Domain-specific knowledge base for building videos with Remotion and React. Covers 30+ rule files spanning animations, audio, assets, 3D content, charts, text, transitions, and composition management Includes specialized guidance for captions, FFmpeg operations, audio visualization, and sound effects integration Provides patterns for dynamic metadata calculation, media inspection (duration, dimensions, frame extraction), and parametrizable compositions Supports advanced features like Lottie animations, light leak effects, Mapbox integration, TailwindCSS styling, and AI-generated voiceovers
Render videos from React/Remotion component code via inference.sh. Pass TSX code, get MP4. Supports all Remotion APIs: useCurrentFrame, useVideoConfig, spring,…
Generate AI videos with Google Veo, Seedance 2.0, HappyHorse, Wan, Grok and 40+ models via inference.sh CLI. Models: Veo 3.1, Veo 3, Seedance 2.0, HappyHorse…
Create AI avatar and talking head videos via inference.sh CLI. Recommended: P-Video-Avatar (fastest, cheapest, built-in TTS). Also: OmniHuman, Fabric,…
Batch image editing with identity preservation, background swaps, and spatial localization via Google Nano Banana 2. Supports up to 20 input images per call for consistent batch edits, with 1–4 outputs per request and locked aspect ratio/resolution for gallery consistency Preserves subject identity, pose, and clothing while enabling targeted edits via spatial language ("background only", "left object", "upper-right corner") Handles localized object removal/addition, background replacement, and text swaps within images; routes to GPT Image 2 edit for multilingual in-image text, Flux Kontext for single-reference precision edits, or Nano Banana 2 t2i for generation from scratch Requires RunComfy CLI ( npm i -g @runcomfy/cli ), a RunComfy account, and publicly-fetchable HTTPS image URLs; set RUNCOMFY_TOKEN env var for CI/container deployments
Smart router matching image edit intent to the right RunComfy model. Routes between four specialized models: Nano Banana Edit (batch up to 20, identity-preserving), GPT Image 2 Edit (multilingual text, multi-reference composition), Flux Kontext Pro (single-shot precise local edit), and Z-Image Turbo Inpaint (mask-driven region replacement). Each route includes documented prompting patterns and schema to optimize results without trial-and-error model selection. Invokes models via runcomfy run <vendor>/<model>/edit CLI; requires RunComfy account and local CLI installation. Triggers on "image edit", "edit image", "i2i", "swap background", "remove object", "rewrite headline", or explicit image editing requests.
Text-to-video generation with native 1080p, synchronized audio, and multi-shot character consistency. Ranked #1 on Artificial Analysis Video Arena; supports 3–15 second clips in five aspect ratios (16:9, 9:16, 1:1, 4:3, 3:4) at 720p or 1080p resolution Generates in-pass synchronized audio (dialogue, ambient, Foley) alongside video in a single pass, eliminating separate audio workflows Maintains character and wardrobe consistency across multi-shot narratives when prompts restate visual anchors Accepts prompts in six languages (English, Chinese, Japanese, Korean, German, French) up to 2,500 characters Invoked via RunComfy CLI; requires RunComfy account and API token; outputs download automatically to specified directory
Cinematic short-form video generation with multi-modal references and native lip-sync. Accepts up to 9 images, 3 videos, and 3 audio tracks as synchronized references; combines them with text prompts to produce 4–15 second clips at up to 720p resolution Generates in-pass audio with natural lip-sync when generate_audio: true ; supports tone direction ("warm friendly", "crisp newsroom") and audio reference matching Invoked via runcomfy run bytedance/seedance-v2/pro CLI; requires RunComfy account and token authentication Best suited for spokesperson ads, branded narratives, and cinematic previs where identity stability (via image refs) and evolving narrative (via prompt) matter; route to HappyHorse 1.0, Wan 2.7, or Kling for other video workflows
Multi-tier video generation with Kling 3.0 across Standard, Pro, and 4K rendering endpoints. Six endpoints spanning three quality tiers (Standard 1080p, Pro 1080p, 4K native) and two modes (text-to-video, image-to-video), each with distinct pricing and motion-fidelity budgets Supports multi-shot sequences within a single generation with consistent character identity across shots, up to 15 seconds per clip, plus synchronized native audio generation Image-to-video mode accepts source images and optional tail images for controlled frame-to-frame transitions; text-to-video generates scenes from prompts alone Invokes via RunComfy CLI with JSON schema inputs including prompt, duration (3–15s), aspect ratio (six options), CFG scale tuning, and audio generation toggle
