wan 2.5 API?
Wan 2.5 is a video model from Alibaba. Model Studio’s release notes list wan2.5-t2v-preview and wan2.5-i2v-preview on 23 September 2025, with “synchronized audio generation with visuals” and “10-second long video generation” (Model Studio release notes, October 2026).
The API is asynchronous. You create a task, receive a task ID, and ask for the result until the status is “SUCCEEDED”. Alibaba says a text-to-video task takes 1 to 5 minutes (text-to-video API reference, October 2026).
Both model names still end in “preview” a year after release. Alibaba’s reference now files them under a “legacy protocol” that covers Wan 2.6 and earlier.
How much does the Wan 2.5 API cost?
Alibaba bills Wan 2.5 on output only. The formula in its reference is unit price by resolution, multiplied by duration in seconds.
| Resolution | Price a second | 5-second clip | 10-second clip |
|---|---|---|---|
| 480p | $0.05 | $0.25 | $0.50 |
| 720p | $0.10 | $0.50 | $1.00 |
| 1080p | $0.15 | $0.75 | $1.50 |
Source: Model Studio pricing, Singapore region, page updated 3 October 2026. The clip columns are our arithmetic from the per-second price.
Text to video and image to video cost the same. The China (Beijing) region lists slightly lower prices for the same models. Read the row for your own region before you budget.
New accounts get a free quota of 50 seconds for each model. Alibaba states that the quota exists in the Singapore region only and is valid for 90 days. Failed calls “do not incur charges or consume free quota”.
What a 100-product run costs
Here is an illustrative budget, not a measured result. Take 100 products and one 5-second clip each at 1080p. That is 500 seconds, or $75. Most teams run more than one take. At three takes for each product, the same job costs $225 and you keep one clip in three.
The price of one second is the small number. The number of clips you reject decides the bill.
What are the limits of the Wan 2.5 API?
These are the limits Alibaba documents for the two Wan 2.5 models.
| Setting | Wan 2.5 limit |
|---|---|
| Duration | 5 or 10 seconds, default 5 |
| Resolution | 480p, 720p or 1080p, default 1080p |
| Aspect ratios | 16:9, 9:16 and 1:1 at every tier; 4:3 and 3:4 at 720p and 1080p |
| Prompt | Up to 1,500 characters |
| Negative prompt | Up to 500 characters |
| First-frame image | Up to 20 MB, each side from 240 to 8,000 pixels |
| Audio file | WAV or MP3, 3 to 30 seconds, up to 15 MB |
| Output | One MP4 with H.264 encoding for each request |
| Result link | Valid for 24 hours |
| Rate limit | 5 requests a second and 5 tasks in progress |
Sources: Alibaba’s text-to-video and image-to-video API references and its rate limit page, October 2026.
Three of these matter more than they look.
The 24-hour link. Alibaba writes that video URLs “are valid for only 24 hours and then automatically purged”. Your code must copy each file to your own storage. The task ID expires after 24 hours too.
The watermark setting. The parameter adds the text “AI Generated” in the lower-right corner. Alibaba’s API reference lists the default as off. Its user guide says the API adds a watermark when you leave the parameter out. Set it in every request, so the result does not depend on a default.
The seed. A fixed seed “improves reproducibility”, and Alibaba adds that “the same seed does not guarantee identical results”. Plan for takes that differ.
Does Wan 2.5 generate audio?
Yes, always. Alibaba’s reference says the Wan 2.6 and Wan 2.5 series “generate videos with audio by default”, and that only the Wan 2.2 and 2.1 models make silent video.
You have two ways to control the sound. Leave the audio field empty, and the model writes sound that fits the picture. Or pass an audio file URL, and the model builds the clip around it. If the file is longer than the clip, only the first seconds are used. If it is shorter, the rest of the clip is silent.
For a product clip this is a cost to plan for. A silent 5-second loop for a product page still arrives with a soundtrack, and you remove it in the edit. Among the first-frame models, only wan2.6-i2v-flash has an audio switch, and its silent price is half its audio price.
Is Wan 2.5 open source?
No. The Wan-AI account on Hugging Face holds open weights for Wan 2.1 and Wan 2.2, and nothing labeled Wan 2.5 or later (Wan-AI on Hugging Face, October 2026). The newest uploads there are Wan 2.2 Animate models from July and August 2026.
