Is Wan 3.0 uncensored for NSFW video?

Yes. We tested it on two-person romance prompts, including a kiss, that are refused at the content filter on some other vendors' video models; both the neon-alley and boxing-gym runs generated normally on the Wan 3.0 line. Boundaries we actually measured are listed in the limits above.

How long can a Wan 3.0 clip be?

Two to thirty seconds, set with duration_seconds as a whole number. The default is five. Longer clips drift more in identity and geometry, so settle the shot at five seconds first.

Does Wan 3.0 generate audio?

Yes, and it is on by default. Picture and sound come back in one file, already aligned. Set generate_audio to false when you plan to add your own score.

What resolutions does Wan 3.0 support?

480p, 720p and 1080p, defaulting to 720p. If you need 2K or 4K, use Wan 3.0 Pro instead — it publishes a different resolution ladder.

What is the difference between Wan 3.0 and Wan 3.0 Prime?

They publish the same tasks and the same fields. Alibaba describes Prime as the high-speed version of Wan 3.0 with the same capabilities, so what changes is turnaround and per-second price, not the look. Nothing in your client has to change except the model name.

Can I set the first and last frame?

On image-to-video, yes. image_url is the opening frame and last_image_url is optional; supplying both pins where the move starts and where it lands, which is what keeps product geometry readable.

How many reference images does reference-to-video take?

Up to ten images, plus up to five reference videos and five reference audio files. Reference videos are limited to fifteen seconds in total, and so are reference audio files; at least one of the three arrays, a reference file or a web link has to be present, while the prompt is optional.

How do I keep the same character across several shots?

Use reference-to-video with the same face and wardrobe plates in every request, hold one seed and one prompt skeleton, and vary only the action line. Text-to-video cannot do this: without a reference it re-invents the face on each run.

Wan 3.0 API — Uncensored NSFW Video, Ten References

alibaba/wan-3.0/video-extend

Continue a clip with Wan 3.0: the last frame of your video opens a new 2 to 30 second segment, and you get back one file with the original and the continuation joined, at 480p, 720p or 1080p.

From $0.045/s

Run model View API

Input

Video extension

1
Pick the clip to continue
Pick a video and we’ll keep it going from its last frame.
2
What happens next
Pick a clip in step 1 and this step opens.
Model
Prompt
Optional. Leave it empty and the model carries the scene on by itself from your final frame. To steer it, describe what happens next: the action that follows, the camera move and the sound. The opening frame already fixes the subject and the set, so spend the words on motion. at least 1 characters
Resolutionaffects price
Resolution the new segment is generated at: 480p, 720p or 1080p, default 720p. Match your source’s resolution so the seam shows no step in sharpness. The joined MP4 keeps your clip’s frame rate and its framing, with no black bars. To line the seam up, the picture may be trimmed or adjusted by a few percent at the edges. If the model returns a noticeably different shape, the new segment is scaled to fill your frame and centered, cropping what spills over. It sets the price.
Durationaffects price
Length of the new segment: any whole number from 2 to 30, or -1 to let the model choose, in which case 30 seconds are held up front and the charge settles on the length delivered. Defaults to 5. Your source is added on top and is not billed. no lower than 2 · no higher than 30
Generate audio
On by default, so the new segment comes with generated sound. Your source keeps its own sound either way; when one side is silent, the joined file is padded with silence.
Closing frame0/1
Asset URL
Drop files or media links, paste, or browse · 1 file · Max 10 MB · JPG · PNG · WEBP · GIF
0 / 1 files
Optional target end frame, an image URL: the new segment finishes on exactly this image, used as the last frame and not as loose guidance. Use it when the continuation has to land on a fixed composition, such as a product back on its hero angle. The start still comes from the last frame of your clip. Same file limits as an image sent to image-to-video: JPEG, PNG without transparency, BMP or WEBP; 240 to 8000 px per side, aspect ratio no wider than 8:1, 20 MB at most.
Deep thinking
Has the model think through the brief and any reference material before generating. It helps with dense documents, web pages and complex reference images, and makes the run slower. Off by default; it switches on automatically when you pass a reference file or link. It does not change the price.
Prompt expansion
Off by default. When on, the prompt is rewritten before generation, which helps thin briefs and fights you on detailed ones.
Seed
0 to 2147483647. Hold it steady to compare one changed field at a time; drop it for variety. no lower than 0 · no higher than 2147483647

Safety checkerOff

SpicyAPI does not run a safety checker on this model. Your prompt and media go to the model as written, and the model itself decides what comes back.

