Libtv Video
nexu-io/nexu
Seedance 2.0 video & image generation via LibTV Gateway - AI text-to-video, image-to-video, video continuation, style transfer, and text-to-image using Seedance 2.0 model.
Agent skill
by dracohu2025-cloud in dracohu2025-cloud/draco-skills-collection
A skill your agent uses when running a Feishu/Lark Base-centered Seedance video production pipeline: inspect/backfill the Base row, create/reference assets, build Row-24-style Chinese Seedance…
$ npx skills add dracohu2025-cloud/draco-skills-collection --skill feishu-seedance-video-pipeline -a claude-codeProject install by default; add -g for ~/.claude/skills/.
$ gh skill install dracohu2025-cloud/draco-skills-collection feishu-seedance-video-pipeline --agent claude-codeProject scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).
$ git clone --depth 1 https://github.com/dracohu2025-cloud/draco-skills-collection.git skills-src && mkdir -p .claude/skills && cp -r skills-src/feishu-seedance-video-pipeline .claude/skills/feishu-seedance-video-pipeline && rm -rf skills-srcUse ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.
Claude Code skills documentation · loads skills from .claude/skills/
Install the "feishu-seedance-video-pipeline" agent skill from https://github.com/dracohu2025-cloud/draco-skills-collection/tree/main/feishu-seedance-video-pipeline into .claude/skills/feishu-seedance-video-pipeline/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "feishu-seedance-video-pipeline", then confirm the skill loads.Claude Code copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$skill-installer install https://github.com/dracohu2025-cloud/draco-skills-collection/tree/main/feishu-seedance-video-pipelineType this inside Codex. $skill-installer <name> installs a curated skill from openai/skills. The installer writes to $CODEX_HOME/skills (default ~/.codex/skills). Restart Codex if the skill does not show up.
$ npx skills add dracohu2025-cloud/draco-skills-collection --skill feishu-seedance-video-pipeline -a codexProject install goes to .agents/skills/; add -g for ~/.codex/skills/.
$ gh skill install dracohu2025-cloud/draco-skills-collection feishu-seedance-video-pipeline --agent codexProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/dracohu2025-cloud/draco-skills-collection.git skills-src && mkdir -p .agents/skills && cp -r skills-src/feishu-seedance-video-pipeline .agents/skills/feishu-seedance-video-pipeline && rm -rf skills-srcUse ~/.agents/skills/ instead of .agents/skills for a personal install.
Codex skills documentation · loads skills from .agents/skills/
Install the "feishu-seedance-video-pipeline" agent skill from https://github.com/dracohu2025-cloud/draco-skills-collection/tree/main/feishu-seedance-video-pipeline into .agents/skills/feishu-seedance-video-pipeline/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "feishu-seedance-video-pipeline", then confirm the skill loads.Codex copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ npx skills add dracohu2025-cloud/draco-skills-collection --skill feishu-seedance-video-pipeline -a cursorProject install goes to .agents/skills/; add -g for ~/.cursor/skills/.
$ gh skill install dracohu2025-cloud/draco-skills-collection feishu-seedance-video-pipeline --agent cursorProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/dracohu2025-cloud/draco-skills-collection.git skills-src && mkdir -p .cursor/skills && cp -r skills-src/feishu-seedance-video-pipeline .cursor/skills/feishu-seedance-video-pipeline && rm -rf skills-srcUse ~/.cursor/skills/ instead of .cursor/skills for a personal install.
Cursor skills documentation · loads skills from .cursor/skills/, .agents/skills/, .claude/skills/, .codex/skills/
Install the "feishu-seedance-video-pipeline" agent skill from https://github.com/dracohu2025-cloud/draco-skills-collection/tree/main/feishu-seedance-video-pipeline into .cursor/skills/feishu-seedance-video-pipeline/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "feishu-seedance-video-pipeline", then confirm the skill loads.Cursor copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ gemini skills install https://github.com/dracohu2025-cloud/draco-skills-collection.git --path feishu-seedance-video-pipeline--scope user (default) or --scope workspace; --path is the subfolder of the repo that holds the skill; --consent skips the security confirmation prompt.
$ npx skills add dracohu2025-cloud/draco-skills-collection --skill feishu-seedance-video-pipeline -a gemini-cliProject install goes to .agents/skills/; add -g for ~/.gemini/skills/.
$ gh skill install dracohu2025-cloud/draco-skills-collection feishu-seedance-video-pipeline --agent gemini-cliProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/dracohu2025-cloud/draco-skills-collection.git skills-src && mkdir -p .gemini/skills && cp -r skills-src/feishu-seedance-video-pipeline .gemini/skills/feishu-seedance-video-pipeline && rm -rf skills-srcUse ~/.gemini/skills/ instead of .gemini/skills for a personal install, then run /skills reload.
