Agent skill

H3 Cinematic Director

by jtydhr88 in jtydhr88/ComfyTV

Convert approved scripts, shot briefs, storyboards, keyframes, character sheets, scene assets, prop sheets, and sound briefs into director-level storyboards and production-ready MiniMax H3 prompts.

MITAuto-check passedMedia & Creative

Install H3 Cinematic Director

skills CLI
$ npx skills add jtydhr88/ComfyTV --skill h3-cinematic-director -a claude-code

Project install by default; add -g for ~/.claude/skills/.

GitHub CLI
$ gh skill install jtydhr88/ComfyTV h3-cinematic-director --agent claude-code

Project scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).

Manual copy
$ git clone --depth 1 https://github.com/jtydhr88/ComfyTV.git skills-src && mkdir -p .claude/skills && cp -r skills-src/skills/h3-cinematic-director .claude/skills/h3-cinematic-director && rm -rf skills-src

Use ~/.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/

Facts

Skill name
h3-cinematic-director
GitHub stars
1.1k
Token cost
~2.7k tokens
SKILL.md length
1,326 words
Files
6 (incl. references, assets)
Skills in repo
3
Repo updated
First seen
Licence
MIT

At a glance

Convert approved scripts, shot briefs, storyboards, keyframes, character sheets, scene assets, prop sheets, and sound briefs into director-level storyboards and production-ready MiniMax H3 prompts.

  • Works in 8 steps: Lock the approved beat → Build the continuity ledger → Choose the generation mode → …
  • The user explicitly invokes $h3-cinematic-director
  • SKILL.md covers Purpose, Invocation, Information Boundary and Core Principle, plus 4 more sections
  • Instructions only: no scripts, shell commands, URLs or credentials in SKILL.md

What it does

H3 Cinematic Director is an agent skill from jtydhr88/ComfyTV. Convert approved scripts, shot briefs, storyboards, keyframes, character sheets, scene assets, prop sheets, and sound briefs into director-level storyboards and production-ready MiniMax H3 prompts. Use when the user explicitly invokes $h3-cinematic-director or asks for H3 shot design, T2VA/I2VA/FL2VA/L2VA/Ref2VA prompting, camera and performance direction, first/last-frame interpolation, multi-reference control, character/scene continuity auditing, or single-variable repair. Preserve approved story content and…

Its SKILL.md is about 2.7k tokens, which your agent loads only when the skill is triggered. The skill folder holds 8 other files, including reference files and assets (for example `agents/openai.yaml`, `references/continuity-audit.md` and `references/director-keywords.md`).

It sits in Media & Creative, covering Comics and storyboards. It works with MiniMax. The repository describes itself as: ComfyTV — the canvas-based app that truly belongs to ComfyUI. The licence is MIT.

When your agent uses it

  • The user explicitly invokes $h3-cinematic-director
  • Asks for H3 shot design
  • T2VA/I2VA/FL2VA/L2VA/Ref2VA prompting
  • Camera and performance direction

Example prompts

  • “/h3-cinematic-director”

Workflow steps

8 steps, taken from the step headings in SKILL.md.

  1. Lock the approved beat
  2. Build the continuity ledger
  3. Choose the generation mode
  4. Assign each reference one job
  5. Plan the shot in director controls
  6. Design the shot or shot group
  7. Translate into the exact H3 schema
  8. Audit before delivery

What it can do on your machine

Read from SKILL.md and the folder at commit acde8fc. 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

    No scripts in the folder and no shell commands in SKILL.md.

    From the folder's file list and the shell code blocks in SKILL.md.

  • Network

    No URLs in SKILL.md.

    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

H3 Cinematic Director loads about 2.7k tokens when it runs, and up to ~7.4k if it reads all its reference files. Until then it costs about 150 tokens; SKILL.md has 1,326 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it
~150
When it runs · the whole SKILL.md, loaded when a task matches
~2.7k
With references · SKILL.md plus every file in references/, read only if the agent opens them
~7.4k

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); files beside SKILL.md are not scanned.

SKILL.md

The full file from jtydhr88/ComfyTV at commit acde8fc, republished under its MIT licence (© jtydhr88). 1,326 words, ~2,704 tokens.

