video-inpainting
๐ฏSkillfrom prime-skills/runcomfy-agent-skills
Installation
npx vibeindex add prime-skills/runcomfy-agent-skills --skill video-inpaintingnpx skills add prime-skills/runcomfy-agent-skills --skill video-inpainting~/.claude/skills/video-inpainting/SKILL.md
|A skill for performing region edits across video frames via RunComfy CLI, supporting object removal, watermark cleanup, and region replacement by routing to Wan 2-7 Edit-Video for prompt-driven edits, Lucy Edit Restyle for identity-stable changes, and Seedream 4-0 for frame-stack treatment.
Overview
This skill handles region-based edits across video frames using the RunComfy CLI. It enables removing objects that appear across many frames, cleaning up wires or watermarks, and replacing regions with motion-matched content. The default route uses Wan 2-7 Edit-Video for prompt-driven region edits that resolve target areas from spatial language. For identity-stable region-aware restyle, it routes to Lucy Edit Restyle. For short sequences that can be treated as frame stacks, it uses Seedream 4-0 Edit-Sequential.
Key Features
- Prompt-driven region edits: Wan 2-7 Edit-Video resolves target regions from spatial language descriptions like "the watermark in the bottom-right corner" or "the cables overhead" without requiring explicit masks
- Identity-stable video restyle: Lucy Edit Restyle handles lightweight outfit or object swaps while tracking identity consistently across frames
- Frame-stack sequential editing: Seedream 4-0 Edit-Sequential applies the same edit instruction across individual frames, useful for short, low-frame-rate sequences
- Single-change-per-call best practice: The skill recommends splitting compound edits into sequential passes for cleaner results
- ComfyUI workflow escalation: When pixel-precise mask propagation is needed (SAM2 segmentation tracking + temporal-aware inpaint), the skill guides users to dedicated ComfyUI workflows like LTX 2-3 video inpaint
- Preservation-first prompting: Built-in guidance to lead prompts with "Preserve all other content exactly" to prevent unintended restyle of surrounding frames
Who is this for?
- Video editors and post-production teams who need to remove watermarks, logos, or unwanted objects across entire clips without frame-by-frame manual work
- Content creators cleaning up footage by removing background distractions like passing pedestrians, overhead wires, or equipment
- Production studios that need region replacement with motion-matched content in their video editing pipeline
Same repository
prime-skills/runcomfy-agent-skills(25 items)
SKILL.md
More from this repository10
A smart intent-routing skill for video editing on RunComfy that automatically selects the best model (Wan 2.7 Edit-Video, Kling 2.6 Pro Motion Control, or Lucy Edit Restyle) based on what the user wants to do, from general restyling and background swaps to precise motion transfer and outfit changes.
A smart intent-routing skill that animates still images on RunComfy by selecting the optimal model: HappyHorse 1.0 I2V for general portrait and product animation with native audio, Wan 2.7 for custom-voiceover lip-sync, or Seedance 2.0 Pro for multi-modal composition with image, video, and audio references.
A skill for generating images with Google Nano Banana 2, the flash-tier text-to-image model in the Gemini family on RunComfy, optimized for rapid iteration, batch ideation, in-image typography rendering, and optional web-grounded context with support for resolutions from 0.5K to 4K.
A smart intent-routing skill for image editing on RunComfy that selects the optimal model from Nano Banana Edit (batch up to 20 images), GPT Image 2 Edit (multilingual text rewrite, multi-ref composition), Flux Kontext Pro (single-shot precise local edit), or Z-Image Turbo Inpaint (mask-driven region edit) based on user intent.
A Claude Code skill that enables precise single-image editing using Black Forest Labs's Flux 1 Kontext Pro model through the RunComfy CLI, with built-in prompting patterns for high-fidelity local edits.
A skill for generating text-to-video with Wan 2.7, Wan-AI's flagship motion model on RunComfy, featuring multi-reference conditioning, audio-driven lip-sync via custom audio tracks, smooth transition physics, prompt expansion control, and up to 1080p output at 15 seconds duration.
A Claude Code skill for generating text-to-video clips with HappyHorse 1.0 via the RunComfy CLI. HappyHorse 1.0 is currently ranked #1 on the Artificial Analysis Video Arena, producing native 1080p video with in-pass synchronized audio and multi-shot character consistency.
A Claude Code skill for generating video with Kling 3.0 via the RunComfy CLI. Covers all six Kling 3.0 endpoints across three rendering tiers (Standard, Pro, 4K) and two modes (text-to-video, image-to-video), with native synchronized audio and multi-shot character consistency.
A Claude Code skill that generates custom OpenAI Codex Pets from a single reference image via the RunComfy CLI. It produces a Codex-compatible spritesheet (1536x1872, 8 columns x 9 rows) and pet.json manifest using GPT Image 2 and ImageMagick, without requiring Codex Pro or an OPENAI_API_KEY.
A Claude Code skill that acts as a smart router for AI video generation across the full RunComfy model catalog. Supports text-to-video, image-to-video, and video-extend modes with models including HappyHorse 1.0, Kling 3.0, Wan 2.7, Seedance, Veo 3.1, and more.