Gemini CLI skills documentation · loads skills from .gemini/skills/, .agents/skills/
Install the "feishu-seedance-video-pipeline" agent skill from https://github.com/dracohu2025-cloud/draco-skills-collection/tree/main/feishu-seedance-video-pipeline into .gemini/skills/feishu-seedance-video-pipeline/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "feishu-seedance-video-pipeline", then confirm the skill loads.Gemini CLI copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ gh skill install dracohu2025-cloud/draco-skills-collection feishu-seedance-video-pipelineInstalls for Copilot at project scope by default; add --scope user for a personal install. Preview a skill first with gh skill preview. Needs GitHub CLI 2.90.0 or later (public preview).
$ npx skills add dracohu2025-cloud/draco-skills-collection --skill feishu-seedance-video-pipeline -a github-copilotProject install goes to .agents/skills/; add -g for ~/.copilot/skills/.
$ git clone --depth 1 https://github.com/dracohu2025-cloud/draco-skills-collection.git skills-src && mkdir -p .github/skills && cp -r skills-src/feishu-seedance-video-pipeline .github/skills/feishu-seedance-video-pipeline && rm -rf skills-srcUse ~/.copilot/skills/ instead of .github/skills for a personal install. Commit .github/skills so cloud agent and code review can use it.
GitHub Copilot skills documentation · loads skills from .github/skills/, .claude/skills/, .agents/skills/
Install the "feishu-seedance-video-pipeline" agent skill from https://github.com/dracohu2025-cloud/draco-skills-collection/tree/main/feishu-seedance-video-pipeline into .github/skills/feishu-seedance-video-pipeline/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "feishu-seedance-video-pipeline", then confirm the skill loads.GitHub Copilot copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ npx skills add dracohu2025-cloud/draco-skills-collection --skill feishu-seedance-video-pipeline -a opencodeOpenCode documents no install command of its own. Project install goes to .agents/skills/; add -g for ~/.config/opencode/skills/.
$ gh skill install dracohu2025-cloud/draco-skills-collection feishu-seedance-video-pipeline --agent opencodeProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/dracohu2025-cloud/draco-skills-collection.git skills-src && mkdir -p .opencode/skills && cp -r skills-src/feishu-seedance-video-pipeline .opencode/skills/feishu-seedance-video-pipeline && rm -rf skills-srcUse ~/.config/opencode/skills/ instead of .opencode/skills for a personal install.
OpenCode skills documentation · loads skills from .opencode/skills/, .claude/skills/, .agents/skills/
Install the "feishu-seedance-video-pipeline" agent skill from https://github.com/dracohu2025-cloud/draco-skills-collection/tree/main/feishu-seedance-video-pipeline into .opencode/skills/feishu-seedance-video-pipeline/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "feishu-seedance-video-pipeline", then confirm the skill loads.OpenCode copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
feishu-seedance-video-pipelineA skill your agent uses when running a Feishu/Lark Base-centered Seedance video production pipeline: inspect/backfill the Base row, create/reference assets, build Row-24-style Chinese Seedance…
Feishu Seedance Video Pipeline is an agent skill from dracohu2025-cloud/draco-skills-collection. Use when running a Feishu/Lark Base-centered Seedance video production pipeline: inspect/backfill the Base row, create/reference assets, build Row-24-style Chinese Seedance prompts, submit/poll/download via Volcengine Seedance 2.0, QA the result, record tokens/cost, and keep asset lineage auditable.
Its SKILL.md is about 5k tokens, which your agent loads only when the skill is triggered. The skill folder holds 15 other files, including scripts, reference files and assets (for example `README.md`, `references/base-asset-ledger-write-guard.md` and `references/director-module.md`).
It sits in Media & Creative, covering Messaging and chat bots and AI video generation. It works with Seedance and Feishu (Lark). The licence is MIT.
3 steps, taken from the step headings in SKILL.md.
Read from SKILL.md and the folder at commit 26e8975. It shows what the files ask for, not the result of running them.
Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.
From allowed-tools in the SKILL.md frontmatter.
Ships 1 file in scripts/ (Python), which the agent can run.
Shell commands in SKILL.md call:
curljqffmpegffprobepython3From the folder's file list and the shell code blocks in SKILL.md.
Hosts in commands or code, which the agent is likely to contact:
ark.cn-beijing.volces.comFrom URLs in SKILL.md, links to its own repository left out.
Names these keys or tokens, usually read from environment variables:
VOLCENGINE_API_KEYFrom names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.
