prime-skills/runcomfy-agent-skills
60 skills11M installs35 starsGitHub
Install
npx skills add https://github.com/prime-skills/runcomfy-agent-skillsSkills in this repo
1Video EditVideo-edit is a RunComfy Pro Pack skill that routes video edit intent to the best catalog model and runs it through the RunComfy CLI. It picks Wan 2.7 Edit-Video for general restyle and background swaps, Kling 2.6 Pro Motion Control for motion transfer, or Lucy Edit Restyle for lightweight outfit changes, each with documented prompting patterns. Triggers on edit video, restyle, motion control, or background swap requests.406kinstalls2Image To VideoThis skill animates still images into video through the RunComfy CLI by routing user intent to the best image-to-video model in the catalog. It picks HappyHorse 1.0 I2V for identity-stable portrait and product motion, Wan 2.7 with audio_url for custom voiceover lip-sync, or Seedance 2.0 Pro when image, reference video, and reference audio must combine in one clip. Each route ships documented prompting patterns, input schemas, and the exact runcomfy run command so agents get sharper output without trial-and-error model selection.405kinstalls3Nano Banana 2This skill generates images with Google Nano Banana 2 through the RunComfy CLI. Nano Banana 2 is the flash-tier Gemini-family text-to-image model optimized for rapid iteration, social-native aspect ratios, and predictable in-image typography when characters are quoted. The skill documents resolution and safety-tolerance knobs, batch num_images workflows, when to route to Nano Banana Pro or Flux instead, and the exact runcomfy run google/nano-banana-2/text-to-image invoke with subject-first prompting patterns.404kinstalls4Nano Banana EditThis skill edits images with Google Nano Banana 2's edit endpoint on RunComfy. It preserves subject identity while swapping backgrounds, localizing changes with spatial language, or running consistent batch edits across up to twenty input URLs in one call. The skill documents preservation-first prompting, aspect_ratio locking for series work, when to route to GPT Image 2 or Flux Kontext instead, and the runcomfy run google/nano-banana-2/edit invoke with full schema coverage.404kinstalls5Image EditThis skill is an intent router for image editing on RunComfy. It matches user goals to Nano Banana Edit for batch identity-preserving work, GPT Image 2 Edit for multilingual in-image text and multi-reference composition, Flux Kontext Pro for single-shot precise local edits, or Z-Image Turbo Inpaint when a grayscale mask defines the region. Each route includes schema fields, preservation-first prompting patterns, and the exact runcomfy run model_id invoke so agents avoid burning iterations on the wrong edit endpoint.404kinstalls6Flux KontextThis skill edits images with Black Forest Labs Flux 1 Kontext Pro through the RunComfy CLI. Kontext excels at single-reference, high-fidelity local edits where one declarative instruction changes a specific element while preserving face, pose, and framing. The skill documents imperative prompting grammar, when to route to Nano Banana Edit or GPT Image 2 instead, and the runcomfy run blackforestlabs/flux-1-kontext/pro/edit invoke with seed-based variant testing guidance.404kinstalls7Wan 2 7This skill generates text-to-video with Wan-AI Wan 2.7 on RunComfy. Wan 2.7 supports multi-reference conditioning, smooth motion physics, and audio-driven lip-sync when an audio_url voiceover is supplied. The skill documents duration, resolution, and aspect-ratio schema, camera and motion prompting, when to route to HappyHorse or Seedance instead, and the runcomfy run wan-ai/wan-2-7/text-to-video invoke for both default and lip-sync workflows.403kinstalls8Gpt Image EditThis skill edits images with OpenAI GPT Image 2's edit endpoint on RunComfy. It excels at preserving identity through targeted edits, rewriting embedded text in any script when characters are quoted, and composing subjects from one reference into scenes from another. The skill documents the images array schema, size auto versus fixed ratios, preservation-first prompting, when to use Nano Banana Edit for larger batches, and the runcomfy run openai/gpt-image-2/edit invoke.403kinstalls9Seedance V2This skill generates video with ByteDance Seedance 2.0 Pro on RunComfy. Seedance combines character images, reference video clips, and reference audio into coherent cinematic shots with native lip-sync when generate_audio is enabled. The skill documents the image