VRGDG H3 Short Film Pipeline
vrgamegirl19/comfyui-vrgamedevgirl
Builds an AI short film in a local ComfyUI with the VRGDG Video Builder, MiniMax H3 scenes, reference images, a music score, QA and a final edit.
Turns a script into a storyboard, assigns MiniMax H3 workflows in ComfyUI, monitors batch generation and builds a Jianying draft of the finished video.
SKILL.md written in Chinese; this summary is our English description.
Install MiniMax H3 Video Director
$ npx skills add TFboy1/oh-my-minimaxh3-director --skill oh-my-minimaxh3-director -a claude-codeProject install by default; add -g for ~/.claude/skills/.
$ gh skill install TFboy1/oh-my-minimaxh3-director oh-my-minimaxh3-director --agent claude-codeProject scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).
Claude Code skills documentation · loads skills from .claude/skills/
Install the "oh-my-minimaxh3-director" agent skill from https://github.com/TFboy1/oh-my-minimaxh3-director/tree/master into .claude/skills/oh-my-minimaxh3-director/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "oh-my-minimaxh3-director", 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.
$ npx skills add TFboy1/oh-my-minimaxh3-director --skill oh-my-minimaxh3-director -a codexProject install goes to .agents/skills/; add -g for ~/.codex/skills/.
$ gh skill install TFboy1/oh-my-minimaxh3-director oh-my-minimaxh3-director --agent codexProject scope by default (.agents/skills/); add --scope user for a personal install.
Codex skills documentation · loads skills from .agents/skills/
Install the "oh-my-minimaxh3-director" agent skill from https://github.com/TFboy1/oh-my-minimaxh3-director/tree/master into .agents/skills/oh-my-minimaxh3-director/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "oh-my-minimaxh3-director", 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 TFboy1/oh-my-minimaxh3-director --skill oh-my-minimaxh3-director -a cursorProject install goes to .agents/skills/; add -g for ~/.cursor/skills/.
$ gh skill install TFboy1/oh-my-minimaxh3-director oh-my-minimaxh3-director --agent cursorProject scope by default (.agents/skills/); add --scope user for a personal install.
Cursor skills documentation · loads skills from .cursor/skills/, .agents/skills/, .claude/skills/, .codex/skills/
Install the "oh-my-minimaxh3-director" agent skill from https://github.com/TFboy1/oh-my-minimaxh3-director/tree/master into .cursor/skills/oh-my-minimaxh3-director/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "oh-my-minimaxh3-director", 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.
$ npx skills add TFboy1/oh-my-minimaxh3-director --skill oh-my-minimaxh3-director -a gemini-cliProject install goes to .agents/skills/; add -g for ~/.gemini/skills/.
$ gh skill install TFboy1/oh-my-minimaxh3-director oh-my-minimaxh3-director --agent gemini-cliProject scope by default (.agents/skills/); add --scope user for a personal install.
Gemini CLI skills documentation · loads skills from .gemini/skills/, .agents/skills/
Install the "oh-my-minimaxh3-director" agent skill from https://github.com/TFboy1/oh-my-minimaxh3-director/tree/master into .gemini/skills/oh-my-minimaxh3-director/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "oh-my-minimaxh3-director", 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 TFboy1/oh-my-minimaxh3-director oh-my-minimaxh3-directorInstalls 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 TFboy1/oh-my-minimaxh3-director --skill oh-my-minimaxh3-director -a github-copilotProject install goes to .agents/skills/; add -g for ~/.copilot/skills/.
GitHub Copilot skills documentation · loads skills from .github/skills/, .claude/skills/, .agents/skills/
Install the "oh-my-minimaxh3-director" agent skill from https://github.com/TFboy1/oh-my-minimaxh3-director/tree/master into .github/skills/oh-my-minimaxh3-director/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "oh-my-minimaxh3-director", 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 TFboy1/oh-my-minimaxh3-director --skill oh-my-minimaxh3-director -a opencodeOpenCode documents no install command of its own. Project install goes to .agents/skills/; add -g for ~/.config/opencode/skills/.
