the UGC fix-loop toolkit — surgically re-render a bad window/beat of a single-take UGC master (stitch_replacement.py, pure FFmpeg) and GPT cross-model review a Seedance prompt before render (vet_seedance_prompt.py, routed through the openai-proxy). Fetch it into a one-shot UGC recipe so both scripts resolve on any machine and the vet call bills the Ads agent.

Assemble an expert/educator motion-graphic LISTICLE video ad from a config — a spoken authoritative voiceover carries a numbered listicle while N web-animated hyperframe beats (HTML plus the Web Animations API, one branded design system of alternating tiles, big hero numerals, and glass-pill callouts) are rendered frame-by-frame via Playwright and anchored to the VO's word-level timestamps, periodic color-graded B-roll windows give visual breath, and captions burn ONLY inside those B-roll windows (2-word chunks, ASS Format header carrying a Name field so none drop) with the VO mixed under a low music bed. This is the FREE deterministic assembly stage (Playwright beat render plus ffmpeg concat plus window-masked caption burn plus VO-and-music mix plus final composite) — the VO, the music bed, and the stock B-roll come from create-vo-elevenlabs, create-music-elevenlabs, and media-proxy. Use for the vo-anchored-motion-listicle format.

Assemble a short-form 'vignette' ad from clean product cutouts composited over a kinetic background video — birefnet cutout, then a cold-open text card + product carousel + annotated specimen-sheet end card, plus loudnorm + separate-pass music mux. FREE assembly (PIL + rsvg + FFmpeg); the recipe supplies the config and gates the paid BG/cutout/music calls to their own capabilities. Use for the vignette format.

Render a designed 'value prop' video from a config — 3-5 noun-phrase benefit claims (<=4 words each) revealed sequentially over per-SKU product visuals, one crisp editorial frame per claim (hook sticker -> N claim beats -> brand end card). Deterministic PIL/HTML beat renderer frame-stepped via Playwright and encoded with FFmpeg, sound-off legible, hard cuts, uniform pacing. FREE (no paid calls); music is added separately (create-music-elevenlabs). Use for the value-prop format.

Assemble a stop-motion hand-swatch-cycle product-demo ad from a config — a sequence of still PLATES (one hand swiping a single-barrel cosmetic across a cream skin-patch, the barrel + swatch changing per plate while the hand, background, crop, and lighting stay locked) is PNG→mp4 loop-encoded at each plate's own stop-motion hold (fast motion frames 150–250ms, per-shade ~380ms, hero beats 1100–1800ms), concat-demuxed with HARD cuts into a silent master, closed on a Playwright HTML-rendered branded end card (serif tagline + sans subtitle + real logo SVG over a hero BG, never AI-rendered text), and muxed with a pre-sourced music track playing under the end card with a fade tail (no VO). This is the FREE deterministic assembly stage (loop-encode + concat-demux + end-card render + music mux); the master-anchor plate, shade plates, and end-card BG come from create-image-gpt-image-fal and the track from create-music-elevenlabs. Use for the stopmotion-hand-swatch-cycle format.

Assemble a split-screen creator ad from a config — a two-zone vertical composite where a supplied AI-creator lip-sync take fills the BOTTOM ~48% while real 16:9 product/demo clips run uncropped in the TOP ~52%, each top clip contain-fit with a darkened blurred cover-scale fill of the same clip (never black bars), a 3px brand-color divider between the zones, the creator slice cover-fit per the per-scene VO timing, scenes hard-concatenated with the body audio being the concatenated creator VO slices, an end card held on the last sharp frame ~3s, then the ASSEMBLED cut transcribed with local Whisper (not the raw VO — concat drops inter-scene silence) and word-level captions burned in the chosen style. This is the FREE deterministic assembly + caption stage (two-zone composite + blurred fill + divider + hard-concat + end card + captions); the VO comes from create-vo-elevenlabs, the anchor from create-image-gpt-image-fal, and the whole-VO lip-sync from a paid VEED Fabric 1.0 take (a no-atom upstream input). Use for the split-screen-creator format.

Assemble a song-driven music-video ad from a config — a generated sung track carries the whole narration across N tableaux (one keyframe -> one i2v clip per lyric beat) with NO separate voiceover, captions synced to the song's OWN word timings (script-window, never Whisper) and the hook word landing on the chorus drop, closed on a PIL brand end card. This is the FREE deterministic assembly stage (clip cut-to-timeline + captions + end card + FFmpeg composite); the song, keyframes, and clips come from create-music-elevenlabs / create-image-fal / create-video-fal. Use for the song-driven-music-video format.

