agentspace-so/runcomfy-skills
12 skills363 installs24 starsGitHub
Install
npx skills add https://github.com/agentspace-so/runcomfy-skillsSkills in this repo
1Image To Videoimage-to-video is a RunComfy Pro Pack skill that acts as a decision layer between your creative intent and the right image-to-video model in the RunComfy catalog. Indie makers and content builders use it when they have a reference still and need motion, optional custom voiceover lip-sync, or richer multi-modal animation from image plus reference video and audio. Instead of guessing model IDs, the skill maps triggers like animate this image or i2v to HappyHorse 1.0 I2V for general high-quality animation, Wan 2.7 when audio_url-driven lip-sync matters, or Seedance 2.0 Pro when combining multiple reference modalities. Each branch carries prompting guidance from vendor documentation so CLI calls through runcomfy run land with parameters that match how those models were trained. It assumes the RunComfy CLI is installed locally and that you are comfortable spending GPU credits per render. The skill does not replace storyboarding or legal clearance for likeness and audio—it accelerates execution once direction is clear.57installs2Nano Banana 2Nano Banana 2 — Pro Pack on RunComfy is an agent skill for solo builders who need fast Gemini-family flash text-to-image without treating RunComfy like a black-box HTTP call. It documents what Nano Banana 2 is good at—ideation loops, social-thumbnail batches, drafts with readable in-image type—and encodes the prompting habits that match Google’s documented patterns so results stay sharper than generic “make an image” chat. You get explicit coverage of resolution-tier pricing, the safety-tolerance dial, and decision rules for when to escalate to Nano Banana Pro, GPT Image 2, Flux 2, or Seedream instead of forcing the flash tier. Installation assumes the RunComfy CLI on your machine; the agent prepares prompts and parameters, then executes the documented run command. Use it whenever someone names nano banana, nano-banana-2, gemini image, or google image gen, or when you are batching marketing visuals and want one repeatable workflow across Claude Code, Cursor, or Codex.54installs3Video EditVideo Edit — Pro Pack on RunComfy is an agent skill for solo creators and indie marketers who already have source footage and need deterministic routing to the correct generative edit endpoint. Instead of guessing which RunComfy model fits “swap background” versus “motion control,” the skill matches intent to Wan 2.7 Edit-Video, Kling 2.6 Pro Motion Control, or Lucy Edit Restyle and ships each vendor’s prompting guidance so fewer iterations burn credits. Invocation is CLI-first: runcomfy run with the documented vendor/model/endpoint path. Triggers align with explicit edit language—video edit, restyle, background swap, motion control, outfit swap. It fits Prism’s Grow journey because edited clips feed distribution and lifecycle content, while Validate-phase builders can use the same router for prototype promo cuts. You need the RunComfy CLI and network access to RunComfy; it does not replace a full non-linear editor for frame-accurate manual cuts.42installs4Gpt Image 2gpt-image-2 is a RunComfy Pro Pack skill for solo builders who need ChatGPT Images 2.0–class generation and editing through a local CLI instead of wiring OpenAI keys. It explains when GPT Image 2 wins on directive precision—logos, in-image text, multilingual labels—and when to switch to sibling models on RunComfy. Use it while building landing pages and UI mocks, or when preparing launch and growth creatives. The skill encodes size constraints, preservation language for edits, and install via npx skills add agentspace-so/runcomfy-skills. Intermediate complexity: you need RunComfy CLI access and comfort with async REST-backed runs. It is an integration skill, not a general design system—pair with your brand guidelines and human review before publishing customer-facing assets.33installs5Seedance V2Seedance 2.0 Pro on RunComfy is an agent skill for solo builders who need cinematic short clips without standing up their own GPU stack. It documents ByteDance Seedance 2.0 Pro’s strengths—multimodal references, refined motion, and native lip-synced audio—and walks your agent through invoking the model with the RunComfy CLI. Use it when someone asks for Seedance, Seedance 2/Pro, or Bytedance-backed video generation, or when you are assembling character images, scene video, and audio into one coherent take. The skill also explains the 4–15 second duration constraints and when sibling models on RunComfy are a better fit for speed, style, or budget. It is aimed at indie marketers, product demo creators, and agent workflows that treat video as a deliverable, not a one-off chat attachment.32installs6Image EditImage Edit is a RunComfy Pro Pack agent skill that acts as a smart router between user intent and the correct edit model in the RunComfy catalog. Solo builders and small teams use it when they need production-quality image-to-image work—background swaps, object removal, localized inpainting, or marketing copy rewritten inside the frame—without manually comparing vendor APIs. The skill documents when to pick Nano Banana Edit for large identity-preserving batches, GPT Image 2 Edit for multilingual text and layout-precise composition, Flux Kontext Pro for single-reference high-fidelity local edits, and Z-Image Turbo Inpaint for mask-driven regions. It encodes prompting patterns bundled per model and drives execution through the local RunComfy CLI (`runcomfy run`). Install it when your agent should recognize edit-oriented language and land on the right model on the first pass rather than