Checking your account before enabling submission and pricing.

1 required field missing: Input video

Output

Ready for a new run

Your result will appear here.

Configure the inputs, review the estimate, then run the model.

{
  "model": "alibaba/wan-3.0/video-extend",
  "input": {
    "enable_thinking": false,
    "resolution": "720p",
    "duration_seconds": 5,
    "generate_audio": true,
    "enable_prompt_expansion": false,
    "prompt": "Continue the scene naturally, keeping the same subject, camera movement and lighting."
  }
}

Introduction

Wan 3.0 Video extend continues a clip you already have. Send it as video_url; a prompt for what happens next steers the continuation, and without one the model carries the scene on by itself. The final frame of your clip becomes the opening frame of a new Wan 3.0 image-to-video shot of 2 to 30 seconds, and the two are joined into one MP4: your original first, then the continuation, with the repeated seam frame removed. The new segment also comes back as its own file, so you can keep extending or edit the pieces separately.

Key features & highlights

Continuity comes from the last frame. Wardrobe, set, light and where everyone stands carry over because the new segment literally starts on your final frame. Motion does not carry over the same way: the model sees one still, not the whole clip, so describe the movement you want to continue instead of assuming it will keep its speed or direction.

Optionally, choose where it lands. Pass last_image_url and the new segment ends exactly on that image, so a shot can return to its hero angle or arrive at a set composition. Without it the model decides how the segment ends. The start is always the final frame of your clip.

One file back, plus the new piece on its own. The joined MP4 keeps your clip’s frame rate and its framing, with no black bars. To line the seam up, the picture may be trimmed or adjusted by a few percent at the edges. If the model returns a noticeably different shape, the new segment is scaled to fill your frame and centered, cropping what spills over. It is encoded as H.264 MP4 with AAC sound. Sound from your source stays; sound generated for the new segment follows it; a silent side is padded with silence so picture and sound stay in step at the seam. The second asset in the output is the new segment alone, ready for another extend or for an editor.

Everything else is Wan 3.0 image-to-video. The new segment runs on the same routes, with the same content range and the same per-second price as the image-to-video endpoint, and a failed generation is refunded the same way. If the generation succeeds but the join itself cannot be built, the task fails and the charge is refunded.

Request parameters

Request parameters for Wan 3.0 · Video Extend
Parameter	Required	Description
`prompt`	No	Optional. Leave it empty and the model carries the scene on by itself from your final frame. at least 1 characters
`video_url`	Yes	The clip to continue: MP4, MOV or WebM, up to 60 seconds. Its final frame becomes the opening frame of the new segment, so cut it on a clean, sharp frame. at least 1 characters
`last_image_url`	No	Optional target end frame, an image URL: the new segment finishes on exactly this image, used as the last frame and not as loose guidance.
`enable_thinking`	No	Has the model think through the brief and any reference material before generating. It helps with dense documents, web pages and complex reference images, and makes the run slower. Default: `false`
`resolution`	No	Resolution the new segment is generated at: 480p, 720p or 1080p, default 720p. Match your source’s resolution so the seam shows no step in sharpness. `480p` · `720p` · `1080p` · Default: `720p`
`duration_seconds`	No	Length of the new segment: any whole number from 2 to 30, or -1 to let the model choose, in which case 30 seconds are held up front and the charge settles on the length delivered. `-1` · `2–30` · Default: `5`
`generate_audio`	No	On by default, so the new segment comes with generated sound. Your source keeps its own sound either way; when one side is silent, the joined file is padded with silence. Default: `true`
`enable_prompt_expansion`	No	Off by default. When on, the prompt is rewritten before generation, which helps thin briefs and fights you on detailed ones. Default: `false`
`seed`	No	0 to 2147483647. Hold it steady to compare one changed field at a time; drop it for variety. `≥ 0` · `≤ 2147483647`

Pricing

Billed per second of the new segment at the resolution you choose; the source clip you send is not billed.

