framesmith

# framesmith: start here framesmith templates are **skills**: a `SKILL.md`, a brief schema, an example brief with assets, and a `scripts/` CLI that renders frames with the shared engine (numpy + Pillow → ffmpeg). It needs no browser, GPU or account. Image generation (gpt-image-2) is optional and uses placeholders when there is no key. ## 1. Route the request <!-- catalog:start --> | Skill | Output | Pick it when | |---|---|---| | [`video-to-template`](../video-to-template/SKILL.md) | a new template skill + the finished video | The user shares a **reference video** ("make one like this", "copy this style"). Runs the whole pipeline: analyse → storyboard → animatic → assets → build → compare → ship | | [`brutalist-product-promo`](../brutalist-product-promo/SKILL.md) | 22.6 s 9:16 MP4, 21 scenes | Hardware/tech product promo, spec-sheet motion graphic, industrial/brutalist look, one accent colour on black and white | | [`devtool-launch-video`](../devtool-launch-video/SKILL.md) | 33.8 s 16:9 MP4, 11 scenes | Launch or "what is X" film for a CLI, dev tool, AI-agent or security product: rogue-terminal problem, brand reveal, verdict-chip proof, features, dashboard, install, CTA | | [`narrated-collage-explainer`](../narrated-collage-explainer/SKILL.md) | ~50 s 1:1 MP4 with voiceover, beats on spoken cues | Narrated explainer or founder story in a paper-white collage style (big pixel keywords, synced subtitles, CRT TV / XP window / cards / starbursts), built from a script plus real screenshots and GIFs | | [`brand-icons`](../brand-icons/SKILL.md) | SVG + PNG brand marks with a licence and attribution table | Any design needs real company/tool logos ("works with" rows, stack grids, partner walls); fetched from thesvg.org with the right variant for the background | | [`keynote-product-film`](../keynote-product-film/SKILL.md) | ~60–90 s 16:9 MP4, music-driven, no VO | Apple-event-style launch or "introducing" film: kinetic sentences with inline icons, real product UI (prompt, agent session, permission prompts, steps, benchmark chart, ticker, masked field), logo reveal on a gradient wash | | [`no-ai-slop`](../no-ai-slop/SKILL.md) | edited copy + what changed (or a findings report) | Before rendering any template: run scripts, captions and on-screen lines through it. Also for any draft that reads as AI-written (by Peter Yang, MIT) | | [`glass-saas-promo`](../glass-saas-promo/SKILL.md) | 16:9 MP4, music-driven | Render a bright, music-driven SaaS promo (about 30–40 s, 16:9) in the frosted-glass style: a lavender-white field, blue-to-violet gradients, gpt-image-2 glass 3D objects (shield, folder, check, sparkle, laptop), real product UI in | | [`blue-story-film`](../blue-story-film/SKILL.md) | 16:9 MP4, music-driven | Render a music-driven SaaS story film (about 60–90 s, 16:9) in the blue-light style. It opens with typewriter captions, a caret and a selection highlight over ghosted UI, then a huge 'Meet' word that collapses into the brand line. | | [`illustrated-explainer`](../illustrated-explainer/SKILL.md) | ~60–100 s 16:9 MP4 with voiceover | Narrated flat-vector corporate explainer: gpt-image-2 illustrations in one house style, pans with small-caps captions, bullets, split screen, whoosh to logo, world map, product cards, CTA | | [`animated-character-explainer`](../animated-character-explainer/SKILL.md) | ~45–90 s 16:9 MP4 with voiceover | Narrated explainer animated from a consistent transparent-PNG cast (on-model poses via gpt-image-2 edits), props, backdrops and icon tiles; pose swaps and icon pops on spoken cues; starter via `init` | | [`archive-launch-film`](../archive-launch-film/SKILL.md) | ~60–90 s 16:9 MP4 with voiceover | Launch film with a documentary cold open (engravings, flooding cutaways, a highlighted document, words synced to the voice) turning into a chrome brand reveal and cursor-driven product UI on a pastel ground; Enterprise cut with fleet wall and holographic reveal | | [`kinetic-launch-film`](../kinetic-launch-film/SKILL.md) | ~60–90 s 16:9 MP4 with voiceover + synthesised SFX | Text-led SaaS/AI launch film: voice-synced spring word rises, stretched letters, words on dashed curves, scatter, "Meet" logo reveal, logo cards, tilted 3D window with detection boxes, live feed with agent cursor, log-to-table, hub with beams | | [`abstract-journey-film`](../abstract-journey-film/SKILL.md) | ~80 s 16:9 MP4, music only (synthesised score) | Conference opener / brand title sequence: a glowing orb travels through 22 abstract worlds (data field, vortex of cards, sphere burst, pastel pillars, frosted sphere, portal rings, illustrated landscape, space) to the logo; gpt-image-2.5 assets plus procedural scenes | | [`cartoon-launch-film`](../cartoon-launch-film/SKILL.md) | ~45 s square 1080 MP4, music only (synthesised score + cartoon SFX) | Hand-drawn boiling cartoon for a social launch or GitHub-stars post: mascot with on-model poses, terminal cold open, notebook slam, ingest gag, notes graph, agent grid, a hand-drawn Chrome window where an extension captures a page and fills a form with sourced answers, cited answer note, gauge, star counter, end card | <!-- catalog:end --> If nothing fits, say so and offer to build a new template (section 3) rather than forcing the wrong one. ## 2. Use a template Each template exposes the same CLI verbs: ``` python3 skills/<name>/scripts/<entry>.py check --brief brief.json # validate python3 skills/<name>/scripts/<entry>.py assets --brief brief.json # gpt-image-2, low/medium quality python3 skills/<name>/scripts/<entry>.py still --brief brief.json 3.4 9.1 python3 skills/<name>/scripts/<entry>.py sheet --brief brief.json # contact sheet python3 skills/<name>/scripts/<entry>.py render --brief brief.json [--audio a.mp3] ``` Always review stills or the contact sheet before a full render, and report which assets are placeholders. ## 3. Build a new template Use [`video-to-template`](../video-to-template/SKILL.md). It holds the full process: reference analysis (`analyze.py`), a timed animatic scaffold (`scaffold.py`), side-by-side review (`compare.py`), the design rules, and a list of pitfalls that each cost a re-render. With no reference video, scaffold a blank template with `python3 tools/new_skill.py <name>` and follow the same phases from Phase 2.