$ gh skill install TFboy1/oh-my-minimaxh3-director oh-my-minimaxh3-director --agent opencodeProject scope by default (.agents/skills/); add --scope user for a personal install.
OpenCode skills documentation · loads skills from .opencode/skills/, .claude/skills/, .agents/skills/
Install the "oh-my-minimaxh3-director" agent skill from https://github.com/TFboy1/oh-my-minimaxh3-director/tree/master into .opencode/skills/oh-my-minimaxh3-director/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "oh-my-minimaxh3-director", 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.
Facts
- Skill name
oh-my-minimaxh3-director- GitHub stars
- 143
- Token cost
- ~2.4k tokens
- SKILL.md length
- 604 words
- Files
- 37 (incl. scripts, references, assets)
- Skills in repo
- 1
- Repo updated
- First seen
- Licence
- MIT
At a glance
Turns a script into a storyboard, assigns MiniMax H3 workflows in ComfyUI, monitors batch generation and builds a Jianying draft of the finished video.
- Works in 5 steps: 自动更新：说明用途（每次使用前自动拉取本 skill 的最新脚本、模板与 → ComfyUI 位置确认：询问用户“ComfyUI 是本地运行还是云端？”—— → 硬件评估：运行 scripts/check_hardware.py 检测… → …
- Turning a screenplay or story text into a generated video
- SKILL.md covers Overview, 首次使用引导（Onboarding）, 输入与项目结构 and 阶段 0：探测 ComfyUI 与可选隧道, plus 8 more sections
- Calls python and npx
What it does
Written in Chinese, this skill automates a script-to-video chain built on ComfyUI. The agent splits a script into segments, with each segment becoming one H3 video holding two or three shots, and assigns a MiniMax H3 template to each segment: Ref2VA, T2V or I2V.
After you confirm the parameters, it submits jobs in batches, monitors and downloads the results, then generates a Jianying assembly script and creates a draft. Prompts can follow the official H3 six-part format, a Chinese director-style mode called wenwu, or a hybrid of the two. Project files such as storyboard.md and storyboard.json go under an outputs folder.
First use is an onboarding conversation: an auto-update choice, whether ComfyUI runs locally or in the cloud (including through a Cloudflare tunnel), a hardware check of NVIDIA VRAM, memory and disk, and guidance for downloading the MiniMax H3 model, about 40GB. A GPU under 12GB or no NVIDIA GPU leads to a suggestion to use a cloud GPU. Version one works at segment level, with no subtitles or background music by default, and runs on Windows.
When your agent uses it
- Turning a screenplay or story text into a generated video
- Producing a game trailer or ad short from a script and a style brief
- Planning shot-to-shot transitions and character reference views before generation
- Batch-generating H3 clips and assembling them into a Jianying draft
Example prompts
- “Take script.md and produce a storyboard, then assign H3 workflows for each segment.”
- “Use hybrid prompt mode for the game trailer script in ./story/trailer.md.”
- “Check whether my GPU is enough to run MiniMax H3 locally, or whether I should use a cloud ComfyUI.”
- “Assemble the finished H3 clips into a Jianying draft.”
Requirements
- ComfyUI running locally or in the cloud
- MiniMax H3 model files, about 40GB
- An NVIDIA GPU with at least 12GB of VRAM, or a remote GPU
- Windows with Python and a project .venv
Workflow steps
5 steps, taken from the first numbered list in SKILL.md.
- 自动更新：说明用途（每次使用前自动拉取本 skill 的最新脚本、模板与
- ComfyUI 位置确认：询问用户“ComfyUI 是本地运行还是云端？”——
- 硬件评估：运行 scripts/check_hardware.py 检测 NVIDIA 显存、内存与
- MiniMax H3 模型：本地 ComfyUI 且未确认过模型时，询问“是否已下载
- 可选依赖：逐个说明用途后询问是否安装——
What it can do on your machine
Read from SKILL.md and the folder at commit 9112661. It shows what the files ask for, not the result of running them.