Feishu Seedance Video Pipeline loads about 5k tokens when it runs, and up to ~9.1k if it reads all its reference files. Until then it costs about 83 tokens; SKILL.md has 1,291 words of instructions outside code blocks.
Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.
The automated check noted patterns worth knowing about, such as sudo or a known installer.
sudo mkdir -p "$OUTDIR"sudo cp "$SRC" "$OUTDIR/reference.png"sudo chmod 644 "$OUTDIR/reference.png"- 不提交 `.env`、token 文件、原始签名 URL 响应。Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); the scripts in this folder are not scanned.
The full file from dracohu2025-cloud/draco-skills-collection at commit 26e8975, republished under its MIT licence (© dracohu2025-cloud). 1,291 words, ~4,989 tokens.
.claude/skills/feishu-seedance-video-pipeline/SKILL.md (or your agent's skills folder). This skill also uses 11 other files; get the full folder from GitHub.这是默认视频产线 skill:以飞书多维表格作为工作流、资产台账和审计源,以 Volcengine Seedance 2.0 作为视频生成引擎。
目标不是“生成一条视频”这么窄,而是完成一条可复用、可追溯、可验收的 Base 行:脚本、角色资产、场景资产、参考图、Prompt、payload、Task ID、成片、抽帧、QA、tokens、成本、Prompt_Output_Map 都在同一行或可追溯链路里。
核心是 Director Module + Base Asset Ledger Write Guard + Reviewer Module:Director 先判断主要角色、角色参考资产、场景环境资产、空间调度和镜头拆分;Base 写入只能由 asset_manifest.json 机械生成并回读审计;Reviewer 在任何 billable Seedance POST 前做强制门禁,核验 Base 物料台账、baseline Prompt 对照、reference_image 绑定、Dialogue Lock、Prompt/Payload 一致性和 ledger audit。Reviewer 不出 PASS_TO_SUBMIT,禁止提交。Seedance API 只是执行层。
Use when:
Do not use when:
doubao-seedance-2-0-260128(Seedance 2.0 标准版 / 非 Fast)。doubao-seedance-2-0-fast-260128。480p,控成本。16:9。record-get 复核。asset://... / Seedance 预置人像库时,不上传自生成 Character Reference Sheet 到该角色附件字段,避免误用。asset://... / Seedance 预置人像库时,必须生成 Wardrobe Reference image,并在每个 Seedance task 的 content[] 里作为 reference_image 输入;Prompt 中写全名 Wardrobe Reference image,不要写缩写。Base 行要回答:
content[] 顺序和 text prompt 是什么?常见字段:
角色参考图(CRS)_角色名_工具
角色参考图(CRS)_角色名_Prompt
输入资产_角色名参考图
角色名参考图_URL
场景环境设定参考图(SES)_工具
场景环境设定参考图(SES)_Prompt
输入资产_场景环境设定参考图
场景环境设定参考图_URL
动作参考图 / 关键帧参考图 / Top-Down Blocking Map_工具
动作参考图 / 关键帧参考图 / Top-Down Blocking Map_Prompt
输入资产_动作时序参考图
Prompt
Seedance视频_Prompt
Seedance Payload文件 / 视频生成 Payload文件
Prompt文件
Reference_URLs
Prompt_Output_Map
视频生成_计划参数
Seedance视频_TaskID
生成视频成片
质量检查抽帧图
QA摘要
Seedance视频_Tokens
Seedance视频_估算成本CNY
Local_Path
状态字段名会变。永远先 field-list,再写。
Director Module 是本 skill 的第一步,也是核心决策层。
职责:
配套文件:
references/director-module.md:Director 的职责、判断规则、输出 schema、门禁。references/base-asset-ledger-write-guard.md:Base 资产台账写入门禁;要求 asset_manifest.json、写入 payload、record-get 回读和机械审计。references/reviewer-module.md:Reviewer 强制门禁;任何 billable Seedance POST 前必须产出并上传 reviewer_report.md,结论为 PASS_TO_SUBMIT 才能提交。templates/director-decision-output-template.md:每次处理新 Base 行时先填写的结构化导演方案模板。templates/reviewer-report-template.md:Reviewer 报告模板。工作原则:先产出 director_plan,再写 CRS/SES/Seedance Prompt。没有导演方案就直接写 Prompt,视为不合格。
Reviewer Module 是提交前强制审查层。它不负责创作,只负责拦截不符合流程和规范的任务。
位置:Director Module、物料准备、Prompt/Payload 构建之后;任何 billable Seedance POST 之前。
硬规则:
reviewer_report.md,上传到当前 Base 行。PASS_TO_SUBMIT 或 BLOCKED。BLOCKED 时禁止提交 Seedance。详见:references/reviewer-module.md 与 templates/reviewer-report-template.md。
BASE='[REDACTED]'
TABLE='[REDACTED]'
REC='rec...'
