render-glassy-matte-grwm

Name: render-glassy-matte-grwm
Author: gooseworks-ai

1.0kstars193forks

How to install render-glassy-matte-grwm?

gooseworks-ai/goose-skills/render-glassy-matte-grwm

$npx -y skills add gooseworks-ai/goose-skills --skill render-glassy-matte-grwm

Installs into the current project.

Summary

Assemble a multi-scene GRWM beauty-demo ad from a config — a locked-identity creator applies ~5 products step by step while a SEPARATE ElevenLabs voiceover narrates and every scene cut is snapped to the VO's product-name word-starts (Whisper word-level timestamps), then ~5 Playwr

Files · 1

View on GitHub

SKILL.md

1	# render-glassy-matte-grwm
2
3	Assemble a multi-scene GRWM beauty-demo ad from a config — a locked-identity creator applies
4	~5 makeup/skincare products step by step at a vanity, a separate ElevenLabs voiceover
5	narrates the routine, and every scene cut is snapped to the VO's product-name word-starts, with
6	~5 Playwright product overlay cards on the product-name beats, a ducked music bed, burned
7	captions, and a flat-lay end card. This capability is the FREE, deterministic assembly —
8	the Whisper-driven re-cut + hard-concat, the Playwright card render + card composite, the VO +
9	music mix, the caption burn, and the flat-lay end card.
10
11	This is the multi-scene beauty demo, distinct from the single-take apparel outfit-reveal
12	(`ugc-grwm`, one Seedance reference-to-video call with native lip-sync and minimal post). Here the
13	timeline is driven by a SEPARATE VO and the scenes are re-cut to its word-starts.
14
15	`scripts/config.example.json` is the worked example (DIBS Beauty "5-Step Glassy Matte Routine",
16	~32s 1080×1920 9:16, 12 VO-snapped cuts + 5 product cards); `scripts/PIPELINE.md` maps every
17	config block to its source step and `scripts/README.md` documents the free assembly.
18
19	## Run
20
21	This is the FREE, deterministic assembly stage — it spends nothing. The paid inputs are
22	separate capabilities — the SEPARATE narration VO (`create-music-elevenlabs`, or a user-supplied
23	mp3; word-level Whisper timestamps set the timeline), ~7 Seedance scene clips one per product step
24	(`create-video-fal`), the ~5 white-bg product cutouts + the flat-lay end-card still
25	(`create-image-gpt-image-fal`), and the ducked music bed. Given the VO + `.words.json` + one clip
26	per step + the ~5 product cutouts + the music bed, `render-glassy-matte-grwm` re-cuts each clip to
27	its VO word-start window, hard-concats on the cut, renders + composites the product cards on the
28	product-name beats, mixes the VO over the ducked music, burns the captions, and appends the
29	flat-lay end card → the master. Re-cuts reuse the existing VO / clips / cutouts and cost $0.
30
31	## Contract (the free assembly)
32
33	- The SEPARATE VO drives the timeline — Whisper it first. The narration is a separate track
34	(not a native take). Its word-level timestamps set every cut; the atempo'd VO ends shorter than
35	the plan expects (a 1.15× VO landed ~27.5s), so time every window to the word-starts, never to a
36	pre-planned grid.
37	- Scene cuts snap to the "step N" word-start; cards snap to the product-NAME word-start. Cut to
38	the next product when its step is announced; the card animates in ~1s later when the NAME is
39	spoken. Both happen. ~12 cuts over ~32s (cuts/10s ≈ 3.75).
40	- Hard-concat with a re-encode. Hard cuts on the VO word-starts, no dissolves; re-encode the
41	concat `-c:v libx264 -crf 20` — `-c copy` corrupts the duration when zoompan/PNG clips are in the
42	chain.
43	- Product cards — Playwright, real cutout, PDP-verified tagline. Playwright renders the card
44	template at 2× scale (real white-bg cutout thumb + name + PDP tagline). The cutout must match the
45	REAL product, not the Seedance scene's hallucinated barrel; the tagline is verified against the
46	brand PDP (AI flat-lays hallucinate sublines). Composite each card onto the master snapped to its
47	product-NAME word-start, 1s fade-in, held until the next product is named. **PNG overlay inputs
48	need `-loop 1 -t <dur>`** — without it the PNG emits one frame at t=0 and the fade/enable filters
49	silently no-op (cards go invisible).
50	- VO leads the ducked music bed. Mix the SEPARATE VO on top of the ducked music (the VO is the
51	lead), `loudnorm I=-14`. If the host ffmpeg lacks a filter, apad/atrim to length before the mix.
52	- Captions — clean-white, override the preset. Clean-white captions from the VO's Whisper
53	words, overridden to 3 words/cue, ~3.0% font, ~20% margin, NO pill, NO shadow (the default
54	5-words/4.5%/18% reads too dense). Burn last. If the host ffmpeg lacks libass, render the cues as
55	timed PIL PNG overlays co

