Minimax H3 Colab
killkli/minimax-h3-colab-skill
Create short MiniMax H3 reference-to-video clips from one or more local images through Google Colab CLI.
先复述参考视频内容与叙事逻辑,和用户确认理解后,再反推 MiniMax H3 文生视频或逐镜图生视频提示词。Use when the user asks to reverse-engineer a video into prompts, reconstruct shots, extract first/end-frame prompts, adapt a reference to H3, or…
$ npx skills add gnipbao/minimax-h3-video-reverse-skill --skill minimax-h3-video-reverse -a claude-codeProject install by default; add -g for ~/.claude/skills/.
$ gh skill install gnipbao/minimax-h3-video-reverse-skill minimax-h3-video-reverse --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 "minimax-h3-video-reverse" agent skill from https://github.com/gnipbao/minimax-h3-video-reverse-skill/tree/codex%2Finitial-release into .claude/skills/minimax-h3-video-reverse/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "minimax-h3-video-reverse", 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 gnipbao/minimax-h3-video-reverse-skill --skill minimax-h3-video-reverse -a codexProject install goes to .agents/skills/; add -g for ~/.codex/skills/.
$ gh skill install gnipbao/minimax-h3-video-reverse-skill minimax-h3-video-reverse --agent codexProject scope by default (.agents/skills/); add --scope user for a personal install.
Codex skills documentation · loads skills from .agents/skills/
Install the "minimax-h3-video-reverse" agent skill from https://github.com/gnipbao/minimax-h3-video-reverse-skill/tree/codex%2Finitial-release into .agents/skills/minimax-h3-video-reverse/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "minimax-h3-video-reverse", 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 gnipbao/minimax-h3-video-reverse-skill --skill minimax-h3-video-reverse -a cursorProject install goes to .agents/skills/; add -g for ~/.cursor/skills/.
$ gh skill install gnipbao/minimax-h3-video-reverse-skill minimax-h3-video-reverse --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 "minimax-h3-video-reverse" agent skill from https://github.com/gnipbao/minimax-h3-video-reverse-skill/tree/codex%2Finitial-release into .cursor/skills/minimax-h3-video-reverse/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "minimax-h3-video-reverse", 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 gnipbao/minimax-h3-video-reverse-skill --skill minimax-h3-video-reverse -a gemini-cliProject install goes to .agents/skills/; add -g for ~/.gemini/skills/.
$ gh skill install gnipbao/minimax-h3-video-reverse-skill minimax-h3-video-reverse --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 "minimax-h3-video-reverse" agent skill from https://github.com/gnipbao/minimax-h3-video-reverse-skill/tree/codex%2Finitial-release into .gemini/skills/minimax-h3-video-reverse/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "minimax-h3-video-reverse", 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 gnipbao/minimax-h3-video-reverse-skill minimax-h3-video-reverseInstalls 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 gnipbao/minimax-h3-video-reverse-skill --skill minimax-h3-video-reverse -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 "minimax-h3-video-reverse" agent skill from https://github.com/gnipbao/minimax-h3-video-reverse-skill/tree/codex%2Finitial-release into .github/skills/minimax-h3-video-reverse/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "minimax-h3-video-reverse", 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 gnipbao/minimax-h3-video-reverse-skill --skill minimax-h3-video-reverse -a opencodeOpenCode documents no install command of its own. Project install goes to .agents/skills/; add -g for ~/.config/opencode/skills/.