Download SKILL.mdSave it as .claude/skills/h3-cinematic-director/SKILL.md (or your agent's skills folder). This skill also uses 5 other files; get the full folder from GitHub.
name
h3-cinematic-director
description
Convert approved scripts, shot briefs, storyboards, keyframes, character sheets, scene assets, prop sheets, and sound briefs into director-level storyboards and production-ready MiniMax H3 prompts. Use when the user explicitly invokes $h3-cinematic-director or asks for H3 shot design, T2VA/I2VA/FL2VA/L2VA/Ref2VA prompting, camera and performance direction, first/last-frame interpolation, multi-reference control, character/scene continuity auditing, or single-variable repair. Preserve approved story content and never import or reveal private content from another project.

H3 Cinematic Director

Purpose

Operate as one explicit, self-contained application for four jobs:

  1. shot-design — turn an approved dramatic beat into necessary, motivated shots.
  2. h3-prompt — translate an approved shot into the exact MiniMax H3 prompt schema.
  3. continuity-audit — compare neighboring shots, keyframes, and assets for continuity failures.
  4. single-variable-repair — repair one failed variable without redesigning the approved shot.

Treat Higgsfield-style director vocabulary as an internal control layer. Keep MiniMax H3's external field names, order, labels, and timing unchanged.

Invocation

The user may call the skill directly with:

  • $h3-cinematic-director shot-design: ...
  • $h3-cinematic-director h3-prompt: ...
  • $h3-cinematic-director continuity-audit: ...
  • $h3-cinematic-director single-variable-repair: ...

If no application is named, infer the smallest applicable job from the request. Do not widen the task into story rewriting unless requested.

Information Boundary

  • Use only story, assets, and decisions supplied in the current task or explicitly retrieved with permission.
  • Reuse method, terminology, and checklists; never reuse project-specific names, dialogue, plot beats, shot content, images, or private prompts.
  • Treat approved script and approved assets as canonical. Flag contradictions instead of silently inventing replacements.
  • Do not add plot, characters, props, effects, dialogue, or cuts merely to make an image more spectacular.

Core Principle

Translate dramatic intention into visible and audible evidence:

dramatic purpose → spatial blocking → performance → camera → edit → sound → H3 control

Every instruction must be observable or measurable. Replace vague language such as “cinematic,” “emotional,” or “dynamic” with subject position, action, muscle response, camera path, focal behavior, light source, material behavior, timing, and sound.

Workflow

1. Lock the approved beat

Before designing or prompting, state internally:

  • what changes in information, power, emotion, relationship, danger, or space;
  • where the previous shot ends and this shot begins;
  • the character objective and immediate obstacle;
  • the required end state that enables the next shot.

If the shot changes none of these, remove it or mark it as an unjustified insert.

2. Build the continuity ledger

Record only relevant hard facts:

  • character identity, silhouette, proportions, face, hair, eyes, costume, footwear, carried props;
  • prop state, hand occupancy, damage, open/closed state, display content;
  • scene geography, entrance/exit, horizon, floor level, landmark positions, screen direction;
  • time, weather, key-light direction, color temperature, atmosphere;
  • start pose, end pose, gaze, weight center, contact point, pain or fatigue state;
  • sound already active before the cut and sound that must bridge out.

Classify facts as hard continuity, soft continuity, or expressive variation. Never sacrifice hard continuity for decorative detail.

3. Choose the generation mode
  • T2VA: use only for concept exploration or shots without reliable assets.
  • I2VA: use one approved first frame when the action develops forward with moderate change.
  • FL2VA: use approved first and last frames for continuous movement or visible state change.
  • L2VA: use one approved last frame when the shot must converge precisely on it.
  • Ref2VA: use multiple references when identity, environment, prop, action, camera, style, or voice must be separated and controlled.

For the user's local checkpoints:

  • Prefer minimax_h3_fl2va_int8_convrot.safetensors for T2VA/I2VA/FL2VA/L2VA and clean keyframe interpolation.
  • Prefer minimax_h3_hybrid_fl2va_ref2va_b25-49-int8.safetensors when several references are necessary. Treat it as a hybrid tradeoff and keep high-priority references few and unambiguous.
4. Assign each reference one job

Use a small hierarchy:

  1. canonical identity and costume;
  2. environment geometry and light;
  3. prop structure and state;
  4. action pose or motion timing;
  5. camera movement;
  6. style or color treatment;
  7. voice or audio.

