Finish a cut-off gesture
The shot stops in the middle of a movement: describe the end of the gesture, add two or three seconds, and the edit gives a complete shot. It is the most common use, and the cheapest.
Extend a video with AI: the continuation of a shot
A five-second shot that stops too early: Grok Imagine generates the continuation from its last frame, in the same format, and the Montage node chains the two. Two possible continuations of the same shot. The whole workflow is visible without an account.
Interactive demo
Real workflow, real results. Pan and zoom freely.
Use this workflowThe real workflow: the starting shot in a Media node, two continuation instructions in two Text nodes, two Video nodes set to Grok Imagine (extend), and a Montage node that chains the shot and its continuation.
At a glance
- What the workflow produces
- 2 videos of 10s in 720p, 3:4
- Canvas structure
- 7 nodes wired by 6 connections
- Models wired in
- Grok Imagine (prolonger)
- Cost of one full run
- about 60 credits, roughly $0.72
A generated shot lasts five seconds, and the gesture is not finished: the character has not reached the camera, the gaze has not had time to settle. Rather than regenerating ten seconds in one go, you extend. The continuation starts from the last frame of the shot, keeps its format, its light and its outfit, and chains with no visible cut.
The Video node set to Grok Imagine (extend) receives the video through the video port and the instruction through the prompt port: what happens next, not what is already there. You choose one to ten seconds to add, and the price follows that number, per second. The result is the added segment alone: the Montage node glues the two, for free.
The demo extends the outfit walk shot in two ways: the model stops and turns to the camera, or keeps going and exits the frame. The same shot, two endings, and the edit of the first one already assembled.
How it works
- 1
Upload or wire the shot to extend
A video of two to eight seconds, uploaded into a Media node or coming straight out of another Video node. The demo starts from a shot generated on the canvas.
- 2
Write what happens next
An action, a movement, a reaction: “she stops, turns to the camera and smiles”. No need to describe the scene again, the model sees it.
- 3
Set the added duration and generate
From one to ten seconds. The price follows the slider and shows before you click. The continuation comes out in the format of the shot, measured on the file.
- 4
Chain in the Montage node
The shot then its continuation, back to back: the cut lands on the last frame, with no fade. Export, or extend the continuation again.
Variations and use cases
The same canvas, tuned for different needs: each variation is two or three lines of prompt away.
The long shot at a low price
Generate five seconds on a premium model for the quality, then extend here, second by second: it is one of the cheapest ways to get a consistent ten or fifteen second shot.
Two endings for the same shot
That is the demo: the same source, two instructions, two continuations. Compare, keep the one that serves the edit, and the original shot was not regenerated.
The reaction after the action
A product is set down, a door opens, a dish arrives: extend on the character's reaction. The cut on the last frame avoids the break a second shot would have imposed.
The frame exit to chain
Have the character exit the frame in the continuation: the next shot, generated separately, then chains naturally in the Montage node.
Extend a shot filmed on a phone
The model also extends a real two to eight second video: the continuation of a take, in its frame and light, to reach the length of a reel.
Common mistakes
Describing the whole scene again
The model sees the shot. An instruction that repeats the set and the outfit dilutes the action: say what changes, in one or two sentences.
A source too short or too long
Under two seconds or over eight: the node refuses before generating, with no charge. Cut first in the Montage node.
Expecting the whole video as output
The continuation comes out alone. If you export it as is, the beginning is missing: go through the Montage node, it is already wired in the template.
Changing the setting in the instruction
The continuation starts from the last frame: asking for another room or another outfit forces a break. To change place, generate another shot.
Recommended models
Grok Imagine (prolonger)
Wan 3.0
Seedance 2.5Frequently asked questions
Does the continuation contain the original shot?
No, only the added seconds. The Montage node assembles the shot and its continuation into a single file, at no cost, and the cut is clean since the continuation starts on the last frame.
Can I extend several times?
Yes: a continuation is a video like any other. Wire it into a new Grok Imagine (extend) node, as long as it lasts between two and eight seconds.
Why is there no resolution choice?
Because the model honors none: it renders at a definition close to the source. Imaginode never shows a setting a model ignores.
Is sound generated?
Yes, the continuation comes out with its soundtrack. For continuous music over the whole thing, lay it in the Montage node over both clips.
How much does an extension cost?
A price per added second, from one to ten, shown on the button as soon as the video is wired and the duration set. A technical failure is refunded.
Is the format of the continuation the same as the shot?
Yes: the node measures the source video and imposes the closest ratio. Vertical, square or wide, the continuation chains with no black bar.
Can I extend a shot with a real face?
Yes, the continuation keeps the person of the shot. For a faithful cut, stay within the action and the light of the original shot.
What is the difference with an end image on another model?
An end image drives a shot toward a precise image. Extension continues an existing shot from its last frame, with no target image: it is the tool to lengthen, not to direct.
Extend a video with AI: the continuation of a shot
Use this workflowFree account, no card required. The workflow opens pre-filled in your canvas.