REPO PACKAGE
inference-sh skills
COMMUNITY85skills
0ready to use
LABSCO SUMMARY
Of those 85, thirty-nine need your own inference.sh account and its per-generation credit balance in our dependency check — every image, video, voice, music, or dubbing skill routes through the same belt CLI and one of the platform's 250+ hosted models (FLUX, Veo, ElevenLabs, and dozens more), so one account and credit balance covers all of them, but none work without it. Three more are plain local tools, and forty-three came back unverified. Not all of it needs an account at all: a run of generic strategy guides — case-study-writing, customer-persona, competitor-teardown, email-design, data-visualization, content-repurposing — read as standalone writing and design knowledge with no inference.sh CLI call anywhere in them, and a handful of others just document the platform's own React chat/agent UI components or its JavaScript and Python SDKs rather than teaching an agent to run a command. The eight ElevenLabs-branded skills alone (dialogue, dubbing, music, sound-effects, stt, tts, voice-changer, voice-isolator) show how the count inflates: one vendor's API, split into eight separate installable skills.
This is for someone who has already bought into inference.sh as their model marketplace and wants an agent that can drive it end to end — pick a model, generate assets, and stay inside one billing relationship. If you don't want another paid AI-generation subscription on top of whatever your coding agent already provides, only the account-free guides and the two SDK skills will be of any use to you; the rest is a 250+-model catalog gated behind one login.
READ THE FULL ANALYSIS
Real churn already. We have seen 110 skills from this repository in total, and 25 are already dead — retired or renamed since we first captured it. That is a higher turnover than most packages this size, and it tracks with how fast the underlying model list moves: several of today's 85 are already a second or third name for the same wrapped model.
Checked 19 September 2026 against inference-sh/skills' own README and skill descriptions; the account-vs-account-free split above is our dependency check's own labeling, not a distinction the README draws itself.
ALSO IN THIS PACKAGE
PLUGINS
1 itemPLUGIN
inference-sh
AI agent skills for models via inference.sh CLI - generate images with FLUX, create videos with Veo, call LLMs, search the web, and more
WHAT'S INSIDE
85 showing · 85 total85 of 85
SETUP
Showing all 85 skills
agent-browser
A web browser an assistant can operate on your behalf — opening pages, clicking, typing, and taking a picture or a video of whatever happens.
agent-tools
One command that runs AI models in the cloud — pictures, video, writing, search, 3D — so nothing has to be installed and no graphics card of your own is needed.
agent-ui
Drops a ready-made AI chat window into a website you are building — one that can act on the page, and that stops to ask permission before it does.
ai-automation-workflows
What it takes to leave an AI job running with nobody watching it, written out as shell scripts you can copy.
ai-avatar-video
A stand-in presenter for a video you would otherwise have to film — one still photo, a script, and the face talks.
ai-content-pipeline
Step-by-step recipes that feed one AI model's output straight into the next, so that what comes out at the end is a finished video and not raw material.
ai-image-generation
Makes or edits a picture from a written description, with more than 50 different image models to choose between behind one command.
ai-marketing-videos
Video made to sell something — cut to the shape each platform expects, and to the beats an advert is expected to hit.
ai-music-generation
Describe the music you want and get an original track back, written and performed by a model rather than by a band.
ai-podcast
A video podcast where nobody was ever filmed: each speaker gets a face and a voice, and the conversation is assembled turn by turn from the script.
ai-podcast-creation
Everything between a written script and a listenable episode — the voices, the pacing, the music at either end — done without a microphone.
ai-product-photography
Product shots without the photo shoot: no studio, no lighting kit, no photographer, just a description of the picture you need.
ai-rag-pipeline
An answer put together from a live search of the web, with links back to where each part of it came from, rather than from whatever the model happens to remember.
ai-social-media-content
Posts that already fit the platform they are going to — the right shape, the right length, and the words to go under them.
ai-video-generation
Turns a written description into a video clip — or brings a still photo into motion — with more than 40 video models sitting behind the one command.
ai-voice-cloning
Turns written words into spoken audio in a voice you choose, with dozens of ready-made voices to pick from and control over how much emotion goes into the delivery.
app-store-screenshots
A playbook for the pictures on your app's App Store or Google Play page — the part of the listing that decides whether someone taps Install.
background-removal
Cuts a person or a product out of a photo and throws away whatever was behind them.
book-cover-design
Romance covers look like romance covers for a reason — this lays out what readers of each genre expect to see, and generates cover art that matches.
building-inferencesh-apps
Takes a program you have written in Python or Node.js and puts it on inference.sh's machines, so it can be run on demand instead of only on your own computer.
case-study-writing
A recipe for the customer success story a sales team hands out, built on hard numbers rather than vague praise about improved efficiency.
character-design-sheet
AI draws a different-looking person every time you ask — this pins one character down so they stay recognisably the same across every new picture.
