Animated Lecture: Turn a Talk into Motion Slides

The finished effect

What it looks like when it is done.

Forty-six seconds of the 22-minute Indian Wisdom lecture used throughout this page (0:53 to 1:39 of the lecture) in the Full Animated Lecture layout: the presenter's voice carries on while the screen is given over entirely to the drawn slide, each point arriving on the phrase that introduces it and the caption band running underneath.

46 s · 1920 × 1080 · with sound

Overview

What it is, and why it exists.

The animated lecture is the deck the rest of the film is built around. It is rendered for the full length of the lecture and then used twice: as the full-frame Animated Lecture layout, and scaled into the slide card beside the presenter in the Presenter + Animated layout. In the 22-minute lecture used throughout these pages it holds the full frame for about two thirds of the runtime.

It solves the problem of a recorded lecture that is only a talking head. Rather than extracting slides, the engine composes a deck for the lecture: chapter headlines that summarise what the speaker is doing, lists in full clauses, definitions that light their key words as they are read, comparisons, sequences and diagrams. Each item is bound to a phrase in the transcript, not to a time, so it cannot drift from the word it belongs to.

It depends on three things made earlier in the pipeline: the cleaned, word-timed transcript (every word with a start and end on the cleaned timeline), the loudness envelope of the studio audio, and the course's template style. It sits after cleanup and studio sound and before composition, which decides where the deck is shown full-frame, where it sits beside the presenter, and where it is cut away from.

The one rule it never breaks: at every frame, what is on screen corresponds to what is being said at that instant. If the speaker departs from his script, the audio wins; the script is a spelling reference, not a timing source.

29 (25 in the deck, 4 drawn-cutaway shapes)Element types

1920 x 1080, 30 fpsFrame

a transcript phrase, never a clock timeCue source

0.22-0.55 s by style (Scholar 0.45 s)Entrance

0.8 s blend at every chapterPalette change

up to 9 words, centred at y 950Caption band

60 s, then a presenter cutLongest full-frame hold

any pause over 0.75 sSilence beat

How it is made

Step by step, in the order the engine runs it.

Build the word timeline

Speech recognition transcribes the cleaned audio in chunks of about 24 s cut at natural pauses, merges sub-word pieces into words, and clamps each word's end to the next word's start. The words are re-timed onto the cleaned timeline, spelling repairs from the lecturer's own script are applied, and the words are grouped into caption phrases. Nothing downstream re-times a word.

Take the loudness envelope

The studio audio is measured in a 55 ms window at 30 fps, normalised to the 92nd percentile of speaking frames and smoothed with a fast attack and slow release. This per-frame curve later scales every pulse, so the screen leans when the lecturer leans on a word.

Compose the deck

The deck is the one authored file in a job. It divides the lecture into chapters, marks each light or dark, and lists the elements for each: type, box, content, and for every part the phrase that introduces it. Headlines are written as summaries, not substrings of the transcript; anchors quote what was actually said, including the recogniser's spelling of an unfamiliar word, because that is what the phrase lookup has to find.

Resolve every cue to a word

Each anchor phrase is found in the transcript, searching forward from the previous cue so repeats resolve to the right occurrence. A phrase that cannot be found stops the build and names the phrase; a mistyped cue can never silently drift. An element's end is the start of its closing phrase minus 0.02 s, or the end of its chapter.

Check the deck before rendering

The deck check refuses any element that ends before it starts, any item cued outside its element's window, and any two elements that are on screen together in the same chapter and share part of their boxes. It reports the number of cue points, the changes per minute and the longest hold with nothing new on screen.

Resolve the style

The course's template style is read from the lecture's settings. Its two typefaces are loaded before the first frame is drawn — a style that names a face the job does not have fails at once rather than rendering thousands of frames in a fallback font — and any pictures are decoded before the first frame is asked for.

Render each frame deterministically

Each frame is a pure function of its time: ground (at 40% speed), content layer, vignette, caption band, then chrome, on a single 1920 x 1080 canvas. No free-running animation and no unseeded randomness, so any frame can be re-rendered identically. Frames are taken five at a time as high-quality stills and encoded in resumable chunks of 900 frames, each marked done when finished, so an interrupted render resumes rather than restarts.

Hand it to composition

The finished deck covers the whole runtime. The composition plan cuts it into full-frame runs of at most 60 s, places it in the slide card beside the presenter elsewhere, gives any stretch where the deck has nothing new to the presenter, and protects its definitions, quotes and heroes from cutaways.

Elements

Everything this effect puts on screen, one piece at a time.

Each element below is shown as the engine actually produced it, in the real lecture, with its parts, and the rules that govern it.

A hero opening a chapter: kicker first, then the title.

hero

Opens a chapter: kicker, then one big title

The hero is the chapter's front door. A small tracked kicker appears first, centred, and the title arrives on its own cue a beat later, so the eye lands on the small line and then drops to the big one. If the title's cue has not been reached yet, the space it will fill is held by a dashed placeholder rather than left empty. Because a hero is one of the moments the deck is written for, the composition planner protects it: no cutaway of any kind may be placed over a hero while it is on screen. In the real lecture below, eleven of the sixty-one elements are heroes, one per chapter except the first.

Anatomy

Box used in practice: 260,350 at 1400x360 px
Kicker: UI face, weight 700, 22 px, tracked 3.4 px, accent colour, centred on the box
Title: display face 700, auto-fitted from 96 px down to 34 px (line height 1.16), centred at 58% of the box height
Entrance: the style's entrance curve stretched by 1.11; title rises 16 px as it lands
Before the title cue: dashed outline (7 px dash) over 76% of the box width, 52% of its height