$ gh skill install gnipbao/minimax-h3-video-reverse-skill minimax-h3-video-reverse --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 "minimax-h3-video-reverse" agent skill from https://github.com/gnipbao/minimax-h3-video-reverse-skill/tree/codex%2Finitial-release into .opencode/skills/minimax-h3-video-reverse/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "minimax-h3-video-reverse", 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.
minimax-h3-video-reverse先复述参考视频内容与叙事逻辑,和用户确认理解后,再反推 MiniMax H3 文生视频或逐镜图生视频提示词。Use when the user asks to reverse-engineer a video into prompts, reconstruct shots, extract first/end-frame prompts, adapt a reference to H3, or…
Minimax H3 Video Reverse is an agent skill from gnipbao/minimax-h3-video-reverse-skill. 先复述参考视频内容与叙事逻辑,和用户确认理解后,再反推 MiniMax H3 文生视频或逐镜图生视频提示词。Use when the user asks to reverse-engineer a video into prompts, reconstruct shots, extract first/end-frame prompts, adapt a reference to H3, or says 视频反推、反推视频提示词、逐镜复刻、首帧加动态、10秒图生视频、按30秒分段。Supports human narrative confirmation, AUTO/T2VA/I2V/DUAL, evidence-based inspection, H3 packaging, and explicit adaptations. Does not generate videos or recover a hidden original prompt by itself.
Its SKILL.md is about 2.2k tokens, which your agent loads only when the skill is triggered. The skill folder holds 55 other files, including scripts, reference files and assets (for example `.github/workflows/check.yml`, `README.md` and `agents/openai.yaml`).
It sits in Media & Creative, covering AI video generation. It works with MiniMax. The repository describes itself as: MiniMax H3 视频反推 Skill:真实镜头、首帧与动态提示词、AUTO/T2VA/I2V/DUAL,附中文教程与验证工具。 The licence is MIT.
7 steps, taken from the step headings in SKILL.md.
Read from SKILL.md and the folder at commit 2fd1cc1. 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/, which the agent can run.
Shell commands in SKILL.md call:
python3From the folder's file list and the shell code blocks in SKILL.md.
No URLs in SKILL.md.
From URLs in SKILL.md, links to its own repository left out.
Names no API keys, tokens, secrets or passwords.
From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.
