Creative Space
Turn One Concept into a Clear Explainer Video

İçerik

Provide: Concept/question, audience/grade/expertise, duration/aspect, style, any textbook/handout/article/syllabus to follow.

Receive: A 1–3-minute video (default 90 seconds, 16:9; optional portrait), script, storyboard and source table.

Help viewers retell: question → example → concept → one-sentence answer, one idea at a time.

Style

Use one style throughout; settings/prompts: . Options: hand-drawn illustration (default for science/humanities/everyday topics); flat illustration (technology/business/processes/dense information); stick figures (abstraction/thought experiments/informal tone); whiteboard (derivations/steps/drawing while explaining); collage (opinions/history/culture); type/graphic motion (data/formulas/comparisons without illustration).

Tools and references

  • Host readers extract facts from supplied sources; optionally search authoritative sources if absent.
  • The host plans shots using the sentence-level fields and continuity checks below; no separate planning skill is required.
  • Flova, required: imagery/scenes, clips needing motion, narration/music.
  • HyperFrames as needed: accurate formulas, numbers, charts and labels as transparent clips.
  • FFmpeg, required: assembly/audio and overlay; do not depend on often-unavailable drawtext/subtitles.
  • , steps 2–3: questions/misconceptions, concept progression, timing, synthesis/quiz, source table.
  • , steps 4–6: six recipes/consistency.
  • , steps 4/7: code layers and data animation.
  • Captions (optional): If requested, offer a suitable HyperFrames, Remotion or FFmpeg post-production workflow.

Boundaries

Record sources for every fact/number/formula/date. Unverified claims are omitted or explicitly attributed as a view. Supplied textbooks/syllabi govern terminology/content. Exact formulas/units/numbers/names/chart labels use code, not generated image text; request text-free visuals. Simplify without falsifying; explain analogy limits. Match grade vocabulary and avoid out-of-syllabus concepts unless requested. Use authorized imagery/music; do not imitate living artists' personal styles. Remind users of platform AI-label requirements.

Workflow

Confirm topic/audience → verify facts → script/confirm → storyboard/confirm → test segment/confirm → batch → diagrams/information layers → assemble/check.

  1. Topic/audience: Confirm one-sentence question, audience, duration, ratio, style, language, textbook/syllabus. Narrow broad topics with two/three options (e.g. quantum mechanics → why electrons do not fall into nuclei). One video answers one question; suggest a series for more.
  2. Facts: Read supplied files/pages/connectors. Otherwise use textbooks, scholarly societies, or official institutions, not just general encyclopedias. Record fact/source/excerpt in docs/facts.md using script reference. Identify one/two common misconceptions.
  3. Script — confirmation: Question → intuition/misconception → concrete example → concept → revisit question → one-sentence answer, optionally quiz. Estimate speech duration; cut content rather than speed up narration.
  4. storyboard — confirmation: Sentence-by-sentence: original narration, subject/action/meaning, code layers, duration, still/moving treatment. One meaning per shot; demonstrate, not merely decorate. Keep one visual metaphor for abstraction. Write shared style/palette/line/character/background settings reused verbatim.
  5. Trial — confirmation: Test the opening plus hardest explanatory shot, usually Quick generation via Flova Skill. Generate imagery/voice, add accurate layers, make a 15–20-second test. Confirm style, voice, pace before batching.
  6. Batch: Repeat shared settings; anchor recurring characters with an initial reference. Request no text/logo/watermark. Default to stills with subtle push/pull; processes (flow/growth/movement) use image-to-video with the still as first frame. Generate narration once then cut by sentence, specifying voice/rate/patient clear tone. Use one light music track beneath intelligible speech. Check consistency, no gibberish, script-to-image meaning; redo only failed shots.
  7. Diagrams and information layers: Render the formulas, numbers, charts and labels required by the storyboard from the script and source table.
  8. Assembly/check: Narration sets timing; mute generated clips and cut at sentence endings. Watch completely: formulas, numbers and units match sources; images demonstrate the spoken idea; text remains readable without pausing; the conclusion answers the opening question. Deliver video, script, storyboard, docs/facts.md and project files. A changed sentence affects only its shot, voice and relevant information layers.
explainer-<topic>/
├── docs/      brief.md, facts.md, script.md, storyboard
├── frames/    Stills and style.md shared settings
├── shots/     Moving clips
├── audio/     Full/sentence narration and music
├── layers/    formula/chart HyperFrames source and clips
└── output/    video and assembly scripts

Painting style recipe

The entire video uses only one style of painting. After selecting, write a unified setting and save it in frames/style.md. Each screen is requested to be reused verbatim; only change "what to draw in this shot".

Unified settings include: painting style and medium, lines (thickness, color), color matching (3-5 colors, write the color number), background, character appearance (if any), light, composition habits, and unnecessary items (text, logo, watermark, realistic photo texture).