versus text division rule, multi-modal reference specs, when to route to HappyHorse or Wan instead, and the runcomfy run bytedance/seedance-v2/pro invoke with camera and audio-direction prompting patterns.403kinstalls10Happyhorse 1 0This skill generates text-to-video with HappyHorse 1.0 on RunComfy. HappyHorse is currently the top blind-vote video model on Artificial Analysis, delivering native 1080p output with synchronized audio generated in the same pass and strong multi-shot character consistency when anchors are restated. The skill documents duration, aspect ratio, and resolution schema, motion-over-time prompting, when to use Wan for audio_url lip-sync instead, and the runcomfy run happyhorse/happyhorse-1-0/text-to-video invoke.403kinstalls11Flux 2 KleinThis skill generates images with Black Forest Labs Flux 2 Klein on RunComfy. Klein is the distilled low-latency Flux 2 variant with 4B for live iteration and 9B for higher-fidelity output, supporting multi-reference brand styling and declarative subject-first prompts. The skill documents step-count strategy, width and height bounds, when to route to Flux 2 Pro or GPT Image 2 instead, and runcomfy run invokes for both blackforestlabs/flux-2-klein/4b and 9b text-to-image endpoints.403kinstalls12Kling 3 0This skill generates video with Kuaishou Kling 3.0 on RunComfy across all six rendering endpoints. Kling 3.0 produces multi-shot cinematic video with synchronized native audio, consistent character identity, and physics-aware motion at Standard, Pro, or native 4K tiers in both text-to-video and image-to-video modes. The skill documents tier selection, shared input schema including prompt_segments, i2v image_url requirements, and runcomfy run kling/kling-3.0 tier mode invokes with cinematic prompting guidance.380kinstalls13Codex PetThis skill builds custom OpenAI Codex desktop pets from a single reference image via RunComfy. It generates one canonical pose with GPT Image 2 edit, then assembles the required 1536 by 1872 sprite atlas with nine animation rows using ImageMagick micro-transforms for idle, running, waving, jumping, and other Codex Pet states. The output pet.json and spritesheet.webp drop into the Codex pets folder and work like built-in pets without Codex Pro, imagegen, or a direct OpenAI API key—only RUNCOMFY_TOKEN and local magick binary.369kinstalls14Ai Video GenerationThis skill is the master router for AI video generation on RunComfy via the runcomfy CLI. It maps user intent to the right model across HappyHorse 1.0, Kling 3.0 and 2.6, Seedance v2, Wan 2.7, Google Veo, MiniMax Hailuo, and other catalog entries for text-to-video, image-to-video, and Veo video-extend workflows. Each route includes when to pick and avoid the model, documented prompting patterns, and the minimal runcomfy run vendor model endpoint invoke so agents produce clips without catalog research.350kinstalls15Ai Image GenerationThis skill is the master router for AI image generation and editing on RunComfy via the runcomfy CLI. It maps user intent to the right model across FLUX 2 Klein and Pro tiers, Google Nano Banana 2 and Pro, OpenAI GPT Image 2, ByteDance Seedream and Dreamina, Alibaba Qwen and Z-Image Turbo, and Wan 2.7 image endpoints. Each route documents when to pick and avoid the model, prompting patterns for typography, photoreal portraits, and sub-second iteration, plus the exact runcomfy run invoke for t2i and i2i workflows.349kinstalls16Runcomfy CliThis skill teaches agents how to use the RunComfy CLI to call any model on RunComfy from the command line. The runcomfy binary provides one authentication flow, schema discovery, request submission, status polling, output download, and scripting via JSON output mode across image generation, video generation, editing, lip-sync, face swap, inpainting, and LoRA training endpoints. It documents install paths, runcomfy login and RUNCOMFY_TOKEN auth, runcomfy run invocation, streaming and no-wait modes, sysexits error codes, and security boundaries for agent Bash runcomfy permissions.347kinstalls17Ai Avatar VideoThis skill creates AI avatar and talking-head videos on RunComfy by routing user intent across ByteDance OmniHuman, Wan 2.7 with audio_url, HappyHorse 1.0 with in-pass audio, and Seedance v2 Pro multi-modal composition. It classifies whether the user has a pre-recorded audio file or only a script, whether they need photoreal portrait or cinematic scene control, and picks the matching route with documented prompting patterns and the exact