WORK=/tmp/feishu_seedance_pipeline
mkdir -p "$WORK"
lark-cli base +field-list --as user \
--base-token "$BASE" --table-id "$TABLE" \
> "$WORK/fields.json"
lark-cli base +record-get --as user \
--base-token "$BASE" --table-id "$TABLE" --record-id "$REC" \
> "$WORK/record.json"读取 data.record,不要假设一定有 data.record.fields。当前 lark-cli base +record-get 可能直接把字段平铺在 data.record 根部(例如 data.record["Prompt文件"]),而不是放在 data.record.fields。Reviewer/回填验证脚本应兼容两种结构:优先 data.record.fields,不存在时把 data.record 本身当字段 dict。field-list 的字段数组在当前 lark-cli 版本里是 data.fields,不是 data.items。
如果用户给的是“基于某行/刚才那条视频”的新脚本,先 record-get 旧行复用可用资产,再创建新行。新行创建可用:
lark-cli base +record-batch-create --as user \
--base-token "$BASE" --table-id "$TABLE" \
--json '{"fields":["资产名","Prompt"],"rows":[["...","..."]]}'返回里不一定有标准 records[]。优先从 data.record_id_list[0] 读取新行 ID;不要用宽松正则直接抓第一个 rec...,会误抓到正文里的 record 字样。随后立刻写入 record_id.txt 并 record-get 复核。
每个主角独立 Character Reference Sheet,除非需求明确要群像表。
本 skill 自带五份可直接调用的模板/参考文件:
references/director-module.md:Director Module 规则,判断主要角色、Character Reference Sheet / Scene, Environment, and Settings reference image / Top-Down Blocking Map / keyframes 需求、镜头拆分和门禁。templates/director-decision-output-template.md:处理每条 Base 行时先填写的结构化 director_plan 模板。templates/character-reference-sheet-prompt-template.md:主要角色 Character Reference Sheet 完整 Prompt 模板与 QA 门禁。templates/scene-environment-settings-prompt-template.md:Scene, Environment, and Settings reference image 完整 Prompt 模板与 QA 门禁。templates/seedance-row24-director-template.md:符合第24行/第28行风格的 Seedance 中文细导演稿模板,要求逐秒时间轴绑定镜头语言、空间锚点、动作、对白/音效。先填写 Director director_plan,再生成或复用具体资产。
Character Reference Sheet 要点:
Scene, Environment, and Settings reference image 要点:
Top-Down Blocking Map / keyframes:
Top-Down Blocking Map,不要只写 TDBM。Seedance API 需要可访问 URL 或 asset://...。
本机 HTTPS 发布例:
SRC="/path/to/reference.png"
OUTDIR="/var/www/example.com/media/seedance"
sudo mkdir -p "$OUTDIR"
sudo cp "$SRC" "$OUTDIR/reference.png"
sudo chmod 644 "$OUTDIR/reference.png"
curl -L -s -o /tmp/ref_check.png -w '%{http_code} %{content_type} %{size_download}\n' \
"https://example.com/media/seedance/reference.png"Feishu 附件上传要求相对路径:
cd /path/to/files
lark-cli base +record-upload-attachment --as user \
--base-token "$BASE" --table-id "$TABLE" --record-id "$REC" \
--field-id '输入资产_场景环境设定参考图' \
--file scene_reference.pngEndpoint:
POST https://ark.cn-beijing.volces.com/api/v3/contents/generations/tasks
GET https://ark.cn-beijing.volces.com/api/v3/contents/generations/tasks/{task_id}
Authorization: Bearer <VOLCENGINE_API_KEY>
Content-Type: application/jsonContent rules:
{"type":"text","text":"完整中文Prompt"}
{"type":"image_url","image_url":{"url":"https://..."},"role":"reference_image"}
{"type":"image_url","image_url":{"url":"asset://asset-id"},"role":"reference_image"}
{"type":"image_url","image_url":{"url":"https://..."},"role":"first_frame"}
{"type":"image_url","image_url":{"url":"https://..."},"role":"last_frame"}不要在 API Prompt 里写 @Image1。参考图绑定靠 content[] 顺序和 role。
默认 payload:
{
"model": "doubao-seedance-2-0-260128",
"duration": 12,
"resolution": "480p",
"ratio": "16:9",
"generate_audio": true,
"return_last_frame": false,
"content": [
{"type": "text", "text": "..."},
{"type": "image_url", "image_url": {"url": "https://example.com/character_or_scene.png"}, "role": "reference_image"}
]
}提交:
curl -sS -X POST https://ark.cn-beijing.volces.com/api/v3/contents/generations/tasks \
-H "Authorization: Bearer ${VOLCENGINE_API_KEY}" \
-H "Content-Type: application/json" \
-d @payload.json | tee post_response.json | jq .轮询:
TASK_ID="cgt-..."