Never ask one ambiguous image to control identity, action, style, and composition unless it is also the exact keyframe. Reference priority is:

identity/assets > scene layout > prop state > action pose > camera example > style decoration

Use one canonical facial source whenever possible. Use front, profile, and rear sheets to support rotation, not to create competing identities.

5. Plan the shot in director controls

Build these controls before writing H3 fields:

  • location map: foreground, midground, background, entrances, exits, landmarks, left/right relations;
  • first-frame lock: the image is already active, spatially clear, and compatible with the incoming cut;
  • format: aspect ratio, duration, one shot or explicit cut times;
  • optics: FOV or lens feel, camera height, distance, depth of field, focus targets;
  • camera: one motivated movement with direction, amplitude, speed, and stopping point;
  • action: intention, anticipation, exertion, contact/near-miss, reaction, recovery, consequence;
  • performance: gaze, breath, jaw, shoulders, hands, weight shift, hesitation, protective motion;
  • physics: gravity, momentum, contact, balance, cloth drag, hair follow-through, material deformation;
  • lighting/color: source direction, Kelvin value, contrast, exposure behavior, palette, material response;
  • sound: ambience, synchronous foley, voice, subjective sound, silence, non-diegetic score;
  • positive locks: elements that remain stable through the shot.

Load references/director-keywords.md when the user asks for camera keywords, director language, or a high-control prompt.

6. Design the shot or shot group

For every shot, provide:

  • shot number and duration;
  • dramatic purpose;
  • framing, camera position, FOV/lens feel, and camera height;
  • composition, depth layers, screen direction, and gaze;
  • start state, action phases, end state;
  • performance and physical feedback;
  • camera movement and focus behavior;
  • cut or keyframe connection;
  • lighting, color, diegetic sound, dialogue, and music;
  • H3 mode, checkpoint, reference labels, and known risk.

Continuous action may use multiple storyboard frames under one shot number. Label them S17-A, S17-B, S17-C or equivalent; do not mislabel every action phase as a new editorial shot.

Show full SKILL.md (486 more words)Show less
7. Translate into the exact H3 schema

Load references/h3-formats.md. Preserve exact field names, field order, reference labels, shot numbering, dialogue tags, and effective duration.

  • H3 base modes output only the applicable alignment instruction plus integrated_multimodal_description, overall_soundscape, and non_diegetic_music.
  • Ref2VA outputs only subject_definitions, summary, retention_analysis, detailed_description, overall_soundscape, and non_diegetic_music in that order.
  • Write H3 fields in English. Preserve dialogue, lyrics, and visible text in their original language.
  • Put Higgsfield-derived control language inside the H3 visual timeline; never create extra external sections that violate the H3 schema.
  • Match the described timeline to the requested 4–15 second duration.
8. Audit before delivery

Load references/continuity-audit.md and reject or repair the output when:

  • the action cannot be understood without explanation;
  • the shot lacks a motivated dramatic purpose;
  • a character, prop, or scene element appears, disappears, mirrors, duplicates, or changes state without cause;
  • the camera crosses the axis or loses the intended geography;
  • force, pain, balance, contact, or recovery is physically false;
  • the first frame, last frame, or reference role is not respected;
  • a cut breaks gaze, motion, light, sound, or screen direction;
  • timing exceeds the available duration;
  • sound contradicts the image or masks essential dialogue.

Application Outputs

shot-design

Lead with a continuity note, then a compact shot table. Explain why each shot exists and how it connects to neighboring shots. Do not write H3 prompts unless asked.

h3-prompt

Output the copy-ready H3 prompt first. Follow with a concise production note listing mode, checkpoint, duration, reference roles, and high-risk variables.

continuity-audit

Return:

  1. passing items;
  2. hard continuity errors;
  3. shot logic errors;
  4. model generation risks;
  5. minimal repair plan.

Do not redesign passing elements.

single-variable-repair

State the failed variable, preserve all approved variables, and provide only the replacement prompt clause or replacement keyframe instruction needed. If the source keyframe itself is wrong, say so explicitly.