runcomfy run invoke for lip-synced spokesperson, UGC voiceover, and virtual presenter workflows.347kinstalls18Face SwapThis skill swaps faces and characters in images and video on RunComfy via the runcomfy CLI. It routes video identity substitution to community Wan 2-2 Animate for audio-driven character animation or Kling 2-6 Motion Control Pro when source motion must be preserved onto a new character. Still-image swaps route through GPT Image 2 Edit, Nano Banana Edit, or Flux Kontext depending on multilingual text, batch size, and single-ref fidelity needs. The skill documents consent requirements, per-route schemas, and runcomfy run invokes for each face and character swap intent.347kinstalls19Video InpaintingThis skill performs video inpainting and region edits across frames on RunComfy via the runcomfy CLI. It routes prompt-driven region changes to Wan 2-7 edit-video by default for object removal, watermark cleanup, and sky replacement using spatial language without explicit masks. Alternative routes include Lucy Edit Restyle for identity-stable outfit swaps and Seedream 4-0 edit-sequential when treating a clip as a frame stack. Each route documents invoke patterns, limitations on temporal consistency, and when to fall back to ComfyUI mask workflows for pixel-precise inpaint.345kinstalls20Image InpaintingThis skill teaches an agent to perform mask-driven image inpainting on RunComfy using the runcomfy CLI. It routes to Tongyi MAI Z-Image Turbo Inpainting when a binary mask is available, and falls back to identity-preserving edit models like Nano Banana 2, GPT Image 2, and FLUX Kontext when the region must be described in prose. Use it for object removal, watermark cleanup, blemish repair, and any controlled local edit where precision matters.345kinstalls21Controlnet PoseThis skill guides agents to condition image or video generation on pose, skeleton, depth, or motion references using RunComfy. It routes across Kling 2-6 Motion Control, Wan 2-2 Animate, and Z-Image Turbo ControlNet LoRA depending on whether the user needs video motion transfer or still-image pose control. The agent picks the right endpoint and constructs the correct runcomfy run invocation.344kinstalls22LipsyncThis skill helps agents lip-sync faces to audio tracks on RunComfy through the runcomfy CLI. It routes across ByteDance OmniHuman, Sync Labs sync v2, Kling lipsync, and Creatify based on whether the user has a portrait still, source video, or needs generate-and-sync from a script. The agent selects the right endpoint and ships the documented invoke pattern.344kinstalls23Video ExtendThis skill teaches agents to extend or continue existing video clips on RunComfy via the runcomfy CLI. It routes to Google Veo 3-1 extend-video and fast extend-video endpoints, taking a source video plus a prompt describing what happens next. Use it when a short Veo clip needs to be longer or when building chained narrative shots from a single seed.344kinstalls24Elevenlabs Music GenerationThis skill guides agents to generate full songs and instrumental tracks with ElevenLabs Music on RunComfy via the runcomfy CLI. It produces studio-quality 44.1 kHz stereo audio from style descriptions and structured lyrics, with section-level control and multilingual vocals. Use it for backing tracks, vocal songs, jingles, podcast intros, and royalty-free music beds.344kinstalls25Image OutpaintingThis skill teaches agents to outpaint still images on RunComfy using the runcomfy CLI. It extends canvas beyond the original borders, changes aspect ratios, and fills in uncropped areas while preserving central content. The skill routes across Nano Banana 2 Edit, GPT Image 2 Edit, FLUX Kontext Pro, and brand edit endpoints based on whether the outpaint is prose-driven, reference-driven, or brand-locked.344kinstalls26RelightThis skill helps agents relight still images on RunComfy via the runcomfy CLI. It changes lighting setup, color temperature, direction, and mood using Qwen Edit 2509 relight LoRA when purpose-built relighting matters, with fallbacks to Nano Banana 2, GPT Image 2, and FLUX Kontext for prose-driven lighting edits. Use it for product relighting, portrait mood shifts, and color-grade changes.344kinstalls27Video OutpaintingThis skill guides agents to outpaint videos on RunComfy using the runcomfy CLI. It extends the spatial canvas, changes aspect ratios such as 9:16 to 16:9, and adds environment beyond the original frame while preserving central action. The skill routes prompt-shaped