Six painting styles

Hand-drawn illustrations (default)
  • Writing: Hand-painted watercolor or colored pencil texture, visible brush strokes and paper texture, soft color matching, and plenty of white space.
  • Suitable for: Popular science, humanities, and life knowledge.
  • Note: Please indicate "a set of illustrations by the same illustrator"; when the characters are to be fixed, please provide a character setting diagram for reference.
Flat illustration
  • Writing: Flat vector style, geometric shape, solid color block with no gradient or light gradient, unified rounded corners, information graphic feel.
  • Suitable for: Technology, business, process and system content with high information density.
  • Note: The color matching is strictly limited to the set color number; the iconized objects must be consistent with the icon style in the layer.
Stickman
  • Writing: White or solid color background, black single-line stick figure, round head, consistent line thickness, the simplest dots and lines for expression, and a small amount of accent color.
  • Good for: Abstract concepts, thought experiments, contrast and contrast, lighthearted tone.
  • Note: Actions should be exaggerated and clear; the same character should be distinguished by the same mark (hat, color), not by face.
Whiteboard
  • Writing method: White board or white paper background, black marker lines, a small amount of red and blue emphasis, like drawing while talking.
  • Suitable for: Derivation, steps, content that needs to be built step by step.
  • Method: The same picture appears in several steps: First, the complete picture is produced, and then the layer mask is used to reveal the pieces one by one following the lecture notes. This is more stable than letting the model draw frame by frame. The stroke animation that appears in hand drawing can be produced by using Tuxedo, but you need to check whether the picture is deformed.
Collage
  • Writing: Handmade paper collage, elements cut from half-tone dot printed pictures, thick black outlines and hand-torn edges, projections under each layer of paper, flat solid color background, one picture only expresses one idea.
  • Suitable for: Opinion, history and cultural topics.
  • Note: There are often short label texts in collages, and the accurate content still depends on the layer; if the label drawn by the model is written incorrectly twice, it will be superimposed later.
Pure graphics motion graphics
  • Writing: No illustrations, all done by layers: large characters, icons, charts, digital animations, with solid color or light texture background (the background can be generated with Flova).
  • Suitable for: Content based on data, formulas, and comparisons.
  • Method: Mainly look at .

Consistency check

After each batch of images comes out, the comparison is set uniformly to see:

  • The painting style, lines, and colors are consistent with the first final draft, and there is no sudden change to realism or 3D;
  • The appearance, clothing, and markings of the same character are consistent;
  • There are no garbled text, fake formulas, or watermarks on the screen;
  • The meaning expressed by each picture matches the storyboard, and is not a general "related picture".

If there is any inconsistency, only redo that one, and pass in the first final version as a reference picture.

Graphics, formulas and numbers

Flova illustrations or scenes form the picture layer. Formulas, numbers, charts and annotations form the information layer, rendered with transparency in HyperFrames and composited with FFmpeg overlay. Information-layer text comes from the script and source table; changing one number rerenders only the affected layer.

Which content goes to the information layer?

  • Formulas, units, chemical formulas, numbers, dates, proper nouns;
  • Diagrams, processes, step numbers;
  • Arrows, circles, and labels pointing to somewhere on the screen;

The request for the screen layer states that no text is required; any text drawn by the model is not used.

First check the user’s local component library

The component library will be updated, and the version installed by each person is different, so the component name is not preset. Before starting work, search the user's machine by purpose, note down the available components and their variables, and write them into layers/components.md:

npx hyperframes catalog --query <关键词> --json
PurposeWhat keywords to search forSelection criteria
Bar chart, line chartchart, graphcan appear gradually in reading order, and finally stop at a given value
Numbernumber, counterThe value that stops exactly in the source table after scrolling
Process, cause and effect, stepsdiagram, flowchart, pipelineNodes can appear in the order of the lecture notes
Layered structurelayer, stack, explodedcan be taken apart and closed again
Arrows and circlesarrow, annotate, markerHand-drawn, whiteboard style hand-painted series
Emphasis on a sentence or wordHighlight, calloutSummarize a commonly used sentence
title cardtitle, titlecardopening question, chapter title

| Map | map | Used when talking about geography and migration; map data needs to be checked |

  • The selected components should be consistent with the style of painting: do not mix hand-drawn series and clean flat series in the same video.
  • When a suitable component cannot be found for a certain purpose, write an HTML or SVG block in the project and align the style with the selected component.
  • Commands, parameters, and variables are subject to native HyperFrames help and component descriptions.

Formula

First search for formula and equation in the local component library; if you cannot find it (usually not), write an HTML block yourself in the HyperFrames project:

  • Render with KaTeX or MathJax (for accurate typesetting), or write directly into SVG;
  • Appear item by item: Come out with the integer first, then highlight each letter in sequence following the lecture notes, and mark its meaning and unit next to it;
  • When deriving step-by-step, each step has one picture, and the previous step fades and stays on top, so that it is not piled up all at once;
  • The formula is written exactly the same as in the source table, including symbols and units.

Illustration principles

  • A picture only says one thing, no more than three or four labels;
  • The order of appearance follows the lecture notes: wherever you talk, draw the diagrams;
  • Leave at least 1.5–2 seconds after the value stops for the audience to finish reading;
  • When the color has meaning, the whole film is unified (e.g., heat is always orange), consistent with the color matching in the style formula;
  • About one-fifth of the bottom and the right side of the vertical version will be blocked by the platform button, and the information will be placed outside.

