konka0812/narrated-explainer-video

Narrated Explainer Video

First, clarify what the user wants the audience to understand, then see what available assets can show the audience. Follow the mutually agreed script, the user-specified final version, and the scope of this delivery. If a real-life shot already explains the process, let it play through coherently. If an asset only shows the result, use it to display the result, then explain the principle with a diagram. If the script's meaning genuinely needs to change, discuss the specific sentence separately.

The flow, storyboard table, and shot examples below can be tailored per project. When choosing a method, keep three justifications: references match the original meaning, subtitle timestamps come from the selected version, and it is clear what was actually checked at delivery.

Starting from the Current Progress

First, examine the user's materials, existing project, and confirmed decisions, then choose the entry point. Do not ask the user to repeat the full workflow:

• Only a topic or viewpoint: Compile a core question, target audience, and conclusion, and help write a natural, speakable script. If the content is too extensive, suggest focusing or splitting, but do not unilaterally split into multiple videos.
• Only a script: Proceed with a visual concept and rough storyboard. Without official audio, timing is only an estimate; you cannot claim synchronization.
• Audio and subtitles already available: Directly check inputs, determine the timeline, and design the storyboard. Do not re-time based on an old script.
• Existing project or final cut: First locate the segment the user is giving feedback on. If the source project is available, modify the corresponding layers; if only an MP4 exists, first check if text is burned into the video. Choose adjustments based on the approach for modifying only a segment of an existing video.
• Release delivery requested: Verify the final inputs and project, and produce the required final cut, thumbnail, or copy. Only create the deliverables needed for this task.

When information is missing, infer from materials and project state. Only ask when it genuinely affects the result, e.g., when the final audio version is unknown. If the user explicitly requests confirmation of a plan, provide a specific plan and wait; if already authorized or confirmed, proceed without repeatedly asking for approval.

Before moving into layout and timeline implementation, clarify the aspect ratio and release scenario, the duration basis, and the deliverable for this task. Prioritize the existing project and confirmed requirements; when a final voiceover is available, arrange based on its actual duration, do not trim audio to fit a target duration. When requirements are unclear, first propose a suitable default and state the assumptions; if a key choice cannot be reasonably inferred and would cause significant rework, then ask the user for confirmation. Work that does not depend on these choices (e.g., topic discussion, scriptwriting) can continue.

Before Production: Check and Prepare the Environment

When the task enters the implementation or preview phase, first read Environment Check and First Setup. If choosing Remotion but the machine has no usable environment, first set up the runtime, initialize the project, install dependencies, and verify preview and test frames before producing the full video. Do not assume that installing this skill already sets up Remotion.

Distinguish between: missing a Remotion guidance skill, missing a runtime like Node.js, missing project dependencies, and missing a project. When an existing project is available, prioritize reusing it and installing via the lock file; avoid reinitializing or upgrading freely. When only a script or visual concept is required, there is no need to install software first.

1. Establish the Content and Time Baseline

Record the current input files, audio duration, subtitle range, aspect ratio, frame rate, project entry point, and target deliverable. The user-specified final version takes precedence over old scripts and old exports.

• Preserve the subtitle text, sequence numbers, and absolute timestamps. Check for missing sentences, empty subtitles, reversed order, overlaps, too-short display durations, and subtitles extending beyond the audio. Overlaps may also be intentional; judge first.
• Use the finalized script to understand sentence meaning, use the final SRT to locate subtitles, and use the actual vocal delivery and pauses to arrange actions. If the user specifies "follow the new subtitle," design accordingly. If the audio materially differs from it, point out the specific sentences and time spans; do not unilaterally change words or shift time codes.
• Verify time-sensitive facts, numbers, policies, and news sources. When making comparisons, first clarify the objects and versions, then check whether definitions, units, and applicability conditions are comparable. If the final recording contains an error that affects the meaning, propose a specific correction; do not let visuals contradict the narration.
• After new audio or subtitles arrive, recalculate affected time spans, subtitles, transitions, and total frame count; check whether subsequent shots remain synchronized.

When an SRT is available, run scripts/inspect_media.py to check structure and duration. When final audio is available and action alignment is needed, you can spot-listen to changed sentences or recognizable phrases before laying out the full storyboard; see Implementation and Synchronization for the approach.

2. First Determine Judgable Visual Direction

When proposing a new direction, pick one sentence from the script to illustrate how the visuals explain it. If existing assets are available, point out which detail is worth magnifying or which pause suits letting the audience read a graphic. When custom diagrams are needed, explain what relationship they should clarify. Color schemes, fonts, and layout decisions only expand as needed for this task.

Consider Both Content and Assets

When the user only describes assets and wants to discuss the concept first, you can propose conditional designs based on the description, e.g., "If the photo shows both the light source and the desktop, frame both together." When it is time to select images, crop, or align operations, then examine the actual files to confirm what they show and which details need to be preserved. If an asset can explain a few sentences, give it enough screen time.

When user-provided images, screenshots, screen recordings, or real-life shots are available, first read Asset Design and Reuse, establish their correspondence with the narration, then choose to display as-is, focus on a part, redraw based on reality, or accompany with an explanatory animation. When no pre-existing assets are available, you can prepare code-based graphics or generated assets based on the topic and audience; characters and expressions are only one optional approach.