Pricing
Specification	Price
`480p`	$0.045 / second
`720p`	$0.09 / second
`1080p`	$0.18 / second

Use cases

Lengthen a shot that ended too early: a product turntable that stops before the label faces camera, or a walk that cuts before the subject reaches the door. Extend it by a few seconds and describe only the remaining motion.

Build a longer sequence step by step. Generate a first shot, extend it, then extend the joined result again; each step bills only the seconds it adds. Because every step starts from a single frame, watch each seam before stacking the next one.

Prompt guide & pro tips

Describe the next beat, not the whole scene. The frame already fixes who and where, so spend the prompt on the action that follows, the camera move and the sound. Repeat the camera language your clip ended on, such as a slow push in or a locked shot, so the seam reads as one take. The prompt is optional; with none, the model simply carries on from the frame, which is enough for ambient or loosely staged shots.

Trim before you extend. The last frame is the only thing the model sees, so a blurred, mid-cut or heavily compressed ending carries straight into the continuation. Cut your source on a clean, sharp frame first. Sources of up to 60 seconds are accepted as MP4, MOV or WebM.

Keep the resolution consistent. The resolution only sets how detailed the new segment is generated, so match your source to avoid a visible step in sharpness at the seam. Set duration_seconds to the length you actually need: the charge follows the new segment only.

API quick start

# Generate the idempotency key once; reuse it when retrying after a lost response.
export SPICY_IDEMPOTENCY_KEY="${SPICY_IDEMPOTENCY_KEY:-$(uuidgen)}"
curl -X POST https://api.spicyapi.ai/api/v1/jobs/createTask \
  -H "Authorization: Bearer $SPICY_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: $SPICY_IDEMPOTENCY_KEY" \
  -d '{
  "model": "alibaba/wan-3.0/video-extend",
  "input": {
    "duration_seconds": 5,
    "resolution": "720p",
    "video_url": "https://example.com/reference.mp4"
  }
}'

# Copy data.taskId from the response into TASK_ID, then poll the status.
TASK_ID="TASK_ID_FROM_RESPONSE"
curl "https://api.spicyapi.ai/api/v1/jobs/recordInfo?taskId=$TASK_ID" \
  -H "Authorization: Bearer $SPICY_API_KEY"

See the API reference for authentication, the complete schema, polling, and error handling.

Related models and next tasks

Missing the model you need?

Tell us what to add and where you would use it. Requests go to the people who decide what gets onboarded next.

UNCENSORED VIDEO API

Turn text, a still or ten references into 2 to 30 seconds with sound.

Key numbers

Endpoints: 4; Text-to-video, image-to-video, video-extend and reference-to-video, each with its own contract.
Clip length: 2–30S; Any whole number of seconds in one request; 5 is the default.
Top resolution: 1080P; 480p, 720p or 1080p, chosen per request.
Sound: AUDIO; Generated with the picture in the same pass, on by default.
References: 10+5+5; Up to 10 images plus 5 videos and 5 audio files in one reference call.
Aspect ratios: 5; 16:9, 9:16, 1:1, 4:3 and 3:4; image-to-video can inherit the source frame.

Before you spend

text-to-video

This is where the uncensored range shows: a two-women kiss in a neon alley came out of wording alone, and the same wording is refused at the content filter on some vendors' models. What text alone will not give you is a specific face or garment — without a reference the model invents both, and reruns drift.

image-to-video

The most literal path. Your opening frame pins wardrobe, figure and set so the prompt only has to carry motion. Add last_image_url when the ending matters. Omit aspect_ratio and the clip inherits the source frame's shape; sending a conflicting ratio crops it.

reference-to-video

At least one reference array, a reference file or a web link must be present; the prompt is optional. Ten images, five videos, five audio files; reference videos and reference audio are each capped at fifteen seconds in total. Reference video seconds are billed together with the output, and the two cannot exceed thirty seconds combined. State what each reference controls — identity, wardrobe, location, voice — or the model averages them into a generic subject.

all

Two to thirty seconds, 480p to 1080p. Thirty seconds at 1080p is the slowest thing this family does; find the look at five seconds and 720p first, then scale. You are responsible for consent, rights, age and local law for everyone depicted and for every file you upload. The platform does not screen content. Public pages and examples stay suitable for general viewing.