When you can’t run

If HyperFrames cannot run, explain the limitation. Render formulas, numbers and annotations as transparent PNGs with a browser or Python and composite them with FFmpeg overlay.

Explanation structure and lecture notes

The lecture notes to explain the video are not to read the knowledge points over and over again. The audience has to keep up without being able to pause or ask questions, so every sentence must answer "Why should we listen to the next sentence?"

Structure

ParagraphProportion (90 seconds as an example)What to do
QuestionAbout 10% (8–10 seconds)Ask a specific question that makes people want to know the answer. Don't say "Today we are going to talk about..."
Intuition or MisunderstandingAbout 15%Name most people’s first reaction, or a common misunderstanding, and point out what’s wrong with it
ExampleAbout 25%A specific example that can be drawn, first look at the phenomenon
ConceptAbout 30%Extract concepts from examples and give names, definitions, and formulas if necessary
Return to questionAbout 10%Use the concepts just discussed to answer the questions at the beginning
SummaryAbout 10%Say the core in one sentence; you can add a quiz or "Next time you see... you will think of..."

Not every piece must have six paragraphs, but three paragraphs - question, example, and return to question - must be included: without questions, the audience will not know why they are listening; without examples, the concept will be empty; without returning to the questions, the audience will not know what the answer is after listening.

Questions starting with ##

  • Specific: "Why do airplanes fly when they are so heavy?" is better than "Let's understand lift."
  • The audience will wonder themselves: Start from an everyday phenomenon, a counter-intuitive fact, a number.
  • No spoilers for answers, and no telling too long: Start giving clues within 10 seconds after the question is asked.
  • When the topic is a definition (e.g. "What is inflation"), replace it with the phenomenon it explains: "Why 100 yuan ten years ago cannot buy the same thing now."

Misunderstanding

  • Pick one of the most common misconceptions on this topic, don’t make a list. Source: "Error-prone points" in textbooks, teachers' experience, and frequently asked questions in search results.
  • First admit that it "sounds reasonable" and then use an example to show where it goes wrong. Don’t laugh at the audience.
  • When you cannot find a definite common misunderstanding, use "Intuitively, we would think..." to elicit it, and do not make up statistics.

How to explain a concept

  • Talk about one thing at a time. Only one new word appears in a sentence; the first occurrence of the new word is followed by an explanation.
  • Examples first, definitions later. The audience sees the phenomenon first and then hears the name.
  • Be clear about the boundaries of analogies: When using an analogy e.g. "electricity is like water flow", add a sentence where it is not true to prevent the audience from leaving with a wrong understanding.
  • The formula only appears as needed: When it appears, first say what each letter is, and then talk about their relationship; use layers to display it on the screen, and highlight each item one by one (see the illustration section).
  • Numbers must have a sense of scale: "150 million kilometers" is not as easy to understand as "light takes more than 8 minutes"; both must have a source.

Writing method

  • Spoken language, short sentences; one sentence has one meaning, suitable for pronouncing.
  • Use less "first, second, and in summary"; use questions and transitions to advance: "Then why...?" "Here comes the question:...".
  • Write proper nouns, names, and numbers the way you want them to be pronounced, and Voiceover will not mispronounce them (e.g., write "carbon dioxide" instead of CO₂, and then display CO₂ on the layer).
  • When facing students, the words used should be consistent with the teaching materials, and no names other than those in the teaching materials should be used.

Speaking speed and duration

Experience value, subject to actual voiceover:

LanguageClear explanation of speaking speedHow long is approximately 90 seconds
ChineseAbout 220–260 words per minute330–390 words
EnglishApproximately 140–160 words per minute210–240 words

If the time is exceeded, the content will be deleted and the speaking speed will not be increased. Leave pauses in complicated places to allow the icons in the screen to complete.

Summary and quiz

  • Summarize in one sentence that can directly answer the opening question.
  • Optional quiz: a multiple choice question or "think about it", the answer is given in the last few seconds of the video or in the release copy.
  • A video series can preview the next question at the end.

Fact source table

docs/facts.md, one line of facts appearing in each speech script:

NumberFacts (original words in the speech)Source (book title or website)Excerpts from the original textCheck status
F01Light travels from the Sun to the Earth in about 8 minutes and 20 seconds(Source)(Excerpt)Checked
  • In the lecture notes, add numbers after the matching sentences, and the storyboard and layers reference the same number.
  • The verification status is written as "Verified", "User provided, not verified", "Reliable source not found"; the last one cannot be entered into the finished video.
  • Give it to the user when it is delivered, so that it can be easily checked by the teacher or reviewer.

Yaratıcı süreç

  1. Flova

    Flova

    Generate explanatory visuals

    Create illustrations, video, and voiceovers from the script.

  2. HyperFrames

    HyperFrames

    Create explanatory diagrams

    Animate formulas, numbers and charts as needed by the explanation.

  3. FFmpeg

    FFmpeg

    Assemble the explainer

    Arrange visuals around the narration, synchronize sound and export.