Alibaba
Wan 2.6 extends clips by chaining its general I2V mode off the source clip's last frame — a budget-friendly approach when Pixverse Extend is overkill or you want creative control over the extension. Drop the source into a frame-extraction tool node, pipe the last frame into Wan 2.6 with a continuation prompt, and the model generates new footage that picks up where the original left off. For high-volume production where credits matter and the extension can absorb a small style shift, Wan is the practical pick.
On Martini's canvas, route the source clip into a frame-extraction tool node. Scrub to the final stable frame — the moment before any motion accent or fade. This frame becomes the seed for Wan 2.6's extension. The cleaner the seed frame, the cleaner the extension joins.
Pipe the extracted frame into a Wan 2.6 image-to-video node. Wan generates 4-8 second clips from a starting frame. The output continues from the seed frame, so when stitched after the source, the cut reads as continuous if the prompt matches the original motion direction.
The prompt should restate the source's shared visual language and add the new motion: "character continues walking forward through autumn forest, soft golden key light, same wardrobe, slow handheld camera." The repetition ("same lighting, same wardrobe") is what holds the extension cleanly to the source.
Wan 2.6 base handles general I2V extension. For tighter character continuity, route through Wan Animate Move (motion reference) — feed the source clip's tail motion as the motion reference and a character anchor as the appearance reference. Animate Move outputs new footage with the same motion vector and the same character. More credits, but tighter control.
On the canvas, send both source and Wan extension into the sequence builder as adjacent clips. The cut should be invisible if the seed frame and the prompt continuity are right. If a small visible jump remains, blend with a 0.5s cross-fade in the NLE export — Wan extensions are typically clean enough that a hard cut works.
Wan's open-weight architecture keeps Sutui costs low. If the first extension does not match cleanly, regenerate at minimal credit cost — try varying the prompt or the seed frame timestamp. For high-volume work (extending 20+ clips for a campaign), Wan beats Pixverse on per-clip cost.
Standard continuation prompt. The "same wardrobe" / "same lighting" repetition holds the extension to the source.
character continues walking forward through autumn forest, soft golden key light, same wardrobe, slow handheld camera, 6 seconds
Camera-led continuation — useful when the source motion is camera-driven rather than subject-driven.
camera continues slow dolly forward through coffee shop interior, warm interior light, ambient cafe sound, 5 seconds
Loop closer — extends a turntable shot to come full circle for a seamless loop.
subject completes the rotation and slowly returns to start position, same studio light, neutral grey backdrop, 4 seconds
Wan 2.6 extends via I2V from the source's last clean frame — use frame extraction first.
Repeat the source's shared visual language in the continuation prompt to anchor continuity.
For tighter character control, switch to Wan Animate Move with both character + motion references.
Wan is open-weight and cheap — iterate freely if the first extension does not stitch cleanly.
For pure motion-preservation with no creative control, Pixverse Extend is tighter; Wan is for budget-conscious extension with prompt control.
Wan 2.6 outputs 4-8 second extensions at 720p-1080p, depending on the chosen tier. Generation runs 90-150 seconds per extension. Open-weight architecture means cheaper Sutui per generation, suitable for high-volume work. The extension joins are slightly looser than Pixverse Extend (which is purpose-built for V2V continuation) but offer prompt control that pure extension models do not. For a campaign extending 20+ clips, Wan's per-clip cost wins; for the cleanest single hero stitch, use Pixverse.
Connect Wan 2.6 with other AI models on Martini's infinite canvas. No GPU required — start free.
Get Started FreeByteDance
Pixverse Extend is the specialist for one job and one job only: take a video clip and seamlessly continue it. It does not generate from text, it does not animate a still image, and it does not edit content — its single purpose is preserving original motion, style, and lighting in the continuation. For an editor who inherits a 5-second AI clip that needs to be 12 seconds, Pixverse drops onto the canvas as a downstream V2V node and extends the footage without re-prompting from scratch.
View guideRunway
Runway Aleph is V2V-only — it operates on existing footage and preserves the original camera and timing exactly. While its primary use is style transfer and reference-driven re-render, Aleph can also be used for extension by pairing the source clip with a continuation reference image and a directional prompt. The output picks up the source's motion vector while honoring the new reference. For an editor who needs a hero shot lengthened with a specific look (a brand seasonal pivot mid-clip), Aleph is the most controllable extend option.
View guide