Choose your music video input by deciding what the next shot must preserve. Use image-to-video when you have an opening composition worth animating. Use reference images when you need visual guidance but want a new composition. Consider video editing when you already have useful movement and need a specific change.
Those are different jobs. Uploading the same portrait into three different controls can produce three different kinds of guidance. Before spending time on another prompt, check what the selected input actually does.
This guide uses an invented lighthouse sequence to show how to make that decision. The examples are planning exercises, not tested presets or promises of model behavior.
Choose a music video input by the detail you need to keep
Write one sentence beginning with “Keep…” before selecting a mode. The answer should name something visible, such as the opening silhouette or the timing of a hand movement.
| What you already have | What needs to stay | A useful starting point | What to review afterward |
|---|---|---|---|
| An approved storyboard still | Opening framing and object placement | Image-to-video | Changes during movement |
| Portraits, clothing, or location images | Appearance across a newly composed shot | Reference images | Identity, outfit, and interpretation of each reference |
| A usable moving clip | Existing action with a requested visual change | Video editing | The entire clip, including apparently unchanged regions |
| A clip with unwanted sound | The picture | Mute or replace its audio in your editor | The final music mix |
| A good shot placed too early | The shot itself | Move or trim it on the timeline | The incoming and outgoing cuts |
The last two rows matter: a generation request is unnecessary when the required change is an ordinary timeline edit.
Provider terminology and supported combinations vary. For a concrete example, xAI's video generation documentation distinguishes a starting image from reference-image guidance and lists separate video-editing and extension routes. Check the documentation for the model and interface you actually use; a similar button label does not establish identical behavior.
Start from a still when the opening frame is already right
Imagine a verse ending with a lighthouse keeper standing beside a circular window. You have approved the still: the keeper is on the left, the sea is visible on the right, and the lower part of the frame has room for lyrics.
That composition is the reason to choose image-to-video. Prepare the crop before uploading it, then describe a small action that fits the image:
The keeper slowly turns toward the window as reflected light passes across the wall.
The camera stays stationary. Keep the circular window and the keeper's opening position.
Use restrained movement and leave the lower wall visually quiet.
This is an illustrative motion brief, not a model-specific syntax. Set duration and other supported controls in the interface rather than assuming that mentioning them in prose configures the request.
Inspect the returned opening, middle, and ending. A correct first frame does not prove that the window keeps its shape or the coat stays consistent once the subject moves. If a hand or face needs a large turn, simplify the action before adding more descriptive adjectives.
Our music video camera movement guide can help you choose a restrained camera instruction that serves the passage.
Use references when the next shot needs a new composition
Now the chorus moves outside. The keeper walks toward the lighthouse, and the camera sees a wider view. Reusing the interior still as an opening frame would start the shot in the wrong place.
Instead, assemble guidance for the new view. Give each image a clear role:
- A portrait identifies the keeper's appearance.
- A clothing image clarifies the dark raincoat and its collar.
- An exterior view establishes the lighthouse entrance.
Remove images that contradict the intended shot. Two different coat colors create a decision the generator has to resolve. A close portrait with no useful costume information may be enough for appearance, but it cannot explain the architecture behind the subject.
An accompanying brief could read:
Create a wide exterior view of the lighthouse entrance using the supplied location reference.
The keeper wears the dark raincoat shown in the clothing reference.
They approach the doorway from the right side of the frame.
Keep the sea visible behind them and use an overcast evening atmosphere.
Verify which uploads are treated as references and which are pinned frames. Some systems support combinations, but that is a capability to check, not something to infer from the number of upload slots. For a provider-specific continuity example, see our Grok reference-image guide.
Edit an existing clip when its movement is worth preserving
Suppose the approach to the doorway works, but a distracting red sign appears on the wall. The clip already fits the chorus. A focused video-editing request may be worth evaluating before rebuilding the action:
Remove the red sign beside the doorway and continue the surrounding stone texture.
Keep the keeper's walk, the doorway position, and the camera movement consistent with the source.
Supply the actual clip through a supported editing input. xAI documents this as a separate video-editing workflow; writing “edit this video” in a text-only generation request does not attach a source video.
Keep the original file. Preservation instructions are requests, so compare the revised clip against the source across its full duration. Watch the walking pace, coat, door edges, shadows, and final pose. If removing the sign introduces a jump at the cut, the revision may be less useful than the original.
When the problem is simply a late arrival at the doorway, try trimming or shifting the clip first. For instructions that refer to moments inside a source clip, our timestamp planning guide explains how clip time differs from song time.
Review the result where it belongs in the song
Place the candidate between its neighboring shots and play the finished recording. Check whether the keeper's screen direction makes sense, whether the chorus has enough visual change, and whether the lyrics remain readable over the moving background.
Keep the recording separate from any sound delivered with a generated clip. Listen to the exported sequence rather than assuming that an apparently silent preview proves the final mix is correct. A singing close-up also needs a separate lip-sync review; matching a portrait does not establish matching mouth movements.
For a video centered on your recording and readable lyrics, LyricVideoMaker offers an audio-upload lyric-video workflow. The external generation and editing modes discussed here are not a claim that every provider control is built into the product.
Questions to settle before another request
Can I use the same picture as both a reference and an opening frame?
Possibly, if your chosen system supports those roles together. Decide whether the picture's exact composition matters or only its appearance guidance, then verify the relevant control. Do not assume the roles are interchangeable.
Will video editing leave everything else unchanged?
Do not rely on that. Preserve the source and compare the full output, especially the action and the frames at either end. Accept the revision only after it works in the sequence.
What should I save with an approved shot?
Keep the source assets, selected input mode, prompt, accepted clip, and intended song placement. Those notes make a later correction easier to reproduce.
For your next music video input, choose one visible requirement to protect. Animate an approved opening, guide a new composition with references, or repair an existing clip. Then judge the result at the actual cut before expanding the sequence.