Tool permissions
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.
Runs code
Ships 1 file in scripts/, which the agent can run.
Shell commands in SKILL.md call:
pythonnpx
From the folder's file list and the shell code blocks in SKILL.md.
Network
Links to these hosts (documentation or services it may open):
comfy.org
From URLs in SKILL.md, links to its own repository left out.
Credentials
Names no API keys, tokens, secrets or passwords.
From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.
Context cost
MiniMax H3 Video Director loads about 2.4k tokens when it runs, and up to ~23k if it reads all its reference files. Until then it costs about 79 tokens; SKILL.md has 604 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.
Safety
Auto-check passed
The automated check found no risky patterns in SKILL.md.
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.
SKILL.md
The full file from TFboy1/oh-my-minimaxh3-director at commit 9112661, republished under its MIT licence (© TFboy1). 604 words, ~2,431 tokens.
Download SKILL.mdSave it as
.claude/skills/oh-my-minimaxh3-director/SKILL.md (or your agent's skills folder). This skill also uses 36 other files; get the full folder from GitHub.- name
- oh-my-minimaxh3-director
- description
- AI 视频导演流水线：剧本自动分镜（含 WenWu 导演级镜头衔接与人物四视图参考图）→ MiniMax H3 工作流分配与参数确认 → 批量生成监控 → 剪映自动拼合成片，可选 Cloudflare 隧道远程访问。提示词支持官方 H3 六段式（official）、WenWu 中文导演模式（wenwu），以及两者兼顾的 hybrid 模式（官方六段式外壳 + WenWu 导演级内容）。用户提到剧本出片、自动分镜、分镜衔接、WenWu、人物四视图、H3 生成、工作流分配、剪映拼合、AI 导演、游戏 PV/宣传片/广告短片/风格与音乐简报，或给出剧本/故事文本要求生成视频时使用。
ComfyUI 剧本视频流水线
Overview
本 skill 把一条完整链路自动化：探测/打通 ComfyUI（可选 Cloudflare 隧道）→ 剧本自动分镜（每段一个 H3 视频，段内 2-3 个分镜）→ 按段分配 H3 模板并覆盖 参数 → 参数确认 → 批量提交/监控/下载 → 生成剪映拼合脚本并建草稿。
v1 只服务 MiniMax H3 的 Ref2VA / T2V / I2V 三种模板，段级生成，默认无字幕无
BGM。所有文件 UTF-8，Python 使用项目 .venv（存在时），Windows 下运行。
首次使用引导（Onboarding）
第一次使用（以用户目录下的 ~/.oh-my-minimaxh3-director.json 是否存在判断）
时，按下面顺序逐项与用户确认，每一项先说明用途再询问，不替用户做决定；
完成后把选择写入该配置文件。非首次且用户启用自动更新时，每次使用前先
npx skills update oh-my-minimaxh3-director -g -y（失败不阻塞流程）。
- 自动更新：说明用途（每次使用前自动拉取本 skill 的最新脚本、模板与 教程），询问是否启用。推荐启用；选项：启用 / 不启用 / 每次询问。
- ComfyUI 位置确认：询问用户“ComfyUI 是本地运行还是云端？”——
- 本地：运行
scripts/probe_comfy.py --workspace <工作区根> --write探测本机 8188；未安装时按 setup-guide.md 第 1 节给出官网桌面版下载教程（https://www.comfy.org/download）。 - 云端：请用户提供网址链接（trycloudflare 隧道地址或 AutoDL 的
seetacloud 地址；还没有 AutoDL 实例时按
