handdrawn-live-video-generator
Skill4 starsChanged 58 days agoApache-2.0View on GitHub
For creators making surreal short videos that blend rough glowing hand-drawn animation with live-action spaces. Users provide a scene idea, contact object or hand, desired mood, and optional language or style constraints. The Skill clarifies the physical contact, designs continuous morphing, escape route, and delayed handheld chase movement, then writes a reusable 15-second 16:9 video prompt in the user's language. After user confirmation it recommends MiniMax H3 generation and checks contact realism, camera delay, rough glowing stroke texture, and non-horror tone. Best for single-scene creative clips, not polished CG, horror jump scares, plush characters, or multi-scene cuts.
- Installs packages
9 other projects use a version of this, 6 of them edited.
Adopted by
Most recently taken up by
- qxryz/workflowgeneratoredited, first seen Oct 25, 2026
- himly0302/use-skillsedited, first seen Oct 13, 2026
- MiniMax-AI/MiniMax-H3edited, first seen Oct 8, 2026
- vllm-project/vllm-omnitheir own version, first seen Oct 8, 2026
- SwotAtmk/infinite-creationedited, first seen Sep 28, 2026
- SuperMate-Ai/SuperMate-Harness-Systemedited, first seen Aug 30, 2026
Other versions9 projects use a version of this.
Word for word · 0
None found.
Edited · 6
- MiniMax-AI/MiniMax-H3★ 11k97% same
- SwotAtmk/infinite-creation★ 5697% same
- qxryz/workflowgenerator★ 4838% same
- GAGAA-HUE/ASHUO-claude-skills★ 595% same
- himly0302/use-skills★ 499% same
- SuperMate-Ai/SuperMate-Harness-System★ 199% same
Wrote their own version · 3
- vllm-project/vllm-omni★ 7.8ksame name
- sjh00/minimax-h3-storyboard-prompt-skills★ 35same name
- badgids/Story-Film-Skills★ 4same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository15 other files it gives its agents.
- openspec-apply-changeSkill
- openspec-archive-changeSkill
- openspec-exploreSkill
- openspec-proposeSkill
- 3d-animation-short-generatorSkill
- brand-promo-video-generatorSkill
- co-op-game-intro-generatorSkill
- h3-prompt-writingSkill
- minimalist-product-ad-generatorSkill
- minimax-multimodal-toolkitSkill
- minimax-music-genSkill
- minimax-music-playlistSkill
- music-video-subtitle-generatorSkill
- paper-collage-explainer-generatorSkill
- papercraft-stop-motion-explainerSkill
The file
We split this skill into the rules it gives an agent
6 rules. 6 are common, 0 are distinctive, and 0 are very specific to this project.
6
Common
Most of the 7 edited versions we read keep these.
0
Distinctive
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Common: line 38
When the user provides any language, understand it as the following workflow requirements and express the final prompt in the dominant …
- Common: line 78
Do not add titles, auxiliary headings such as 舞台设置 or 色味, or explanatory notes.
- Common: line 94
Keep the prompt executable as a video-generation prompt, not an essay.
- Common: line 97
Do not generate a video, image, audio, storyboard, or intermediate asset until the user explicitly confirms the H3 generation step.
- Common: line 102
Use soft, emotional, comic, cute beats: tiny stumbles, shy gestures, a dot falling behind, a petal sticking to lens, a star dropping into a …
- Common: line 103
Avoid any horror-coded anatomy or threatening motion.
“Common” is measured: kept by at least 6 in 10 of the 7 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Tool mapping (mmx-cli edition)
This skill is the mmx-cli adaptation of the MiniMax-H3 Hub skill of the same name: the creative methodology is unchanged, while Hub canvas tools are replaced by the following portable equivalents.
| Hub tool | mmx-cli / portable equivalent |
hub_generate_image | mmx image generate --prompt "..." --out <path> |
hub_generate_video | mmx video generate --prompt "..." --out <path> (Hailuo; see mmx video generate --help for duration/model flags) |
hub_generate_audio / hub_synthesize_speech | mmx speech synthesize --text "..." --out <path.mp3> |
hub_generate_music | the minimax-music-gen skill (MiniMax music API; mmx has no music command) |
hub_video_edit | ffmpeg assembly (concat / xfade / audio mix) — deterministic, offline |
hub_analyse_media | ffprobe + mmx vision describe <file> |
hub_image_search | user-provided assets first; mmx search query "..." for public references |
hub_canvas_get_node / hub_canvas_group_recent_outputs | the episode working directory: assets-manifest.md, outputs/, decisions.md ledger files |
| choice cards | numbered options presented in chat, choice recorded in decisions.md |
Prerequisites: npm install -g mmx-cli and mmx auth login --api-key sk-xxx
(region auto-detected; mmx quota shows the Token Plan balance). ffmpeg/ffprobe
must be on PATH for assembly and inspection.
