SpecReel: Make 3D Explainer & Product Videos with Claude Code and Codex

Descripción

SpecReel is an agent skill for Claude Code and Codex that turns any topic into a narrated, captioned 3D video — product launches, processes, data stories and explainers.

Ask for a video. Get a finished MP4.

Write one prompt: topic, audience, length, language. Your agent plans the piece, writes the script, voices it with your local text-to-speech or a studio voice from Fish Audio, builds 3D scenes and camera moves timed to every spoken word, adds music and sound effects, and renders an MP4 with karaoke captions and an .srt file.

No timeline editing. No per-minute video credits. It all runs on your computer.

New in 2.0

• Cinematic camera: 16 moves (push-in, orbit, crane, dolly-zoom, whip-pan, rack focus and more) and 8 transitions, each anchored to a spoken word.
• Brand kits: start from your logo or HEX colors, or pick one of three brand-inspired palettes the agent suggests. 13 built-in palettes in bold, dark, pastel and light moods, with WCAG contrast checks.
• Windows, macOS and Linux: the Doctor detects your OS, GPU and hardware encoder, and one command installs the rest. Fast on NVIDIA or Apple Silicon, works on CPU.
• Studio voice in two minutes: a setup assistant connects Fish Audio (a separate paid service) and keeps your API key out of the chat. Free local voices are still included.
• Cleaner sound: automatic whooshes and risers, a closing chime and a true-peak limiter that lands every mix at −14 LUFS.

One workflow, any topic

• Product & promo: turntable hero shots, parts that lift away, feature callouts, an end card with your CTA.
• Process / how it works: numbered steps that appear as they are spoken, with connections and moving packets.
• Data stories: bars and lines that grow on the words that name them, count-ups and result chips.
• Explainers: science, health, tech, finance and education, with hero objects, macro zooms, cross-sections and data cards.
• Tutorials & brand stories: mix the recipes in your brand colors, with your logo and your end card.

Without SpecReel → with SpecReel

• The agent writes a one-off animation that breaks at render time → a deterministic, frame-by-frame engine built for narrated video.
• Voice, captions and motion drift apart → every event is anchored to a spoken word with word-level timestamps.
• You find mistakes after a long render → two approval checkpoints, and the agent renders stills, looks at them and scores them before the final render.
• A crash or closed laptop means starting over → progress lives in specs/STATE.md and the loop resumes where it stopped.
• Unclear voice licenses → a Doctor script flags non-commercial models before you use them for clients.

What's inside (v2.0)

• The SpecReel agent skill (open SKILL.md format) plus 11 reference guides (new: cinematography) the agent loads on demand.
• 3D engine (Three.js): scene director with a camera rig (16 moves, 8 transitions), depth of field, labels, scale bars, callouts and chips.
• Brand kit tool: suggests three brand-inspired palettes with a preview sheet, builds a kit from your colors or your logo, checks contrast and styles the captions to match.
• 7 scene recipes: product hero, process flow, data story, end card, hero object, micro and data card.
• Audio pipeline: 13 voice adapters (Kokoro, Chatterbox, Piper, XTTS, Fish Speech, OpenAI-compatible servers, any CLI, Windows and macOS voices, eSpeak, ElevenLabs, Fish Audio), an ASR round-trip check, word alignment, procedural music and SFX, and a mix at −14 LUFS.
• Resumable multi-lane renderer, a cross-platform encoder that uses your hardware encoder when available, and a one-command installer for Windows, macOS and Linux.
• Loop runners: /ralph-loop for Claude Code, and loop.ps1 and loop.sh for unattended runs with Claude Code or Codex.
• Spec templates: constitution, brief, spec, script, storyboard, plan, tasks and state.
• 4 example productions with their specs: Showreel (English, 21 s), a brand-kit bottle demo (vertical, 17 s), Cell demo (Spanish, 15 s) and a Microneedling explainer (Spanish, 47 s).
• Docs in English and Spanish: setup for Windows, macOS and Linux, a spec-driven master plan and an audio migration guide.

What you get out

• MP4 (H.264, 30 fps) in 9:16 (1080×1920), 1:1, 4:5 or 16:9.
• Burned-in karaoke captions plus an .srt file.
• A cover frame, CREDITS.md and a QA REPORT.md for every video.
• Narration in any language your voice engine supports.

Requirements

• Claude Code or Codex (billed separately by Anthropic or OpenAI).
• Node.js 20+, Python 3.10+, ffmpeg and Playwright Chromium (the included installer sets them up).
• Windows 10/11, macOS or Linux (auto-detected). An NVIDIA GPU or Apple Silicon is recommended; CPU-only works, just slower.
• A voice: free local Kokoro (Apache-2.0, runs on CPU, set up by the Doctor), or Fish Audio or ElevenLabs with your own API key.
• Basic comfort with a terminal. The agent does the rest.