• Let the presentation method serve the topic: mechanisms suit schematic animations, processes suit object flows, commentary suits clear separation of evidence and opinion. Do not force all content into the same card layout.
• When the user does not specify, you may propose 16:9, 1920×1080, 30fps; prioritize existing projects, vertical format requirements, and platform mandates. Choose fonts that fit the content's character and language; do not default to rounded fonts or a specific brand.
• Start the video by letting the audience see the problem and subject as early as possible. When introducing multiple terms to beginners, first state what the video helps the audience understand or decide. When presenting a list, group by problems the audience can understand, then place the terms in their corresponding contexts.
• When the style is unsatisfactory, check composition, asset language, shot rhythm, and action logic; changing only the color scheme often does not solve the template feeling.

First Validate the Direction with Representative Frames

For layout issues, you can try the most challenging frame as a still and judge. To validate a new style or a major pacing change, pick two segments covering different explanatory tasks, produce a short sample, and play them consecutively with transitions. Check whether actions follow the narration, whether the audience's gaze is drawn to key information, whether the result leaves enough reading time, and whether composition and entry methods are repetitive. When the user requests a plan first, wait for confirmation before expanding; when the direction is already clear or only local modifications are needed, proceed directly.

3. Organize Continuous Shots According to Sentence Meaning

First group together several sentences of narration that explain the same point, and pick a subject the audience can follow. Determine sentence by sentence what the audience should look at at that moment; the visual can respond through object changes, gaze shifts, or resting on a result. When moving to a new point, reselect the subject, composition, and movement style based on the explanatory need. Simple segments can be covered with a single shot description. When the user requires per-sentence effects or complex shot transitions, then mark each sentence's corresponding change with subtitle ID and time. Storyboard and Semantic Animation contains a customizable table and continuous shot examples.

Prioritize changes that express meaning: appearance reason, process, state comparison, accumulation, decomposition, trial and error, feedback, and result. The same object can run through the whole video, but it must evolve with the narrative; repeating a static asset cannot replace development.

Establish only one primary attention center per shot. Subtitle, title, illustration, and camera movement should not compete for attention. Complex mechanisms can be revealed gradually and allowed to settle; the number of actions is not a quality metric.

4. Create Graphics and Animation

When using Remotion, also use the official remotion-best-practices skill and read Implementation and Synchronization. If that skill is missing, refer to Environment Check and First Setup to obtain it, loading guidance relevant to this implementation; when installation is blocked, explain the limitation and proceed with what can be done using the official documentation. An existing non-Remotion toolchain can be maintained when appropriate; do not force migration solely to use this skill.

• Graphics, arrows, values, charts, and state changes should be made into editable code or vector elements first. Only generate or select bitmap assets when illustrations, textures, or characters are needed.
• Subjects that need independent movement should be on separate layers. Generated images provide visual assets; shots and actions still must be designed around the timeline. Real screenshots and screen recordings show facts; supporting animations explain relationships. Do not use decorative icons to replace key operations, and do not fix all assets to the same side.
• News and evidence use authentic materials and cite actual sources; generated images must not impersonate news screenshots, test results, or statistical evidence. Screenshot zooms and highlights should accurately point to the relevant region while preserving necessary context.
• Keep on-screen text according to its purpose. Titles raise questions or signal transitions, labels help identify objects, data, and conditions, and subtitles relay narration. Storyboard descriptions should be converted into visual actions or state changes. When reviewing, extra text can be temporarily hidden; if the image is still understandable without it, remove it. Keywords, formulas, or key conclusions can appear in two places, but with a clear hierarchy and, if necessary, staggered timing. For text on complex backgrounds, use stable background colors, local masks, or separate layouts to ensure contrast.
• Titles and labels can enter with the narration; after they are read, redirect attention to the subject or the next shot. See Controlling Text and Layout for specifics.
• Check titles, figure captions, screenshot original text, and subtitles at the target viewing size. Magnify the subject, tighten the layout, and crop key content before piling on decorations to fill space.
• Reserve space for subtitles and platform UI; key actions, sources, and explanations should not hide behind subtitles. Fonts and assets must be loadable stably within the project.

5. Preview, Modify, and Deliver

Choose the review method based on this deliverable, and distinguish three types of evidence:

1. Still frames: composition, text size, contrast, occlusion, screenshot highlight positions.
2. Continuous playback: synchronization of narration, subtitles, actions, and transitions; whether repetitions and pauses are reasonable.
3. File inspection: resolution, frame rate, duration, audio tracks, encoding, and decodability.

Do not replace full dynamic review with "build succeeded" or "looked at a few screenshots." Record the actual scope checked; honestly state what could not be viewed or heard, and avoid claiming full approval.

When receiving feedback, first locate the specific frame, e.g., "small text" might be due to an insufficient title font size or because a full-page screenshot was scaled down too much. After modification, review the changed segment and transitions; when shared components are involved, also check other affected shots.

When a final cut is needed, export based on the authorized scope and verify the final file. Thumbnails, publication titles, and copy are only created on demand, with content based on the final video; titles should be attractive but not overstate the conclusion. Creating release materials does not mean publishing on external platforms on the user's behalf.

Keep the final audio, subtitle, editable project, referenced assets, final cut, and necessary sources. Use topic or project names for file naming; do not hardcode episode numbers, personal directories, or machine environments. Before cleaning up, check project references and shared dependencies; do not delete old files by default.

Deliver the plan, segment preview, or final cut as agreed. When delivering a video, include position, specifications, and actual inspection results; if an editable project is available, provide the entry point.