Build the deterministic PIL pill overlays for a 'perfect-score + proof-points' UGC video ad (white 10/10 score header with a medal + orange sub with a finger-down + 3-4 green-check proof pills) and composite them onto a base clip in a diagonal L->R->L->R cascade, then mux the music bed into master-final.mp4. Config-driven (config.json), 1080x1920 9:16, FREE and deterministic (no paid calls, text stays pixel-crisp). Use for the overlay-proof-points format; the base clip + music come from separate paid capabilities.

Assemble a two-host fake-podcast skit ad from a config — per-line lipsync clips hard-concatenated in script order, scaled/padded to 1080×1920, WHITE bottom-center captions (up to 5 words per cue, broken on sentence punctuation, word-wrapped to stay in-frame, held at least 0.9s) built from each line's OWN ElevenLabs char-level timestamps (offset by cumulative clip start, never Whisper), and closed on a Playwright/PIL brand end card composited from the real wordmark — never AI-rendered text. This is the FREE deterministic assembly stage (concat + white captions + end card + crf28 encode); the per-line VOs, photoreal gpt-image-2 base stills, expression variants, and lipsync clips come from create-vo-elevenlabs / create-image-gpt-image-fal / create-video-fal. Use for the podcast-skit format.

Render a 'photo-grid promo card' video from a config — a real-DOM card with a FIXED header/footer and a CONTINUOUSLY SCROLLING 2-row grid of MIXED tiles (video clips, product/lifestyle stills, big-serif %-OFF, dark promo-code or CTA) + feature chips, frame-stepped via Playwright with real-time <video> playback and encoded with FFmpeg — deterministic assembly, FREE (the clips + music come from create-video-fal + create-music-elevenlabs), text stays pixel-crisp. Use for the photo-grid-promo-card format.

Render a punchy ~12s vertical (9:16) music-only direct-response OFFER ad as a 4-beat kinetic-typography film — HEADLINE slam → real PRODUCT drop → CLAIM/proof → CTA pill — from one config of copy slots, a real product photo, a brand palette, fonts, bpm, and beat split. DETERMINISTIC + FREE (a bundled Remotion project; springs + interpolate, no AI-gen for visuals). Backgrounds are engine gradient divs off the palette, props are inline SVG, the ONLY composited bitmap is the REAL product photo (objectFit:contain, never stretched), and ALL headline/claim/CTA/URL/wordmark text is typeset in the engine — never AI-rendered (the format's credibility guard). A driver binds the config to Remotion input props, renders the 9:16 master, and derives a 1:1 center-crop with ffmpeg. Two gating checks run before render (claim verbs must match the product's physical format; the claim beat needs an edge-entry mechanism prop). Use for the motion-graphics-offer-ad format.

Assemble a narrated-UGC "stitch reply" ad from a config — a single spoken VO carries a verbatim testimonial while ~30 per-cut i2v clips (one creator across ~5 wardrobes in ~3 worlds, plus product B-roll) are each trimmed to their EDL window built from the VO's Whisper word boundaries and hard-concatenated via filter_complex concat (never the demuxer, which drops audio on a duration mismatch), the VO mixed over an optional sidechain-ducked instrumental bed (−20dB, 20 to 1) so the VO stays on top, karaoke-pop captions burned on every word throughout (VEED Whisper preset, re-spelled against the locked script), a landing-page scroll rendered as FFmpeg zoompan over a Playwright PNG (not i2v), and closed on the brand's real end-card PNG — never AI-rendered text. This is the FREE deterministic assembly stage (trim-to-EDL + filter_complex concat + VO and music mix + karaoke captions + landing-page zoompan + end-card append); the VO, creator, start-frames, and clips come from create-vo-elevenlabs / create-image-gpt-image-fal / create-image-fal / create-video-fal. Use for the narrated-ugc-wardrobe-stitch format.