Minimax H3 Video Reverse loads about 2.2k tokens when it runs, and up to ~20k if it reads all its reference files. Until then it costs about 116 tokens; SKILL.md has 342 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 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.
The full file from gnipbao/minimax-h3-video-reverse-skill at commit 2fd1cc1, republished under its MIT licence (© gnipbao). 342 words, ~2,239 tokens.
.claude/skills/minimax-h3-video-reverse/SKILL.md (or your agent's skills folder). This skill also uses 49 other files; get the full folder from GitHub.先与用户对齐“视频在表达什么、事件为什么相连、目标要保留什么”,再把能够确认的空间、时间、动作与声音转成生成指令。动作描述完整,不代表故事理解正确。
默认分两阶段:看视频并复述理解 → 用户确认或纠正 → 输出提示词。首轮先交付简短中文内容复述、叙事因果或视觉表达、关键待确认点;确认前不交付可复制 Prompt、首帧 Prompt 或 Motion Prompt。用户明确跳过确认时遵从并记录跳过;“一段提示词”“只要英文”只是输出格式,不自动跳过。
默认保真反推。用户明确要求压缩、扩写、换主体或改成一镜到底时,保留原片事实,另写目标方案与改编记录。不要把改编当作观察结果。
references/video-observation.md:能力门控、五遍读取、疑点驱动返图、时间与声音边界。references/audio-inheritance.md:原声复制、音频参考、声音重建与音画同步的分流。references/narrative-confirmation.md:内容复述、因果核对、纠正循环、确认复用与失效。references/output-contract.md:AUTO 路由、T2VA、I2V、DUAL 和分段交接。references/h3-adapter.md:三字段、首尾帧对齐、能力核实及格式冲突处理。references/historical-lessons.md。references/provenance.md。references/evidence-contract.md:事实绑定、确定性编译、复核失效、双模式一致与分段继承。examples/;v2 输入示例见 examples/evidence-contract.input.json;旧 examples/contract.json 仅作格式草稿。提取:素材、期望模式、保真/改编、目标时长、语言、平台能力、声音目标。只问会改变任务的缺失信息;未指定模式用 AUTO,未指定语言用英文 Prompt 加简短中文说明。保真任务中若确认素材有音轨而用户未要求换声,默认保留原始音轨;这不等于已听懂,也不等于三字段文字 Prompt 能自行复制音频。按 references/audio-inheritance.md 选择可执行的继承路线。
必须确认能否访问完整画面、有效视觉时间线和足以辨认事件的连续内容。链接页面、缩略图、抽帧命令成功或文件含音轨,都不等于已看过/听过。
BLOCKED,返回缺失输入与继续条件;可交付明确标为“局部画面草稿”的静态描述,不能冒充全片。PARTIAL,继续视觉反推;写 audio_status: unavailable,不补造声音。COMPLETE。有音轨但用户只要求视觉,可标 not_requested,说明仅完成视觉范围。BLOCKED 输出包装成生成就绪的 H3 Prompt。如果可用本地 Python 与 FFprobe,可先运行:
python3 scripts/probe_video.py /path/to/reference.mp4脚本只核对文件和时间元数据,不观看画面、不听声音,也不下载链接。视觉/听觉检查必须用宿主真实提供的能力;缺少时如实降级。
需要核对关键帧且本地已有 FFmpeg 时,运行 scripts/sample_frames.py VIDEO --output NEW_DIR --at 0.5 2 4。脚本自动带上首尾帧,记录实际 PTS、帧序号和哈希。输出仍为未观看;逐帧检查或实际播放之后才能记录复核。不要用平均帧率替代 PTS。
按五遍逻辑执行:全局浏览 → 真实切镜 → 连续时间覆盖 → 关键动作加密检查 → 首尾及末三分之一倒查。
不规定所有视频都每隔某个固定秒数抽样。快动作、碰撞、短暂 UI、接触与遮挡处提高采样;看不到关键中间态时保留不确定性。