People who liked this also viewed

6 picks

Type

render-glassy-matte-grwm

1.0kstars193forks

View on GitHub

How to install render-glassy-matte-grwm?

gooseworks-ai/goose-skills/render-glassy-matte-grwm

$npx -y skills add gooseworks-ai/goose-skills --skill render-glassy-matte-grwm

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# render-glassy-matte-grwm
2
3	Assemble a multi-scene GRWM beauty-demo ad from a config — a locked-identity creator applies
4	~5 makeup/skincare products step by step at a vanity, a separate ElevenLabs voiceover
5	narrates the routine, and every scene cut is snapped to the VO's product-name word-starts, with
6	~5 Playwright product overlay cards on the product-name beats, a ducked music bed, burned
7	captions, and a flat-lay end card. This capability is the FREE, deterministic assembly —
8	the Whisper-driven re-cut + hard-concat, the Playwright card render + card composite, the VO +
9	music mix, the caption burn, and the flat-lay end card.
10
11	This is the multi-scene beauty demo, distinct from the single-take apparel outfit-reveal
12	(`ugc-grwm`, one Seedance reference-to-video call with native lip-sync and minimal post). Here the
13	timeline is driven by a SEPARATE VO and the scenes are re-cut to its word-starts.
14
15	`scripts/config.example.json` is the worked example (DIBS Beauty "5-Step Glassy Matte Routine",
16	~32s 1080×1920 9:16, 12 VO-snapped cuts + 5 product cards); `scripts/PIPELINE.md` maps every
17	config block to its source step and `scripts/README.md` documents the free assembly.
18
19	## Run
20
21	This is the FREE, deterministic assembly stage — it spends nothing. The paid inputs are
22	separate capabilities — the SEPARATE narration VO (`create-music-elevenlabs`, or a user-supplied
23	mp3; word-level Whisper timestamps set the timeline), ~7 Seedance scene clips one per product step
24	(`create-video-fal`), the ~5 white-bg product cutouts + the flat-lay end-card still
25	(`create-image-gpt-image-fal`), and the ducked music bed. Given the VO + `.words.json` + one clip
26	per step + the ~5 product cutouts + the music bed, `render-glassy-matte-grwm` re-cuts each clip to
27	its VO word-start window, hard-concats on the cut, renders + composites the product cards on the
28	product-name beats, mixes the VO over the ducked music, burns the captions, and appends the
29	flat-lay end card → the master. Re-cuts reuse the existing VO / clips / cutouts and cost $0.
30
31	## Contract (the free assembly)
32
33	- The SEPARATE VO drives the timeline — Whisper it first. The narration is a separate track
34	(not a native take). Its word-level timestamps set every cut; the atempo'd VO ends shorter than
35	the plan expects (a 1.15× VO landed ~27.5s), so time every window to the word-starts, never to a
36	pre-planned grid.
37	- Scene cuts snap to the "step N" word-start; cards snap to the product-NAME word-start. Cut to
38	the next product when its step is announced; the card animates in ~1s later when the NAME is
39	spoken. Both happen. ~12 cuts over ~32s (cuts/10s ≈ 3.75).
40	- Hard-concat with a re-encode. Hard cuts on the VO word-starts, no dissolves; re-encode the
41	concat `-c:v libx264 -crf 20` — `-c copy` corrupts the duration when zoompan/PNG clips are in the
42	chain.
43	- Product cards — Playwright, real cutout, PDP-verified tagline. Playwright renders the card
44	template at 2× scale (real white-bg cutout thumb + name + PDP tagline). The cutout must match the
45	REAL product, not the Seedance scene's hallucinated barrel; the tagline is verified against the
46	brand PDP (AI flat-lays hallucinate sublines). Composite each card onto the master snapped to its
47	product-NAME word-start, 1s fade-in, held until the next product is named. **PNG overlay inputs
48	need `-loop 1 -t <dur>`** — without it the PNG emits one frame at t=0 and the fade/enable filters
49	silently no-op (cards go invisible).
50	- VO leads the ducked music bed. Mix the SEPARATE VO on top of the ducked music (the VO is the
51	lead), `loudnorm I=-14`. If the host ffmpeg lacks a filter, apad/atrim to length before the mix.
52	- Captions — clean-white, override the preset. Clean-white captions from the VO's Whisper
53	words, overridden to 3 words/cue, ~3.0% font, ~20% margin, NO pill, NO shadow (the default
54	5-words/4.5%/18% reads too dense). Burn last. If the host ffmpeg lacks libass, render the cues as
55	timed PIL PNG overlays co

People who liked this also viewed

6 picks

Type