spatial extension through Wan 2-7 edit-video and points to ComfyUI outpaint workflows when seam quality matters.344kinstalls28Ai MusicThis skill is a smart router for AI music generation on RunComfy via the runcomfy CLI. It routes to ElevenLabs AI Music for premium 44.1 kHz vocal tracks and ACE Step for cheaper tag-driven composition, plus ACE Step audio-inpaint and audio-outpaint for repairing or extending existing tracks. The agent picks the right model for vocal hooks, background beds, multilingual songs, or track edits.336kinstalls29Ace StepThis skill teaches agents to generate, inpaint, and outpaint music with ACE Step on RunComfy via the runcomfy CLI. ACE Step is StepFun-AI open-weights music with tag-driven composition, multilingual lyrics, and stereo output at very low per-second cost. Four endpoints cover text-to-audio, ACE Step 1.5 lyrics, audio-inpaint for chorus repair, and audio-outpaint for lengthening tracks.335kinstalls30Gpt Image 2This skill guides agents to generate and edit images with OpenAI GPT Image 2 on RunComfy through the runcomfy CLI. It documents GPT Image 2 strengths including embedded text, logos, multilingual typography, and instruction precision, plus its three fixed sizes and edit-with-preservation language. The skill also explains when to route to sibling models like Flux 2 or Nano Banana Pro instead.51kinstalls31Ace StepRoutes ACE Step open-weights music generation API calls via runcomfy CLI with tag-driven composition support. Developers specify genre, mood, instruments, and multilingual lyrics; the skill generates 5s-4min stereo tracks at 27x lower cost than premium alternatives.0installs32Ai Avatar VideoCreate AI avatar and talking-head videos via RunComfy CLI. Routes across OmniHuman for portrait avatars, HappyHorse for script-only videos with in-pass audio, and Seedance for cinematic compositions.0installs33Ai Image GenerationGenerate and edit images with 11+ AI models via the RunComfy CLI - text-to-image and image-to-image, one auth, one command. This skill picks the right model for the user's intent (typography precision, photoreal portraits, sub-second iteration) and ships each model's documented prompting patterns plus the minimal runcomfy run invoke.0installs34Ai MusicSmart router for music generation via runcomfy CLI that picks between ElevenLabs (premium $0.0083/sec stereo vocals) and ACE Step (open-weights $0.0002-0.0003/sec). Developers request audio output; the skill routes and returns the generated track.0installs35Ai Video GenerationAI Video Generation is a RunComfy agent skill that routes video requests across the full model catalog through one runcomfy CLI interface. It classifies intent into text-to-video, image-to-video, or Veo extend paths, then picks models like HappyHorse 1.0 for default Arena-leading clips with in-pass audio, Wan 2-7 for audio-driven lip-sync, Seedance v2 for multi-modal cinematic shots, Veo 3-1 for physics-accurate product motion, and Kling 3.0 for multi-shot character identity at up to 4K. Each route documents schema fields, example --input JSON, prompting tips, and when to avoid overpowered or costly tiers. Common patterns cover social vertical reels, brand product spins, cinematic ads, dialog lip-sync with audio_url, and chained Veo extends via the video-extend skill. Prerequisites are RunComfy CLI login or RUNCOMFY_TOKEN and user-provided reference URLs treated as untrusted. Security notes cover token storage, JSON input boundaries, and a 2 GiB download cap. Use when users ask to generate video, animate a still, make something move, or extend an existing clip.0installs36Codex PetCodex Pet on RunComfy builds custom OpenAI Codex desktop companions from a single source image without Codex Pro or the internal $imagegen skill. It produces the exact artifact Codex expects: pet.json plus a 1536x1872 spritesheet.webp with 8 columns and 9 animation rows of 192x208 cells covering idle, running, waving, jumping, failed, waiting, review, and related states. The pipeline makes one runcomfy run openai/gpt-image-2/edit call for a canonical chibi pose on magenta chroma-key, then assembles all 72 frames programmatically with ImageMagick micro-transforms. Prerequisites are RunComfy CLI, RUNCOMFY_TOKEN, ImageMagick, and a public HTTPS source image URL. Output installs to ${CODEX_HOME}/pets/<name>/ where Codex picks it up beside eight built-in pets. The skill documents prompting for exaggerated chibi proportions, chroma-key cleanup, per-row frame counts, and tuning animation