You generated a good AI video clip in 16:9. Now the social team wants it vertical, and the paid team wants a square. The reflex is to regenerate the clip in each format and hope the result matches. That costs credits, and it rarely gives you the same shot twice.
You can often reframe AI video to 9:16 without regenerating it. The tools generate the missing edges around the frame you already approved, so the subject, the motion and the timing stay as they are. It works well on some shots and badly on others. This post explains which is which.
What does reframing actually do?
A plain crop cuts away part of the picture. Going from 16:9 to 9:16 by cropping throws away most of the frame, and the subject may end up half out of shot.
Reframing with outpainting works the other way. The tool keeps your original pixels and invents new content above and below, or to the sides, to fill the new shape. Several video tools now offer this: a guide on HackerNoon names Runway's Expand Video, Luma's Reframe and Higgsfield's Reframe as examples. Vendors describe their own tools in glowing terms, so treat any quality claim as a claim until you have tested it on your own footage.
The value is simple: reframe when the subject is right but the frame is wrong. Regenerate when the subject itself has to change.
Which direction is easier: wide to tall, or tall to wide?
Wide to narrow is easier. Making a 16:9 clip into 9:16 by cropping keeps the center and discards the sides. Making a 9:16 clip into 16:9 means inventing a lot of new picture on both sides, which raises the risk of visible seams and odd geometry.
So plan the delivery formats before you generate. If a campaign needs 16:9, 9:16 and 1:1, generate the master with the subject centered and with some safe room around it, then reframe from there. It is the same logic as shooting wide on set and cropping in later.
Where does reframing break?
We see the same failure points again and again, and the same ones are described in public guides:
- Edge content. Generated background can shift geometry enough to break an eyeline or a movement toward something in frame.
- Static, centered shots. If nothing important sat near the edges, outpainting adds background with little benefit, and a simple crop may do.
- Long clips. Most tools cap input length at around a minute, so longer clips are split, and every split is a continuity risk.
- Hidden inconsistencies. A wall detail that does not fit the room, or a light that changes at the seam, can slip past a quick look.
- Dialogue. Audio is not changed, but a new background can make a performance stop matching what is on screen.
- Resolution. Reframing changes the canvas, not the sharpness. Upscaling is a separate step.
Reframing is usually cheaper than regenerating, but that advantage disappears if it takes several rounds of fixes.
How should you review a reframed clip?
Watch it three times. First at full speed on a phone, since that is where a 9:16 version will live. Second, frame by frame at the seam lines, where the original ends and the invented area begins. Third, with the sound on, to check that the action and the audio still agree.
Look for flicker in the generated area, for a background that moves differently from the original, for hands or props that cross the seam, and for product edges. If a product is in the shot, check the label and the logo against the real pack. A reframe can quietly redraw a corner of the bottle.
This review is where post-production earns its keep. We often fix what the tool leaves behind with masks, tracked patches and grading, rather than going back to the generator. Our article on how to fix AI video artifacts goes through the common problems and the repair route for each.
What should the brief say?
If you are commissioning a clip that has to work in several formats, put this in the brief:
- The delivery ratios: for example 16:9, 9:16 and 1:1.
- The primary format, the one the clip is generated in.
- A safe area where the subject, the product and any text must stay.
- Which shots may be reframed and which must be regenerated.
- How many review rounds are included.
When is regenerating the better choice?
Regenerate when the problem is the content, not the frame. If the subject's pose is wrong for a vertical, if the product sits too low to survive a tall crop, or if a gesture only works in the wide version, no amount of edge painting will fix it. The same is true when a clip needs a different camera move for each format. A slow push-in on a wide shot is a different shot in a tall frame.
A useful rule from our own work: if you can fix it with a mask, a patch or a grade, reframe and repair. If you would need to change what the subject does, go back to the generator and ask for it in the new format from the start, using the approved frame as a reference image.
Practical takeaway
Decide your formats first, generate wide, keep the subject in the center and reframe only shots that are already approved. Review each version on its own, on the device where it will be watched, and expect to repair a few frames by hand.
If you want help turning one AI clip into a full set of formats, you can see how we work at 35milimetre.com.