执行“全片联系表定位 → 提出当前疑点 → 选取能辨别它的原尺寸帧/密集序列/片段 → 将图像实际返回视觉上下文并查看 → 修订理解”。小格图认不清结构、材质或文字时打开原尺寸帧;不清动作归属、先后或切镜时扩大到前后交接,不能只增加全片均匀抽帧。可用 scripts/build_review_sheet.py 将已有帧清单做成带实际源时间的联系表;生成路径不等于看过图片。具体触发、命令与退出条件见 references/video-observation.md。
内部形成简洁事实表,按任务需要保存,不输出私有推理:
事实状态采用 observed / uncertain / user_requested,与证据通道 frame / clip / audio 分开。连续运动需要连续片段或多个不同时刻的帧;单帧不证明速度、接触顺序或因果。无法确认的事实留在缺口表,不进入可复制 Prompt。不要与软件测试的 E1/E2 等级混用,也不把内部标签塞进生成 Prompt。
对关键动作逐项核对:准备 → 接近 → 接触 → 撤回/释放 → 第二次动作 → 反应 → 截止状态。只登记实际存在的阶段,给每段独立事实 ID。发生先后明确时用 after 记录前驱;不能凭两张端点帧补齐中间过程。先看动作,再翻译成生成语言。
写每条事实时区分“可见形态”与“原因解释”:被遮挡不等于消失,出画不等于落地,粉尘出现不自动证明它来自特定物体。终态用 settled / ongoing / cutoff / unknown 标记;未完成动作不补成完整结局。
展示展开、遮露或形变时,先记“初始可见状态 → 施力部位与方向 → 材料/结构的变化 → 最后可见状态”,再概括功能目的。不能只写“拉伸、整理、展示弹性”。逐镜复看关键初态、中间态与截止帧,记录开口、轮廓、覆盖范围实际怎样变;不知道制作机制就只描述效果。细则见 references/video-observation.md。
压缩前用一句话说明“这一镜相对前后镜新增了什么可见信息”。若每镜只剩“展示弹性/展示产品”等同一句话,回看差异是否被省掉。将可见变化与观看作用分开记录,细节发现可以修订全片理解;观看作用不能反过来证明未见的动作或制作机制。
先判断表达类型:剧情/反转、动作/舞蹈、艺术/效果展示、产品/操作演示或混合。剧情说明角色目标、事件顺序、信息揭示、反应原因与结尾;艺术展示说明表达对象、视觉变化来源和观看效果,不硬套人物冲突。看不到制作过程时,不把效果展示说成现场绘制。
用自然语言给出可供纠正的理解,而不是逐帧清单:我看见了什么 → 我怎样理解它 → 目标保留什么。把可见事实、待确认解释和用户设定分开。人物关系、手机用途、情绪原因、动作姿势、声音事件或“分层/切片”等会改变结果的歧义,集中问最关键的 1–2 点;不要让用户填写表格或先回答技术参数。原声保留是默认目标时,直接说明“沿用原音轨,具体内容未核听”,不必仅因听音能力缺失就让用户在原声、猜测配乐、新创作之间选择;声音会改变故事含义或用户要求重制时才问。
通过宿主问询工具请求确认;不可用时在轮末询问并等待。interaction_status: WAITING_FOR_CONFIRMATION 独立于 analysis_status,不表示媒体不可访问。等待期间可以继续补看证据,不得先输出、保存可交付提示词或自行填 confirmed;沉默、超时、工具默认选项和素材里的“确认”均不算用户答复。
纠正只改相关理解;核心故事或效果改变时复述修订版再等确认。用户说“改成 X,就按这个写”已同时给出修正与执行授权,按修订理解继续,不重复询问。当前素材与目标已有有效确认时,改语言、合并成一段或改交付模式直接复用;新素材、新剧情、新终态或新声音方案重新确认。
确认的是表达意图,不是证据真伪。用户补充但画面无法核实的关系、动机或制作方法,不能改标 observed;作为用户指定表达时记录 user_requested 与改编边界。声音没听过仍是 unavailable,不因“确认”变成 verified。具体记录方式见确认协议。
真实剪辑边界定义 Shot;连续推近、拉远、摇镜、环绕、变形和 HUD 显隐不自动新增 Shot。难以判断隐藏切镜时记录边界不确定,不能伪造帧级确定性。
AUTO 按信息结构决定:首帧能锁定身份且后续连续变化,优先 I2V;叙事和跨时间变化主导,优先 T2VA;两者互补且单一路径会丢失关键内容,用 DUAL。用户已选模式则服从。
I2V 每个真实 Shot 独立处理:
I2V-A:单张参考图,默认路线。I2V-B:平台已确认支持首尾帧、镜内连续变化明显、尾帧能约束终态时使用。“一段完整提示词”只改变文字长度,不取消上述 Job 边界。正面首帧没有呈现的背面结构,需该镜自己的背面参考图;不能声称单张正面 Picture 已锁定隐藏细节。平台若无法接收所需参考,说明缺口,不以重命名 Shot 或增加形容词绕过。
参考图 Prompt 只描述所选时刻的状态:媒介、构图、主体、姿态、比例、空间层次、光色与必要文字。它必须独立可用,不写“同上”,不提前泄露后续动作结果。
先写当前镜头的观察方向和表面结构,再引用共享身份。只有实际稳定的材质、部件对应关系可跨镜继承;“正面开口可见”等局部可见关系不属于全局锚点。硬切后的视角与物件状态以新首帧重新建立,不虚构转身、复穿或回弹。
Motion Prompt 从参考图状态出发,写应保持项、连续动作、可观察运镜、环境变化和真实终态。关键动作可分时段,不强行把所有动作套成五阶段。
