Reframe video

Take a finished clip and place it on a new canvas at a different aspect ratio, then let the model extend the scene to fill the space the original didn't cover. One horizontal master becomes a vertical cut for social or a square cut for a feed, without cropping away the subject, because the model generates the new edges. This is outpaint-style edge generation, distinct from edit-video (which reworks the look and holds the source aspect ratio) and from a plain crop (which throws pixels away).

Inputs to collect

• The source video (URL or upload). This is the clip being reframed.
• The target aspect ratio / canvas (e.g. 16:9 → 9:16, or 1:1). This sets the new width and height.
• What fills the new space. A short description of the surroundings to extend ("sky above and sea below", "more street to the left"). The prompt drives the extension.
• Optional: where the source sits in the new frame, if not centered (bias the new space toward one edge).
• Optional: target resolution tier (360p through 1080p).

Models

• Luma Ray 3.2 (luma:ray@3.2) - the model for this outcome. It supports inputs.video with a new canvas at a different aspect ratio and settings.sourcePosition for placing the source. Confirm it is live and inspect its schema via the runware-models + runware-run skills before calling. Never hardcode a stale choice.
• No general fallback for reframing today. If reframing is unavailable, the alternative is a crop (loses the subject) or frame extension on a same-aspect canvas (see Technique).

Workflow