yichen-jianying-edit

Headless Jianying Editing

By default, deliver a native draft that can continue to be edited. Run native export only when the user explicitly requests a finished video. Headless does not mean Jianying does not need to be installed; this Skill does not include the official program library or a complete editing backend. The main version is 11.5.0, compatible with 11.4.2; 11.6.0 supports creating new drafts, editing copies of existing drafts, and local fonts (including font changes), and saving and reading back have been calibrated against the measured rules of 11.6.0; compound clips can be built and read offline, but homepage registration and frozen snapshot delivery are still unavailable; headless native export is still rejected, so when a finished video is needed, use ui-export(Jianying UI rendering + script verification, a different tier of evidence). Select the configuration based on the exact installed version and fingerprint; the export snapshots of the two versions must not be mixed. The HD black-and-white filter and orange-outline stylized text have been removed and are no longer listed as features pending verification; when an old plan or snapshot contains these two items, explicitly reject it and do not silently remove the effects.

First confirm dependencies and the scope of this session

1. According to installation instructions , locate the authorized core project checkout directory. When the Skill and the core project are in the same repository, it is located automatically; for a standalone installation, set JIANYING_HEADLESS_ROOT. Do not collect access credentials from the user's other projects, browsers, or chats, and do not automatically copy the official library.
2. Run python3 SKILL/scripts/headless_draft.py doctor. If the core project is missing or the version/hash does not match, stop native operations; fix dependencies according to the error, and do not change the pinned version and hash to pass the check.
3. Determine materials, create or modify, speed, sound effects/BGM, and editing intensity from the request. Use the parameters explicitly specified in this session; do not carry over speed and cut preferences from historical projects; ask about ambiguous semantic cuts in a consolidated manner.
4. Transcription is a separate optional dependency. When $yichen-asr exists, follow its instructions; when a word-by-word transcript has already been provided, use it directly. Paid transcription requires authorization for this session; do not automatically top up, switch providers, or resubmit.

Choose an entry point

• Create new: read headless plan and commands. Create an independent draft directly; there is no requirement to prepare a template in the UI first; online templates are outside the scope of this release.
• Hypit project handoff: read material and font handoff, and write the materials, timeline, and actual font files of the selected finished video into the new plan; for text clips, you can use font_path to specify the font directly.
• Modify an existing draft: read independent copy editing, and keep the original project.
• Explicitly request a finished video: read native export, and export only a verified frozen snapshot. First check the version of doctor : 11.5.0 / 11.4.2 use headless export；11.6.0 has no reviewed export configuration, so switch to the ui-exporton the same page (Jianying UI rendering + script verification, a different tier of evidence, and it will modify the draft being exported).
• Spoken-word semantic editing: read semantics and audio quality check and spoken-word plan format, with scripts/edit_plan.py uniformly compiling the time mapping.
• Spoken-word automatic sound effects: when the user wants sound effects and has not provided local audio, write English keywords in the spoken-word plan openverse event;from-compiled then use scripts/openverse_sfx.py add to select and write independent audio tracks, see steps Headless plans and commands。

Execution and acceptance

• Source footage remains untouched; use authorized local assets. Subtitles are required by default; adjust style to this project's visuals. Do not add BGM unless requested. Do not remove hard-to-confirm numbers, product names, or core judgments.
• build Write only to this isolated working directory.publish / resume-publish is registered on the local CapCut home page, not an online publish; before executing, have the user save their work and fully quit CapCut, do not force-kill processes.
• All outputs use a new directory under this run's work/ . Asset copies, hashes, consistency, and directory locks are verified by tools; do not mistake consistency parameters such as --expect-* for approval questions requiring item-by-item user input.
• Effect resources may only come from matching caches legally obtained on this machine, preserving original license fields and byte verification. Do not download, unlock, or forge authorization; cache presence or successful technical rendering does not count as a membership/commercial license.
• openverse_sfx.py goes online, accessing only the Openverse API and Freesound preview addresses; draft build/publish still do not go online. Sound effects retain their respective CC licenses: by default CC0, CC BY, and CC BY-NC are all usable; CC BY-NC is selected only when no suitable CC0 / CC BY exists. When the request indicates a commercial order, ad, product promotion, traffic-driving, or monetization-enabled video, add --commercial, use only CC0 / CC BY.
• When no qualified sound effect is found for a keyword or rate limiting is encountered, the entire step fails and no new plan is written. Change the keyword in the voiceover plan and recompile, or rerun later with a new --work ; do not switch to unreviewed sources, bypass rate limits, or remove the user's requested sound effects just to succeed.
• When Openverse sound effects are used, the delivery notes must relay this run's credits.md: whether CC BY-NC is included (only for non-commercial videos), which sound effects require attribution, and include ready-to-paste attribution text. Selection is based only on title, tags, duration, and license, with no preview listening; listen to the sound effects during native acceptance.
• Compound clips may only be used as experimental offline builds, modifications, and frozen snapshot exports; formal home page registration is still blocked. Do not deliver them as nested drafts that have passed native save persistence acceptance.
• Drafts must be checked in CapCut for visuals, subtitles, volume, and cuts; save, quit, cold reopen, and read back. When UI conditions are lacking, continue the offline portion, but clearly mark native acceptance as incomplete.
• After export, check the actual visuals and audio.encoded-and-decoded does not mean all content is correct; the default draft delivery does not generate a temporary MP4. On failure, preserve the scene, do not auto-retry, remove effects, or replace with an old snapshot.
• Preserve original assets and existing drafts; cleanup is performed only per this run's explicit authorization, and files are never automatically permanently deleted.

Installation scope and licensing

This Skill and the public core project are installed separately; installing only the Skill does not include the full backend. Code licenses, CapCut terms, asset licenses, and account rights apply separately. See the relevant statements at Runtime dependencies and sources。