render-absurdist-explainer

Name: render-absurdist-explainer
Author: gooseworks-ai

1.0kstars193forks

How to install render-absurdist-explainer?

gooseworks-ai/goose-skills/render-absurdist-explainer

$npx -y skills add gooseworks-ai/goose-skills --skill render-absurdist-explainer

Installs into the current project.

Summary

Assemble an absurdist animated-explainer video ad (~38s, 9:16) from per-scene i2v clips + their measured VO windows — retime each clip to its VO, re-encode every segment to identical 30fps/libx264/yuv420p so the concat demuxer never drops frames, concat, build a REAL-product PIL

Files · 1

View on GitHub

SKILL.md

1	# render-absurdist-explainer
2
3	The free, deterministic renderer for the absurdist-explainer video ad format — the
4	bright Pixar/Disney 3D spot where a personified villain (the problem) narrates the whole
5	ad in one voice, teaches the product's ownable mechanism through cartoon biology, lists
6	the damage, then watches its own scheme collapse when the product arrives. This
7	capability is the FREE assembly stage only. All generative work (nano-banana
8	keyframes, Seedance i2v clips, ElevenLabs VO + music) happens upstream in the recipe and
9	is handed to this capability as files.
10
11	It ports the validated compose recipe from two reference runs (HUM "Big Chill" cortisol
12	absurdism and Soteri "Eczema, the pH villain"). The recipe is deterministic — iterate the
13	cut for free, re-roll only the offending paid beat.
14
15	## What it does (the deterministic recipe)
16
17	1. Per-scene retime. Each i2v clip is retimed to its measured VO window
18	(`scale=1080:1920:force_original_aspect_ratio=increase,crop=1080:1920,fps=30,setsar=1`,
19	then `tpad=stop_mode=clone` if the VO is longer than the clip, else `-t` trim).
20	2. Identical re-encode. Every segment is re-encoded `libx264 -crf 18 -pix_fmt yuv420p
21	-r 30` even if already correct — a framerate mismatch makes the concat demuxer silently
22	drop frames.
23	3. Concat all scene segments + the end card via the concat demuxer (`-c copy`).
24	4. Real-product end card. `build_endcard.py` composites the REAL retail product photo
25	over the brand palette (flat, or sampled from the photo's own edge pixel) with a typeset
26	wordmark + claim rows + CTA pill in PIL `ImageDraw.text` — **never an AI cartoon bottle,
27	never AI-rendered brand text**. `compose.py` Ken-Burnses it 1.00 → 1.04 over the dwell.
28	5. Mix. VO bus `loudnorm I=-14 TP=-1.5`, music bus `loudnorm I=-26 TP=-3` then
29	`volume=0.62`, `amix inputs=2 duration=first normalize=0` → master lands at
30	-14.5..-13.5 LUFS with the music ducked under the VO.
31	6. Captions last. `make_captions.py` emits a libass `.ass` (one cue per scene, Arial 64
32	white / 6px outline / MarginV=330, `start = scene_start + 0.08s`, suppressed on the end
33	card). `compose.py` burns it as the final filter so captions sit on top.
34
35	## Scripts (free — Python + ffmpeg + PIL, no bash, no paid calls)
36
37	- `scripts/build_endcard.py` — PIL composite of the real product photo + typeset brand
38	layer (wordmark / product line / claim rows / accent CTA pill). Reads the same
39	`config.json`. Run this FIRST so `end_card.image` exists before `compose.py`.
40	- `scripts/make_captions.py` — emits the per-scene libass `.ass` from the SAME scene table
41	compose reads, so caption windows stay in lockstep with the cut. Run before `compose.py`
42	(or point `config.captions_ass` at nothing to skip captions).
43	- `scripts/compose.py` — the assembler: per-scene retime + identical 30fps re-encode →
44	concat → Ken-Burns end card → VO/music loudnorm mix → burn captions → master mp4.
45	- `scripts/config.example.json` — the shape of the `config` the recipe binds (the
46	brand-neutralised Soteri values as a worked reference).
47
48	## Inputs (all via `--config` + a runtime work dir — NO hardcoded paths)
49
50	`config.json` carries: `scenes[]` (each `{id, clip, target_sec, vo, caption, atempo?}`
51	where `target_sec` is the measured VO window), `end_card{product_image, image,
52	dwell_sec, zoom_to, wordmark, product_line, claims[], cta, background?}`, `brand_palette
53	{primary, primary_lite, accent, grey}`, `music_bed`, `music_volume` (default 0.62),
54	`atempo` (compose-stage VO speed-up, default off; the reference runs used 1.3 when the VO
55	read slow), `captions_ass`, and `caption_style`. See `config.example.json`.
56
57	## Craft rules (load-bearing — faithful to the source molecule)
58
59	- The end card is the REAL product photo, composited — never an AI cartoon bottle.
60	Both reference runs shipped an AI bottle first and had to re-shoot with the real photo.
61	- No AI-rendered brand text anywhere. Wordmark, claims, CTA, motif — all PIL
62	`ImageDraw.text`. AI draws the world + characters only.
63	- Re-encode every segment to 30fps before concat, even if already correct, or the
64	concat demuxer silently drops frames.
65	- **`target_sec` is the ME