Generate Codex-compatible custom desktop pets from a single image via RunComfy. Produces a 1536x1872 PNG/WebP spritesheet with 9 animation rows (idle, running, waving, jumping, failed, waiting, running, review) and pet.json manifest ready to drop into ${CODEX_HOME:-$HOME/.codex}/pets/<name>/ Calls OpenAI GPT Image 2 edit once to generate a canonical chibi-proportioned pose, then assembles all animation frames programmatically with ImageMagick micro-transforms (1–2 px shifts, blinks, tilts) Requires only RUNCOMFY_TOKEN , RunComfy CLI, and ImageMagick — no Codex Pro, no $imagegen , no OPENAI_API_KEY Triggers on "codex pet", "create codex pet", "hatch", "spritesheet.webp", or explicit requests to build custom Codex Pets
Search, download, and manage Lark Minutes recordings with transcript and metadata extraction. Search minutes by keyword, owner, participants, or time range; retrieve basic metadata including title, cover, duration, and owner information Download audio/video media files or generate shareable download links valid for one day Upload local audio/video files to generate new minutes, with optional conversion to transcripts, summaries, action items, or chapters via the vc +notes command Rename minutes titles and replace speaker attributions in transcripts by rebinding statements from one user to another Requires lark-cli binary and proper authentication; refer to shared Lark skill documentation for permission handling and domain boundaries
Search, retrieve, and analyze Feishu video meeting records, notes, and participant snapshots. Search ended meetings by keyword, time range, organizer, participant, or room; query future schedules using lark-calendar instead Retrieve meeting notes (AI summaries, action items, chapters) and verbatim transcripts; choose products based on user intent (AI summary vs. original transcript for re-analysis) Download meeting artifacts including transcripts, minutes, and cover images organized by minute_token directory Query participant snapshots for any meeting using vc meeting get --with-participants , including who attended and entry/exit times Use three primary shortcuts ( +search , +notes , +recording ) for common operations; agent join/leave and real-time meeting events require lark-vc-agent skill
Object removal and region editing across video frames via prompt-driven spatial language. Routes across three models: Wan 2-7 Edit-Video (default, prompt-driven edits), Lucy Edit Restyle (identity-stable region swaps), and Seedream 4-0 Edit-Sequential (frame-stack treatment for short clips) Invoked via runcomfy run CLI with video URL and natural-language prompt describing the region and desired edit (e.g., "remove watermark in bottom-right") Handles watermark removal, object deletion, and region replacement with motion continuity; for pixel-precise mask propagation, directs to ComfyUI workflows with SAM2 segmentation tracking Triggers on phrases like "video inpaint," "remove from video," "clean up video," or explicit region-edit requests
Extend video canvas spatially—swap aspect ratio, uncrop vertically or horizontally while preserving central action. Routes prompt-shaped canvas extension through Wan 2-7 Edit-Video for quick aspect-ratio swaps (9:16 to 16:9, square to wide), and points to dedicated ComfyUI LTX 2-3 outpainting workflows when hero-grade seam quality is required Invoked via runcomfy run wan-ai/wan-2-7/edit-video with video URL and spatial extension prompt; accepts TikTok-to-YouTube, Instagram-to-banner, and old-footage-to-modern-ratio use cases Seam quality varies with Wan 2-7; multi-pass outpainting degrades after two iterations; ComfyUI workflows recommended for strict temporal consistency and motion-aware fill in premium deliverables
Create AI avatar, talking-head, and lip-sync videos on RunComfy via the `runcomfy` CLI. Routes across ByteDance OmniHuman (audio-driven full-body avatar),…
Generate cinematic short-form video with ByteDance Seedance 2.0 Pro on RunComfy. Documents Seedance 2.0 Pro's strengths (multi-modal references — up to 9…
Extend or continue an existing video clip on RunComfy via the `runcomfy` CLI. Routes to Google Veo 3-1's `extend-video` and `fast/extend-video` endpoints —…
Generate cinematic short-form video with ByteDance Seedance 2.0 Pro on RunComfy. Documents Seedance 2.0 Pro's strengths (multi-modal references — up to 9…
Install the CLI. Run your first Skill in 30 seconds. Scale when you're ready.