Communitygithub.com

handdrawn-motion

Shows an agent how to plan, build, or cut "handdrawn motion" so the concrete output is a finished video cut. The procedure moves through Bundled Scripts & `BrandConfig` Parameterization, Four-Layer Anti-AI & Anti-Homogenization Architecture (v11.1 Spec), Layer 3: Continuous World & Breath-Valley Camera (`一镜连续世界摄影机`). Use it when a new assignment calls for this craft.

handdrawn-motion とは?

handdrawn-motion is a Claude Code agent skill that shows an agent how to plan, build, or cut "handdrawn motion" so the concrete output is a finished video cut. The procedure moves through Bundled Scripts & `BrandConfig` Parameterization, Four-Layer Anti-AI & Anti-Homogenization Architecture (v11.1 Spec), Layer 3: Continuous World & Breath-Valley Camera (`一镜连续世界摄影机`). Use it when a new assignment calls for this craft.

対応~Claude Code~Codex CLI~Cursor
npx skills add https://github.com/xeronsh/handdrawn-video-skills/tree/HEAD/skills/handdrawn-motion

お気に入りのAIに質問する

このエージェントスキルを事前に読み込んだ状態で新しいチャットを開きます。

ドキュメント

Handdrawn Motion (v11.1 — Chibi Watercolor Studio Craft & Four-Layer Anti-AI Architecture)

Build tactile, signature-craft 1920x1080 @ 60fps (CRF 19) hand-drawn watercolor animations starring any parameterized creator mascot IP (BrandConfig) without naive programmer-art primitives, uncanny stretched bicycle legs, mechanical lerp twinning, or slide-deck cross-fades.

Bundled Scripts & BrandConfig Parameterization

All three scripts live under scripts/ and resolve sibling imports automatically via Path(__file__).resolve().parent:

  1. scripts/cairo_handdrawn.py — BrandConfig (dataclass + from_file / from_dict / from_env + cross-OS font resolver), PyCairo dip-pen (HandDrawnStage, brush_stroke, pencil_underdraw, marker_wash_poly, soft_shadow), 12fps boil_offset_xy, viewer_read_envelope, ContinuousCamera2D, shape_morph_poly, draw_plush_mascot, draw_brand_signature, apply_paper_grain, and lint_and_render_3zoom_qa.
  2. scripts/chibi_watercolor_kit.py — Warm #FAF8F3 cold-press cotton paper (make_paper_stage), pre-densified wet-edge watercolor washes (wet_watercolor_wash), 2.5-head Chibi mascot rig (draw_chibi_mascot), draftsman callouts (draw_draftsman_callout), corner Easter eggs (draw_debug_duck_egg, draw_peek_mini_mascot_egg, draw_sticky_badge), and tactile studio props (draw_glass_kettle, draw_skill_cartridge, draw_poker_chip_stack, draw_compounding_flywheel).
  3. scripts/motion_rig_kit.py — CleanCanvas renderer, fluid 3-joint C/S-spine mascot rig (draw_mascot_actor, draw_mascot_at_computer_desk), official app/social icons (draw_brand_icon, draw_official_app_icon), AI UI/method/slang props (draw_ai_ui_prop, draw_ai_method_prop, draw_ai_jargon_prop), Karplus-Strong stereo Lo-Fi guitar BGM synth (synth_studio_lofi_bgm, mix_stereo_master), raw-VO vowel-nucleus subtitle alignment (align_subs_on_raw_vowel_peaks, align_raw_vo_subtitles), and multi-shard ffmpeg H.264 encoder (encode_turbo_film).

Configure your IP in one call before rendering:

from cairo_handdrawn import BrandConfig, set_brand_config
cfg = BrandConfig.from_file("templates/ip-profile.example.yaml")  # or BrandConfig.from_env()
set_brand_config(cfg)

Four-Layer Anti-AI & Anti-Homogenization Architecture (v11.1 Spec)

Layer 1: Pure 2.5-Head Chibi Silhouette & Natural Watercolor Media (纯正二头身造型与水彩笔刷介质)