One price, one license

• Price: US$36.90, one-time payment. No subscription, no membership.
• License: 1 creator. Unlimited videos for your own brand, your channels and your clients: commercial and agency work included.
• Delivery: instant download of one ZIP with everything, plus free 2.x updates.
• The license covers the videos you make. Reselling or sharing the skill files is not allowed.

FAQ

Do I need to know how to code? Not to get started. You describe the video in plain language; the agent writes and fixes the scene code. You will run a few terminal commands during setup.

What kinds of videos can it make? Explainers, product and promo videos, how-it-works processes, data stories and tutorials, built from illustrated 3D scenes with labels and captions. It does not make photoreal people, avatars or lip-sync, and it does not edit camera footage.

Does it work with both Claude Code and Codex? Yes. Install steps for both are included, and the skill follows the open SKILL.md format.

Is there a cost per video? Not from SpecReel. With a local voice you only pay for your agent usage. Cloud voices bill you directly.

Does it work on a Mac? Yes. Windows, macOS and Linux are detected automatically, and one command installs the dependencies. Apple Silicon renders fast.

How long does a video take to render? It depends on your hardware and the length. The 21-second showreel rendered in about 8 minutes on a 2-core cloud machine without a GPU. A GPU is typically much faster, and a low-resolution animatic lets you iterate quickly.

Which languages? Narration in any language your voice engine supports. Kokoro covers English, Spanish, French, Italian, Portuguese, Japanese, Chinese and Hindi; Chatterbox Multilingual covers 23 languages; ElevenLabs and Fish Audio cover more. Docs come in English and Spanish.

Which voice sounds best? The free local voices are good. For studio quality, the included assistant connects Fish Audio (a separate paid service) in about two minutes.

Can I use my brand? Yes. Give the agent your logo or HEX colors and it builds a brand kit: palette, fonts, caption style and an end card with your CTA. No kit yet? It suggests three brand-inspired palettes with a preview.

Can I use the videos commercially? Yes, for your own brand and for clients. Check the license of the voice model you choose; the Doctor flags non-commercial ones.

Do I get updates? Every 2.x update is free: you get an email with the new download link.

Refunds? 14 days. If you can't get the included demo to render after we've tried to help, you get your money back.

SpecReel is an independent product, not affiliated with Anthropic or OpenAI. Claude Code and Codex are trademarks of their owners and are named only to describe compatibility. The Fish Audio link in the voice assistant is an affiliate link; Fish Audio is a separate service with its own pricing.

After you download: start here

• Unzip specreel-v2.0.0.zip and open README.md (Spanish: README.es.md).
• Install the skill: Claude Code → copy skills/specreel to ~/.claude/skills/ (Windows: %USERPROFILE%\.claude\skills\). Codex → copy it to ~/.agents/skills/.
• Set up and render the demo: run python scripts/setup.py from the skill folder (it detects Windows, macOS or Linux and installs what is missing), then follow "First video in 5 minutes" in README.md.

Then open your agent in an empty folder and paste a kickoff prompt from docs/en/SDD-PLAN.md.

Optional, studio voice: inside your video project run python tools/tts/fish_setup.py. It opens Fish Audio, asks for your key with hidden input and stores it safely in .env.

Need help? Reply to your receipt email or write to contact@tonyciencia.com with your license key, the file .svs/env.json (it contains no secrets) and the output of the failing command.

Refund policy

14-day "it runs or your money back" guarantee. If you can't render the included demo on a supported system (Windows 10/11, macOS or Linux with Node 20+, Python 3.10+ and ffmpeg) after we've tried to help, email us within 14 days of purchase with your license key and .svs/env.json and we'll refund you in full.

Descripción en español

SpecReel: crea videos 3D explicativos y de producto con Claude Code y Codex. Es la misma oferta y los mismos archivos; el ZIP trae documentación en español e inglés.

SpecReel es una skill para Claude Code y Codex que convierte cualquier tema en un video 3D narrado y con subtítulos: lanzamientos de producto, procesos, historias con datos y explicativos.

Pide un video. Recibe un MP4 terminado.

Escribe una instrucción con el tema, el público, la duración y el idioma. Tu agente planifica la pieza, escribe el guion, lo locuta con tu TTS local o con una voz de estudio de Fish Audio, construye escenas 3D y movimientos de cámara sincronizados con cada palabra, agrega música y efectos, y renderiza un MP4 con subtítulos karaoke y su archivo .srt.

Sin editar línea de tiempo y sin créditos por minuto. Todo corre en tu computador.