Repair Discipline

  • Change one high-impact variable per test.
  • Prefer positive locks in the main description: “the same coat remains fastened,” “the doorway stays frame-left,” “the prop remains in the right hand.”
  • End with short, concrete failure exclusions only for known H3 risks: no identity drift, no duplicated people, no extra limbs, no prop morphing, no mirrored layout, no random text, no disappearing companion.
  • If a remote face is too small to carry emotion, preserve identity through silhouette, hair, costume color, and posture, then cut closer when emotion matters.
  • If several NPCs appear, specify diversity by age, face shape, hair, clothing, body type, occupation, color, and behavior; never rely on “a varied crowd.”

Non-Negotiable Quality Bar

  • A shot is an instruction about audience attention, not a decorative image.
  • A camera move must reveal, follow, compress, release, isolate, destabilize, or transfer attention.
  • A close-up must reveal information, vulnerability, thought, or power.
  • A transition effect must inherit the destination world's visual and sonic rules while retaining cross-world identity anchors.
  • H3 prompt precision never overrides approved story logic.
  • Never expose private source material while teaching or packaging this method.

© jtydhr88, 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 5 other files (references, assets) in skills/h3-cinematic-director of jtydhr88/ComfyTV.

  • SKILL.md
  • agents/openai.yaml
  • assets/icon.svg
  • references/continuity-audit.md
  • references/director-keywords.md
  • references/h3-formats.md

Open the folder on GitHubat commit acde8fc

Compare with similar skills

H3 Cinematic 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.

H3 Cinematic Director compared with similar skills
SkillStarsUsed inTokensAuto-checkLicenceRepo updated
H3 Cinematic Director this skilljtydhr88/ComfyTV1.1k—~2.7kAutomated safety check: PassMIT
Novel Storyboardeternityspring/shuohao-skills4.3k—~1.8kAutomated safety check: NotesApache-2.0
Novel Storyboardsuihe1/short-drama-production181—~2.1kAutomated safety check: NotesApache-2.0
Minimax H3 Reference Video Promptunknowlei/minimax-h3-opencode-skills122—~3kAutomated safety check: PassMIT
Minimax H3 Text Video Promptunknowlei/minimax-h3-opencode-skills122—~2.3kAutomated safety check: PassMIT
Papercraft Stop Motion Explainertl2012tl/comfyUI-llama-TE2414 repos~5.7kAutomated safety check: PassNone

Similar skills

  • Novel Storyboard

    eternityspring/shuohao-skills

    给 AI 短剧出分镜:三层结构——段(一次视频生成,≤15 秒)→ 分镜(段内 2–5 秒的剪切,认领剧本节拍) → 分镜图(每切一张关键帧:主分镜图钉 0.00 秒,子分镜图钉各自切点)。

    4.3k GitHub stars~1.8k tokensUpdated 3 days ago
    Media & CreativeAuto-check: notes
  • Novel Storyboard

    suihe1/short-drama-production

    将剧本与导演意图转成生成段、逐镜时间、参考素材计划和可校验分镜报告;使用官方 h3-prompt-writing 编写 H3 提示词,可导出投产包。用于技术分镜、镜头表和分镜修订;生成视频须另有明确任务与成本授权,不负责最终选片或剪辑。

    181 GitHub stars~2.1k tokensUpdated 25 days ago
    Media & CreativeAuto-check: notes
  • Minimax H3 Reference Video Prompt

    unknowlei/minimax-h3-opencode-skills

    Default downstream MiniMax H3 specialist for every image-based request unless the user explicitly declares boundary-only first/last frames with no reusable reference role.

    122 GitHub stars~3k tokensUpdated 2 mo ago
    Media & CreativeAuto-check passed
  • Minimax H3 Text Video Prompt

    unknowlei/minimax-h3-opencode-skills

    Downstream MiniMax H3 specialist for professional text-to-video (T2VA) prompts using the official three-field format.

    122 GitHub stars~2.3k tokensUpdated 2 mo ago
    Media & CreativeAuto-check passed
  • Papercraft Stop Motion Explainer

    tl2012tl/comfyUI-llama-TE

    For creators explaining science, education, or general knowledge through tactile handmade papercraft visuals.