autodl-cloud.md 帮其开通云端算力），用
scripts/probe_comfy.py --workspace <工作区根> --url <链接> --write验证并把该地址写入.config/pipeline-config.json的base_url； 链接失效时回退到自动探测。 把comfyui_location: local | cloud与地址记入~/.oh-my-minimaxh3-director.json，后续运行不再重复询问。
- 本地：运行
- 硬件评估：运行
scripts/check_hardware.py检测 NVIDIA 显存、内存与 磁盘；本机无独显但已有远程 ComfyUI（隧道/AutoDL）时改用--remote-url <地址>直接评估远程 GPU。若判定不适合（显存 < 12GB 或 没有 NVIDIA GPU），明确告知用户，并询问是改用云 GPU（AutoDL，教程见 autodl-cloud.md / Comfy Cloud）还是仍想 尝试；远程可用时流水线不受影响。 - MiniMax H3 模型：本地 ComfyUI 且未确认过模型时，询问“是否已下载
MiniMax H3 模型”；未下载则按 setup-guide.md
第 2 节给 ModelScope 教程（仓库
Comfy-Org/minimax-H3，含文件清单、 放置目录与约 40GB 空间提示）。 - 可选依赖：逐个说明用途后询问是否安装——
h3-prompt-writing：专业改写 H3 六段式提示词（仅服务 official 与 hybrid 的六段式外壳；wenwu 纯中文模式不依赖它）；jianying-editor：剪映草稿自动拼合（本 skill 拼合阶段依赖）；comfy-mcp：让 Codex 直接管理/运行 ComfyUI，可选增强。 需要安装时按 setup-guide.md 第 4 节执行； 明确拒绝的记录下来，之后不再反复询问。
输入与项目结构
输入：剧本文件（.md / .txt，或对话中的故事文本），可选参考图目录
（refs/）与角色设定。输出项目放在 outputs/<项目名>/（或用户指定目录）：
<项目>/
storyboard.md # 人读分镜
storyboard.json # 机器可读分镜（build_workflows 输入）
prompts/seg_01.txt … # H3 提示词（official / wenwu / hybrid）
workflows/seg_01_api.json …
jobs/params.json # 参数清单（确认对象）
jobs/seg_01_job.json … # 提交/下载状态（断点恢复）
clips/raw/<batch>/seg_01_00001-audio.mp4 …
assemble_jianying_<标题>.py # 生成的拼合业务脚本（放项目根）
edit/jianying-draft-report.json分镜 JSON 结构与提示词格式见 storyboard-schema.md；模板路由与字段映射见 workflow-routing.md；隧道操作见 cloudflared.md；首次安装与硬件评估见 setup-guide.md；导演级分镜与 WenWu 引擎见 wenwu-director.md；hybrid 完整示例见 hybrid-example.md；无人值守与资源监控见 resource-monitoring.md。 ComfyUI API 调用与 comfy-mcp 见 api-and-mcp.md。
阶段 0：探测 ComfyUI 与可选隧道
- 运行
scripts/probe_comfy.py --workspace <工作区根> --write探测地址。 探测顺序：本机 8188 → 已记录隧道 URL → AutoDL 配置；全部失败时用--interactive询问用户。 - 询问用户是否需要远程访问（这是唯一必须问的选项）。需要时运行
scripts/ensure_cloudflared.py --workspace <工作区根>：复用可达的已有 隧道，否则自动下载 cloudflared 并启动 trycloudflare 临时隧道。 - 把可用
base_url写入.config/pipeline-config.json，后续脚本自动读取。 隧道失败不阻塞：回退原地址并继续。
阶段 1：剧本 → 分镜
- 询问故事要求：片长、风格、目标平台、角色/参考图、是否指定镜头结构； 用户意图足够明确时直接创作，不连续追问。
- 创建
outputs/<项目名>/，按 storyboard-schema.md 生成衔接分镜： 把剧本切段（默认每段 10 秒，5-15 秒可调），每段 2-3 个分镜，上下镜头之间 写清衔接锚点（动作方向 / 视线 / 光线 / 道具 / 轮廓 / 声桥 / 受力），禁止 无理由跳切。 - 生成人物参考图四视图：模型支持生成图片时直接生成；角色是网上已有角色
时先搜索下载参考图再生成——每角色三张不同视角全身 + 一张脸部特写，存到
refs/并在storyboard.json的characters登记。 - 先做提示词模式决策，再写提示词：写提示词前必须判断简报复杂度，并把