deltas. It targets users who want batch pet generation, contest entries, or a RunComfy-backed alternative to official hatch-pet.0installs37Controlnet PosePose-conditioned generation via RunComfy CLI. Routes across Kling Motion Control for video motion transfer, Z-Image ControlNet LoRA for image generation from pose skeletons or depth maps. Picks the right model by intent: single still vs video.0installs38Elevenlabs Music GenerationRoutes ElevenLabs Music API calls via runcomfy CLI to generate studio-quality songs and instrumentals from text prompts. Developers supply style descriptions and structured lyrics; the skill returns 44.1 kHz stereo tracks with section markers.0installs39Face SwapSwap a face into a still or video via the RunComfy CLI. Routes across Wan 2-2 Animate for video character swap, Kling Motion Control for motion transfer, and image-edit models for still images. Picks the right model by intent.0installs40Flux 2 KleinGenerates images with Black Forest Labs' Flux 2 Klein, the distilled low-latency variant of Flux 2, through the local RunComfy CLI. A developer uses the 4B variant for sub-second concepting and live art-direction, then switches to the 9B variant for a polished final pass with the same prompt grammar. It documents the step-count strategy and routes to Flux 2 Pro or Seedream 5 when maximum resolution or detail is needed.0installs41Flux KontextEdits a single source image with Black Forest Labs' Flux 1 Kontext Pro model, invoked through the local RunComfy CLI. A developer reaches for it for targeted local edits that keep the source identity intact, like adding an object to a portrait or swapping brand text on a label. It bundles the model's documented prompting patterns and tells you when to route to a sibling model for multi-image or embedded-text edits.0installs42Gpt Image 2gpt-image-2 from agentspace-so/runcomfy-agent-skills routes image generation and editing to OpenAI GPT Image 2 (ChatGPT Images 2.0) via the local RunComfy CLI using runcomfy run openai/gpt-image-2/text-to-image or /edit. The skill documents GPT Image 2 strengths—embedded text, logos, multilingual typography, and instruction precision—and its three fixed output sizes plus edit-with-preservation prompting patterns. Developers reach for gwt-image-2 when triggers include gpt image 2, gpt-image-2, ChatGPT Images 2, or explicit generate-or-edit requests for this model. The skill also explains when to route to sibling models Flux 2, Nano Banana Pro, or Seedream instead. MIT-licensed and hosted on runcomfy.com, it suits agent pipelines that need branded visuals, UI mock imagery, or localized text-in-image assets without managing OpenAI API credentials directly. Requires RunComfy CLI installed locally.0installs43Gpt Image EditEdits images with OpenAI's GPT Image 2 /edit endpoint, run through the local RunComfy CLI. A developer uses it to localize an ad headline into another script, swap a CTA in place, or compose a subject into a new scene from multiple references. It documents the model's preservation-first prompting and routes to Nano Banana Edit or Flux Kontext when batch consistency or single-shot fidelity matters more.0installs44Happyhorse 1 0Generates text-to-video with HappyHorse 1.0 through the local RunComfy CLI. A developer uses it for multi-shot brand stories that keep one character consistent, talking-head explainers that need in-clip voiceover, or multilingual short-form ads. It outputs native 1080p with synchronized audio in the same pass and routes to Wan 2.7 or Seedance for audio-driven lip-sync from an external track.0installs45Image EditEdits images on RunComfy by routing the user's intent to the right model: Nano Banana Edit for identity-preserving and batch edits, GPT Image 2 Edit for multilingual in-image text and multi-reference composition, Flux Kontext Pro for single-shot precise local edits, or Z-Image Turbo Inpaint for mask-driven region replacement. A developer uses it to edit one image or a batch without guessing the model. It invokes the RunComfy CLI with each model's documented body.0installs46Image InpaintingMask-driven region edits on still images via RunComfy CLI - remove objects, fill gaps, replace masked areas. Routes to Z-Image Turbo Inpainting when a mask is available, falls back to instruction-driven models when the region must be described.0installs47Image OutpaintingRoutes image outpainting API calls via runcomfy CLI to extend still-image canvas beyond original bounds. Developers specify desired