People who liked this also viewed

6 picks

Type

render-absurdist-explainer

1.0kstars193forks

View on GitHub

How to install render-absurdist-explainer?

gooseworks-ai/goose-skills/render-absurdist-explainer

$npx -y skills add gooseworks-ai/goose-skills --skill render-absurdist-explainer

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# render-absurdist-explainer
2
3	The free, deterministic renderer for the absurdist-explainer video ad format — the
4	bright Pixar/Disney 3D spot where a personified villain (the problem) narrates the whole
5	ad in one voice, teaches the product's ownable mechanism through cartoon biology, lists
6	the damage, then watches its own scheme collapse when the product arrives. This
7	capability is the FREE assembly stage only. All generative work (nano-banana
8	keyframes, Seedance i2v clips, ElevenLabs VO + music) happens upstream in the recipe and
9	is handed to this capability as files.
10
11	It ports the validated compose recipe from two reference runs (HUM "Big Chill" cortisol
12	absurdism and Soteri "Eczema, the pH villain"). The recipe is deterministic — iterate the
13	cut for free, re-roll only the offending paid beat.
14
15	## What it does (the deterministic recipe)
16
17	1. Per-scene retime. Each i2v clip is retimed to its measured VO window
18	(`scale=1080:1920:force_original_aspect_ratio=increase,crop=1080:1920,fps=30,setsar=1`,
19	then `tpad=stop_mode=clone` if the VO is longer than the clip, else `-t` trim).
20	2. Identical re-encode. Every segment is re-encoded `libx264 -crf 18 -pix_fmt yuv420p
21	-r 30` even if already correct — a framerate mismatch makes the concat demuxer silently
22	drop frames.
23	3. Concat all scene segments + the end card via the concat demuxer (`-c copy`).
24	4. Real-product end card. `build_endcard.py` composites the REAL retail product photo
25	over the brand palette (flat, or sampled from the photo's own edge pixel) with a typeset
26	wordmark + claim rows + CTA pill in PIL `ImageDraw.text` — **never an AI cartoon bottle,
27	never AI-rendered brand text**. `compose.py` Ken-Burnses it 1.00 → 1.04 over the dwell.
28	5. Mix. VO bus `loudnorm I=-14 TP=-1.5`, music bus `loudnorm I=-26 TP=-3` then
29	`volume=0.62`, `amix inputs=2 duration=first normalize=0` → master lands at
30	-14.5..-13.5 LUFS with the music ducked under the VO.
31	6. Captions last. `make_captions.py` emits a libass `.ass` (one cue per scene, Arial 64
32	white / 6px outline / MarginV=330, `start = scene_start + 0.08s`, suppressed on the end
33	card). `compose.py` burns it as the final filter so captions sit on top.
34
35	## Scripts (free — Python + ffmpeg + PIL, no bash, no paid calls)
36
37	- `scripts/build_endcard.py` — PIL composite of the real product photo + typeset brand
38	layer (wordmark / product line / claim rows / accent CTA pill). Reads the same
39	`config.json`. Run this FIRST so `end_card.image` exists before `compose.py`.
40	- `scripts/make_captions.py` — emits the per-scene libass `.ass` from the SAME scene table
41	compose reads, so caption windows stay in lockstep with the cut. Run before `compose.py`
42	(or point `config.captions_ass` at nothing to skip captions).
43	- `scripts/compose.py` — the assembler: per-scene retime + identical 30fps re-encode →
44	concat → Ken-Burns end card → VO/music loudnorm mix → burn captions → master mp4.
45	- `scripts/config.example.json` — the shape of the `config` the recipe binds (the
46	brand-neutralised Soteri values as a worked reference).
47
48	## Inputs (all via `--config` + a runtime work dir — NO hardcoded paths)
49
50	`config.json` carries: `scenes[]` (each `{id, clip, target_sec, vo, caption, atempo?}`
51	where `target_sec` is the measured VO window), `end_card{product_image, image,
52	dwell_sec, zoom_to, wordmark, product_line, claims[], cta, background?}`, `brand_palette
53	{primary, primary_lite, accent, grey}`, `music_bed`, `music_volume` (default 0.62),
54	`atempo` (compose-stage VO speed-up, default off; the reference runs used 1.3 when the VO
55	read slow), `captions_ass`, and `caption_style`. See `config.example.json`.
56
57	## Craft rules (load-bearing — faithful to the source molecule)
58
59	- The end card is the REAL product photo, composited — never an AI cartoon bottle.
60	Both reference runs shipped an AI bottle first and had to re-shoot with the real photo.
61	- No AI-rendered brand text anywhere. Wordmark, claims, CTA, motif — all PIL
62	`ImageDraw.text`. AI draws the world + characters only.
63	- Re-encode every segment to 30fps before concat, even if already correct, or the
64	concat demuxer silently drops frames.
65	- **`target_sec` is the ME

People who liked this also viewed

6 picks

Type