Handdrawn Live-Action Fusion Video Generator
Use this Skill when the user wants a finished 15-second, 16:9 live-action-and-hand-drawn fusion video. The output must preserve the structure: a flat hand-drawn luminous animation appears in a real space, clearly contacts live-action hands or objects in the first 0-3 seconds, continuously morphs as one single entity, escapes, and a handheld phone camera follows slightly late.
This Skill organizes the prompt in the user input language and recommends MiniMax H3 as the confirmed generation step. Do not generate video until the user explicitly confirms H3 generation. Do not route to planner or executor for prompt-writing. The final prompt language must follow the dominant language of the user input: Chinese input produces Chinese, English input produces English, Japanese input produces Japanese; for mixed input, use the dominant language; when unclear, use the current conversation language. Only user-required proper nouns, model names, or literal parameters may remain unchanged.
Step 1: Understand the user intent as same-language constraints
When the user provides any language, understand it as the following workflow requirements and express the final prompt in the dominant language of the user input:
- 基于参考 prompt 创作一个全新的 15 秒视频生成 prompt,不要表面模仿原句。
- 必须保留影像结构:实拍空间中出现平面的手绘发光动画;动画与真实手或真实物体接触;同一个存在连续变形并逃跑;相机总是慢半拍追赶。
- 手绘动画质感必须像蜡笔、粉笔、彩色铅笔、粉彩、粗糙笔刷;线条轻微抖动,有涂抹不均、毛边和逐帧重画感。
- 禁止 3DCG、毛绒玩具感、均匀矢量线、平滑霓虹、恐怖怪物、巨大眼睛、裂口、牙齿、威吓、扑咬、突然黑屏、跳吓。
- 0-3 秒必须出现实拍手与手绘动画的清晰接触,例如缠住手指、落在掌心、被抓时逃跑、从指尖诞生。
- 动画必须作为同一个实体连续变形,可在“线条、生物、记号、植物、交通工具、生活小物”等形态之间变化,并保留前一形态的痕迹。
- 不允许突然出现另一个全新角色。
- 全片在同一空间或相邻范围内连续展开,不用剪辑跳到另一个地点,像拍摄者真的边走边追。
- 每个区间 0-3、3-6、6-10、10-13、13-15 秒都要有新的变形、移动、接触、发现、恶作剧或惊喜。
- 拍摄者也必须参与:伸手、抓、追、打开门或盒子、接住、后退、被恶作剧等。
- 调性是可爱、生活感、怀旧、温柔、略带切感,不是恐怖喜剧。
- 13-15 秒必须有空间级变形:之前的线扩散到墙、地板、天花板、窗户、水槽或通道,变成巨大花、星空、夕阳、云、丝带、涂鸦小镇等;结尾要有感动余韵和一点可爱笑点。
Step 2: Invent all creative content fresh
For every run, create a new combination. Do not reuse the reference prompt's dark room, PC, fridge, stars, hearts, blue vortex, butterfly, snake, octopus, hamburger, giant eye, teeth, or dark ending.
Choose new values for all of these:
- 实拍空间,但仍是生活化且可连续追踪的相邻范围;
- 手绘实体或初始形态;
- 中心色彩;
- 连续变形链路;
- 与真实手或实物的接触方式;
- 在空间中的追逐路线;
- 拍摄者反应;
- 最后的空间级变形与可爱笑点。
Good example directions include: 雨天厨房水槽、旧阳台晾衣角、清晨玄关、小书店走廊、火车窗边小桌、浴室镜柜、手作桌、温室通道、自助洗衣店长椅、旧餐桌. Use only one or adjacent connected areas.
Step 3: Required output format
The final prompt must begin with a sentence pattern matching the dominant language of the user input. For Chinese input, use this pattern:
15秒,16:9横版视频。将实拍的〇〇与手绘发光动画融合的影像。
Replace “〇〇” with the newly invented live-action space or everyday scene. For non-Chinese input, use an equivalent opening in the same language and include duration, 16:9 landscape format, live-action space, and hand-drawn glowing animation fusion.
Do not add titles, auxiliary headings such as 舞台设置 or 色味, or explanatory notes. Keep this paragraph order:
- 开头句;
- 实拍空间与手机拍摄质感;
- 0-3 秒;
- 3-6 秒;
- 6-10 秒;
- 10-13 秒;
- 13-15 秒;
- 手绘质感;
- 相机追随方式;
- 禁止事项;
- 环境音。
Step 4: Prompt writing rules
- Keep the prompt executable as a video-generation prompt, not an essay.
- Unless the user asks for explanation, the final answer should contain the prompt text in the dominant language of the user input first, followed by one short next-step recommendation in the same language.
- The recommendation must invite the user, in the same language, to use MiniMax H3 to generate a 15-second 16:9 video from this prompt. Chinese example:
下一步建议:如果你确认这个 prompt,我可以继续用 H3 模型生成 15 秒 16:9 视频。 - Do not generate a video, image, audio, storyboard, or intermediate asset until the user explicitly confirms the H3 generation step.
