AI Video Editing vs. AI Video Generation
Same models, different jobs: why editing an existing clip and generating a new one are separate workflows, and how to pick the right entry point.
Both "AI video editing" and "AI video generation" end with a video file, and sometimes the same underlying models power both. They are still different jobs with different inputs, different promises, and different honest descriptions of the output. Picking the wrong one wastes credits and produces footage you cannot use.
The core difference: is there a source clip?
Editing starts from a clip that already exists. You upload it, describe one change, and the model re-renders that clip with the change applied. The output aims to keep the source's duration, framing, and camera move, so the two can be put side by side: "this is the same footage, before and after." The correspondence is close but not pixel-exact — part of checking an edit is watching for drift in what should have stayed unchanged.
Generation starts from a description — or from one or two images. There is no source clip. The output is new footage that did not exist before, and it has no frame-for-frame relationship with anything. It cannot honestly be presented as an "edit" of anything.
That last point matters more than it sounds. Presenting two independently generated clips as a before/after pair is misrepresentation, and presenting a generated clip as "your footage, improved" is the same mistake in softer words.
When to edit
Editing is the right call when the footage itself is the point:
- you shot the clip, or have the rights to it, and want a variant of that exact shot
- the deliverable is a comparison (product color options, background alternatives, style tests)
- you need the original audio preserved
- the clip is short (1–15 seconds) with one clear subject
Typical tasks: changing a color, replacing a background, applying a style. One change per pass; iterate with drafts.
When to generate
Generation fits when the footage does not exist:
- B-roll for a concept that has not been shot
- animating a still image or product photo (image-to-video)
- exploring what a described scene could look like (text-to-video)
- first drafts where "roughly this" beats "exactly that"
Generation does not give you control over a specific shot. If you need "our product, from this angle, with this logo" — that is an edit of footage you already have, not a generation.
What each workflow gives you
| Editing | Generation | |
|---|---|---|
| Input | your clip (MP4, 1–15 s) | text prompt, first/last frames, or reference images |
| Output | re-render of your clip, one described change | new footage |
| Duration/framing | intended to follow the source — verify drift | chosen in the form |
| Audio | source audio can be preserved | generated or absent, model-dependent |
| Comparability | direct before/after comparison | nothing to compare against |
| Failure mode | drift in what should stay unchanged | the scene simply not matching the intent |
Neither mode is pixel-exact. Both are iterative: check the output, adjust one factor, run again.
Credits and mode
Both workflows are billed in credits, quoted before submission. Editing has a draft mode that costs less for prompt testing; generation cost scales with duration and resolution. Failed or cancelled tasks are refunded. See pricing for plans and the studio for live quotes.
One studio, two jobs
This site keeps both in one workspace with one credit balance: the editor for changing clips you have, the generation tools for creating footage you do not. The prompt library covers the editing side; generation prompts are written per scene in the studio.