canvas expansion or aspect ratio; the skill selects the optimal model and returns the outpainted image.0installs48Image To VideoAnimates a still image on RunComfy by routing intent to the right image-to-video model: HappyHorse 1.0 I2V for general portrait or product animation, Wan 2.7 with an audio_url for custom-voiceover lipsync, or Seedance 2.0 Pro for multi-modal composition from image, reference video, and reference audio. A developer uses it to turn a photo into motion without picking the model by hand. It invokes the RunComfy CLI with each model's documented body.0installs49Kling 3 0kling-3-0 is an agent skill wrapping Kling 3.0 video generation capabilities for use in Claude agents. It enables agents to generate videos from text prompts and images via the RunComfy AI media skills framework.0installs50LipsyncDrive a face's mouth from an audio track via RunComfy CLI. Routes across Sync Labs sync v2 for hero-quality mouth-swap on existing video, OmniHuman for portrait avatars, and Kling Lipsync for script-only videos.0installs51Nano Banana 2Generates images with Google Nano Banana 2, the Gemini-family flash-tier text-to-image model, hosted on the RunComfy Model API. A developer uses it for rapid drafts, social-thumbnail batches, and posters that need predictable in-image typography. It calls runcomfy run google/nano-banana-2/text-to-image and documents resolution tiers, a safety-tolerance dial, and optional web-grounded generation.0installs52Nano Banana EditEdits images with Google Nano Banana 2's image-to-image edit endpoint hosted on the RunComfy Model API. A developer uses it to preserve subject identity while swapping backgrounds or clothing, localize edits with spatial language, and run consistent batch edits across up to 20 input images. It calls runcomfy run google/nano-banana-2/edit through the RunComfy CLI.0installs53RelightRoutes image relighting API calls via runcomfy CLI to adjust lighting, color temperature, direction, or mood in product photos and portraits. Developers describe desired lighting changes in natural language; the skill routes to Qwen Edit 2509 or fallback models.0installs54Runcomfy CliRun any model on RunComfy from the command line via the runcomfy CLI. One binary, one auth, hundreds of model endpoints - image generation, video, edit, lip-sync, face swap, ControlNet, and more. Submit a request, poll for status, download the output.0installs55Seedance V2Generates cinematic short-form video with ByteDance's Seedance 2.0 Pro through the local RunComfy CLI. A developer uses it for lip-synced spokesperson ads, brand-consistent multi-language narratives, or previs that combines character images, scene videos and reference audio into one shot. It emphasizes putting stable identity in image references and evolving action in the prompt, and routes to HappyHorse or Wan 2.7 for other needs.0installs56Video EditEdits existing video on RunComfy by routing the user's intent to the right model: Wan 2.7 Edit-Video for general restyle and background or packaging swaps, Kling 2.6 Pro Motion Control for transferring motion from a reference clip, or Lucy Edit Restyle for lightweight identity-stable outfit swaps. A developer uses it to transform a source video without guessing which model fits. It invokes the RunComfy CLI with each model's documented schema.0installs57Video ExtendRoutes Google Veo 3-1 extend-video API calls via the runcomfy CLI to continue short clips with consistent motion and lighting. Use when developers have a seed video and need temporal continuation beyond per-call duration limits.0installs58Video InpaintingRegion edits across video frames via RunComfy CLI - remove objects that appear across many frames, clean up wires or watermarks. Routes to Wan 2-7 Edit-Video for prompt-driven edits, Lucy Edit for identity-stable restyle.0installs59Video OutpaintingRoutes video outpainting API calls via runcomfy CLI to extend video canvas and change aspect ratio. Developers specify desired frame expansion or aspect conversion; the skill routes through Wan 2-7 or ComfyUI workflows depending on quality needs.0installs60Wan 2 7Generates text-to-video with Wan-AI's Wan 2.7 model through the local RunComfy CLI. A developer uses it for short clips that lip-sync to a supplied voiceover track, multi-language dub variants, or motion-controlled shots up to 15 seconds. It documents the duration, resolution and aspect-ratio schema and points to sibling models like HappyHorse or Seedance for in-pass voice generation.0installs