render-food-product-sizzle is the deterministic, $0 assembly stage of the food-product sizzle\nformat. The paid stages (the ~4 photographic macro keyframes, the ~4 locked-off i2v clips, the\nnon-diegetic music bed) are separate capabilities — create-image-fal, create-video-fal,\ncreate-music-elevenlabs. This capability spends nothing: it takes the ~4 clips + the music bed +\nthe diegetic SFX + the real logo PNG + the real product PNG and stitches the finished master.\nRe-cuts (new callout timing, a swapped end card, re-timed SFX cues, a grain-pass toggle) reuse the\nexisting keyframes / clips / music and cost $0.
config.example.json is the worked example (Lineage Provisions \"Beef Sticks Sizzle\", ~14s\n1080×1920). PIPELINE.md maps every config block to its source step. This README documents the FREE\nassembly pieces that render-food-product-sizzle owns.
The end card is composed entirely in PIL (NO AI): the ivory (or brand-color) bg + the real logo\nPNG (upper third, ~62% width) + the real product PNG (centered, with a soft shadow) + a serif\nheritage headline + a smaller CTA line. The brand text is never AI-rendered — a diffusion model\ngarbles a wordmark and the packaging. On macOS pick a serif with the em-dash / middle-dot glyph\n(use ·). It holds ~3s after the body with the music still playing under it.
Two short diegetic hits are synthesized with ffmpeg lavfi anoisesrc (NO AI): a crisp ~120ms snap\n(white noise, fast envelope, highpass / lowpass, acompressor) on the fiber tear, and a ~180ms tear\n(pink noise, low-end roll) on the box-open. They mix in at their measured cue points (~1.5s and\n~11s in the demo), not round numbers — a couple of hits, not a wall of sound.
The supplied / generated acoustic-bluegrass bed (no vocals — no VO to duck under) opens sparse; the\nupstream step trims the ~2.5s intro so it kicks in from frame 0. The assembly loudnorms it and\nfades it in / out to the master length, playing it UNDER the end card with a fade tail (no silent\ntail).
FFmpeg stitches the master: normalize fps / SAR across the ~4 body clips, concat them in scene order,\napply the anti-AI grain pass GLOBALLY (eq=contrast=1.06:saturation=0.93,hqdn3d=1.5:1.5:3:3,noise=alls=8:allf=t+u\n— load-bearing for the tactile / photographic read), composite the audio (music bed + snap.wav +\ntear.wav at their cues), append the PIL end card, burn the OPTIONAL serif stat-callout pills at\ntheir beats, mux with a fade tail, and loudnorm. Output is a 1080×1920 h264 + aac master (~14s).\nDeterministic, no paid calls, no keys.
% in a callout — ffmpeg drawtext reads a raw % as a strftime spec; write the callout string\nto a textfile and use textfile= + expansion=none (see the source build_master_v2.sh c1.txt /\nc5.txt).
No libass needed — the sizzle body carries no burned captions (it is wordless); the stat-callout\npills are drawtext / drawbox overlays, not subtitle events.