Can H3 Max Extend a Video? How Continuation Actually Works

H3 Max can extend an existing video, but extension is a separate upstream workflow—not another setting inside the standard text and image generator. fal currently publishes a dedicated minimax/h3-max-turbo/extend-video endpoint. It accepts a source clip, reads the ending, and generates the next 1–15 seconds from a prompt about what happens next.

The H3 Max generator on this website does not expose that endpoint or accept a video upload. It starts from text or still images. That distinction matters: generating a neighboring shot from the final frame can be useful, but it is not true source-video continuation.

Pick the workflow that matches the edit

An extension is most useful when the viewer should believe the camera never stopped: a runner keeps moving through the same street, a product turntable completes another quarter turn, or a boat drifts from shadow into sunlight. If the next moment changes location, angle, or time, a deliberate cut is usually cleaner than forcing one continuous take.

What the Extend endpoint actually does

The current official H3 Max Turbo Extend schema defines a specific input contract:

• the source video must be 1.625–60 seconds, no larger than 50 MB, with an aspect ratio between 0.4 and 2.5;
• the prompt describes the continuation, not a summary of the source;
• the requested duration adds 1–15 seconds of new footage;
• auto aspect ratio follows the source, while choosing another ratio crops the video to fit;
• this dedicated endpoint lists 480P, 768P, 1080P, and 2K output—unlike this site's standard generator, which offers 480P and 768P;
• the original source segment keeps its soundtrack, and optional reference audio can guide sound in the newly generated section.

The endpoint also offers two different outputs:

• Extended returns the source and generated footage together as one video.
• Continuation returns a separate clip that begins with a 1.625-second repeat of the source ending, followed by the new footage.

That repeated section is an editing handle, not extra story time. If you place a continuation file directly after the untouched source, the audience will see the final 1.625 seconds twice, and its audio may duplicate too. Align the repeated frames with the end of the source, find the cleanest seam inside that overlap, and keep only one copy.

Write forward from the final moment

The easiest way to weaken a continuation prompt is to spend half of it retelling a scene that is already visible. Start at the next event instead. Repeat only the details that must survive the seam: subject identity, screen direction, camera movement, lighting, and the most important sound.

Imagine the source ends with a red paper boat leaving the shadow beneath a stone bridge. The camera tracks beside it at water level while rain falls. A continuation prompt could read:

The same red paper boat clears the bridge shadow and enters warm
late-afternoon light. It catches briefly against the wet curb, turns
clockwise once, then continues downstream. Keep the camera at water
level and preserve its forward tracking speed, the steady rain, the
boat's folded-paper shape, and its left-to-right direction.
Audio: continue the rain and water ambience, then add one distant
bicycle bell as the boat enters the light. No cut, no new boat,
no music.

This prompt gives the new section one readable event: clear the bridge, catch, turn, continue. It asks the continuation to keep the camera height and speed instead of introducing a new reveal. The audio instruction asks for the existing ambience to continue, then ties one small cue to the change in light.

Treat it as a writing template and adapt the anchors to the final two seconds of your own source. A locked camera needs different continuity language from a handheld pan; a quiet room needs a different audio bridge from a passing train.

Build the seam before polishing the shot

Use this order when preparing a continuation:

1. Watch the final two seconds several times. Note where the subject sits in frame, which direction it moves, how fast the camera moves, where the light comes from, and what sound is active at the cut.
2. Choose one next beat. A short extension can complete an action or introduce one turn. It should not change the location, reveal three objects, deliver dialogue, and start a chase at once.
3. Write the next action first. Then add only the continuity anchors the model could otherwise reinterpret.
4. Keep the aspect ratio on auto unless a crop is intentional. A new ratio changes the frame at the same moment you are trying to hide a seam.
5. Choose the output for the job. Use the combined extended file when you want one ready sequence. Use continuation when you need control over the edit, remembering that its first 1.625 seconds repeat the source ending.
6. Review picture and sound separately. First watch the seam muted for a jump in position, scale, camera speed, or light. Then listen without watching for doubled words, an ambience reset, or a sudden level change.

Optional reference audio can guide the generated part, but it does not turn the endpoint into a soundtrack editor. The source audio remains part of the extended result. If an exact music edit, approved voice line, or final mix matters, finish that work in an editor after the visual seam is stable. The H3 Max synchronized-audio guide explains how to keep sound instructions tied to visible events.

When a clean cut is the better answer

Continuation is not automatically more cinematic. It is valuable when uninterrupted time carries the idea. A cut is often stronger when:

• the next shot changes from wide to close-up;
• the story jumps forward in time;
• a new subject enters from an unrelated location;
• the camera needs to cross the action line;
• the source already ends on a settled composition.

In those cases, use the ending as context for the edit rather than asking the model to preserve every pixel. Match one or two visual ideas—direction, color, or shape—then let the new shot have its own composition. A cut from the paper boat turning clockwise to a bicycle wheel turning clockwise may feel more intentional than a technically continuous extension.

A practical fallback in this generator

If you want to make the next H3 Max shot here, export a clean still from the end of your source video and use it as the first image. Describe the next action, camera behavior, and sound, generate the new clip, then join the clips in an external editor.

This method passes a still composition, not the source video. The generator cannot read the incoming camera velocity, movement immediately before the frame, occlusion history, or soundtrack timing. Expect to treat the join as an editorial cut or to hide it with a motivated transition. Do not present it as the same result as the dedicated Extend endpoint.

If you know exactly where the new shot should end, add a compatible ending image and follow the first-and-last-frame guide. If the goal is a longer film rather than one continuous take, plan separate panels with the storyboard workflow.

The useful question is not simply whether H3 Max can make a video longer. It is whether the next moment should feel like the same take. Use true extension when the seam must disappear; use a new shot when the cut can do the storytelling.