Invoked byVideo Master Controlafter the user explicitly selects the browser route. The project uses composition-v2, episode.output_target=web. First readStoryboard ProjectandAsset Contract。
Plan and Implementation
Image prompts and image generation followHand-drawn Image Generation Rules. When an image type is specified, directly generate the complete image-and-text frame according to the original image type; do not use the default video or picture-book process to split it into textless assets. Do not repeatedly add the same copy to a full-page image; independent subtitles must still be checked for occlusion.
Read by shot ST numberStoryboard Asset Preparation Directoryand the corresponding documents, and verify the frame inputs and file formats. After the plan is confirmed, deliver specific prompts, actual reference images, and return requirements for generatable missing items; change real screen recordings and data, etc. into collection requirements. Only after all inputs pass review, generate the animation implementation, drawing arrangement, and actual voiceover subtitles.
ST-001–007 may be used; one main type per shot, with necessary auxiliary layers. ST-001 has only full-screen or picture-book mode, with optional environmental effects. Remotion and the two whiteboard types use their respective locked runtimes to generate per-shot caches, then are composited uniformly; do not describe all types as the same kind of browser animation.
Picture-Book Production
When the user requests a picture book, you must readPicture-Book Production Guide. Organize story pages with multiple ST-001 shots, using paper-page whitespace, centered illustrations, and short sentences below; separate the original illustration and editable copy on each page, and measure typography according to the actual frame size and font. Basic pages use the existing image/text/region/keyframes plus shared subtitles and audio, and do not require calling the whiteboard or Remotion. Slight camera movement, reveal, and cross-page dissolve are arranged according to the plan; raindrops, ripples, watercolor masks, and paper texture are optional, and may be reported as passing only after they are actually implemented and reviewed in the current episode; do not force copying old case code. Any programmatic effect is recalculated per frame, and the result is consistent after seeking. Page story short sentences and actual narration subtitles are laid out separately to avoid overlap or duplicate display.
Only after the complete plan is confirmed and all frame assets have actually passed review, write the animation, voiceover, and rendering. Flow tasks, asset strategy, voiceover, and subtitles still useShared StageandExternal Video Process, and do not create a looser threshold separately.
All motion is determined per frame; retain renderFrame(t) and arbitrary time seeking. When assets, runtime, or implementation are missing, report clearly and do not silently substitute. New types are defined only in the current episode, and after actual verification the user decides whether to share them.
Execution and Delivery
When delivering or modifying a preview, executeShot Review and Delivery Checklist. Actually view the cases and assets specified by the user, and distinguish textless illustrations from already-typeset full pages; clearly label screenshots/final videos, asset cards, and static layout sketches. If page screenshots have not been obtained, visual pass cannot be marked; check the actual output keyframes, font size, line breaks, subject boundaries, and subtitle panel, then synchronize the preview, asset references, and verification records.
Execute through the master control hvp entry: export-plan check; render and mix or export/build to output MP4. export does not generate voiceover; build follows the user-specified service. Do not hand shots with render_backend=jianying to the browser for automatic fallback.
Volcano voiceover uses the same packageVolcano Voice Skilland hvp tts, does not read external local CLI. User accounts are stored in local configuration outside the repository, and updating the Skill does not overwrite them; if actual voiceover times out, reshoot the shots first, and do not change speed to pad the duration.
Deliver the final film, preview, cover, subtitles, and production records. Check frame count/duration, complete decoding, transparent edges, transitions, subtitles, and audio; technical pass and dynamic aesthetic acceptance are recorded separately.