remotion_paper
Deliver actual Remotion-rendered paper explanation videos, along with the project and source records needed for continued editing. By default, new complete works deliver a single video with British English voiceover and Chinese subtitles; Chinese subtitles faithfully correspond to the actual English narration. This skill provides only the workflow and experience, not templates, execution scripts, or assets.
Continue from the current task
- First determine whether this is new video production, content modification, voice change, subtitle fix, or catalog organization. Check existing scripts, timelines, assets, final videos, and review records. Only advance relevant work; do not redo the entire video just because the skill was invoked.
- User-specified voice, speaking speed, language style, and production scope remain in effect. Only ask for choices that cannot be determined from files or conversation and would significantly affect the final product.
- By default, only produce British English voiceover and Chinese subtitles; do not additionally generate a Chinese voice version or bilingual subtitles unless the user explicitly requests it in the current session. Updating the skill does not trigger redoing, deleting, or migrating historical dual-version videos; local modifications reuse existing qualified content.
- Simultaneously check
candidate_paper/,paper/, and existing ledgers to locate papers and projects by ID, original title, DOI, or PDF hash. When old chat paths are invalid, consult registration records and migration mappings to avoid duplicate entries. Prefer existing projects; do not restructure directories without a migration requirement. - For multi-paper workspaces, use
candidate_paper/,paper/, andremotion/: the candidate directory receives original-filename PDFs and retains pending and in-progress papers;paper/stores PDFs of completed video generation and review; projects and videos are saved inremotion/. After registration, PDFs useP0001__YYYY-MM-DD__FullEnglishTitle.pdf, and projects use the same basename; the ID is permanent, the date indicates the start of interpretation, and usependingif not started. The same paper reuses one project. - Before first production, complete registration and record the actual start date. The PDF remains in the candidate directory during production; after review, move it to
paper/. Preserve the ID, start date, filename, and PDF content, synchronizing source citations and ledger. Revisions of completed papers continue to usepaper/and the original project.paper/catalog.jsonuniformly registers candidate, in-progress, and completed papers;paper/README.mdis the generated reading entry. See Paper Library Management for candidate directory compatibility and archiving rules.
Production Mainline
- Establish common evidence. Read the user-specified version of the paper; verify the title, authors, publication venue, and date. Map key claims, formulas, and chart values to their original positions. Before writing the script, create English main notes, saved by default as
docs/paper-notes.mdwithin the paper project, recording the paper understanding and video design. Update continuously during production and deliver with the project for cross-paper review. Source records indicate the reading scope; do not write a full-paper interpretation from an abstract-level reading. - Reconstruct and verify research background and motivation. Sort out the actual contributions of prior work, their applicability conditions, and their relationship to the research gap, motivation, and the response of this paper. Do not forcibly assign a flaw to every citation. By default, selectively read key original studies supporting the core gap. Distinguish the paper authors' evaluations, original literature verification results, and the interpreter's summary, recording sources and evidence limitations. The scope should be limited to the research path necessary to understand this paper.
- Design explanations and write the English script; generate voiceover. First determine the core conclusions, the prerequisite knowledge and decisive evidence needed to understand them, then arrange the main visuals, content selection, and length. Write a natural British English script directly from the study results. After a short opening, proceed to specific questions; explain background and fundamentals when the relevant mechanism is first needed, so the audience can follow the method, evidence, and boundaries. Generate British English voiceover by complete explanation paragraphs, saving the actual word timing information. Do not add new gates requiring manual approval unless the user explicitly requests a preview.
- Produce Chinese subtitles and animation. Translate the actual English narration into natural, accurate Chinese subtitles, preserving the argument order, key values, conditions, and conclusion strength. Subtitles are translations; do not write a separate set of Chinese explanations. Arrange subtitles and semantic actions according to the real time information of the English speech. Check the main visual state and the explained object. Paper titles, author names, methods, and metrics may retain the English original; mixed Chinese-English layout is allowed in visuals.
- Review and deliver a single video. Within the paper project, provide the final video, Chinese SRT, cover image, English script, chapters, English main notes, source records, and review records. Check the background argument, core content, subtitle translation, and semantic synchronization. Provide a playback entry and start/stop methods. After first production review passes, move the source PDF from the candidate directory to
paper/per the paper library rules, verify the hash, and update effective citations. Synchronize the actual delivery status and file entry in the ledger. Reuse unchanged content during modification and migration. Clean up only within the current authorization scope.
Production Requirements
- Opening tour totals about 30 seconds. The basic paper introduction and content overview share this time, usually planned for 25–35 seconds; opening, pauses, and transitions are all included. User explicit requests take priority for the current session. Bibliographic information is primarily carried by visuals; the narration briefly describes the research topic and explanation route, then quickly enters the main content. See Short Opening Tour for specific trade-offs and actual time verification.
