Every-format delivery

Every-format delivery

One finished landscape edit becomes three deliverables: the 16:9 master, a 9:16 vertical and a 1:1 square (4:5 on request). Same cut, words and graphics, each composed for its own frame: the picture fills the frame and follows the subject, titles sit where that frame has room, and type is sized for how that frame is watched. A small 16:9 picture floating in a band is the failure this skill exists to prevent: never letterbox, never fill the rest with a blur or gradient.

Changing a copy's canvas centre-crops the footage (whoever isn't in the middle is lost) and leaves every overlay at its old pixel size and percentage position, so most fall off the frame or render distorted. This skill is the pass that fixes that. It is not a Short: every copy keeps the whole video (use tella-short to pick the best 30 s).

Read the master, then ask once

Read the master's timeline with layouts, overlays, text overlays, settings and words, a filmstrip of the whole thing with subtitles, and a full-width frame of each graphic. Find where the subject is over time: get_camera_faces for a camera clip; for uploaded footage it returns nothing, so read it off a filmstrip. Note the footage's own camera moves; the reframe should pan with them, not fight them.

Then one ask_form, prefilled: formats (9:16 1080×1920 and 1:1 1080×1080 checked, 4:5 1080×1350 offered), a sign-off board (yes), export when done (yes, captions burned in or not). Never edit the master: duplicate_video first, then set the copy's canvas with update_video. A filmstrip of each fresh copy is the breakage list; name each break ("speaker cut off at 2.5 s", "headline off the right edge").

Frames and safe zones

Reframe the footage to fill and follow

• Camera clips: update_clip keepCameraFaceCentered on each clip of the copy keeps the face centred as it moves (cameraFocusPoint when the camera has no face detections).
• Uploaded footage: crop sections. Each is a fullscreen cover layout whose crop leaves a full-height window in the target aspect, centred on the subject and off film borders or black bars, with transitionStyle: "spring" so Tella animates the change as a pan: update_layout the base, then add_layout per section. Start a new section only when the subject drifts more than ~15 % of the window's width, about 0.3 s before they arrive; merge sections under ~0.9 s and tiny moves. Calm beats busy: in the example the 9:16 needed five sections over 11.6 s, the square one. Soft 4× archive footage on 9:16 is the footage, not a reason to letterbox.
• Screen recordings: don't crop UI to a sliver. Screen and camera on 9:16 goes side-by-side style: even (position: bottom) or camera-bubble; screen-only follows the action with crop sections and must stay readable at phone size.

The look

When a graphic can't survive the new frame (a wide card, a corner callout that would land on the face), re-set it in this system rather than shrinking it. Masters made with this skill are built in it from the start.

• Palette. Paper #F1E9DA, ink #15120F, accent #E2512C, muted #9C9184.
• Edge tabs. A headline is one snug tab per line, flush to one margin, not a card: a paper rect with ink Archivo 800 at −0.01 em. Each tab is its words plus 0.3 × size each side, 1.14 × size tall, 0.06 × size apart. Above them a kicker tab in the accent with ink Roboto Mono 500 caps at +0.14 em: its words plus 0.55 × kicker size each side, 1.75 × kicker size tall, 0.18 × size above the first line. Ragged edges are the point: the shape is the sentence. A line that wraps in a narrow format just becomes another tab.
• Series bug. An accent square, the series name in Roboto Mono 600 and the issue number in Roboto Mono 400, on one line in a top corner, held the whole video, with a soft shadow (y 2, blur 10, opacity 0.6) because footage turns bright behind it. On 9:16 the head owns the top, so the bug turns 90° and runs down the right margin from about y 240.
• Captions (update_video): backdrop style, #15120FE6 behind paper text, spoken word #FF8A5B, Archivo 700, one line per block.

Words follow tella-motion-design: one to four of the speaker's words per line, at most two lines plus a kicker.