So “Wan 2.5 download” pages cannot give you the model. If you need a model you can run on your own hardware, Wan 2.2 is the newest one Alibaba has published. Our guide to Seedance alternatives covers the other open-weight video models and their licenses.
Where can you call the Wan 2.5 API?
There are two kinds of place: the model owner, and hosts that resell the same model.
Alibaba Cloud Model Studio. This is the first-party route. The pricing page lists Wan 2.5 video in two regions, Singapore and China (Beijing). The model, the endpoint and the API key must be in the same region, because “cross-region calls fail”. New Wan versions appear here first.
Model hosts. fal lists Wan 2.5 text-to-video and image-to-video endpoints at the same three per-second prices as Alibaba (fal.ai, October 2026). Replicate lists Wan 2.5 text-to-video, image-to-video and a fast image-to-video model (replicate.com, October 2026). Other hosts and aggregators list it too. A host gives you one key for many models and its own queue. The model underneath is the same.
Check four things on any host before you build on it:
- Which resolutions and durations the endpoint accepts.
- Whether the audio file input is available.
- How long the result link lasts.
- Which Wan version the endpoint name points to, and whether that can change.
If you want to compare models before you choose a host, Runway alternatives shows how one developer API resells several vendors’ models at each model’s own price.
Wan 2.5 against Wan 2.6, 2.7 and 3.0
Wan 2.5 is no longer the current model. Alibaba’s image-to-video reference says of Wan 2.7: “We recommend this version”.
| Version | On Alibaba’s Model Studio since | Clip length | What changed | Price at 720p and 1080p |
|---|---|---|---|---|
| Wan 2.5 preview | 23 September 2025 | 5 or 10 seconds | Sound with the picture | $0.10 and $0.15 a second |
| Wan 2.6 | 3 December 2025 | 2 to 15 seconds | Multi-shot video, no 480p tier | $0.10 and $0.15 a second |
| Wan 2.7 | 3 April 2026 | 2 to 15 seconds | First and last frame, video continuation, a new request format | $0.10 and $0.15 a second |
| Wan 3.0 | 6 August 2026 | Up to 30 seconds | Reference, editing and several input types in one model | $0.10 and $0.20 a second list price |
Sources: Alibaba’s Model Studio release notes, API references and pricing page, Singapore region, October 2026.
Two details change the budget. Wan 2.6 and Wan 2.7 text to video cost the same per second as Wan 2.5, so the older model is not the cheaper one on Alibaba’s own page. Wan 3.0 bills input video as well as output video, and on 3 October 2026 its standard model carried a limited-time discount on the list price.
Wan 2.7 uses a different request format from Wan 2.6 and earlier. Code written for Wan 2.5 will run on Wan 2.6. It needs changes for Wan 2.7.
Alibaba also retires models. Its policy gives three months of notice for a mainline model and 30 days for a dated snapshot (model decommissioning policy, October 2026). Write the model ID as a setting, never as a fixed value in code.
What does the API leave for you to build?
One call returns one clip. A product video job needs more than that.
- The first frame. Image to video starts from a still. Someone has to make that still for each product, in the same light and crop. Our guide to AI image to video for ecommerce covers what a good first frame needs.
- The queue. With 5 tasks in progress at a time and up to 5 minutes for each task, a few hundred clips take hours. You need retries and a record of each task ID.
- Storage. Every file must be copied before the 24-hour link expires.
- Review. Someone must watch each clip and reject warped products and wrong labels. The AI video prompting guide lists the camera moves that cause most distortion.
- The brand rules. The API has no brand record. Each request carries its own prompt, and nothing links request 1 to request 200.
- The next model. Four Wan versions arrived in under a year. Each change is a migration.
The same list applies to image endpoints. We counted it for stills in GPT Image 2 API pricing. For Google’s and Runway’s video endpoints, see our guide to the AI video generation API.
Product video as a workflow
Anyone can make an AI picture. Making hundreds that still look like your brand is the hard part.
DesignerBox is AI creative production for brands and agencies. Scale your images, ads and video with AI and keep your brand on every piece. You build a workflow once with your brand rules: the first frame, the motion, the cut and the caption. The workflow picks a video model for each step, and you see the cost before each run. Batch runs it over a whole sheet of products. You keep or discard each row, and you re-run one row alone.
That covers the first frame, the queue, the storage, the review and the brand rules without code. The full workflow from the first product photo to the finished ad, in one subscription. The current list of video models is on the models page. Check it before you assume a specific Wan version is there.