- Script is direct and natural. English narration, Chinese subtitles, and visual explanatory text should directly state the object, method, evidence, and logical relationships, reducing transitions, previews, and audience instructions that add no information. This is an important qualitative editing standard. Retain necessary concept distinctions, step sequences, and navigation; do not set forbidden phrases or occurrence counts.
- Aimed at beginners. The default audience has no professional background in the field or core methods. Build understanding from specific problems and intuitive examples, then introduce necessary terms, symbols, and methods. Explain inputs, outputs, and metric meanings. Foundation building should focus on the knowledge required to understand this paper.
- Focused explanation; control length based on understanding needs. Prioritize organizing into about 15 minutes, typically aiming for under 20 minutes; this is an editorial preference. Actual length is determined by the paper's information content and necessary explanations, not as a hard review threshold. User explicit requests take priority for the current session. Merge repeated groundwork and similar results; retain core mechanisms, decisive evidence, important qualifications, and examples needed by beginners. Maintain a natural speaking speed; do not fix the number of shots, page duration, or sections. See Core Content and Duration.
- Final video focuses on the paper. Narration and visuals do not actively display production tags such as British English, English voiceover, or Chinese subtitles. The producer's literature search, download, verification, and production process remain in the project records. The video directly introduces necessary research content, preserving scientific conditions and evidence limitations that affect understanding.
- Main visuals carry specific explanations; support pause reading. Present the research object, approach, relationship, evidence, and necessary conditions. Keywords may serve as titles or labels but cannot bear the core explanation alone. Choose tables, mechanism diagrams, examples, or data graphs based on information relationships; retain comparable content; avoid duplicating narration outlines. When relevant research needs comparison, prefer tables. The mechanism of prior work necessary to understand this paper may be independently animated and connected to this paper's design. Show citations with content; label original results, teaching examples, and interpreter summaries separately. See Visuals and Assets.
- Use British English voice preference. Unless otherwise specified, start with Ryan
en-GB-RyanNeuralas the voice base. Adjust the speaking speed according to the current task; do not hardcode parameters and style from one paper. - Build a real timeline. Generate voiceover by complete explanation paragraphs; organize subtitle cues and animation actions separately. One paragraph of narration may correspond to multiple actions; triggers and visual state must match the actual object and step being discussed in the English narration, not just proceed by paragraph number. See Semantic Actions and Animation Triggers. Chapters and total duration also rely on the real time information of the English voiceover. Chinese subtitles appear with the corresponding English content. In-video subtitles and the Chinese SRT share text, time, and line breaks; preserved English words, abbreviations, and hyphenated terms are not split.
- Actual rendering and verification. Use Remotion CLI or Render API; do not substitute code for the final video. First check representative frames and short samples, then export the full video. Verify decoding, audio track time synchronization, subtitle alignment, animation semantic correspondence, output parameters, and final duration. Video length, number of chapters, and number of animations are determined by the content and user requirements.
Read References On Demand
- When registering papers, creating projects, correcting start dates, or organizing directories, read Paper Library Management; reuse existing IDs and tools; distinguish between daily registration and historical migration.
- When creating a new work, read "English Main Notes" in Production Workflow to establish a shared record of paper understanding and video design; subsequent modifications maintain related content.
- When creating a new work or modifying research background, read "Paper and Script" and "Research Background, Research Gap, and Motivation" in Production Workflow; perform key original research verification and use the results for script and content review; see "Visuals and Assets" for literature display.
- When writing or modifying scripts or visual explanatory text, read "Paper and Script" and "Visuals and Assets" in Production Workflow; first determine the core explanation and length, then check introductory explanations, visual information, and natural expression; complete related modifications before voiceover.
- When creating a new work, read "English Voiceover and Chinese Subtitles" and "Multi-Paper Directories and Cleanup" in Production Workflow; read corresponding sections when handling voice, subtitles, rendering, or incremental modifications.
- To understand the rationale behind conventions, or when encountering keyword-outline visuals, length imbalance, unnatural voiceover, English word splitting, animation semantic mismatch, browser timeout, or directory migration issues, read Production Experience. There, versions, durations, counts, and styles are case information, not requirements for all papers.
Distinguish Evidence States
Construction pass, actual rendering, full decoding, audio track time synchronization, subtitle alignment, animation semantic correspondence, main visual information check, representative frame layout check, and manual listening review are different conclusions. Only report items actually completed and their coverage. Technical synchronization and semantic correspondence passing still require checking whether the main visuals adequately explain the content. Do not claim reproduction without running paper experiments; user acceptance of a trial listen does not equal acceptance of all new content. Normal production and modifications within clear authorization do not add repeat confirmation steps. A request for a new video does not imply permission to delete other papers or existing versions.