Confirm Displayed Orientation Before Choosing the Canvas
Open each source in a player and note its displayed orientation and essential content. A phone recording may rely on rotation metadata to appear upright, so encoded width and height alone do not establish what viewers see. The skill reads rotation and pixel aspect information when deriving displayed geometry; the editorial decision still requires looking at the picture.
If the creator specifies portrait delivery, check whether each selected shot can fit that requirement. Otherwise compare which canvas makes the content easiest to read: is the subject central, or must both people at opposite sides remain visible? The orientation of the first file in a list should not decide the whole film.
“Canvas” here means the finished image’s width-to-height ratio. “Padding” is the space left around a fitted source to preserve its whole image. These decisions are separate from selecting one story across several videos.
Compare What Fitting and Cropping Lose
Do not stretch a source into a different aspect ratio. Compare two tradeoffs:
| Approach | Check |
|---|---|
| Fit proportionally with padding | Preserves the source edges, but a landscape shot inside a portrait canvas can become small. Check faces, hand actions, and text already in the image |
| Enlarge to fill and crop | Makes the subject larger but removes side or vertical content. Check whether movement leaves the frame and whether a receiving hand, reacting person, or visible outcome disappears |
A film can use different choices by shot if position and action remain understandable. When a source cannot support the crop, use an available close-up, retain the full image with padding, or reconsider the delivery canvas. Choosing another shot cannot supply a reaction that was never filmed.
This is an editorial comparison. The current skill’s multi-source cut uses proportional scaling and padding. It does not automatically track people or decide a narrative crop for you.
Worked Example: A Handoff Across a Table
This is a fictional editing exercise; nothing has been rendered. The required delivery is portrait. Three clips exist: a landscape two-shot with Chen He on the left and Lin Xia on the right, showing Chen handing a blue-cord folder into Lin’s hands; a portrait close-up of Lin checking its cover after receiving it; and another landscape shot of Lin placing it in a box beside the table. The story must show actual receipt and storage, beyond an attractive cover.
| Clip | Decision for this version |
|---|---|
| Two-person handoff | Compare a centered portrait crop. If either hand or the receipt disappears, retain the whole landscape image with padding. If the fitted image is too small to understand, look for an existing shot showing the receipt. If none exists, reconsider delivery requirements; a cover close-up cannot stand in for the handoff |
| Cover check | Use the portrait close-up to focus on the cover. It shows what was received, rather than proving how the previous handoff happened. Keep subtitles away from content viewers need to identify |
| Placement in the box | Keep the full interval from the folder leaving her hand to entering the box. Starting a reach does not prove storage. If filling the canvas removes the box, fit the whole image or choose an existing valid angle |
Padding makes the people smaller in the first shot; the close-up makes Lin larger in the second. These are framing choices. Check whether viewers can follow the same object and action across the cut, without requiring every person to occupy a fixed screen height. If the cover remains unreadable on a phone, narrow the script’s claim or choose clearer material rather than treating enlarged blur as readable evidence.
Understand the Skill’s Automatic Canvas Choice
In the cited version, multi-source cutting checks sources actually used in the selected clips. Used duration determines the leading landscape, portrait, or square orientation; within that orientation it selects a source with a larger displayed area. Frame-rate buckets are also weighted by used duration. It does not simply take the first listed file or choose portrait for a platform automatically.
Each segment is then scaled proportionally into one canvas with padding, and pixel aspect, frame rate, and pixel format are normalized before concatenation. Compatible parameters make the segments joinable; they do not establish that people, props, or text are legible.
The cut produces the intermediate edited_source.mp4. Narration, subtitles, and final packaging follow later. If its automatic canvas differs from the creator’s required delivery, identify that difference and check whether later framing can remain readable. Do not assume /video-recap exposes an undocumented automatic crop or canvas-lock option.
Review the Sequence at Phone Size
Watch the actual delivery at phone size and normal speed. First check for stretched people or objects, incorrect rotation, and missing content. Then check action across cuts, text in the source, and subtitles. A complete fitted source may still be too small; a filled canvas may still have lost essential evidence.
Replay each cut: does a framing change distract, repeat or omit an action, or alter the understood positions? Review again after adding subtitles, especially padded segments. Subtitles and logos must not obscure the receiving action you preserved. See subtitle line breaks and frame review for subtitle layout.
Write a concrete revision: “In shot one, retain both hands and the folder entering Lin’s hands; in shot three, keep the box visible.” This is easier to implement and check than “make every shot fuller.”
FAQ
Must landscape footage be cropped for a portrait video?
No. Proportional fitting preserves the whole image, while cropping can enlarge the subject but loses edges. Choose by the people, objects, and actions each shot must show, then review at phone size. When suitable material is unavailable, change the selection or reconsider the delivery canvas.
Do it with the skill
In a host with Video Recap Skills installed, invoke /video-recap with landscape and portrait sources you may use. Specify the required delivery canvas and people, props, and actions that must remain visible. First check displayed orientation and compare fitted versus cropped readability by clip, including the current multi-source cut’s automatic canvas choice. Report proposed framing and actual rendered checks separately.
Read the method: Multi-source cut and intermediate output · Display geometry and canvas selection · Proportional fitting, padding, and concatenation · Action evidence and playback review