Home › Photo to Video › Edit Video
How to edit a video by describing the change
Upload a clip, write what should change, and the AI re-renders the video with that change applied — restyle the whole scene, or alter one element in it while the rest holds. In the Photo to Video app for iPhone this is Edit Video mode. Clips run 3 to 15 seconds at 720p or 1080p, and you can add reference images to steer the result.
Steps
- Open Edit Video. In the app, go to Modes and choose Edit Video.
- Add the clip to edit. Pick a video from your camera roll.
- Optionally add reference images. Useful when the change is a look you can show but not easily describe.
- Describe the change. Be specific about what changes — and what must stay.
- Generate and save. The result is a new clip; the original is untouched.
What it supports
| Input | One video, plus optional reference images |
|---|---|
| Output | Video, 720p or 1080p |
| Length | 3–15 seconds |
| Aspect ratio | Follows the source clip |
| Models | Happy Horse |
| Prompt | Required — it is the edit instruction |
| Platform | iPhone (iOS). No Android or web version. |
| Pricing | Credits per second, and this mode is the most expensive per second in the app — keep edits short. |
Which model to use
This mode runs on Happy Horse only, so there is no model decision. What does need a decision is length: video-to-video editing bills for the clip going in as well as the clip coming out, which makes it the priciest mode per second here. Trim before you edit rather than after.
Getting a better result
- Trim to the seconds you actually need first. Cost scales directly with length, and this mode is the most expensive per second in the app.
- State what must not change. “Keep the subject and framing, change the season to winter” holds together far better than “make it wintry”.
- One change per pass. Stacked instructions get applied partially.
- Test the wording on a 3-second cut before committing to the full clip.
What people use it for
Restyling footage to match a look
The most reliable use of this mode: change the overall grade, era or medium of a clip rather than one element inside it.
Prompt restyle this clip as 1970s 16mm film, warm grain, softer contrast
Changing the season or time of day
A whole-scene change the model applies consistently across frames, which is where video-to-video beats editing stills one by one.
Prompt change the scene to winter: bare trees, snow on the ground, overcast light; keep the subject and camera move
Replacing a background you cannot reshoot
Say explicitly what holds still. Without that, the subject gets reinterpreted along with the background.
Prompt keep the subject and their movement exactly; replace the background with a quiet city street at night
Fixing one distracting element
Possible, but the least reliable use — the model works the whole frame, so a local change can pull the rest with it. Trim to the shortest cut containing the problem.
Prompt remove the bright sign in the top right; keep everything else identical
Where it struggles
This is the most expensive mode per second in the app, because video-to-video bills the clip going in as well as the one coming out — trim before you edit, not after. Localised changes are unreliable next to whole-scene ones, and the maximum length is 15 seconds.
Questions
Can AI edit an existing video from a text instruction?
Yes. Edit Video re-renders your clip with the described change applied, and leaves the original untouched.
How long a clip can it edit?
3 to 15 seconds, at 720p or 1080p.
Why is this mode more expensive?
Video-to-video editing bills for the input clip as well as the output, so cost scales with length in both directions. Trim before editing.
Can I show it a style instead of describing one?
Yes — add reference images alongside the clip.
Does it change the whole video or just part?
Either. You can restyle the entire scene or ask for one element to change while the rest stays.