Assemble a silent, music-led 3-world product-tour ad — trim and hard-cut-concat the per-world WIDE-arrival + top-down-macro clips, composite the HTML/Playwright brand end card ("FIND YOUR DAILY." + handwritten scent labels + arrows) over an AI flat-lay background, and mux one music bed into a 720x1280 web master. FREE, deterministic assembly (Playwright + PIL/HTML + FFmpeg); the recipe supplies the config and gates the paid clip/background/music calls. Use for the multiworld-product-tour format.

Render a 'mosaic-grid-reveal' video from a config — a real-DOM FULL-BLEED N×N mosaic of real product tiles that pops in one tile at a time (scatter order, ease-out-back overshoot), the grid clears, then the brand wordmark builds line-by-line followed by a sub-label, tagline, and CTA; frame-stepped via Playwright and encoded with FFmpeg — deterministic assembly, FREE (the music bed comes from create-music-elevenlabs), so the wordmark, tile captions, and CTA stay pixel-crisp. Use for the mosaic-grid-reveal format.

Render a 'model comparison grid' video from a config — a fal-style "same prompt, N contenders" showcase — a dark real-DOM stage where per beat a monospace prompt fades in centered, docks to a small top strip, then a labeled 2-4 panel grid (static images OR muted video clips, mixable per cell) staggers in and holds for comparison, plus a minimal end card — frame-stepped via Playwright (video cells are frame-seeked deterministically) and encoded with FFmpeg. Deterministic assembly, FREE (cell media comes from create-image-fal / create-video-fal, music from create-music-elevenlabs), text stays pixel-crisp. Use for the model-comparison-grid format.

Assemble an iMessage chat-reveal video ad from a thread JSON — one continuous Playwright recording of the conversation animating in (typing dots, composer typing, bubble pops, auto-scroll) crossfaded into a designed end card, with iMessage send/receive SFX and an optional ducked music bed. FREE assembly (Playwright + ffmpeg); the recipe supplies the per-brand thread + product + end-card config and gates the paid product-image/music calls to their own capabilities. Use for the imessage-chat format.

Assemble an iMessage notification-cascade video ad (≈14s, 9:16) from a phone-on-desk plate + 3–5 messages — authentic Apple Messages banners composited in PIL (SF Pro text, green Messages icon, warm translucent-greige fill, soft shadow) spring in one-by-one at the BOTTOM and push the stack UP, a right-aligned Show-less/X pill rides above, the X clears the stack, then a serif end card resolves. FREE assembly (PIL + ffmpeg); the recipe supplies the per-brand plate, notifications, and end-card config and gates the paid plate-clean/music calls to their own capabilities. Use for the imessage-notification-cascade format.

Render an 'Instagram-Live social-proof gallery' video from a config — ~5 real brand product stills each framed as an Instagram-LIVE card (IG gradient-ring avatar, username, verified check, red LIVE badge, viewer count, close X, a short claim-free live-comment feed, an empty 'Add a comment…' bar, and reaction hearts floating up the right edge), with a brand-approved benefit sentence building one phrase per slide and a clean logo endcard, rendered deterministically with PIL frames plus FFmpeg — FREE (the music bed comes from create-music-elevenlabs), products and wordmark stay crisp. Use for the ig-live-gallery format.

Assemble the FREE steps of the product-hypermotion + kinetic-typography video format — dice ONE Seedance 2.0 hypermotion clip into 5-6 segments and intercut them with PIL kinetic-typography spec/CTA cards (italic skew, 1.08x outline echo, 3D extrusion, slam-with-shake, inversion flash) plus a real-logo PIL end card (base64-decoded from the brand SVG), center-crop 1:1 to 9:16, and explicit-map mux a music bed. Deterministic, FREE (PIL + FFmpeg), no paid calls, the real logo and spec typography stay pixel-crisp. The paid steps (the ONE Seedance i2v, the music) are separate capabilities; the recipe orchestrates them. Use for the product-hypermotion format.

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 Playwright product overlay cards (real PDP-verified taglines) are composited onto the master each on its product-NAME word-start, the SEPARATE VO is mixed on top of a ducked music bed at loudnorm I=-14, clean-white 3-words/cue captions are burned, and the video closes on a flat-lay end card. This is the FREE deterministic assembly stage (re-cut to the VO word-starts, hard-concat, Playwright card render + card composite, VO plus music mix, caption burn, flat-lay end card); the VO, scene clips, product cutouts, and music come from create-music-elevenlabs / create-image-gpt-image-fal / create-video-fal. Use for the glassy-matte-grwm format.