meta.prompt_mode与meta.prompt_mode_reason写入 storyboard.json， 禁止静默走默认。三种模式：hybrid（推荐）：创意简报 / 游戏 PV / 宣传片 / 广告 / 短片默认。保持 官方 H3 六段式英文外壳（subject_definitions / summary / retention_analysis / detailed_description / overall_soundscape / non_diegetic_music），内容按 WenWu 导演标准写（逐秒镜头脉冲、生命核、 八条生命通道、表演/状态肌理、音轨编排、结尾constraints:风格与 负向约束块），规范见 wenwu-director.md 的「hybrid：官方 六段式 × WenWu 导演深度」，完整示例见 hybrid-example.md。项目还可附带 视听签名audiovisual_signature、声音事件表sound_events、 7 列镜头卡shots（含机位库/运镜库/转场库/高级技法）、节奏统计、 世界锚点world_anchors、参考片样本reference_films、climax_segment与total_duration，规范见 storyboard-schema.md。wenwu：用户明确提到导演 / 分镜衔接 / 镜头设计 / 表演 / WenWu，或 要求纯中文导演分镜提示词时使用（WenWu 成片书写法）。official：仅用于快速批量、无强风格要求的段级生成。 判断规则：简报含风格系统、色彩系统、音乐编排、逐秒分镜或负向约束， 或任务属性是 PV / 宣传片 / 广告 / 短片 → 必须hybrid。h3-prompt-writing只用于official与hybrid的六段式外壳；wenwu纯中文模式不依赖它。 参考图在提示词中用<Picture N>标签与refs数组对应。
- 分镜汇报（必须按秒段）：分镜与提示词写完后，向用户汇报剧本时必须
逐段给出精确时间轴——「几秒到几秒 → 什么画面 → 什么对白」，禁止用
“前面是开场、中间打斗、结尾反转”这类宽泛概括。汇报格式见
storyboard-schema.md 的「分镜汇报
格式」；hybrid/wenwu 模式直接采用镜头卡里的
time/duration/content/sound与台词。用户据此确认或指出要改的镜头后再进入阶段 2。
Show full SKILL.md (259 more words)Show less
阶段 2：工作流扫描、选择与构建
铁律：优先复用用户已有的工作流，禁止自行创建新工作流。 步骤如下：
- 扫描用户工作流：
python scripts/scan_workflows.py --project <项目目录> --workspace <工作区根> 扫描范围：<项目>/workflows、<项目>/templates、<工作区>/workflows、
<工作区>/pv1min_workflows 与 pipeline-config.json 的 templates_dir。
输出每个候选的格式（API/UI）、模式（ref2va / i2v / t2v / batch）、参考图
槽位、Turbo 标记与模型文件。
2. 展示候选并询问用户：把扫描结果整理成表格（编号 / 文件 / 模式 / 适合
场景）给用户，询问“用哪个工作流”。用户选定后把映射写入
storyboard.json：
- 整批统一：
meta.workflow_map = {"<mode>": "路径或模板名"}； - 按段指定：
meta.workflow_map = {"<段号>": "路径或模板名"}，或直接在 该段写segments[].template。
- 只有以下情况才用内置模板（
assets/templates/的 ref2va / t2v / i2v）：扫描结果为空，或用户明确表示没有合适工作流、愿意用内置模板。 内置模板也必须经用户确认，不允许静默替换。 - 构建：
python scripts/build_workflows.py --project <项目目录> --workspace <工作区根> 模板解析优先级：segments[].template > meta.workflow_map[段号] >
meta.workflow_map[mode] > 内置模式推断（有参考图 → ref2va；首尾帧 →
i2v；纯文本 → t2v）。按 class_type 覆盖提示词、参考图、时长、步数、
scheduler、种子、分辨率与输出前缀，做后端校验（时长 5-15s、步数 4-40、
种子范围、参考图存在、提示词非空），写出 workflows/seg_XX_api.json 和
jobs/params.json，并打印参数汇总表。
用户工作流是 UI 格式时脚本会明确报错，要求先用 convert_ui_workflow.py
转换——脚本绝不自动改写用户工作流文件。
提示词深度也做后端校验：六段齐全、镜头标记与时间码、hybrid 的
constraints: 块；meta.strict_prompt_validation: true 时不达标直接拒绝。
hybrid meta 另做后端校验：视听签名字段、段时长累加、高潮位置与声音事件表。
阶段 3：参数确认
- 把
params.json汇总成表格展示给用户：段号 / 时长 / 模式 / 模板 / steps / scheduler / seed / 比例 / 参考图 / 输出前缀。 - 用户可修改