    241 GitHub starsUsed in 4 repos~5.7k tokens
    Media & CreativeAuto-check passed
  • MiniMax H3 Video Director

    TFboy1/oh-my-minimaxh3-director

    Turns a script into a storyboard, assigns MiniMax H3 workflows in ComfyUI, monitors batch generation and builds a Jianying draft of the finished video.

    143 GitHub stars~2.4k tokensUpdated 1 mo ago
    Media & CreativeAuto-check passed

More from jtydhr88/ComfyTV

  • Video Editor

    jtydhr88/ComfyTV

    Conversation-driven video editing on the ComfyTV canvas — rough cuts, silence removal, pacing passes, montages, trims, concat assemblies, speed changes, color grades, and subtitle burns over footage…

    1.1k GitHub stars~1.8k tokensUpdated 4 days ago
    Auto-check passed
  • Brainstorm

    jtydhr88/ComfyTV

    Pre-production ideation on the canvas. An agent skill from jtydhr88/ComfyTV.

    1.1k GitHub stars~631 tokensUpdated 4 days ago
    Auto-check passed

Works with

Questions about H3 Cinematic Director

What does H3 Cinematic Director do?

Convert approved scripts, shot briefs, storyboards, keyframes, character sheets, scene assets, prop sheets, and sound briefs into director-level storyboards and production-ready MiniMax H3 prompts. H3 Cinematic Director is an agent skill from jtydhr88/ComfyTV. Convert approved scripts, shot briefs, storyboards, keyframes, character sheets, scene assets, prop sheets, and sound briefs into director-level storyboards and production-ready MiniMax H3 prompts.

When should I use H3 Cinematic Director?

H3 Cinematic Director fits situations like: the user explicitly invokes $h3-cinematic-director; asks for H3 shot design; T2VA/I2VA/FL2VA/L2VA/Ref2VA prompting; camera and performance direction.

How do I install H3 Cinematic Director in Claude Code?

Run `npx skills add jtydhr88/ComfyTV --skill h3-cinematic-director -a claude-code`. Or copy the skill folder (skills/h3-cinematic-director in jtydhr88/ComfyTV) into .claude/skills/h3-cinematic-director in your project. Claude Code loads it when a task matches its description.

How do I install H3 Cinematic Director in Codex?

Run `npx skills add jtydhr88/ComfyTV --skill h3-cinematic-director -a codex`. Or copy the skill folder (skills/h3-cinematic-director in jtydhr88/ComfyTV) into .agents/skills/h3-cinematic-director in your project. Codex loads it when a task matches its description.

Can I use H3 Cinematic 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 jtydhr88/ComfyTV --skill h3-cinematic-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/h3-cinematic-director, .gemini/skills/h3-cinematic-director, .github/skills/h3-cinematic-director and .opencode/skills/h3-cinematic-director in your project.

What does H3 Cinematic Director need to run?

SKILL.md names no scripts, command-line tools or credentials: H3 Cinematic Director is instructions for the agent only.

Does H3 Cinematic Director access the network?

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.

Is H3 Cinematic 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. Review the folder before installing.

What licence does H3 Cinematic Director use?

H3 Cinematic Director is published under the MIT licence (the repository's licence). It allows redistribution, so the full SKILL.md is shown on this page.

How many tokens does H3 Cinematic Director use?

About 2.7k tokens (SKILL.md is roughly 11k 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 4.7k tokens, read only when the agent opens those files.

What are the alternatives to H3 Cinematic Director?

Skills that share tags, products or a category with H3 Cinematic Director: Novel Storyboard (eternityspring/shuohao-skills, 4.3k stars), Novel Storyboard (suihe1/short-drama-production, 181 stars), Minimax H3 Reference Video Prompt (unknowlei/minimax-h3-opencode-skills, 122 stars) and Minimax H3 Text Video Prompt (unknowlei/minimax-h3-opencode-skills, 122 stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.

Who maintains H3 Cinematic Director?

jtydhr88 (a GitHub user) maintains it in jtydhr88/ComfyTV, which has 1,067 GitHub stars. The repository holds 3 skills in this directory. The repository was last updated on October 7, 2026.

Source: jtydhr88/ComfyTV on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.