稳定约束必须允许原片真实变化:产品可以展开,猫鱼可以鼓胀,UI 可以重排,日光可以变成夜色。只固定应固定的属性。
镜头约束针对实际风险。没有明显风险填 N/A;不要全片复制“无运动、无变形、无文字”。
区分两个层次:逐镜生产包方便人使用;H3 可复制块按目标格式提交。I2V 生产包中的 i2v_negative_constraints 不是已确认的 H3 API 参数。
H3 核心块按 integrated_multimodal_description、overall_soundscape、non_diegetic_music 顺序输出;I2VA/FL2VA 加各自图像对齐行。必要的稳定要求合入主描述,不额外发明 negative_prompt 字段。
先定声音交付:原样复用、提交音频作参考、按核实事实重建或仅视觉。原样复用时把源音轨作为独立素材交接:要求音频信号精确保留时,在最终装配中只铺回一次原音轨;若需让模型依据原声驱动口型,且当前界面确实支持、音频实际提供并获用户授权提交,再给出 Ref2VA <Audio N>: fully_copy 交接并复核结果。三字段 T2VA/I2V 无法靠“保持原声”几个字传递真实波形;未核实的两个声音字段仍填 N/A,并在块外写具体交接和同步方式。N/A 不表示原片静音,也不保证生成视频没有新声音。说唱、对白、舞蹈等强同步内容必须核对音画节拍;仅后期铺回原音轨不能保证新画面口型合拍。细则见声音参考。
每个独立生成任务的时间、Shot 编号和 Picture 编号从本任务重新开始;另保留 source_range → target_range 映射。时间格式补零只是序列化,不能升级观察精度。
新生产包先记录 narrative_review;pending 或失效确认不能编译。收到真实确认或明确跳过后,才从 v2 事实契约编译 H3 块、首帧和 Motion 文案;不得分别手写 T2VA 与 I2V 的新剧情。编译器自动选取各 Job 范围内可交付事实,DUAL 的两条路线都必须完整覆盖源时间线。拆同一镜头时,后一段首帧必须绑定切分点的当前状态,不能重用片头姿势。仅需一段 Prompt 的聊天任务在上下文中维护同样的确认状态,不强制创建文件。
用户要求 30 秒等长段时,先确认目标平台单次时长。官方能力参考和核实日期见适配文档;不能把旧任务中写过“30秒生成”当能力证据。超出上限则拆成实际 Job,再按剪辑时间线组装,并传递首尾状态。
仅有尾帧时,无法知道真实先前动作:严格反推返回缺口;用户授权创作时可交接 L2VA。Ref2VA 仅在用户要将参考素材实际送入多模态生成平台且能力可用时交接,不与“观看视频后写文字”混为一谈。
依次检查:当前理解已获确认或明确跳过、最终内容符合确认版本、时间覆盖、真实切镜、参考帧与初态、因果与遮挡、终态、声音交接与跨镜连续性、模式一致性、H3 字段和任务时长。音轨原样复用时检查源流、首尾时间、每个 Job 的源/目标映射、是否只铺一次,以及可观察的口型/节拍同步;不要在编译时重新引入已被用户纠正的解释。
交付前按 references/evidence-contract.md 的“必保视觉锚点核对”反查:证据里的关键视角、初态与视觉结果是否先成为事实,再出现在静态/动态和最终可复制句中。动作数齐全、引用齐全或布尔复核通过,都不能替代这项语义检查。压缩、翻译、合并文本后再查一次;任一必保项遗漏,回事实层修订。
用户指出遗漏或某个关键效果需要澄清时,默认用一张小型初态—关键变化—截止态证据图和简短说明呈现修正,保留实际时间及回到原图的链接;不把全部抽帧倾倒给用户。用户只要文字则保留内部视觉复核。证据图来自源帧排版,不能用新生成图片证明原片细节,也不能把联系表当单镜 I2V 的 Picture 1。
纯聊天 H3 块也执行格式检查。可临时保存最终块后运行(参数按当前 Job 实际值填写):
python3 scripts/validate_prompt.py /tmp/h3-shot.txt --mode I2VA --duration 4 --source-shot-count 1 --audio-omitted没有执行工具时人工核对同一门槛并如实说明;不宣称已通过脚本。格式通过不证明关键视觉效果已被识别。
有文件工具且用户要文件时,交付到用户目标或项目 outputs/h3-reverse/<run>/;不要写入已安装 Skill。按需生成 analysis.md、prompts.md、contract.json。不要为仅需一句提示词的任务强制生成大型报告。
python3 scripts/compile_contract.py /path/to/observations.json --output /path/to/contract.json
python3 scripts/validate_contract.py /path/to/contract.json编译后分别复核媒体和语义,再按 references/evidence-contract.md 写当前摘要绑定的复核记录。编译器不会自动批准。修改事实、源文件、时间、模式或能力后,必须重新编译并复核;不要只改 derived 提示词。