Novedades de la 2.0

• Cámara de cine: 16 movimientos (push-in, órbita, grúa, dolly-zoom, whip-pan, rack focus y más) y 8 transiciones, cada uno anclado a una palabra del guion.
• Kit de marca: parte de tu logo o de tus colores HEX, o elige entre tres paletas inspiradas en marcas que te propone el agente. 13 paletas incluidas en estilos intenso, oscuro, pastel y claro, con contraste revisado (WCAG).
• Windows, macOS y Linux: el Doctor detecta tu sistema, tu GPU y tu codificador de video, y un comando instala el resto. Rápido con NVIDIA o Apple Silicon; en CPU también funciona.
• Voz de estudio en dos minutos: un asistente conecta Fish Audio (servicio aparte, de pago) sin que tu clave pase por el chat. Las voces locales gratis siguen incluidas.
• Mejor sonido: whooshes y risers automáticos, cierre con campanita y un limitador true-peak que deja cada mezcla en −14 LUFS.

Un solo flujo para cualquier tema

• Producto y promo: el producto girando, piezas que se separan, etiquetas de características y un cierre con tu llamado a la acción.
• Procesos («cómo funciona»): pasos numerados que aparecen al nombrarlos, con conexiones y paquetes en movimiento.
• Historias con datos: barras y líneas que crecen con las palabras que las nombran, contadores y resultados destacados.
• Explicativos: ciencia, salud, tecnología, finanzas y educación, con objeto principal, zoom macro, cortes y tarjetas de datos.
• Tutoriales e historias de marca: combina las recetas con los colores de tu marca, tu logo y tu cierre.

Por qué es distinto

• Planifica antes de renderizar: brief, spec, guion y storyboard, con dos aprobaciones tuyas.
• Todo sincronizado: cada animación, etiqueta y sonido se ancla a una palabra hablada.
• Revisa su propio trabajo: renderiza fotogramas, los mira, los puntúa con una rúbrica y corrige.
• No pierde el hilo: el estado queda en specs/STATE.md y el loop se reanuda donde quedó.
• Licencias claras: el Doctor detecta tus voces locales y avisa de los modelos no comerciales. Puedes pasar a ElevenLabs o Fish Audio cambiando una línea.

Qué incluye (v2.0)

• La skill SpecReel (formato abierto SKILL.md) y 11 guías de referencia (nueva: cinematografía).
• Un motor 3D (Three.js) con director de escenas, cámara de cine (16 movimientos y 8 transiciones), profundidad de campo, etiquetas y barras de escala, más kits de marca con 13 paletas o tus propios colores.
• 7 recetas de escena: producto, proceso, datos, cierre, objeto principal, micro y tarjeta de datos.
• Audio con 13 adaptadores de voz, verificación por ASR, alineación por palabra, música y efectos, y mezcla a −14 LUFS.
• Render reanudable en varios carriles, codificación con aceleración por hardware, instalador de un comando para Windows, macOS y Linux, y loops para Claude Code y Codex.
• 4 producciones de ejemplo con sus specs: Showreel (inglés), demo v2 de la botella con kit de marca (vertical), Demo de la célula y Microneedling (español).
• Documentación en español e inglés.

Requisitos

• Claude Code o Codex (se pagan aparte).
• Node 20+, Python 3.10+, ffmpeg y Playwright Chromium (el instalador incluido los configura).
• Windows, macOS o Linux (se detecta solo). GPU NVIDIA o Apple Silicon recomendados; en CPU también funciona, más lento.
• Una voz: Kokoro local y gratis (la instala el Doctor), o Fish Audio o ElevenLabs con tu propia clave.

Precio y licencia

• Precio: US$36,90, pago único. Sin suscripción ni membresía.
• Licencia: 1 persona, videos ilimitados para tu marca, tus canales y tus clientes (uso comercial incluido).
• Entrega: descarga inmediata de un ZIP con todo, y actualizaciones 2.x gratis.
• La licencia cubre los videos que haces; no permite revender ni compartir los archivos de la skill.
• SpecReel es un producto independiente, sin afiliación con Anthropic ni OpenAI. Claude Code y Codex son marcas de sus dueños y se nombran solo para indicar compatibilidad. El enlace de Fish Audio del asistente de voz es de afiliado; Fish Audio es un servicio aparte con sus propios precios.

Política de reembolso

Garantía de 14 días «funciona o te devolvemos el dinero». Si no logras renderizar el demo incluido en un sistema compatible (Windows 10/11, macOS o Linux con Node 20+, Python 3.10+ y ffmpeg) después de que intentemos ayudarte, escríbenos dentro de los 14 días posteriores a la compra con tu clave de licencia y el archivo .svs/env.json, y te devolvemos el 100 %.

Soporte: contact@tonyciencia.com