Sizes and placement per format

Type scales up as the frame narrows, because a phone shows the frame smaller. If a line won't fit, break it in two; never shrink the type.

Always choose the spot from frames of that copy: the empty part of the frame, clear of the whole person, inside the safe zones.

Motion and timing

Each tab wipes open from its anchor edge in 0.5 s on one ease, bezier (0.16, 1, 0.3, 1): kicker first, each line 0.07 s after the last. Its words trail by a few frames and slide 0.4 × size into place behind the wipe. On exit the tabs collapse toward the far edge with easeInCubic over the last ~0.4 s, taking their words. No fades, no bounce, never an empty tab. The bug's square scales in, then its words slide off it (about 0.3 s each, 0.12 s apart) and hold to the end. A headline enters just before its first spoken word, and keeps the master's start and duration in every format.

Building it in Tella Motion

Draw every graphic on a full-frame document (overlay at 0,0, dimensions = the canvas) with all its nodes in one group, gfx, so a new format is canvas.set to the new size, delete gfx, and rebuild at that format's sizes and positions. Each tab is a rect, its title, and a hidden copy of the rect the title is matted to. Check each rect hugs its words in a full-width get_motion_frame.

Other graphics: a Tella Motion one gets canvas.set, then its nodes moved and resized, side-by-side parts stacked top to bottom on 9:16; one whose shape is the frame's (a full-frame title card, a diagram) gets its own layout per format, never a scaled-down 16:9 one. Text overlays get a new point and a bigger fontSize on vertical. Image and video overlays can't reflow: re-place them keeping their aspect, or ask whether to drop one from a format.

Gotchas

• The copy inherits the master's per-layout crop, which cover then crops again (a 16:9 headroom crop on 4:3 footage became a 196 px sliver in 9:16). Always replace it.
• apply_video_edits rejects crop on layouts: reframe sections go through single add_layout / update_layout calls, one at a time. After moving a section's time, list_layouts again; it can rewrite the base layout's crop.
• An overlay whose document canvas changed renders mirrored or stretched, even after resizing the overlay. Remove the overlay, duplicate_motion_document, and add the duplicate as a new overlay. A removed overlay keeps its document claimed, so remove and add in separate batches.
• Wipe a tab by keying its rect geometry (field:geometry/rect/rect/x and …/width), not scale, which is uniform and squashes the type.
• Matte the words to the hidden copy of their tab; matting them to the visible tab hides the tab. Deleting gfx on a document that doesn't have it fails the whole batch.
• Captions re-chunk per format width: read them in frames, don't assume the master's blocks.
• Library footage placed with create_video had no transcript after several minutes (so no captions); the same file through create_source + upload_clip was transcribed within a minute.

Check every format with frames

For each copy, filmstrips with subtitles (tiles 300 wide for 9:16, 400 for 1:1) at each graphic's entry, middle and exit and at 250 ms steps through every pan: nothing clipped or in a safe zone, nothing over the face, the head in frame through each pan, type readable at phone size, the tabs opening and collapsing with their words, kicker first. Fix, then re-check that range.

Sign-off board

A 1920×1080 Tella Motion clip on the ink ground: each format's preview (get_video_preview with subtitles, uploaded with create_source) plays side by side at equal height (512 px for 16:9 + 9:16 + 1:1, 64 px margins, 40 px gutters), with the format in Archivo 800 and its size in muted mono under each, a short title and series line above, and one accent playhead under all of them. The first panel keeps its sound; mute the rest. addMedia fits each video to the canvas height and centres it, so its scale is panel height ÷ 1080 and x/y are offsets from the centre. Name it "<master name> — all formats" and check a mid-way frame: the same moment in every panel.

Export and report

When agreed, export each copy (and the master if it has no current export), with captions burned in if asked, and wait for every export. Report one row per format: link, reframe sections and their times, each graphic re-set (old → new placement and size), caption position, anything dropped. End with the board's link.