# Crop Desk > A 9:16 CROP OF A 16:9 FRAME KEEPS 31.6% OF THE PICTURE. Delivering one film in several shapes is > not the same film several times - each shape is a different rectangle cut out of the same frame, > and how much survives is one division: `area kept = min(a,b) / max(a,b)`. 16:9 into 1:1 keeps > 56.3%, into 4:5 keeps 45%, into 9:16 keeps 31.6% - and the 68% that goes was framed months > earlier by somebody who was not thinking about the vertical cut. Five lanes over one master frame > and the shapes it is cut into, plus a free browser-side engine doing exact rectangle geometry. URL: https://crop-desk.skillsafe.ai/ API: https://crop-desk.skillsafe.ai/api.html Derived from kostja94/marketing-skills (https://github.com/kostja94/marketing-skills), MIT: its `video-optimization` skill covers planning video for short-form and long-form and optimising it for mobile viewing. Not affiliated with or endorsed by kostja94. ## The one division everything follows from ```text area kept = min(a, b) / max(a, b) a, b the two aspect ratios ``` 16:9 into 9:16 keeps 31.6%. Into 1:1, 56.3%. Into 4:5, 45%. Into 2.39:1, 74.4%. WHICH AXIS GOES DEPENDS ONLY ON WHICH RATIO IS LARGER. A delivery narrower than the master takes the sides. A delivery WIDER than the master - 2.39:1, 21:9 - takes the top and bottom instead, and every horizontal instinct about the frame is irrelevant to it. Say which axis before saying anything about position. ## Subjects and the crop window The window can move. A window of width `w` has its centre somewhere in `[w/2, 1 - w/2]` - it cannot leave the frame. A subject spanning `[x - s/2, x + s/2]` is contained when the window's edges are outside the subject's, which reduces to an INTERVAL of permitted centres. That interval can be empty two different ways and THE DIFFERENCE DECIDES WHAT YOU DO: - THE SUBJECT IS WIDER THAN THE WINDOW. No position holds it. This is not a reframe done badly - there is no reframe. Accept a partial subject, cut to different footage, or re-shoot. - THE SUBJECT IS AGAINST THE EDGE OF THE FRAME. A window would hold it if it could go further, and it cannot. The subject survives only jammed to one side of the delivery. FOR TWO SUBJECTS THE QUESTION COLLAPSES TO ONE COMPARISON: is the span from the leftmost edge to the rightmost edge wider than the window? If it is, no crop holds them both - and that is a decision about the CUT rather than the export. Two singles, or a different take. No amount of care in the reframe session produces a position that does not exist. A CENTRED CROP IS WHAT AN AUTOMATIC REFRAME PRODUCES, and it is what a batch export gives everybody by default. A subject that survives only when somebody moves the window is a subject the batch will lose. ## The geometry does not change with resolution. The pixels do. A 9:16 crop keeps 31.6% of the width whatever the master is. 31.6% of a 4K master's 3840 pixels is 1215, which is more than a 1080-wide delivery needs. 31.6% of an HD master's 1920 is 608, and the same delivery is then a 1.78x UPSCALE. This is the fault that is invisible at every step. The export succeeds, the file is exactly the right dimensions, nothing reports a problem, and the picture is soft in a way that is easy to blame on the codec. The only thing that was ever going to catch it is a division done before the shoot. THE PUNCH-IN BUDGET is `delivery width / master width`, and it is what every later decision spends - a push-in, a stabilise, a horizon correction. Somebody spends it in week three without knowing it went in week one. RESOLUTIONS ARE THE SHORT EDGE. `1080p` is 1920x1080 landscape, 1080x1920 vertical and 1080x1080 square, because the number people say is the short side in all three. ## Safe areas Surviving the crop is necessary and NOT SUFFICIENT. Text that survives and then lands in the outer tenth of the delivery frame is still wrong: that is where captions, interface chrome, progress bars and rounded corners live. The trap is that safe areas are measured in the DELIVERY frame, not the master - so a lower third laid out comfortably inside 16:9 title safe can survive a 9:16 crop and still be against the edge of it, because the frame it is measured against got three times narrower. Graphics laid out for the master are the single most reliable casualty of a vertical cut. ## What this page cannot do Everything here is geometry over numbers you supply. THIS PAGE HAS NOT SEEN THE PICTURE. It does not know what is in the frame, where the attention goes, or whether the thing at 0.3 is the subject or a lamp. So never say a crop looks good or bad: `31.6%` is what survives, not "unusable"; `1.78x` is the scale factor, not "soft as mud". And never invent a position - a subject with no coordinates is unplaced, not centred. SUBJECTS ARE HORIZONTAL ONLY. A delivery wider than the master crops the HEIGHT, and this page holds no vertical positions, so for those shapes it reports the geometry and says plainly that it cannot judge survival. Naming the gap is the honest move; answering anyway would be inventing data. ## The sheet ```text FRAME source | 16:9 shapes | 9:16, 1:1, 4:5 master | 4k delivery | 1080p SUBJECTS s01 | 0.42 | 0.18 | | the presenter s02 | 0.15 | 0.24 | text | the lower third ``` A subject row is `id | centre | width | kind | description`. Centre and width are fractions of the FRAME WIDTH. A percentage works too. `kind` is `text` or blank. A field that is missing stays missing - a subject with no position is in no crop test and the page says so. ## The lanes - `plan` - says how to frame so the crop is a choice later rather than a rescue - `check` - says what every shape takes out of the frame and which finding changes the job - `subjects` - the interval of window positions per subject, and who can share one - `resolution` - the punch-in budget in pixels, and what upscales - `deliver` - sorts findings into reframe, re-cut and do not make, which severity does not decide ## What it does not do It crops nothing and renders nothing. Every figure is geometry over numbers you supply, so it is only as good as the positions you measure.