Is Wan 2.6 NSFW? Not on Alibaba's own API, which checks both prompts and outputs. On our API, five of the six Wan 2.6 endpoints (base, Spicy and Flash) carry no content filter; the one that holds back is base Wan 2.6 reference-to-video.
Short answer: Wan 2.6 has no open weights, so every copy you can use is hosted by someone, and the host decides how strict it is. In our tests the base text and image endpoints, Spicy and Flash delivered nearly everything as written. Pick by input:
- From a still, cheapest: Wan 2.6 Flash. 5 seconds at 720p costs $0.1125, silent.
- From a still, with sound or lip-sync to your own audio: Wan 2.6 Spicy or base Wan 2.6.
- From text only: base Wan 2.6 text-to-video.
- From reference stills or clips: Wan 2.6 Flash reference-to-video, not the base reference endpoint.
Prices checked 21 October 2026. A refused or failed clip costs nothing; a clip that finishes but is toned down is billed at the normal price (refund policy).
Published Wan 2.6 Spicy example, image-to-video, 5 seconds. Not a test output.
What Alibaba actually released
Alibaba released Wan 2.6 on 16 December 2025 through Alibaba Cloud's API and the Wan website. The series added a reference-to-video model, clips up to 15 seconds, multi-shot storytelling and synchronized sound.
Unlike Wan 2.1 and 2.2, it has no open weights. The Wan-AI organization on Hugging Face lists nothing newer than the 2.2 line, so pages claiming that Wan 2.6 is open source are wrong. In practice you can't run it locally and you can't add LoRAs. If LoRAs are what you're after, our Wan 2.2 LoRA guide covers the version that does support them.
Alibaba's API checks both input and output and rejects content it flags as inappropriate. We haven't tested Alibaba's own app, so we can't say where its line sits. What we can say is how the copies we host behave.
The six Wan 2.6 endpoints we host
| Endpoint | Input | Length | Sound | Page label | 720p per second |
|---|---|---|---|---|---|
| Wan 2.6 text-to-video | Text | 5, 10 or 15 s | Yes | Uncensored | $0.09 |
| Wan 2.6 image-to-video | Still | 5, 10 or 15 s | Yes | Uncensored | $0.09 |
| Wan 2.6 reference-to-video | Up to 3 clips, 2 stills | 2 to 10 s | No | Filtered | $0.065 |
| Wan 2.6 Spicy image-to-video | Still | 5, 10 or 15 s | Your own audio drives lip-sync | Uncensored | $0.095 |
| Wan 2.6 Flash image-to-video | Still | 5, 10 or 15 s | Off by default; on doubles the price | Uncensored | $0.0225 |
| Wan 2.6 Flash reference-to-video | Clips and stills, up to 5 references | 2 to 10 s | Off by default; on doubles the price | Uncensored | $0.0225 |
"Uncensored" is our page label for an endpoint where we add no content filter; "Filtered" means requests can be refused. All six accept shot_type (single or multi) and negative_prompt; picking multi switches prompt expansion on.
How far each Wan 2.6 endpoint goes
We test on five intensity levels, from Mild to Maximum, with fictional adults only; the test prompts aren't published. Results come from our September and October 2026 API runs, mostly at 720p and the shortest length each endpoint allows. Each cell is clips delivered as written out of runs. A miss is a toned-down clip unless the cell says refused.
| Endpoint | Mild | Moderate | Strong | Very strong | Maximum |
|---|---|---|---|---|---|
| Wan 2.6 text | not tested | 1/1 | 1/1 | 1/1 | 2/2 |
| Wan 2.6 image | not tested | not tested | 1/1 | 0/1 | 1/1 |
| Wan 2.6 reference (Filtered) | not tested | 0/1 | refused 0/1 | not tested | not tested |
| Wan 2.6 Spicy image | 3/3 | 3/3 | 3/3 | 2/3 | 7/7 |
| Wan 2.6 Flash image, September | 3/3 | 3/3 | 3/3 | 3/3 | 7/7 |
| Wan 2.6 Flash image, since October update | not tested | not tested | 1/1 | 1/2 | 1/1 |
| Wan 2.6 Flash reference | not tested | 2/2 | 1/1 | 1/1 | not tested |
Base endpoints got one or two runs per cell. Read the table as where the line is, not as a pass rate. A toned-down clip finishes and is billed, but the model drops or softens part of what you asked for.
What each endpoint did:
- Wan 2.6 text-to-video delivered every level we ran as written, from Moderate up to Maximum (5 runs). We haven't run Mild on it.
- Wan 2.6 image-to-video delivered Strong and Maximum as written; one Very strong run came back toned down. Small sample: 3 runs.
- Wan 2.6 reference-to-video is the one Wan 2.6 endpoint that holds back. It toned down a Moderate prompt and refused a Strong one. Use it for ordinary casting work.
- Wan 2.6 Spicy delivered Mild through Strong and every Maximum run as written; one of three Very strong runs was toned down (19 runs in all). It can also go further than an ordinary prompt asks, so check every clip before you publish it.
- Wan 2.6 Flash image-to-video delivered every level in 19 September runs. Since its October update, one Very strong run of two was toned down, while Strong and Maximum came back as written. Results from October differ slightly from September, which is why the table splits them.