- Mention Seedance only as a target usage context when the user asks; do not add model parameters inside the creative prompt unless requested.
- The first 0-3 seconds must make real/hand-drawn fusion obvious through contact.
- The camera must not neatly center the animation. It should lag behind, pan/tilt/advance after the entity already leaves the frame edge.
- The entity should remain traceable: each new form preserves a line, tail, color smear, body curve, or motif from the previous form.
- Use soft, emotional, comic, cute beats: tiny stumbles, shy gestures, a dot falling behind, a petal sticking to lens, a star dropping into a palm, confetti sneeze, one small creature lagging behind.
- Avoid any horror-coded anatomy or threatening motion.
- The final prompt must not randomly mix in vocabulary, transition sentences, or writing systems from outside the user input language; when the user input language is Japanese, natural Japanese is allowed.
Step 5: Delivery
Output the final prompt in the dominant language of the user input first. After the prompt, add exactly one short same-language recommendation line inviting the user to continue with MiniMax H3 video generation. Do not include a title, checklist, model call summary, filename, or delivery note. If the user later confirms H3 generation, generate the 15-second 16:9 video with MiniMax H3 and check that the result preserves contact, continuous morphing, delayed camera chase, and non-horror hand-drawn texture.
<!-- QUALITY_BASELINE_V1 -->When to use(什么时候使用)
当用户需要 在明确输入、预算和交付约束后执行生成或写入操作 时加载本技能。先从请求中提取目标、输入、约束、交付格式和验收标准;描述摘要为:For creators making surreal short videos that blend rough glowing hand-drawn animation with live-action spaces. Users provide a scene idea, contact object or hand, desired mood, and optional language or style constraints. The Skill clarifies the physical contact, designs continuous morphing, escape route, and delayed handheld chase movement, then writes a reusable 15-second 16:9 video prompt in the user's language. After user confirmation it recommends MiniMax H3 generation and checks contact realism, camera delay, rough glowing stroke texture, and non-horror tone. Best for single-scene creative clips, not polished CG, horror jump scares, plush characters, or multi-scene cuts.。
Rules
- 先读后写:先确认当前状态与真实能力,再执行会改变外部状态的动作。
- 权限最小化:只使用完成当前步骤所需的文件、工具、账户与网络范围。
- 证据优先:运行结果、资源 ID、版本、哈希或测试输出缺失时,明确标记为
NOT_VERIFIED。 - 幂等优先:保留请求标识与阶段状态;结果不明确时先查询,不进行盲目重试。
- 隐私安全:日志、示例、回执和错误信息不得包含 token、cookie、密钥或个人敏感数据。
Workflow
Step 1:澄清意图
确认本技能是否匹配目标;若只是相邻需求,交给更精确的技能。
Step 2:执行预检
校验输入、模型/工具能力、输出路径、预算上限和审批状态;任一关键条件未知时停止在只读阶段。
Step 3:形成计划
列出将调用的工具、会改变的对象、成功标准以及失败后的安全退出方式。
Step 4:执行动作
按一次批准执行并记录请求标识;模糊结果先查询而不是重提;每个外部调用均保留可关联的状态或回执。
Step 5:验证交付
验证产物存在性、格式、哈希/标识、成本状态和质量门禁,并把事实、推断和未验证项分开陈述。
Validation checklist
- 技能触发条件与用户意图一致,没有把相邻任务误路由到本技能。
- 输入、目标对象、版本和输出位置均已明确,且没有使用猜测值替代必填值。
- 所有写入、付费、发布或不可逆动作都在用户授权范围内。
- 结果已用独立检查验证;仅有“命令成功”或“文件存在”不算完整验收。
- 输出包含实际证据、失败/跳过项、剩余风险和可执行的下一步。
Gotchas
- 把计划当结果:文档或提示词不等于真实执行;必须标明实际运行层级。
- 错误重试:超时或响应丢失可能已经产生远端状态,先查询再决定是否重试。
- 隐式扩大范围:批量、全量、发布、覆盖和付费不是普通读写的自然延伸。
- 版本漂移:引用外部资源时记录版本、tag 或提交;不要把可变分支当发布证据。
- 证据过期:缓存、旧截图和历史测试不能证明当前环境;在交付前刷新关键证据。
不适用与边界
付费、发布、覆盖、上传或外部写入必须使用当前任务的显式授权;不自动扩大次数和预算。 如果请求需要别的技能,不复制其正文;按技能名称进行交接,并保留当前任务上下文。
Progressive disclosure
- 需要确定输入/输出、状态和授权点时,读取
references/workflow-contract.md。 - 需要交付前自检时,读取
references/validation-checklist.md。 - 遇到超时、部分成功或恢复场景时,读取
references/error-recovery.md。 - 首次运行、拒绝越权和失败恢复分别参考
examples/happy-path.md、examples/boundary-refusal.md、examples/failure-recovery.md。
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

