Make videos longer with prompt-guided AI. Continue the shot and add 5 to 15 seconds of new footage.
Real output: 5.1-second source plus a 10-second continuation. Prompt: “Camera crash-zooms out fast, then tilts and rises to the roof, pointing down at the boxer KO on the ring floor as the camera starts rotating. Shot on ARRI Alexa.”
The source limits, output behavior, and price are clear before generation.
This guide highlights five ways to continue a clip. Choose by the added duration, available output resolution, and whether the continuation can include generated audio. The signed-in tool can show additional compatible models, and this table makes no quality ranking.
Model controls checked in Melies on 26 August 2026.
| Model | How it extends | Added time | Resolution | Continuation audio | Choose when |
|---|---|---|---|---|---|
| Veo 3.1 | Native continuation | 7 seconds | 720p | Available | You want one exact 7-second pass with native video extension and optional generated audio. |
| HappyHorse 1.0 | Guided by the final frame | 5, 10, or 15 seconds | 720p or 1080p | Not available | You need the longest selectable single pass or a 1080p option. |
| Kling O3 Pro | Guided by the final frame | 5 or 10 seconds | No visible choice | Available | You want a 5- or 10-second continuation with generated audio available. |
| Kling O3 Standard | Guided by the final frame | 5 or 10 seconds | No visible choice | Available | You want a 5- or 10-second continuation with generated audio available. |
| Luma Ray 3.2 | Guided by the final frame | 5 or 10 seconds | 540p, 720p, or 1080p | Not available | You want to choose among three visible resolution options. |
The join is checked, and failed attempts are regenerated or refunded. That check does not guarantee consistent motion, subject, lighting, or style throughout the generated footage; results vary by source clip and model.
The exact credit cost appears in the tool before generation. Model availability and controls can change as providers update their endpoints.
Drop your clip
MP4, WebM, MOV, or MKV
Upload a supported MP4, WebM, MOV, or MKV file up to 120 seconds or 200MB. You can also right click a clip in the AI Video Generator and select "Extend video".
Optionally describe what happens next: "camera pulls back" or "she turns around." Leave it blank to request a continuation without added direction; results still vary by source and model.
The AI generates new frames that continue your clip, and every result passes an automatic continuity check before you get it. Extend again to keep going.
Melies uses an uploaded source clip to generate new footage after it, with five highlighted routes and visible controls for duration, resolution, and continuation audio.
Veo receives the full source clip. Other routes receive its extracted final frame. Every result is checked frame-to-frame at the join. If a continuation drifts from your footage, we regenerate once and refund automatically if it still fails.
A model can follow camera and subject motion into the generated portion, but consistency varies with the source, prompt, and selected model.
Describe how the video should continue. 'Camera pulls back,' 'character walks forward,' or 'scene transitions to sunset.'
Choose Veo 3.1 native extension or a workflow guided by the final frame with HappyHorse 1.0, Kling O3 Pro, Kling O3 Standard, or Luma Ray 3.2.
Extended videos integrate with your Melies timeline. Build scenes by extending, chaining, and composing AI-generated clips.
Extending video is different from stretching footage. It creates new frames that continue the shot, giving editors more usable material.
AI video extension works best when the last frames give the model clean visual context. Use it to add breathing room, finish motion, or create a natural transition without looping the same frames.
Clips with readable motion, steady lighting, and a clear subject extend more smoothly than hard cuts or heavy motion blur.
Veo 3.1 uses the uploaded source video. Other models use an image extracted from its final frame. End the source on readable motion you want continued: a camera move mid-push, an actor mid-gesture, or weather mid-drift.
Short prompts like "camera slowly pushes in" or "the character turns toward the window" help steer the generated continuation.
Generate a few extra seconds, review the result, then extend again. Iterating keeps continuity stronger than asking for one long jump.
A short direction changes everything: the hero demo was made with the first prompt below. Name the camera move, the action, and the mood, then let the model continue the shot.
Used in the demo above
“Camera crash-zooms out fast, then tilts and rises to the roof, pointing down at the boxer KO on the ring floor as the camera starts rotating. Shot on ARRI Alexa.”
Cinematic reveal
“Continue the shot as the camera slowly pulls back, revealing the full city skyline while keeping the same blue-hour lighting.”
Character motion
“The character finishes turning toward the door, pauses for a beat, and the warm hallway light spills across their face.”
Product shot
“Extend the close-up with a slow clockwise camera move, keeping the product centered and leaving clean space for a logo.”
Nature and B-roll
“Let the wave continue rolling toward the beach, preserving the same handheld camera motion and soft sunset color.”
From short AI generations to film scenes, video extension adds new footage after a supported clip that ends too early.
Add 5 to 15 seconds to an AI-generated clip in each pass. Turn short generations into usable scene lengths.
Build longer scenes by chaining extensions. Start with a shot, extend the motion, and compose a complete sequence.
Extend short B-roll clips into usable lengths. More footage from less source material.
Create slow reveal shots by extending a zoom or pan to build tension and visual interest.
Extend and loop video clips for backgrounds, ambience, and digital displays.
Extend shots that ended too early. Additional footage for editing flexibility and scene pacing.
Extending is not the same as generating a new clip from a text prompt. Here is the pipeline your video goes through, and why the result continues your shot instead of replacing it.
The tool reads the resolution, frame rate, and duration of your upload, and extracts the final frames. If the file cannot be read or exceeds the limits (120 seconds, 200MB), you find out before any credits are charged.
Veo 3.1 receives your full video through its native extension route. HappyHorse, Kling O3, and Luma Ray receive an image extracted from the final frame, then generate a continuation from that image.
Before you see the result, the continuation is compared frame-to-frame against your source. A continuation that drifts is regenerated with stronger conditioning, and refunded if it drifts again.
The final output includes your original clip and preserves its audio. For routes guided by the final frame, Melies appends the generated continuation. Veo can return a complete extended clip, so Melies checks the duration before deciding whether concatenation is needed.
Last real frame
First generated frameIf a generation fails, drifts from your clip, or comes back shorter than promised, credits are refunded automatically and the reason is shown on the asset.
The exact output duration and credit cost appear before every generation. No surprises after the fact.
Prompts and clips are screened before credits move, so safety-filter rejections almost always cost nothing.
Uploads are private to your account and are never used to train AI models. Extended videos have no watermark and are yours to use commercially.
Practical answers before you start creating.
Upload an MP4, WebM, MOV, or MKV clip within the 120-second and 200MB limits, describe the continuation, and review the model-specific result.