- Wan 2.6 Flash reference-to-video delivered all 4 runs as written, from Moderate up to Very strong, using ordinary reference stills.
What trips it up
- Say the characters are adults and give an age range, such as 30s. Every prompt in our runs did.
- On the strongest prompts, Flash may leave your opening frame's setting after the first second. If the set matters, describe it again in the prompt.
- Spicy can add things you didn't ask for. Review each clip, including ones from ordinary prompts.
- Multi-shot switches prompt expansion on. That rewrites your prompt before it runs; use single shot when you need control over the wording.
Is Wan 2.6 uncensored? Wan 2.6 Spicy vs Flash vs base
Yes on five of six endpoints, with differences in how reliably they follow a strong prompt:
- Base Wan 2.6 is the only one of the three with a Filtered endpoint (reference-to-video). Its text and image endpoints delivered nearly everything we ran, but with only 8 runs between them, treat that as a first look.
- Spicy delivered 18 of 19 runs across all five levels; the one miss was a toned-down Very strong run. It costs $0.475 for 5 seconds.
- Flash is about a quarter of the price per second, and its reference endpoint takes up to five references without a filter. Since its October update it toned down one of two Very strong runs, so check the strongest prompts.
If every level must come back as written from a still, Wan 2.7 Spicy delivered all five in our tests, at $0.1235 per second at 720p.
Which Wan to pick
| You need | Use | 5 s at 720p |
|---|---|---|
| Cheapest Wan 2.6 clip from a still | Wan 2.6 Flash | $0.1125 |
| A still, with sound or your own audio | Wan 2.6 Spicy | $0.475 |
| Text only, close to your script | Wan 2.6 text-to-video | $0.45 |
| Casting characters from reference images | Wan 2.6 Flash reference-to-video | $0.1125 |
| Every level from a still | Wan 2.7 Spicy | $0.6175 |
| Clips over 15 seconds | Wan 3.0, 2 to 30 s, from $0.045 per second at 480p (duration test) | – |
| Cheapest still-to-video | Wan 2.2 Spicy, from $0.019 per second at 480p; Maximum was mostly toned down (Wan 2.2 test) | – |
Choose 2.6 over the others when you want a multi-shot cut of up to 15 seconds in one run, or a low-cost way to put characters from reference images into a new scene without a filter (Flash reference-to-video). For the older 2.5 build and the full family grid, see our Wan versions test. More options across families are in the best uncensored AI video models roundup.
Price and what you're charged for
Prices checked 21 October 2026.
| Endpoint | 720p per second | 5 s at 720p |
|---|---|---|
| Wan 2.6 text-to-video and image-to-video | $0.09 | $0.45 |
| Wan 2.6 reference-to-video | $0.065 | from $0.13 for 2 s of stills |
| Wan 2.6 Spicy image-to-video | $0.095 | $0.475 |
| Wan 2.6 Flash image-to-video, silent / with sound | $0.0225 / $0.045 | $0.1125 / $0.225 |
| Wan 2.6 Flash reference-to-video, silent / with sound | $0.0225 / $0.045 | from $0.045 for 2 s of stills |
1080p costs 1.5 times the 720p rate. On reference endpoints, seconds of reference footage count toward the bill, up to 5 seconds in total; reference stills are free. Refused or failed jobs cost nothing: the hold goes back to your balance. A toned-down clip completes, so it's billed at the normal price. Details are in the refund policy and the Wan 2.6 reference-to-video guide.
A prompt that works
Keep it to a camera move, a lighting note and one clear action. For example, with Flash image-to-video or reference-to-video:
A woman in a long black coat walks along a rain-soaked harbour quay at night, crane lights reflecting in the puddles. Slow tracking shot, she pauses and looks back at the camera. Cinematic, shallow depth of field. Characters are in their 30s.
Published Wan 2.6 text-to-video example, 5 seconds, 720p. Not a test output. For more on phrasing, see the video prompt guide and our longer clips guide.
FAQ
Is Wan 2.6 NSFW?
Not on Alibaba's own API, which checks inputs and outputs. On ours, five of the six Wan 2.6 endpoints carry no content filter and delivered what we asked in most runs. The sixth, base reference-to-video, is Filtered.
Is Wan 2.6 open source?
No. There are no open weights for Wan 2.6; Hugging Face has Wan 2.1 and 2.2 only. That means no local install and no LoRAs.
What is the difference between Wan 2.6, Wan 2.6 Spicy and Wan 2.6 Flash?
Base Wan 2.6 does text and image input with sound, plus a reference endpoint that is Filtered and silent. Spicy is image-to-video only and accepts your own audio for lip-sync. Flash is the cheapest, silent by default, and takes both stills and reference footage.
Why was my Wan 2.6 reference-to-video request toned down or refused?
It's the only Filtered Wan 2.6 endpoint. For the same job with no content filter, use Flash reference-to-video.
Do I pay for a refused clip?
No. The hold on your balance is released. A clip that completes but is toned down is billed at the normal price.
Can Wan 2.6 make clips longer than 15 seconds?
No. A single run is 5, 10 or 15 seconds (2 to 10 for reference-to-video). For longer clips use Wan 3.0 or see the longer clips guide linked above.
Use fictional adults only (18+). Get consent before using anyone's likeness, and only upload material you have the rights to.