burning credits on mismatched capabilities.30installs7Nano Banana EditNano Banana Edit is a RunComfy Pro Pack agent skill for solo builders who need Google's Nano Banana 2 image-to-image endpoint without guessing schema or model siblings. It explains strengths—keeping subject identity, swapping backgrounds, localizing edits with spatial language, and batching up to twenty reference images—and documents the request shape and CLI invocation. The skill steers you away from wrong routes: text-to-image-only work, cases better served by GPT Image 2 edit or Flux Kontext, or flows that need generation rather than edit. Install via the runcomfy-skills repo and invoke when users say nano banana edit or ask to edit images with this model. It fits agent-driven creative pipelines where the agent selects the correct RunComfy model and runs the command locally, reducing wrong-endpoint errors during product thumbnails, marketing variants, or user-upload edit features.25installs8Flux 2 Kleinflux-2-klein is a RunComfy-backed agent skill that turns Black Forest Labs’ distilled Flux 2 Klein model into a repeatable image pipeline for indie makers who need speed without abandoning prompt discipline. It documents when Klein’s sub-second latency and multi-reference styling beat heavier models, how to trade 4B for throughput versus 9B for quality, and how step counts should drop for exploration then rise for final renders. The skill calls the local RunComfy CLI against documented blackforestlabs/flux-2-klein routes and includes explicit escalation guidance to Flux 2 Pro, Seedream 5, or GPT Image 2 when fidelity or text rendering demands it. Triggers align with plain language requests for flux klein or BFL flux 2 generation. It suits solo builders producing launch graphics, app store shots, or branded social content who want the model’s pro pack baked into the agent procedure rather than rediscovering parameters on every session.21installs9Flux Kontextflux-kontext is a RunComfy integration skill for Black Forest Labs’ Flux 1 Kontext Pro—an image-to-image model aimed at precise local edits from one reference image with strong prompt adherence. Solo builders use it when they need marketing visuals, UI mock tweaks, or product shots without opening a separate design toolchain for every iteration. The skill documents when Kontext is the right choice versus alternatives named in the pack (Nano Banana Edit, GPT Image 2 edit, Flux 2 Klein), the input schema of image plus prompt, and how to invoke the model through `runcomfy run` on the local CLI. Triggers cover explicit names like “flux kontext”, “BFL kontext”, and “flux 1 kontext”. It is optimized for agents that should not guess API flags: follow the bundled prompting patterns for sharper edits than generic chat instructions. You still need RunComfy credentials and network access; the skill does not host weights itself.18installs10Gpt Image EditGPT Image Edit is a RunComfy Pro Pack agent skill for OpenAI GPT Image 2’s image-to-image `/edit` endpoint (ChatGPT Images 2.0). Solo and indie builders install it when they want an agent to edit screenshots, mockups, ads, or UI visuals without guessing API fields or weak prompts. The skill encodes how to phrase preservation, typography, layout, and multilingual in-image text changes, and how to attach multiple reference images (up to ten) for consistent composites. It also explains when the same job should switch to Nano Banana Edit, Flux Kontext, or GPT Image 2 text-to-image so you do not fight the wrong model. Triggers include explicit asks for “gpt image edit”, “chatgpt image edit”, or editing with GPT Image 2. You need the RunComfy CLI and network access to RunComfy; outputs are edited image files suitable for landing pages, social posts, or design iterations inside your repo workflow.18installs11Wan 2 7Wan 2.7 is a RunComfy-hosted agent skill for Wan-AI’s flagship text-to-video model, aimed at solo builders who need cinematic clips without maintaining GPU stacks. The skill explains when Wan 2.7 wins—supplying an audio track for lip-sync, tightening motion with multiple reference inputs, and leaning on prompt expansion for smoother transitions—and documents the parameter surface for length, resolution, and aspect ratio. It also gives an honest sibling comparison so you do not burn credits on the wrong model when Seedance, Kling, LTX 2, or HappyHorse 1.0 fits better. Installation is via the skills.sh pattern against agentspace-so/runcomfy-skills, then generation through the RunComfy Model API and CLI. Use it while wiring generative pipelines in your repo, producing launch and growth creatives, or letting an agent route natural-language video requests to the correct RunComfy endpoint.17installs12Happyhorse 1 0happyhorse-1-0 is a RunComfy Pro Pack skill for solo builders who want agent-invoked text-to-video without building their own GPU pipeline. It documents HappyHorse 1.0 on the RunComfy Model API—native 1080p output, in-pass synchronized dialogue and ambient audio, multi-shot character consistency, and multi-language prompts—and gives the exact CLI entry point `runcomfy run happyhorse/happyhorse-1-0/text-to-video`. The SKILL.md also explains parameter schema for duration, aspect ratio, and resolution and when to route elsewhere to Wan 2.7, Seedance 2, or LTX 2 for different quality or cost tradeoffs. Install via `npx skills add agentspace-so/runcomfy-skills --skill happyhorse-1-0 -g`. Triggers include happyhorse, happy horse, happyhorse 1.0, happyhorse video, or explicit requests to generate video with this model. Ideal for short-form promos, demo clips, and creative iteration from Claude Code or Codex with network and shell access to RunComfy.16installs