Know the limits before you choose this route:
- DesignerBox does not have a public API. It has 68 tools over MCP, so an AI chat such as Claude, ChatGPT or Cursor can run your workflows. If your own software must call a video endpoint, Alibaba’s Model Studio or a model host is the right route.
- You do not get the per-second price. An 8-second clip costs 40 to 560 credits, depending on the model.
- Results are downloaded, or sent with a webhook or an S3 step. Nothing is published to a store for you.
- Plan gates apply. Uploading your own photos and the commercial license start on the Pro plan. AI video, virtual try-on, upscaling, the image editor and the video editor start on the Premium plan. Plans and credits are on the pricing page.
How to decide
Read down the list and stop at the first line that fits.
- Video generation is a feature of your own product. Build on the API. Start with Wan 2.7, as Alibaba recommends, unless you need the 480p tier.
- You already run Wan 2.5 in production. Keep it, set the watermark in every request, and test Wan 2.6 first, because it accepts the same request format at the same price.
- You must run the model on your own hardware. Use Wan 2.2. Wan 2.5 cannot be downloaded.
- You want to test several vendors’ models with one key. Use a model host, and check the four points above.
- You are a brand or an agency with no engineers. Run video as a step in a saved workflow, and review by row.
Whichever route you take, count accepted clips. The price of a generated second tells you what the model costs. The price of an accepted clip tells you what the catalog costs.
Scale content without limits.
Generate content for your products or services with AI, starting today, and keep your brand on every piece.
FAQ
Does Wan 2.5 have an official API?
Yes. Alibaba Cloud Model Studio is the first-party route. The model IDs are wan2.5-t2v-preview and wan2.5-i2v-preview. The API is asynchronous: you create a task and ask for the result with the task ID.
How much does the Wan 2.5 API cost?
In the Singapore region of Alibaba’s Model Studio it costs $0.05 a second at 480p, $0.10 at 720p and $0.15 at 1080p, for text to video and image to video. A 10-second clip at 1080p is $1.50. Only finished video is billed.
Can I use the Wan 2.5 API for free?
Partly. Alibaba lists a free quota of 50 seconds for each Wan 2.5 model. It applies in the Singapore region only and is valid for 90 days. After that, every second is billed.
Is Wan 2.5 open source?
No. Alibaba has not published Wan 2.5 weights. The Wan-AI account on Hugging Face stops at the Wan 2.2 family, which is the newest version you can run yourself.
Is Wan 2.2 or Wan 2.5 better?
They suit different needs. Wan 2.5 adds sound and 10-second clips, and it runs only as a hosted API. Wan 2.2 is silent and fixed at 5 seconds on Alibaba’s Model Studio, and it has open weights you can run on your own hardware.
How long can a Wan 2.5 video be?
5 or 10 seconds. Wan 2.6 accepts any whole number from 2 to 15 seconds, and Alibaba’s release notes say Wan 3.0 makes videos up to 30 seconds long.
Can Wan 2.5 make a silent video?
Not through a setting. Alibaba’s reference says Wan 2.5 makes video with audio by default, and that only Wan 2.2 and 2.1 make silent video. Remove the audio track in your edit if you need a silent clip.
Sources
- Wan text-to-video API reference, parameters, limits and the 24-hour link: Alibaba Cloud Model Studio, October 2026
- Wan image-to-video first frame API reference, Wan 2.1 to 2.6: Alibaba Cloud Model Studio, October 2026
- Wan text-to-video user guide, watermark and billing notes: Alibaba Cloud Model Studio, October 2026
- Model pricing, Wan video rows, page updated 3 October 2026: Alibaba Cloud Model Studio
- Release notes, Wan 2.5 on 23 September 2025 and Wan 3.0 on 6 August 2026: Alibaba Cloud Model Studio, October 2026
- Rate limits for the Wan series: Alibaba Cloud Model Studio, October 2026
- Model decommissioning policy: Alibaba Cloud Model Studio, October 2026
- Wan open weights: Wan-AI on Hugging Face, October 2026
- Wan 2.5 host listings: (fal.ai and replicate.com, October 2026)
- DesignerBox plans and models: (DesignerBox pricing and models pages, October 2026)
Wan 2.5 API prices and limits verified from Alibaba Cloud Model Studio documentation and Hugging Face as of October 2026. Prices change, so check the pricing page for your region. Individual results vary.
