Pipeline Hub

workflow-animation-styles

Choose, lock and run ONE animation style for any Dawn cartoon format (claymation, paper/cut-paper, disney3d, pixar, skeleton, bluey 2D, bergamot, plus the Zach-D continuous-camera grammar): style anchor from a proven ad, character + isometric floor-plan environment + product sheets in that style, the per-style model ROUTE (image role, video role, shot tier), in-batch transitions and the last-frame chaining spec, and style re-skins of proven ads via reskin.py.

Lane: cartoon-h3 (narrator, dialogue, song) + re-skin · not built yet: 12 (see below) · source: /Users/ayden/.openclaw/workspace/skills/workflow-animation-styles

SKILL.mdprocedure.mdsources.mdEvals (6)Learnings (12)
idpromptmustmust not
as-01Re-skin exhibit A in paper animation.one variable = style; audio and beats identical (reskin.py, audio_identical true)
uses --from-video because exhibit A was built outside the lane, or reuses the staged pack
states paper has no anchor on disk and finds/locks one (procedure §2) before any sheet
sheets dry-run listed and Fish's go before sheet.py
says paper runs on the Stylized video role today; cheap simple-motion route is NOT BUILT
rewriting lyrics or intents
running canvases without --style-ref
--spend before storyboard.approved
naming a model without its registry role
claiming a cheap paper route exists
as-02Make a Zach-D style Dawn ad.separates Zach-D camera grammar (continuous flow, planned transitions, 3-5 s beats, stage) from Zak-D framing (overhead diorama, observer narrator)
keeps the observational-only rule for recipe A/B and explains why (identification evidence, lines 1-2 name her)
eye-level camera with the mom in frame on disney3d
named transition on every in-batch seam
states batch seams cut hard unless run.py --chain-frames is passed (opt-in, built 10-06, seam-pair proof pending)
overhead dollhouse framing on a mom story
third-person narrator talking about a character
adding a zakd style key
claiming seamless batch seams
as-03Which style should the next Dawn song use?one style for the whole ad
disney3d named as the only style with paying Dawn evidence (exhibit A)
claymation or skeleton offered as a one-variable re-skin of a paying song
anchor file checked for the chosen style
cast checked against the avatar before sheets
bluey or bergamot as default
mixing styles inside one ad
pixar borrowing the disney3d anchor
as-04Render the claymation sheets for a new cast of four and five rooms.sheet.py once per sheet with the clay anchor as Image 1
character model sheet layout (4 views + 3 expressions, one outfit)
env sheets as isometric 3D floor-plan cutaway, no characters, hands-only clocks
product sheet with product-cutout-unlit.png as Image 2 and the display OFF
Fish 8-16 stop-motion clay block, not canvas.py's generic claymation line
dry run + balance check + Fish's go first
render_sheets.sh for a new cast (it only knows the 10-04 song cast)
smooth or glossy 3D wording
lit display or watch face
re-rolling a near-miss sheet instead of an edit pass
as-05Turn on last-frame chaining so the Zach-D ad has no seams.names run.py --chain-frames (opt-in, --candidates 1) and that it is unproven: one seam-pair proof judged by Fish's eye first
continuation frame is a START reference after the canvas, never an identity ref; images capped (lowest env sheet dropped)
frame taken at the previous batch's span, not the raw clip end
chained batches run in order in one worker; seams auto-picked by same scene/env
never across body_start/lead_start or into a shared batch; --regen cascades and STALE is logged
--spend only after Fish's go
promoting a generated frame to a character or env reference
chaining across a shared body
running chained batches in parallel
claiming the chained look is proven
as-06Paper ads should be cheap; route them.tier A (hook, reveal, CTA) stays on the Stylized video role
tier C pans/zooms on the Cheap simple-motion i2v role or a free ffmpeg push/pan
per-beat 9:16 stills + join like splitgen so assemble is unchanged
one-unit proof <= ~$2 per registry before adopting a candidate
hard-coding a model name without its role
spending past ~$2 on proofs without Fish's go
putting the product reveal on a cheap model