for i in {1..80}; do
curl --max-time 30 -sS "https://ark.cn-beijing.volces.com/api/v3/contents/generations/tasks/${TASK_ID}" \
-H "Authorization: Bearer ${VOLCENGINE_API_KEY}" \
| tee result_latest.json | jq '{status, usage, error, content}'
sleep 15
donePrompt 不合格的典型病:只写剧情,不写镜头。
必须包含:
content[] 顺序说明,不写 @Image。generate_audio=true/false,对白逐句列明。时间轴模板:
[0-2秒] 镜头构图与运镜:中广角建立镜头,镜头从桌面空间锚点轻轻推近。角色动作:...。空间锚点/道具连续性:...。对白/音效:...。
[2-4秒] 镜头构图与运镜:切到中近景,保持角色左右关系不变。角色动作:...。空间锚点/道具连续性:...。对白/音效:...。提交前复核:
Prompt == Seedance视频_Prompt == payload.content[0].text == 本地 prompt 文件
模型 == doubao-seedance-2-0-260128,除非用户另指定 Fast 或 A/B 对比
无 @Image
无未解释 CRS / SES / TDBM / BGM 等缩写
每个时间段都有镜头语言
未擅自提交生成当项目输入给出“多幕/多场/短剧”脚本,且要求每一幕独立生成时:
asset://... / Seedance 预置人像库时,必须生成 Wardrobe Reference image:同一套服饰的前/侧/后/细节多视图;每个 act / 每个 Seedance task 的 content[] 都必须加入这张 Wardrobe Reference image 作为 reference_image。Prompt 中写全名 Wardrobe Reference image,不要写缩写。Seedance视频_TaskID 写入多行清单,例如 Act1: cgt-...\nAct2: cgt-...。生成视频成片 字段上传每幕成片和最终拼接成片;质量检查抽帧图 字段上传每幕抽帧和最终抽帧。飞书附件字段会 append,同一字段多次上传是预期行为。Seedance视频_Tokens 和 Seedance视频_估算成本CNY 写合计值;QA摘要里列每幕 tokens/成本和总计。Reference_URLs 字段实测会把 JSON/多 URL 字符串自动改写成奇怪 Markdown 链接;优先写短说明或资产标签清单,真实可访问 URL 以 payload 文件、附件和本地 reference_urls.json 为准。Prompt_Output_Map 必须说明:共享 CRS/SES、每幕 payload 文件、每幕 Task ID、最终拼接文件。find(..., previous_pos+1),不要用 prompt.index() 列表,否则重复句会误判顺序失败。多幕拼接命令:
cat > "$WORK/run/concat_list.txt" <<EOF
file '$WORK/run/act1.mp4'
file '$WORK/run/act2.mp4'
EOF
ffmpeg -y -hide_banner -loglevel error \
-f concat -safe 0 -i "$WORK/run/concat_list.txt" \
-c copy "$WORK/run/final_concat.mp4" \
|| ffmpeg -y -hide_banner -loglevel error \
-f concat -safe 0 -i "$WORK/run/concat_list.txt" \
-c:v libx264 -c:a aac -movflags +faststart "$WORK/run/final_concat.mp4"任何 billable POST 前:
record-list / record-get 拉取该行 Prompt、Prompt文件、Prompt_Output_Map、Reference_URLs 和资产附件,对比后再写新 Prompt;不得凭记忆重写。未使用。record-get 已复核。asset_manifest.json、写入 payload、post-write record-get、audit result,且结果为 PASS。result=PASS_TO_SUBMIT 才允许提交。长任务不要静默。用后台进程跑,阶段性 poll。
task_id,并立即回填 Base 的 Task ID / 状态,避免上下文断掉后丢任务。post_response.json / result_latest.json / Task ID;不要重复 POST。curl --max-time 30 循环验证;不要盲等。原因可能是 stdout 缓冲、网络请求卡住、Hermes process preview 没刷新,或 urllib 请求没有按预期输出。running 且无 error,不要判失败、不要重复 POST。可以把轮询改成后台进程 + notify_on_complete,同时定期 record-get/状态回填,避免用户以为卡死。result_*_latest.json 的 mtime/status,或用 process log / ps 验证。必要时让脚本同时写日志文件。succeeded 后下载卡住,使用已保存 video_url 做 Range 分片下载;不要重提。QA:
ffprobe -v error -show_entries stream=index,codec_type,width,height,duration,nb_frames,r_frame_rate,codec_name -of json output.mp4 > ffprobe.json
ffmpeg -y -hide_banner -loglevel error -i output.mp4 -vf "fps=1,scale=240:-1,tile=5x3" -frames:v 1 contact_sheet.jpg对白/配音自动转写检查【暂时禁用】:
# 暂停使用 Whisper 做中文对白 QA:中文识别准确率不足,容易误判。
# 仅保留音轨存在性/规格检查;对白内容以人工验收为准。
# 未来替换为更高准确率模型后再恢复自动转写模块。
# ffmpeg -y -hide_banner -loglevel error -i output.mp4 -vn -ac 1 -ar 16000 audio.wav
# whisper audio.wav --language Chinese --model tiny --output_dir whisper --output_format txt --fp16 FalseQA 摘要至少包括:
QA 注意:
browser_navigate 在本 VM 可能因 Chrome sandbox 失败;需要可视检查时可先用 agent-browser open <file-or-url> --args "--no-sandbox" 打开,再截图/人工核对。成功任务 GET 结果通常含 usage.total_tokens。
公开估算口径:
46 元 / 1,000,000 tokens;含视频输入约 28 元 / 1,000,000 tokens。37 元 / 1,000,000 tokens;含视频输入约 22 元 / 1,000,000 tokens。video_url。TOKENS=$(jq -r '.usage.total_tokens // .usage.completion_tokens' result_final.json)
MODEL=$(jq -r '.model // "doubao-seedance-2-0-260128"' payload.json)
HAS_VIDEO=$(jq '[.content[]? | select(.type == "video_url")] | length > 0' payload.json)
python3 - <<PY
tokens = int("$TOKENS")
model = "$MODEL"
has_video = "$HAS_VIDEO" == "true"
rate = 22 if ("fast" in model and has_video) else 37 if "fast" in model else 28 if has_video else 46
print(f"tokens={tokens:,}")
print(f"rate_cny_per_million={rate}")
print(f"estimated_cost_cny={tokens * rate / 1_000_000:.2f}")
PYlark-cli base +record-batch-update 用:
{"record_id_list":["rec..."],"patch":{"字段名":"值"}}不要用 records: [{record_id, fields}]。
写入后必须 record-get 复核,尤其是长 Prompt、附件、Prompt_Output_Map。
附件清理坑:Feishu Base 附件字段不要乱 PATCH 清空/去重。实测可能 append 成 double,或被 MOBILE_ONLY / UploadAttachNotAllowed 拒绝。删除旧附件优先让用户在 UI 手动删;未复核消失前,不要声称已删除。
asset://... 官方/授权人像素材。asset://... / Seedance 预置人像库时,必须先生成 Wardrobe Reference image;每个相关 Seedance task 都把它作为 reference_image 输入。[REDACTED]。.env、token 文件、原始签名 URL 响应。把产线拆成“API 调用”和“补表”两件事。 实战里它们是一条链:不补表就烧钱,后面必丢上下文。
短 Prompt 直接提交。 此工作流建议是 Row-24 式细导演稿。
时间轴只写动作。 每段都要有镜头语言。
重复 POST。 超时先查 Task ID;有 ID 就只 poll/download。
把 @Image1 写进 API Prompt。 API 不认这个绑定。
官方 asset 角色又上传自生成 CRS。 后续容易误用。
用 API 硬删附件。 容易 append 或失败,先 UI。
标准版误计 Fast 成本,或 Fast payload 却写标准版计划参数。 模型、payload、Base、成本必须同步。
不要基于 Whisper 中文转写自动判废。 Whisper 中文准确率不足;配音/对白内容先由人工验收,未来替换更高准确率模型后再恢复自动门禁。
field-list / record-get JSON 结构猜错。 当前 lark-cli 返回字段列表在 data.fields,不是 data.items;record-get 可能把业务字段直接平铺在 data.record 根部,不一定有 data.record.fields。Reviewer脚本必须兼容两种结构,否则会误判 Base Material Ledger / Pre-submit State 为 BLOCKED。
record-batch-create 新行 ID 抓错。 响应里优先读 data.record_id_list[0];不要 regex 抓第一个 rec...,可能抓到普通英文单词前缀导致 record-get 失败。
Reference_URLs 多行字符串被飞书当成奇怪 Markdown 链接。 实测 JSON 字符串也可能被自动改写成 [...](http://...)。Base 里优先写短说明/资产标签,真实 URL 放 payload 附件、本地 reference_urls.json 或审计文件。
后台轮询无输出还一直等。 先杀掉,改前台 curl --max-time 30 loop 复核任务状态;成功后再单独下载视频。
字段名混淆。 当前表使用 Seedance视频_Tokens / Seedance视频_估算成本CNY,不是 视频生成_Tokens / 视频生成_估算成本CNY;写入前用 field-list 复核。
视觉 QA 失败就编结论。 provider 403 时只说明自动视觉审查失败,回填抽帧并请求人工验收。
多幕 Dialogue Lock 用 index() 验证重复台词。 同一幕可能重复同一句台词;用 prompt.index() 会每次返回第一次出现的位置,导致误判顺序失败。用顺序扫描:pos=-1; pos=prompt.find(quote, pos+1)。
多幕短剧拆成多行。 需求要求共享 CRS/SES 的短剧测试时,默认同一 Base 行;每幕作为独立 Task/附件记录在同一行,最终拼接视频也回填同一行。
官方/预置人像未加 Wardrobe Reference image。 官方 asset:// 负责身份,不负责稳定服装;任何使用 Seedance 预置人像库的任务都必须生成 Wardrobe Reference image,并在每个 payload 作为 reference_image 传入。Prompt 里写全称,不写缩写。