storyboard.json中的参数（时长、步数、种子、模板、参考图等） 后重跑阶段 2；不要做前端表单，脚本后端校验非法值并拒绝。 - 用户确认后锁定参数，进入提交（
params.json保持为提交依据）。
阶段 3.5：无人值守确认（每次批量开始前必做）
每次开始批量提交前，询问用户本次是否无人值守：
- 是：与用户约定关机时间（本机 Windows 自动关机，或 AutoDL/云实例的到期
时间——云端关机按 autodl-cloud.md 用 API
power_off或在云控制台设置）；把约定写入jobs/run_plan.json；按 resource-monitoring.md 启动scripts/monitor_resources.py后台监控 GPU 显存与系统内存（默认 warn 90% / stop 95%），到达约定时间且用户已授权时执行本机关机。 - 否：正常人工值守流程，跳过关机约定；资源监控可选。
monitor_jobs.py 每轮会读取 jobs/resource_state.json，达到 stop 阈值时暂停
本轮并报告，防止显存/内存爆掉。
阶段 4：提交、监控与下载
python scripts/submit_jobs.py --project <项目目录> --workspace <工作区根>
python scripts/monitor_jobs.py --project <项目目录> --workspace <工作区根>submit_jobs.py先上传本地参考图到远程 input 目录（同批次缓存去重），再 POST/prompt；node_errors非空时中止并报告缺什么。已提交/已下载的段 自动跳过（--force可重提），支持--segments 1,3部分提交。- API 报错先查 api-and-mcp.md 排查表：不要
盲目重试同一条命令，
node_errors/ 超时 / SSL EOF / 任务 error 各有对应 处理；连续 3 次无变化就停下来向用户报告。 monitor_jobs.py轮询/history/{prompt_id}：错误提取报错并按重试上限 （默认 2 次）重新提交，完成后把 MP4 下载到clips/raw/<batch>/seg_XX_00001-audio.mp4；状态落盘，可断点续跑。 监控长任务时放后台运行并定期查看jobs/monitor_state.json。- 单段重试后仍失败：不要静默跳过，向用户报告失败段与原因，由用户决定 重跑、改参数或换模板。
阶段 5：剪映拼合
- 确认片段齐全后生成业务脚本（遵守 jianying-editor“业务脚本放项目目录”规则）：
python scripts/generate_assembly.py --project <项目目录> --title <片名> \
--width <宽> --height <高>- 运行生成的
assemble_jianying_<标题>.py（内部 bootstrap jy_wrapper， 按段序add_media_safe，相邻段加“叠化 0.3s”转场，save()后写edit/jianying-draft-report.json）。无字幕无 BGM。 - 告知用户草稿名称与路径，让用户在剪映中打开微调；仅当用户显式要求
自动导出时，调用 jianying-editor 的
auto_exporter.py（Windows + 剪映 5.9 及以下）。导出前可先检查报告 JSON 的draft_name/drafts_root。
失败恢复与边界
- 断点：
jobs/下所有状态为 JSON，重跑任意阶段都幂等跳过已完成段。 - 隧道抖动：探测失败自动回退；
ensure_cloudflared.py --stop可停隧道。 - 模板：用户已有工作流优先（scan_workflows.py 扫描 + 用户确认）；内置
三套 H3 紧凑模板只是兜底；大模板（如 270KB 多图 Ref2VA）先经
convert_ui_workflow.py转 API 格式再引用，不自动改写用户文件。 - 本 skill 不做前端验证；所有合法性检查由脚本后端完成。
- 不在 skill 目录内创建任何业务脚本/项目文件；业务产物一律放用户项目目录。
脚本速查
| 脚本 | 用途 |
|---|---|
probe_comfy.py | 探测可用 ComfyUI 地址并写入配置 |
ensure_cloudflared.py | 复用/启动/停止 trycloudflare 隧道 |
convert_ui_workflow.py | UI 格式模板转 API 格式 |
build_workflows.py | 分镜 → 每段 API 工作流 + 参数表（含提示词深度校验） |
scan_workflows.py | 扫描用户已有工作流并给出候选清单（模式/格式/参考图） |
submit_jobs.py | 上传资产并批量提交 |
monitor_jobs.py | 轮询下载、失败重试、断点续跑 |
generate_assembly.py | 生成剪映拼合业务脚本 |
check_hardware.py | 检测 GPU/内存/磁盘并给出本地运行建议 |
monitor_resources.py | 无人值守时监控 GPU 显存/内存，支持约定时间关机 |
© TFboy1, MIT. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file
