DojoCodingLabs/remotion-superpowers v2.3.0

v2.3.0 — built on the official skills: Remotion, ElevenLabs, Replicate

on GitHub

Changed

• Built on Remotion's official skills. remotion-dev/skills is now the base for all Remotion markup, preview, rendering and captions. This plugin owns the media: generating, sourcing, analyzing and reviewing it. remotion-production says which side handles what, so the two skill sets no longer give conflicting instructions (e.g. the official voiceover guide asking for an ElevenLabs key that KIE already covers).
• Preview first. /create-video and /create-short open Remotion Studio as soon as the project runs and build live. They render when you ask; /review-video renders on invocation and checks a few frames before the full TwelveLabs review.
• Captions use the official flow. Word-level transcription and <BasicCaptions> per clip, with our TikTok-style presets and short-form placement on top. generate_subtitles stays for SRT export.
• /setup installs all 12 official Remotion skills and scaffolds with the official create-video flags.

Changed (providers)

• Audio comes from ElevenLabs' official skills. Voiceover (eleven_v4), sound effects and music (Eleven Music music_v2_5, cleared for commercial use on paid plans) go through elevenlabs/skills. ELEVENLABS_API_KEY is now required. KIE audio (generate_speech, generate_music, generate_sound_effect) is labelled drafts-only, because KIE resells Suno and the other models without their owners' licence.
• ElevenLabs MCP is deprecated upstream (archived 2026-08-22). It's pinned at 0.12.2 for one more release, with ELEVENLABS_MCP_BASE_PATH=./public so it can read and write project files. Before, it rejected every input outside public/audio.
• Pexels: switched to @hanoak/pexels-mcp-server 1.0.2. The previous server crashed on startup with mcp 2.x and is unmaintained. The new one runs on npm (no Python/uv) and has the pexels_* tools. Thanks @fabiantresarroyos for diagnosing the crash in #10.
• TwelveLabs: works again. Pegasus 1.2 was removed from the API, which broke the index-first flow. Analysis now uses create-asset → sync-analyse-video/async-analyse-video (pegasus1.5) → delete-asset, and /review-video gets a structured JSON rubric. Pinned at twelvelabs-mcp 1.0.0.
• Replicate: the official local replicate-mcp 0.9.0 with REPLICATE_API_TOKEN, layered on replicate/skills. The hosted server was OAuth-only and timed out headless. Our old tool names didn't exist, and two model IDs were invalid. There's a dated model shortlist and a "search and check the schema first" rule.
• remotion-media pinned at 1.2.2 and used for images and clips only. Its subtitles tool is no longer used: it built a shell command from unquoted input. Captions fall back to ElevenLabs Scribe via @remotion/elevenlabs when there's no GPU for local Whisper.
• Every MCP server is pinned to a tested version.
• The README and setup guide list official add-ons (fal, Runway, Gemini skills) without bundling them.

Fixed

• Outdated Remotion APIs across the rules and skills: <OffthreadVideo>, <Audio> from remotion, startFrom/endAt, <Loop>, @remotion/light-leaks, npm run dev / localhost:3000 → @remotion/media, trimBefore/trimAfter, loop, @remotion/effects, npx remotion studio.
• Word-by-word caption highlight couldn't work from line-level SRT; caption spans used CSS transitions, which don't render.
• Glitch effect used Math.random() (flickers across render threads); non-integer durationInFrames.
• ElevenLabs MCP tool names (voice_clone, search_voices, isolate_audio) and its ~/Desktop default output.
• generate_speech doesn't return a duration; durations are now measured in calculateMetadata.
• One asset naming convention for MCP outputs and downloads.

Removed

• rules/3d-content.md (superseded by the official remotion-markup 3D guide).

Added

• CI guard against outdated Remotion APIs in the docs.

Upgrading: set ELEVENLABS_API_KEY (now required for voice, music and SFX), run /setup to install the official Remotion and ElevenLabs skills, and make sure you have Node.js 22.18+.