需求中出现“开始生产”就直接烧。 错。开始生产不等于跳过物料门禁;必须先把当前行的 CRS / Scene, Environment, and Settings reference image / Wardrobe Reference image / Prompt文件 / Payload文件 / Prompt_Output_Map 补齐并 record-get 复核。若当前行的输入资产字段为空,不得提交 Seedance。
没有 Reviewer 报告就提交。 严重违规。任何新视频必须在提交前产出并上传 reviewer_report.md;没有 PASS_TO_SUBMIT 就不得 POST。Reviewer 必须核验当前 Base 行物料台账、baseline 对照、Prompt/Payload一致性、Dialogue Lock 和附件回填。
director_plan。asset://... / Seedance 预置人像库,已生成 Wardrobe Reference image,并已作为每个相关 Seedance task 的 reference_image。asset_manifest.json 或等价清单;Base 写入由清单机械生成,不是手写摘要。record-get 与 ledger audit,结果 PASS;Reviewer 报告引用 manifest / write payload / record-get / audit result。Prompt == Seedance视频_Prompt == payload.content[0].text == 本地 prompt 文件。reviewer_report.md,结果为 PASS_TO_SUBMIT;若为 BLOCKED,不得提交。© dracohu2025-cloud, MIT. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file
SKILL.md and 11 other files (scripts, references, assets) in feishu-seedance-video-pipeline of dracohu2025-cloud/draco-skills-collection.
Open the folder on GitHubat commit 26e8975
Feishu Seedance Video Pipeline next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.
| Skill | Stars | Used in | Tokens | Auto-check | Licence | Repo updated |
|---|---|---|---|---|---|---|
| Feishu Seedance Video Pipeline this skilldracohu2025-cloud/draco-skills-collection | 227 | — | ~5k | Automated safety check: Notes | MIT | |
| Libtv Videonexu-io/nexu | 3.3k | — | ~3.5k | Automated safety check: Pass | MIT | |
| FeedgrabiBigQiang/feedgrab | 614 | — | ~2k | Automated safety check: Pass | MIT | |
| Seedance Storyboard Generatorliangdabiao/Seedance2-Storyboard-Generator | 2.6k | — | ~2.2k | Automated safety check: Pass | None | |
| Analyze Videokrusemediallc/arcads-claude-code | 1.6k | — | ~4.2k | Automated safety check: Pass | MIT | |
| Imagesmixs/visual-skills | 494 | — | ~2.1k | Automated safety check: Pass | CC-BY-4.0 |
nexu-io/nexu
Seedance 2.0 video & image generation via LibTV Gateway - AI text-to-video, image-to-video, video continuation, style transfer, and text-to-image using Seedance 2.0 model.
iBigQiang/feedgrab
Universal content grabber — fetch any URL and return structured Markdown.
liangdabiao/Seedance2-Storyboard-Generator
专业的Seedance 2.0平台AI视频脚本和分镜生成器。当用户要求:(1) 将文章/故事转换为视频脚本,(2) 生成Seedance 2.0分镜提示词,(3) 规划多集AI视频系列,(4) 为GPT-Image-2、Seedream、Nano Banana…
krusemediallc/arcads-claude-code
Analyze a reference video and reverse-engineer its style into a reusable Seedance 2.0 prompting template.
smixs/visual-skills
Image prompting skill for Nano Banana (NBP/NB2) and GPT Image 2.5 (Flare/Sunburst).
cclank/lanshu-waytovideo
使用剪映(Jianying/小云雀)的 Seedance 2.0 模型自动生成AI视频。支持文生视频(T2V)、图生视频(I2V)、参考视频生成(V2V)和向后延伸(Extend)四种模式。当用户需要生成AI视频、使用Seedance模型创作短片、基于参考图像/视频进行风格转换,或对已有结果继续延长时使用此技能。需要预先配置 cookies.json 登录凭证。
dracohu2025-cloud/draco-skills-collection
A skill your agent uses when parsing a DingTalk daily news document and syncing title, date, Top 3, summary, conclusion, document link, and status into a DingTalk multidimensional table.