Files
SKILL.md and 36 other files (scripts, references, assets) in the repository root of TFboy1/oh-my-minimaxh3-director.
- SKILL.md
- .gitignore
- CONTRIBUTORS.md
- LICENSE
- README.md
- agents/openai.yaml
- assets/templates/i2v.json
- assets/templates/ref2va.json
- assets/templates/t2v.json
- banner.svg
- docs/README_DE.md
- docs/README_EN.md
- docs/README_FR.md
- docs/README_JA.md
- evals/evals.json
- references
- … and 21 more
Open the folder on GitHubat commit 9112661
Compare with similar skills
MiniMax H3 Video Director 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 |
|---|---|---|---|---|---|---|
| MiniMax H3 Video Director this skillTFboy1/oh-my-minimaxh3-director | 143 | — | ~2.4k | Automated safety check: Pass | MIT | |
| VRGDG H3 Short Film Pipelinevrgamegirl19/comfyui-vrgamedevgirl | 765 | — | ~4.2k | Automated safety check: Pass | Custom licence | |
| Open Videoagent-next/video-agent | 120 | — | ~3.2k | Automated safety check: Pass | Apache-2.0 | |
| Hong Kong Comic Fighter for H3karuvanan/MiniMax-H3-Director-Cut-Studio | 132 | — | ~4.4k | Automated safety check: Pass | Custom licence | |
| Long-Form H3 Directorkaruvanan/MiniMax-H3-Director-Cut-Studio | 132 | — | ~1.5k | Automated safety check: Pass | Custom licence | |
| Scripting And Storyboardingsocial-media-skills/skills | 134 | — | ~2.1k | Automated safety check: Pass | MIT |
Similar skills
Open Video
agent-next/video-agent
Generate, edit, or direct videos via open-source models (MiniMax H3 baseline; Wan2.2 / LTX future).
Hong Kong Comic Fighter for H3
karuvanan/MiniMax-H3-Director-Cut-Studio
Special skill that turns loaded Hong Kong comic panels into photoreal MiniMax H3 martial-arts sequences with readable attack and defence action.
Long-Form H3 Director
karuvanan/MiniMax-H3-Director-Cut-Studio
Plans long-form MiniMax H3 video productions as Sequence, Shot, and Segment contracts that stay continuous across generation windows.
Scripting And Storyboarding
social-media-skills/skills
The pre-production system — turn a video idea into a shootable, editable plan: the two-column AV script, the storyboard as a decision document, the numbered shot list, the batch-shoot plan, and the…
H3 Video
agent-next/video-agent
OpenVideo skill (v0.1.0): generate high-quality local video with the OpenVideo product (MiniMax H3 backend).
