Alibaba Wan Video Watermark on Free Tier: What to Do

May 20, 2026 · Editorial Team · 6 min read · wan-video troubleshooting error-fix

You generated a video with Alibaba's Wan 2.1 model, downloaded it, and found a watermark burned into the lower right corner. If you're using the free tier of Wan's official web interface (wanx.aliyun.com or the international version), this is expected behavior, though Alibaba doesn't make it especially obvious upfront. The watermark is an "Alibaba Cloud" or "Wan AI" identifier that appears on all free-tier outputs and is applied at the render stage, before the video is delivered to you.

Here's what matters: this watermark is baked into the video pixels. You can't cleanly remove it from the existing file without visible artifacts. But you have several legitimate options for getting watermark-free output, including running the open-source model locally (which Alibaba has released for exactly this reason) and using commercial tiers that remove the watermark entirely.

What this error actually means

There's no error here in the technical sense. The watermark is an intentional feature of Alibaba's free tier, applied to distinguish commercial from non-commercial usage and to prevent free-tier output from being used in professional contexts without a paid license.

Wan 2.1 is both a commercial web service (with tiered pricing that removes the watermark) and an open-source model release. The open-source weights are available on Hugging Face and through Wan's GitHub repository. When you run the open-source weights locally, there is no watermark. The watermark exists only in Alibaba's hosted service's free tier.

So the "fix" depends entirely on what you need: quick access without setup (pay for a tier that removes it), or free, watermark-free output with some technical setup (run the model locally).

Quick fix (when you need it working in 60 seconds)

Check the Wan pricing page at wanx.aliyun.com or the international equivalent. The Basic commercial tier removes the watermark and costs significantly less than comparable Western AI video services.
If you're using Wan through Alibaba Cloud's API, confirm your API key is tied to a paid project. Free API keys apply the watermark; paid project API keys do not.
For a one-time need, some third-party video platforms (like Wan integrations on other creative tools) provide their own commercial licensing that removes the watermark. Check the specific platform's terms.
If you're running Wan 2.1 locally via Hugging Face weights, confirm you're loading the correct weights. The Wan-AI/Wan2.1-T2V-14B weights from Hugging Face produce no watermark. Watermarks only come from Alibaba's hosted inference endpoint.
There is no legitimate way to remove the watermark from an already-generated video without degrading quality. Inpainting tools can attempt to fill the area, but results are usually detectable at full resolution.

Why this happens

Alibaba's Wan web service applies the watermark server-side, during the final compositing step before video delivery. It's not added by the model itself (the model has no knowledge of watermarks). It's added by the serving infrastructure that wraps the model on Alibaba's cloud.

This is a standard practice for commercial AI video tools that also release open-source versions: the hosted service adds value through convenience and interface but uses the watermark as the incentive to upgrade to paid tiers. Stable Diffusion's commercial deployments use the same model; Midjourney's free trial uses watermarks; Wan follows the same pattern.

The watermark position (lower right, semi-transparent) is deliberate. It's visible enough to identify the footage as free-tier Wan output but positioned to be less distracting during review. For professional use, the position doesn't matter: it's still there.

The common frustration arises because Wan 2.1's quality is genuinely excellent, and users who encounter the watermark on their first generation may not have known it was coming. Alibaba's international web interface makes the tier difference clearer than the Aliyun Chinese-language version, but neither is prominently upfront about the watermark before generation.

Permanent fix

Option 1: Upgrade to a paid Wan tier.

Go to your account settings on wanx.aliyun.com or the international platform you're using.
Select a commercial plan. The entry commercial tier (as of May 2026) removes the watermark and allows commercial use of generated content.
Existing free-tier generations will still have watermarks. Only new generations under the paid plan will be clean.
If you generated something specific under the free tier that you need watermark-free, regenerate it under the paid plan with the same prompt and settings.

Option 2: Run Wan 2.1 locally (no watermark, no cost per generation).

Install the required dependencies:

conda create -n wan python=3.10 -y
conda activate wan
pip install torch==2.3.1 torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
pip install diffusers transformers accelerate

Download the model from Hugging Face:

huggingface-cli download Wan-AI/Wan2.1-T2V-14B --local-dir ./wan-weights

Run generation using the provided inference script from Wan's GitHub:

python generate.py --model-dir ./wan-weights --prompt "your prompt" --output-dir ./output

The local inference produces clean, watermark-free MP4 files. No account required.
For consumer GPU users with less than 24GB VRAM, use the 1.3B parameter distilled version Wan-AI/Wan2.1-T2V-1.3B instead of the 14B model. The 1.3B runs on 8GB VRAM and produces decent quality at reduced resolution.

Option 3: Use a third-party platform that has a commercial Wan license.

Several video creation platforms (including some listed on Agentbrisk) have licensed Wan's API commercially and offer watermark-free Wan 2.1 outputs through their own subscription tiers. Check the platform's terms to confirm they hold a commercial license rather than running free-tier API calls.

Prevention

If you're doing any professional or commercial work, do not use the free tier for final outputs. Use it for testing prompts and settings, then switch to a paid tier or local install for your actual deliverables.

For local install users, maintain your conda environment carefully and don't update package versions between projects without re-testing. Wan's inference scripts can be sensitive to version changes in diffusers and transformers.

Document your working generation commands. Wan 2.1 local inference has several quality-affecting parameters (guidance scale, number of steps, resolution). Once you find settings that work well for your use case, save them to avoid rediscovering them for each project.

If you're evaluating Wan for a team or agency, request a commercial license quote directly from Alibaba Cloud rather than using the per-generation pricing on the consumer platform. Volume licensing is significantly cheaper for high-volume use cases.

When the fix doesn't work

If you've upgraded to a paid Wan tier and your videos still show a watermark, log out and log back in. The tier change sometimes doesn't propagate to your session immediately. If the problem persists after a new login, contact Wan support: the issue may be that your account change didn't fully process.

For local install users seeing a watermark: you're likely not running the open-source weights. You may be running a third-party wrapper that calls Alibaba's hosted API and passes your generations through the free-tier endpoint. Check whether your inference script is loading local weights or calling an external API.

If you've tried inpainting or AI-based watermark removal and the results look bad (they usually do on video, even with good tools), the only path to clean output is regeneration under a plan that doesn't apply the watermark.