dracohu2025-cloud/draco-skills-collection
Production pipeline for Motion Canvas — TypeScript-based programmatic vector animation with real-time preview.
dracohu2025-cloud/draco-skills-collection
A skill your agent uses when collecting public news candidates from Hacker News, GitHub Trending, Hugging Face papers, and other web sources for briefings and daily reports.
dracohu2025-cloud/draco-skills-collection
A skill your agent uses when mining Open Design visual systems and converting them into Open Slide React template kits or standalone 20-page template albums, with build, deployment, screenshot QA…
dracohu2025-cloud/draco-skills-collection
A skill your agent uses when generating a daily 24h AI / Agent / AIGC top-news briefing, publishing it as a DingTalk document, validating it, and optionally archiving it to a DingTalk…
dracohu2025-cloud/draco-skills-collection
可独立运行的 GPT-Image 增强版 EPUB2Podcast:在本地把 EPUB 转成双人中文音频、GPT-Image/Smart Slide 视觉页、最终 MP4,并生成 YouTube 发布素材。
Works with
A skill your agent uses when running a Feishu/Lark Base-centered Seedance video production pipeline: inspect/backfill the Base row, create/reference assets, build Row-24-style Chinese Seedance…. Feishu Seedance Video Pipeline is an agent skill from dracohu2025-cloud/draco-skills-collection.0, QA the result, record tokens/cost, and keep asset lineage auditable.
Feishu Seedance Video Pipeline fits situations like: running a Feishu/Lark Base-centered Seedance video production pipeline: inspect/backfill the Base row; create/reference assets; build Row-24-style Chinese Seedance prompts; submit/poll/download via Volcengine Seedance 2.0.
Run `npx skills add dracohu2025-cloud/draco-skills-collection --skill feishu-seedance-video-pipeline -a claude-code`. Or copy the skill folder (feishu-seedance-video-pipeline in dracohu2025-cloud/draco-skills-collection) into .claude/skills/feishu-seedance-video-pipeline in your project. Claude Code loads it when a task matches its description.
Run `npx skills add dracohu2025-cloud/draco-skills-collection --skill feishu-seedance-video-pipeline -a codex`. Or copy the skill folder (feishu-seedance-video-pipeline in dracohu2025-cloud/draco-skills-collection) into .agents/skills/feishu-seedance-video-pipeline in your project. Codex loads it when a task matches its description.
Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add dracohu2025-cloud/draco-skills-collection --skill feishu-seedance-video-pipeline -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/feishu-seedance-video-pipeline, .gemini/skills/feishu-seedance-video-pipeline, .github/skills/feishu-seedance-video-pipeline and .opencode/skills/feishu-seedance-video-pipeline in your project.
Going by SKILL.md and its folder, Feishu Seedance Video Pipeline needs Python for the scripts in its folder, the command-line tools its instructions call (curl, jq, ffmpeg, ffprobe and python3) and credentials named VOLCENGINE_API_KEY. Our summary lists: Python 3; A credential in VOLCENGINE_API_KEY.
SKILL.md names 1 domain. In commands or code: ark.cn-beijing.volces.com; the agent is likely to contact it when it follows the instructions. This is read from the text; nothing was executed.
Our automated static check of SKILL.md found notes only (runs commands with sudo; mentions a .env file), nothing it rates as a warning. It is not a guarantee. The check reads SKILL.md only: the scripts in the folder are not scanned, so read them before running anything.
Feishu Seedance Video Pipeline is published under the MIT licence (declared in SKILL.md). It allows redistribution, so the full SKILL.md is shown on this page.
About 5k tokens (SKILL.md is roughly 20k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full. Its references folder adds about 4.1k tokens, read only when the agent opens those files.
Skills that share tags, products or a category with Feishu Seedance Video Pipeline: Libtv Video (nexu-io/nexu, 3.3k stars), Feedgrab (iBigQiang/feedgrab, 614 stars), Seedance Storyboard Generator (liangdabiao/Seedance2-Storyboard-Generator, 2.6k stars) and Analyze Video (krusemediallc/arcads-claude-code, 1.6k stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.
dracohu2025-cloud (a GitHub user) maintains it in dracohu2025-cloud/draco-skills-collection, which has 227 GitHub stars. The repository holds 27 skills in this directory. The repository was last updated on September 17, 2026.
Source: dracohu2025-cloud/draco-skills-collection on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.