Related
Works with
Categories
Questions about MiniMax H3 Video Director
What does MiniMax H3 Video Director do?
Turns a script into a storyboard, assigns MiniMax H3 workflows in ComfyUI, monitors batch generation and builds a Jianying draft of the finished video. Written in Chinese, this skill automates a script-to-video chain built on ComfyUI. The agent splits a script into segments, with each segment becoming one H3 video holding two or three shots, and assigns a MiniMax H3 template to each segment: Ref2VA, T2V or I2V.
When should I use MiniMax H3 Video Director?
MiniMax H3 Video Director fits situations like: turning a screenplay or story text into a generated video; producing a game trailer or ad short from a script and a style brief; planning shot-to-shot transitions and character reference views before generation; batch-generating H3 clips and assembling them into a Jianying draft.
How do I install MiniMax H3 Video Director in Claude Code?
Run `npx skills add TFboy1/oh-my-minimaxh3-director --skill oh-my-minimaxh3-director -a claude-code`. Or copy the skill folder (the TFboy1/oh-my-minimaxh3-director repository) into .claude/skills/oh-my-minimaxh3-director in your project. Claude Code loads it when a task matches its description.
How do I install MiniMax H3 Video Director in Codex?
Run `npx skills add TFboy1/oh-my-minimaxh3-director --skill oh-my-minimaxh3-director -a codex`. Or copy the skill folder (the TFboy1/oh-my-minimaxh3-director repository) into .agents/skills/oh-my-minimaxh3-director in your project. Codex loads it when a task matches its description.
Can I use MiniMax H3 Video Director in Cursor, Gemini CLI or GitHub Copilot?
Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add TFboy1/oh-my-minimaxh3-director --skill oh-my-minimaxh3-director -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/oh-my-minimaxh3-director, .gemini/skills/oh-my-minimaxh3-director, .github/skills/oh-my-minimaxh3-director and .opencode/skills/oh-my-minimaxh3-director in your project.
What does MiniMax H3 Video Director need to run?
Going by SKILL.md and its folder, MiniMax H3 Video Director needs the command-line tools its instructions call (python and npx). Our summary lists: ComfyUI running locally or in the cloud; MiniMax H3 model files, about 40GB; An NVIDIA GPU with at least 12GB of VRAM, or a remote GPU; Windows with Python and a project .venv.
Does MiniMax H3 Video Director access the network?
SKILL.md names 1 domain. As links in the text: comfy.org. This is read from the text; nothing was executed.
Is MiniMax H3 Video Director safe to install?
Our automated static check of SKILL.md found no risky patterns, such as piping downloads into a shell, reading credential files or hidden Unicode. 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.
What licence does MiniMax H3 Video Director use?
MiniMax H3 Video Director is published under the MIT licence (from the LICENSE file in the skill folder). It allows redistribution, so the full SKILL.md is shown on this page.
How many tokens does MiniMax H3 Video Director use?
About 2.4k tokens (SKILL.md is roughly 9.7k 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 21k tokens, read only when the agent opens those files.
What are the alternatives to MiniMax H3 Video Director?
Skills that share tags, products or a category with MiniMax H3 Video Director: VRGDG H3 Short Film Pipeline (vrgamegirl19/comfyui-vrgamedevgirl, 765 stars), Open Video (agent-next/video-agent, 120 stars), Hong Kong Comic Fighter for H3 (karuvanan/MiniMax-H3-Director-Cut-Studio, 132 stars) and Long-Form H3 Director (karuvanan/MiniMax-H3-Director-Cut-Studio, 132 stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.
Who maintains MiniMax H3 Video Director?
TFboy1 (a GitHub user) maintains it in TFboy1/oh-my-minimaxh3-director, which has 143 GitHub stars. The repository was last updated on August 16, 2026.
Source: TFboy1/oh-my-minimaxh3-director on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.