Manifesto type sequence

Manifesto type sequence

A statement as a poster that builds itself: heavy capitals in one justified column, the first word of each line flush left and the last flush right, and in the space between them a strip of moving footage, cut exactly to the height of the capitals. The words arrive one at a time, then the strips open line by line and the sequence holds. It works as a 7–10 s Tella Motion clip (an opener, a closing statement) or as full-frame b-roll over the stretch where the speaker says the line. Load tella-motion-design first. For spoken words over the footage itself use tella-kinetic-type; for a plain title card, tella-title-cards.

Ask once

Work out yourself, without asking: the statement (the speaker's own lines from get_transcript, or the user's message), the line breaks, where it goes, the brand colours from the video's presets, and which footage fits the strips (the video's own b-roll and screen recordings, attachments on the message). Then ask one ask_form, every field prefilled, so the user can just submit:

• If the lines are missing (no transcript, nothing in the message), ask with ask_question: "What should the manifesto say? 3–5 short lines, one line per row, e.g. DO WORK / THAT WILL / LIVE ON", suggesting up to three versions from the video's title.
• Strip media can't go in a form. One clip or photo per line. Look first at the message's attachments, then the video's own clips (a source placed with addMedia sourceId; list_library scope video lists its media), then the media library. If the user picked "Attached" and none are there, ask_question: "Attach 3–5 wide, calm clips or photos (a skyline, a landscape, hands at work, a screen scrolling). Each shows as a thin strip about 105 px tall, so texture matters more than detail."
• Fallbacks: Tella's archive film (list_library scope vintage with a word from the line: "city", "ocean", "train", "clouds"), then generate_video with a wide, slow, low-detail prompt.
• A talking head makes a poor strip (a slice of a face). Use it only if the user asks.
• If the user skips the form, build with the prefilled values.

Timing

• To the spoken words: each word starts 0.1 s before the transcript's start time for it. Keep at least 0.25 s between words; if the speaker rushes, spread them evenly across the phrase.
• Even beats: the first word at 1.0 s, then one every 0.35 s (the example: six words from 1.0 to 2.75 s).
• Music beat: get_music_beats, one word a beat, starting on the first real beat (curated tracks can open with 1–2 s of intro).
• The strips open after the last word: the first strip 0.3 s after it, the next ones 0.12 s apart, top to bottom (example: 3.05, 3.17, 3.29 s). On speech, start them on the last word's end instead.
• Hold the finished poster at least 2.5 s after the last strip is fully open (the example holds 4.6 s, to 8.5 s).

The look

Values from the example (1920 × 1080, 30 fps, 8.5 s):

• Type: Archivo 800, 150 px, capitals, line height 1, letter spacing 0. The accent colour goes on one word only, set in that colour from the start (no recolour).
• Column: x 210 to 1710 (1500 px wide). Every word is a title whose box is the whole column (x 210, width 1500): the first word aligned left, the last aligned right. In a three-word line the middle word joins the left side: its own left-aligned title, its x the first word's right edge plus about 30 px.
• Lines: box tops 170 px apart, centred on the frame: top = 455 + (i − (n − 1) / 2) × 170. Three lines are 285, 455, 625; four are 200, 370, 540, 710; five are 115 to 795.
• Width: at 150 px a capital is about 100 px wide (W and M about 125, I about 45). The words of a line need at most about 1240 px, which leaves a strip of at least 200 px. When a line is longer, shorten the words first; else drop the size for all lines and scale every number here with it (pitch 1.133 × size, strip offset 0.113 × size, strip height 0.70 × size).
• Strip: as tall as the capitals, from the box top + 17 px, 105 px tall. Left edge 30 px past the left word's ink, right edge 30 px before the right word's ink. Example strips: (469, 302, 723 × 105), (654, 472, 649 × 105), (591, 642, 855 × 105).
• Footage: fit: "cover", desaturated with a luminance colorMatrix (0.2126, 0.7152, 0.0722 on each colour row), no grade, playing from its start.

Motion

• Words: opacity 0 → 1 over 0.18 s, easeOutCubic. Nothing slides or scales; the column stays still while it fills.
• Strips: a rect mask, x fixed at the strip's left edge, width keyed 1 → full over 0.6 s with the curve (0.16, 1, 0.3, 1), so each strip wipes open left to right from the left word.
• No exit in its own clip (the cut ends it). As b-roll, the cut back to the speaker is the exit.

Building it

1. create_motion_document (or create_clip type: "motion" with the length), canvas.set duration. As b-roll, place the document through add_layout over the stretch. A b-roll layout belongs to one clip (clipId, times on that clip's timeline with cuts applied) and can't run into the next clip: keep the sequence inside the clip where the words are said, shortening it or starting it earlier so it ends by the clip's end. Only when the spoken lines truly run across a cut, split there (untested): one document and one add_layout per clip, each transitionStyle: "hardCut", the second document starting on the frame the first ended on, keys on each clip's own clock. Then list_layouts and make the layout after each time-ranged b-roll a hard cut too.
2. One batch of node.insert: the field (a full-frame roundedRect shape with zero corners), then one text node per word (content.type: "paragraph", position, width: 1500, align, style as above) with common.start on its time and common.opacity keyed {time: 0, value: 0, easeOutCubic} → {time: 0.18 s, value: 1} (times in ticks). The Archivo 800 font asset needs asset.set (hostId: "tella.font.archivo.regular", weight 800) first.
3. Measure the gaps; never guess. get_motion_frame after the last word, startTimeMs, width: 1920, format: "png"; download it and in each line's band (box top + 17 to + 122) mark the columns where any pixel differs from the field colour by more than about 60 in any RGB channel (this finds dark words on Paper and the accent word as well as light words on the dark fields). Join runs less than 25 px apart into words; the gap is between the last left-side word's run and the right word's run, inset 30 px. Measure before adding the strips, or the footage counts as ink.
4. Per strip: addMedia with sourceId, nodeId, start and duration to the end, then one node.patch with destination {x, y, width, height: 105}, fit: "cover", audio: null, common.masks (one rect, width keyed as above) and common.effects (the colorMatrix, skipped when the strips stay in colour).

Check

get_filmstrip of the document at the first word + 0.1 s, the middle, the last word + 0.2 s, each strip's start + 0.3 s (half open) and the last frame. Look for a strip touching or overlapping a word, a three-word line whose middle word crowds the strip, a line wider than the column, the accent on the wrong word, and every word against the transcript. Then get_video_preview at 30 fps to check the timing. Report the lines, the accent word, the strip footage and the time of each word.

Gotchas

• get_motion_frame takes startTimeMs; anything else silently returns frame 0.
• get_video_preview fails with preview_render_failed when a silent archive film keeps the audio block addMedia adds: patch audio: null on every strip.
• Masks are in canvas pixels and don't follow the node; key the mask's rect, not the node.
• An archive film first becomes a source with add_library_item (from: "vintage", scope: "private"); most are 320 × 240, fine for a strip.
• A strip of fast or busy footage flickers; wide, slow shots read best.

Variations

• Portrait (untested): 1080 × 1920, Archivo 800 at 110 px, column x 90 to 990, 5–6 lines of 2 short words.
• Signature: a muted Roboto Mono 24 px line (name, date) about 85 px under the column, 0.5 s after the last strip.