What it does

Two-person scenes

Kisses other models refuse

The clip on this page is a single five-second text-to-video run: two women under one umbrella in a neon alley, ending on a kiss. The same wording is refused at the content filter on some vendors' models. Wan 3.0 accepted it, held both faces through the dolly-in, and returned rain on the soundtrack in the same pass.

Sound with the picture

The sound lines up with the picture

Audio generation is on by default, so one request returns one file with footsteps landing on the frame they belong to and rain rising as the camera pushes in. There is no second job to run and no alignment pass afterwards. Switch it off only when you intend to score the shot yourself.

Reference control
Ten images, five clips, five voices
Reference-to-video takes up to ten images, five videos and five audio files in one request, so a face, a wardrobe, a location plate and a voice can all enter the same shot. Say in the prompt what each reference is for; an unlabelled stack gets averaged into something that resembles none of them.

Made with Wan 3.0

5s
5s
text-to-video · 5s · 720p · 16:9 · audio generated in the same pass
5s
text-to-video · 5s · 720p · 16:9 · audio generated in the same pass
5s
5s
5s
5s
5s
10s
8s
5s

Where it fits

Uncensored scenes that get refused elsewhere
Couples of any gender, a kiss in the rain, two people sharing one frame. Prompts written this plainly are rejected at the content filter on several competing video models. Wan 3.0 runs them, and your own opening frame is the fastest way to get output that matches your wording.
Vertical cutdowns at volume
Set aspect_ratio to 9:16, keep duration at five or six seconds, and sweep the same shot list across seeds. Because the audio bed comes back inside the file, a clip is postable straight out of the API with no music pass, which is what makes a forty-clip week realistic.
Product films from one packshot
Image-to-video takes the packshot as the opening frame and last_image_url as the closing one, so a bottle can turn, catch a highlight and land back on the hero angle inside five seconds. Label text and silhouette survive because both ends of the move are pinned.
Character-locked episodic work
Feed the same face plate and wardrobe stills into reference-to-video for every shot of an episode, reuse one seed and one prompt skeleton, and change only the action line. The cast stays recognisable across a dozen separately rendered clips.

Model reference

Wan 3.0 API reference

Wan 3.0 API specifications

Producer: Alibaba
Published endpoints: text-to-video, image-to-video, video-extend, reference-to-video
Clip length: 2–30 seconds, default 5
Resolution: 480p, 720p, 1080p — default 720p
Aspect ratios: 16:9, 9:16, 1:1, 4:3, 3:4
Audio: Generated with the picture, on by default
Reference limits: 10 images, 5 videos, 5 audio files; reference videos and reference audio each capped at 15 seconds total
Content: Uncensored; no per-request content declaration

How to call the Wan 3.0 API

1. Pick the endpoint from what you already have
Only a written idea → text-to-video. A frame you want moved → image-to-video. A face, wardrobe, location or voice that has to survive across shots → reference-to-video. Each is a separate contract and the family name does not merge their fields.
2. Write it as subject → action → camera → light and sound
One subject, one action, one camera move, one note each for light and sound. This is the exact text behind the clip on this page, run at duration_seconds 5, resolution 720p, aspect_ratio 16:9 with audio on: “Two adult women in a rain-soaked neon alley at night share one umbrella, shoulders pressed together. One tucks a wet strand of hair behind the other's ear, they look at each other and kiss, the umbrella tilting to half-hide them. Slow dolly-in, pink and green neon bleeding across wet ground, anamorphic flare, 35mm grain, rain on the umbrella.”
3. Submit asynchronously, then poll or take the callback
Send the create call with an Idempotency-Key, keep the task id, and either poll it or register a callback URL. Change one field per rerun and hold the seed — that is the only way to tell whether resolution or wording moved the result.

The short version

Wan 3.0 is Alibaba's mainline video family on this platform, published as independent APIs: text-to-video, image-to-video, video-extend and reference-to-video. Every endpoint takes 2 to 30 seconds at 480p, 720p or 1080p, five aspect ratios, an optional seed and synchronised generated audio; reference-to-video adds up to 10 reference images, 5 reference videos and 5 reference audio files, and image-to-video pins both ends with image_url and last_image_url. Work is created asynchronously and billed per second of delivered video — plus, on reference-to-video, the seconds of any reference video — quoted in US dollars before you submit. Wan 3.0 is uncensored: two-person romance scenes that other vendors' models refuse at the filter generate here, and the boundaries we measured are listed above.