Skip to main content

Wan 2.2 Spicy LoRA video-extend continues a source clip you supply, guided by a prompt, with the same LoRAs attached. Each call adds 5 or 8 seconds at 480p or 720p, and reusing the same weights keeps the look steady across the seam.

From $0.038/s5% below the Spicy list price

Workbench

Input

Video extension
  1. Pick the clip to continue

    Picks up from here

  2. 2

    What happens next

    Required. Describe the next beat only — the subject and the look are already carried forward from the source clip.
    Low-noise LoRAs0/3
    Up to three LoRAs applied only in the low-noise stage, where detail and texture resolve. The right place for line art, grain or a rendering treatment. at most 3 items
    High-noise LoRAs0/3
    Up to three LoRAs applied only in the high-noise stage, where composition and motion are decided. The right place for a camera or movement style. at most 3 items
    Style LoRAs0/3
    Up to three LoRAs, and they belong here as well as on the first pass. Repeat the same URLs and strengths, or the style shifts at the join. Each path can be at most 500 characters in every LoRA list on this endpoint. at most 3 items
    Durationaffects price
    58
    Exactly 5 or 8 seconds of new material. Anything else is rejected rather than rounded.
    affects price
    480p or 720p, default 480p. Match the source clip unless you want the change to be visible at the seam.
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.

Output

Ready for a new run
Example result

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.

These previews show selected examples. Try your own prompt and settings in the Playground.

Continue a clip with the same LoRAs as the first pass

Wan 2.2 Spicy LoRA's video-extend endpoint takes an existing clip plus a prompt for what happens next, and adds 5 or 8 seconds of new footage at 480p or 720p. It accepts the same loras, high_noise_loras and low_noise_loras fields as the image-to-video endpoint on this family, so repeating the exact LoRA set from the first pass gives the render treatment its best chance of holding across the seam.

Open the playground
Continue a clip with the same LoRAs as the first pass
Required input
video_url (source clip) and prompt
Duration
5 or 8 seconds of new material
Resolution
480p or 720p
LoRA slots
Up to 3 each in loras, high_noise_loras, low_noise_loras; 500-character cap per path
Output length
Source footage plus the new seconds, not the new seconds alone
Content policy
Unrestricted, no platform filter layer
Input
A clip, not a still, sets the starting pointvideo_url replaces image_url here: the extension picks up from the source clip's final frame, so there is no separate opening image to supply.
Continuity
The LoRA set has to be repeated, not inheritedloras is a fresh field on every call; nothing carries over automatically from how the source clip was generated. Passing the same URLs and strengths again carries the look into the new segment, though every extension still renders fresh, so check each join before building on it.
Output shape
The response includes the source footageThe returned video starts with the original clip and then continues it, so it runs longer than the seconds requested — trim the source portion before splicing the new footage onto anything else.

What Wan 2.2 Spicy LoRA does

The same three LoRA fields as the first pass

The same three LoRA fields as the first pass

loras, high_noise_loras and low_noise_loras use the same entry format here as on the first pass: a direct file URL plus an optional 0–4 strength, up to three per field. Repeating the first pass's values helps a chained scene read as one piece.

5 or 8 seconds of new footage per call

5 or 8 seconds of new footage per call

duration_seconds governs only the material being added, not the total length of the returned clip, which also carries the source footage from video_url ahead of it in the same file, so the file on disk always runs longer than the number requested.

  • Match the source resolution or upgrade deliberately

    resolution can repeat whatever the source clip was generated at, or step up from 480p to 720p. Either way it is billed as its own row alongside duration_seconds, and each endpoint on this family carries its own rates, so read the quote shown before submitting; the first pass's number does not carry over.

  • seed for comparing continuations, not guaranteeing one

    Fixing video_url, prompt, the LoRA set and seed makes two continuation attempts comparable, though an identical seed still doesn't guarantee identical output, so treat it as a control for testing and expect small differences between two otherwise identical calls.

  • No re-upload for LoRA files already in use

    The hosted URLs from the first pass can be sent again here without re-uploading or restaging anything, since loras only points at wherever each file already lives. The one condition is length: every path on this endpoint must fit within 500 characters, a limit image-to-video does not apply.

Who reaches for Wan 2.2 Spicy LoRA

  • Keep the same trained look past the first beat

    Creators continuing a LoRA-styled clip

    Keep the same trained look past the first beat

    If the opening beat already used a LoRA, this endpoint is where that same file and strength get passed again; leave them out and the style shifts at the join, partway through the sequence you are building.

  • Carry one look through several extensions

    Character work across a longer sequence

    Carry one look through several extensions

    Chaining video-extend calls with the same LoRA set carries a recurring character's visual treatment into each new segment, so the look is not rebuilt from the base model at every hop, even though small rendering shifts between segments remain possible.

  • Compare LoRA sets on the same source clip

    Studios testing continuations side by side

    Compare LoRA sets on the same source clip

    Running the same video_url through this endpoint with two different LoRA combinations costs the same either way, since loras doesn't move the price, which makes it cheap to test which set continues a scene better.

  • Grow a scene past 8 seconds

    Teams building past the single-clip ceiling

    Grow a scene past 8 seconds

    Neither endpoint on this family generates more than 8 seconds in one call. Chaining image-to-video into repeated video-extend calls, each with the same LoRA set, is how a scene grows past that ceiling, one reviewed segment at a time.

How to call the Wan 2.2 Spicy LoRA API

  1. Start from a clip you're satisfied with

    This endpoint continues what's already there; it can't fix a source clip's composition or pacing, only build on top of it, so a flawed opening beat carries its flaws into every extension that follows.

  2. Write the next beat, not a recap

    Describe what happens after the source ends. The subject and look are already carried forward from the footage in video_url, so restating them in the prompt spends words that could be steering the new action.

  3. Repeat the first pass's LoRA set exactly

    Same URLs, same strengths in loras, high_noise_loras and low_noise_loras as the call that produced the source clip; a partial match is enough to shift the render treatment at the seam. Each path is capped at 500 characters here, so a long presigned link may need shortening first.

  4. Review each segment before chaining another

    Every extension is a fresh generation, so small drifts in a character's look or motion can compound across several hops. Check continuity and hands before feeding one continuation into the next call.

Wan 2.2 Spicy LoRA worked examples

  • Extending an opening beat into a second action

    Take the 8-second output of an image-to-video call and continue it with a new prompt, the same LoRA set, and the same 720p resolution.

    The character exhales, rises from the chair, and crosses toward the window; camera follows with a slow lateral pan, light shifts from lamp-warm to blue dusk
  • A cheaper 480p pass before a final render

    Continue at 480p and 5 seconds to check pacing, then resubmit the same video_url and LoRA set at 720p once the beat reads right.

    The character turns back toward camera mid-stride, pausing for a beat, expression softening
  • Chaining two extensions from one opening frame

    Feed the output of a first video-extend call back into a second one with the same LoRA set, building an 8-plus-8 second sequence from a single starting still.

    She settles into the armchair, drawing a knee up, gaze drifting toward the fireplace as the amber light dims another shade

Wan 2.2 Spicy LoRA API questions

Treat image-to-video and video-extend as one pipeline: the first produces an opening beat, the second carries it forward. Repeat the exact loras, high_noise_loras and low_noise_loras values from the first pass on every continuation, and review each segment before chaining another — drift compounds quietly, and a small shift two calls back is harder to trace once a third has layered on top.

All Wan models22

Video models18