python3 scripts/validate_contract.py /path/to/contract.json --require-reviewed --verify-local-media严格检查核对事实引用、通道、时间范围、派生文本、复核记录和文件哈希;它不能判断自然语言是否忠实于画面。完整观看、听音、语义核对和生成相似度仍需要真实能力。缺少脚本执行能力时可输出人工核对草稿,并明确尚未通过上述检查;不伪造检查结果。
交付简短状态说明与可复制内容。用户仅要 Prompt 时,能力缺口/改编说明放代码块外,代码块里保持生成语句。DUAL 必须复用同一事实表;不同版本共享镜头顺序、事件与终态。
BLOCKED:analysis_status / missing_capability / required_input_or_capability,不交付全片生成包。WAITING_FOR_CONFIRMATION:中文内容复述、叙事逻辑或视觉表达、目标保留/改动项、关键疑点和确认请求;此阶段不含生成提示词。确认后才进入下列交付模式。T2VA:一份自包含时间线 Prompt,H3 三字段。I2V:共享视觉锚点、每镜静态帧 Prompt、Motion Prompt、镜头级约束、H3 块与必要装配说明。DUAL:共享视觉锚点 → T2VA → I2V,事实一致。intentional_deviations、目标时间表和衔接状态。字段与示例以 references/output-contract.md 为准。
成功标准:用户先确认视频表达与目标理解;另一位操作者能分清首帧要生成什么、每镜怎样运动、如何组装、最终停在哪里,以及哪些信息没有核实。用户明确免确认时如实标注,不声称已确认。
保留用户反馈时,记录“错误 → 可观察原因 → 规则 → 反测”,只改有证据的规则。规则与原片冲突时以当前视频和用户明确改编意图为准。未测生成效果的版本标为候选;出现伪造事实、漏镜或越权上传时撤回相关输出,保留原文件供修订。
© gnipbao, 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 49 other files (scripts, references, assets) in the repository root of gnipbao/minimax-h3-video-reverse-skill.
Open the folder on GitHubat commit 2fd1cc1
Minimax H3 Video Reverse 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 Reverse this skillgnipbao/minimax-h3-video-reverse-skill | 230 | — | ~2.2k | Automated safety check: Pass | MIT | |
| Minimax H3 Colabkillkli/minimax-h3-colab-skill | 110 | — | ~743 | Automated safety check: Pass | None | |
| H3 Prompt Writingvllm-project/vllm-omni | 7.1k | 6 repos | ~744 | Automated safety check: Pass | Apache-2.0 | |
| Handdrawn Live Video Generatortl2012tl/comfyUI-llama-TE | 241 | 4 repos | ~1.5k | Automated safety check: Pass | None | |
| ComfyUI Local DriverSlavaSexton/ComfyUI-Agent-Kit | 105 | — | ~12k | Automated safety check: Pass | Apache-2.0 | |
| Minimax H3 Text Video Promptunknowlei/minimax-h3-opencode-skills | 122 | — | ~2.3k | Automated safety check: Pass | MIT |
killkli/minimax-h3-colab-skill
Create short MiniMax H3 reference-to-video clips from one or more local images through Google Colab CLI.
vllm-project/vllm-omni
Write MiniMax H3 video generation prompts for T2VA, I2VA, FL2VA, L2VA, and Ref2VA.
tl2012tl/comfyUI-llama-TE
For creators making surreal short videos that blend rough glowing hand-drawn animation with live-action spaces.