  • Pure 2.5-Head Chibi Mascot (chibi_watercolor_kit.py -> draw_chibi_mascot): Render the protagonist in pure 2.5-head Chibi proportions (大圆头 + 小短身 + 小短爪, palette driven by BrandConfig: default #0D0D0D sumi-ink body, #6E6378 muted inner ears, flat-topped half-lidded #2EE59D aloof geek eyes). Never place a Chibi actor on props like two-wheeled crank bicycles that force uncanny human-length legs.
  • Natural Pigment, Pressure & Paper Grain: Render onto warm #FAF8F3 cold-press cotton paper (make_paper_stage, configurable via BrandConfig.paper_hex), pre-densified wet-watercolor washes (wet_watercolor_wash, max_px <= 10.5px), variable-pressure dip-pen contours (brush_stroke), hand-hatched contact shadows (soft_shadow), light draftsman leader-arrow callouts (draw_draftsman_callout), and corner Easter eggs (draw_debug_duck_egg, draw_peek_mini_mascot_egg, draw_sticky_badge).
  • CJK Font Safety & Isolated 12fps (onTwos) Line Boil: Drive primary motion at silky 60fps while locking organic hand-boil to 12fps (onTwos) per element. Fonts resolve automatically via HANDDRAWN_FONT_REGULAR/BOLD/SERIF -> Debian Noto -> Linux/macOS/Windows fallbacks. Avoid unsupported NotoSansCJK glyphs (↻, ✗, ♠, ♦ — draw vector icons or brush strike-throughs instead so zero ☐ tofu boxes appear).

Layer 2: Viewer Single-Read Timing Budget & Metaphor-First Logic (观众单拍读点与声画同构)

  • Metaphor-First 1:1 Audio-Visual Alignment & Single Scale: Align spoken copy and visual props to the exact same physical metaphor (e.g., boiling glass kettle 20°C -> 88°C -> 99°C -> 100°C) and never mix conflicting dual scales (% vs °C).
  • One Read at a Time (viewer_read_envelope): Sequence every beat across Lead Eye (0.30–0.40s) -> Reverse Anticipation (0.15–0.20s) -> Fast Action Snap (0.22–0.28s) -> Payoff Hold (0.85–1.20s), staggering secondary motion (eyes -> head +3f -> torso +6f -> paws +8f -> ears/tail +12f) and reserving accent_hex (#2EE59D) strictly for threshold/payoff states.
  • Draftsman Callouts vs. PPT Clutter: Ban opaque PPT thesis banners, slide boxes, and pop-up glossary cards; use light, curved-arrow draw_draftsman_callout margin annotations (2–6 chars) and corner Easter eggs to reinforce visual causality without repeating full narration.

Layer 3: Continuous World & Breath-Valley Camera (一镜连续世界摄影机)

  • Kill PPT Slide Cross-Dissolves: Never glue isolated slide functions with opacity fades. Place zones across a continuous workbench/stage and glide horizontally (0.48–0.60s smootherstep pan via ContinuousCamera2D or world-X offset) inside <short pause> breath valleys, or relay silhouettes via shape_morph_poly.

Layer 4: Multi-Act 3-Zoom Pre-Render QA Gate (全幕切片三倍率审片门禁)

Never encode a full film blind. Enforce pure-function deterministic f(t) rendering and run lint_and_render_3zoom_qa before full encoding:

  1. Subtitle Safe-Zone Ink Lint: Verify stage props/floor stay above y = 850 so y = 956 40px subtitles (y >= 938 band) have < 16% background ink density.
  2. All-Act Visual Inspection (Look Frames + 0.1s Motion Strip + 1:1 Crops): Inspect exported PNGs (view_file / multimodal read) to verify zero ☐ tofu glyphs, zero ear/arrow text collisions, < 5.0% +16.7ms frame-step stability, and crisp Chibi contact silhouettes.

End-to-End Execution Steps

  1. Load Toolkit & Profile: Load scripts/cairo_handdrawn.py, scripts/motion_rig_kit.py, scripts/chibi_watercolor_kit.py, and references/director-and-style-rules.md; initialize BrandConfig from ip-profile.yaml or env vars.
  2. Synthesize Single-Take <short pause> Raw Voiceover: Author one unified spoken script (<= 5 min) with explicit <short pause> breath cues and spaced acronyms (S O P), generate in a single TTS call (NEVER chop breaths or apply ffmpeg atempo), align all clauses directly on the unedited WAV via align_subs_on_raw_vowel_peaks / align_raw_vo_subtitles, and mix tactile foley + stereo Lo-Fi guitar BGM (0.28 speech / 0.58 pause) via mix_stereo_master.
  3. Pass All-Act 3-Zoom QA Gate -> Parallel Multi-Shard 60fps Encode: Inspect QA contact sheets and 1:1 crops, fix any micro-collisions, and encode 1920x1080 @ 60fps (CRF 19) via encode_turbo_film.

Individual skills in this repo

This repo contains 1 individual skill — each has its own dedicated page.

関連スキル