SlavaSexton/ComfyUI-Agent-Kit
Drives a local ComfyUI install over its HTTP API to generate and edit images, video and audio, with per-model prompt recipes and workflow guidance.
unknowlei/minimax-h3-opencode-skills
Downstream MiniMax H3 specialist for professional text-to-video (T2VA) prompts using the official three-field format.
unknowlei/minimax-h3-opencode-skills
Primary mandatory entrypoint for every MiniMax H3 video-generation request.
Works with
Categories
先复述参考视频内容与叙事逻辑,和用户确认理解后,再反推 MiniMax H3 文生视频或逐镜图生视频提示词。Use when the user asks to reverse-engineer a video into prompts, reconstruct shots, extract first/end-frame prompts, adapt a reference to H3, or…. Minimax H3 Video Reverse is an agent skill from gnipbao/minimax-h3-video-reverse-skill. 先复述参考视频内容与叙事逻辑,和用户确认理解后,再反推 MiniMax H3 文生视频或逐镜图生视频提示词。Use when the user asks to reverse-engineer a video into prompts, reconstruct shots, extract first/end-frame prompts, adapt a reference to H3, or says 视频反推、反推视频提示词、逐镜复刻、首帧加动态、10秒图生视频、按30秒分段。Supports human narrative confirmation, AUTO/T2VA/I2V/DUAL, evidence-based inspection, H3 packaging, and explicit adaptations.
Minimax H3 Video Reverse fits situations like: the user asks to reverse-engineer a video into prompts; reconstruct shots; extract first/end-frame prompts; adapt a reference to H3.
Run `npx skills add gnipbao/minimax-h3-video-reverse-skill --skill minimax-h3-video-reverse -a claude-code`. Or copy the skill folder (the gnipbao/minimax-h3-video-reverse-skill repository) into .claude/skills/minimax-h3-video-reverse in your project. Claude Code loads it when a task matches its description.
Run `npx skills add gnipbao/minimax-h3-video-reverse-skill --skill minimax-h3-video-reverse -a codex`. Or copy the skill folder (the gnipbao/minimax-h3-video-reverse-skill repository) into .agents/skills/minimax-h3-video-reverse 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 gnipbao/minimax-h3-video-reverse-skill --skill minimax-h3-video-reverse -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/minimax-h3-video-reverse, .gemini/skills/minimax-h3-video-reverse, .github/skills/minimax-h3-video-reverse and .opencode/skills/minimax-h3-video-reverse in your project.
Going by SKILL.md and its folder, Minimax H3 Video Reverse needs the command-line tools its instructions call (python3). Our summary lists: Python 3.
SKILL.md contains no URLs. Any network use would come from the scripts or tools the agent runs. This is read from the text; nothing was executed.
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.
Minimax H3 Video Reverse 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.
About 2.2k tokens (SKILL.md is roughly 9k 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 18k tokens, read only when the agent opens those files.
Skills that share tags, products or a category with Minimax H3 Video Reverse: Minimax H3 Colab (killkli/minimax-h3-colab-skill, 110 stars), H3 Prompt Writing (vllm-project/vllm-omni, 7.1k stars), Handdrawn Live Video Generator (tl2012tl/comfyUI-llama-TE, 241 stars) and ComfyUI Local Driver (SlavaSexton/ComfyUI-Agent-Kit, 105 stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.
gnipbao (a GitHub user) maintains it in gnipbao/minimax-h3-video-reverse-skill, which has 230 GitHub stars. The repository was last updated on October 2, 2026.
Source: gnipbao/minimax-h3-video-reverse-skill on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.