chat-ui
Prebuilt parts of a chat window — the scrolling message area and the box you type into — ready to drop into a React app.
competitor-teardown
A structured way to size up the companies you compete with — seven things to look at, and where to find each one.
content-repurposing
One blog post or podcast episode is worth ten smaller posts — this is how to get them out of it without pasting the same text everywhere.
customer-persona
A written portrait of the one customer you are actually selling to — their age, their income, what frustrates them, what makes them buy — built from research rather than a guess in a meeting room.
data-visualization
Picking the chart that fits your numbers and making it readable — starting with why a pie chart is almost never the right one.
dialogue-audio
Write out a two-person conversation and hear it spoken back, with each speaker's voice, timing and mood coming from how you wrote the lines.
elevenlabs-dialogue
Casts a scripted conversation with named premium voices — george, aria, brian — instead of one flat narrator reading both parts.
elevenlabs-dubbing
Takes a video or a podcast and re-voices it in another language, keeping each speaker's own voice and timing so it still sounds like them talking.
elevenlabs-music
Describe the music you want in plain words — the mood, the instruments, the tempo — and get back an original track up to ten minutes long.
elevenlabs-sound-effects
Type "glass shattering on a concrete floor" and get that sound back as an audio file, anywhere from half a second to twenty-two seconds long.
elevenlabs-stt
Turns a recording of people talking into a written transcript, in any of ninety-odd languages.
elevenlabs-tts
Reads your text aloud in one of twenty-odd studio-quality voices — British, American, Australian, warm or authoritative — rather than a robotic one.
elevenlabs-voice-changer
Keeps what you said and the way you said it, but makes it come out in somebody else's voice.
elevenlabs-voice-isolator
Pulls the voice out of a noisy recording and leaves the traffic, the hum and the café behind.
email-design
How to build a marketing email that still looks right in Gmail and Outlook instead of collapsing into a column of broken boxes.
explainer-video-guide
Making the short video that explains what your product does, from the first line of the script to the last second of the edit.
flux-image
FLUX is a family of AI image generators — this is which one to reach for when you want the best picture, and which when you want fifty cheap ones.
google-veo
Google's Veo makes a video clip out of a written description, with no camera and no footage involved.
gpt-image
OpenAI's picture-making models, for both inventing an image from scratch and altering one you already have.
happyhorse
Alibaba's video model, built for movement that obeys physics — fifteen seconds at most, at 720p or 1080p.
image-to-video
A still photo turned into a few seconds of movement, with the camera moving the way you describe it.
image-upscaling
Makes a small or blurry image bigger and sharper instead of just stretching it into mush.
infsh-cli
One command that runs any of the AI models hosted on inference.sh — pictures, video, chatbots, web search — on their hardware instead of yours.
javascript-sdk
Puts inference.sh inside a JavaScript or TypeScript project, so the AI calls happen in your own app rather than at a terminal.
landing-page-design
A sales page a stranger can understand before they scroll, and that leaves them one obvious thing to click.
linkedin-content
Writing a LinkedIn post that people actually stop for, rather than one that reads like a press release.
llm-models
Claude, Gemini, Kimi and a hundred more chatbots behind a single command, so switching from one to another is a change of one word.
logo-design-guide
Designing the little symbol that stands for a brand — the swoosh, the apple, the bird — from first rough concepts through to a finished mark.
nano-banana
Google's own image generator, in two versions — one slower and sharper, one fast — reachable from a command line.
nano-banana-2
Google's newest Gemini image model, the one after Nano Banana — same picture-making, driven from a command line or from Python.
newsletter-curation
A newsletter is a pile of links plus a reason to trust whoever picked them — this covers the picking, the writing and the sending.
og-image-design
The picture that shows up when somebody pastes your link into Slack, Twitter or a group chat.
p-image
Pruna takes existing image generators and re-engineers them to run quicker — these are its four, for making pictures and for editing them.
p-video
Three video models from Pruna, tuned so that a clip comes back in less time and for less money than the originals cost.
p-video-avatar
An on-camera presenter made out of one photograph — the cheap, fast way to avoid filming a person or paying for HeyGen.
pitch-deck-visuals
A pitch deck that does not look homemade — what belongs on each slide, and what the picture on it should be.
press-release-writing
The announcement you send to journalists, in the format newsrooms have used for a century, so it reads as news rather than as an advert.
product-changelog
The what's-new list your users read after an update, written so they can tell at a glance whether any of it affects them.
product-hunt-launch
Getting a new product onto Product Hunt's front page — the listing, the pictures, and what to do on the day itself.
product-photography
The photographs a product needs before it can sell online, generated rather than shot in a studio with a real camera.
prompt-engineering
How to ask an AI for what you actually want — the wording that works on a chatbot, and the quite different wording that works on an image or video generator.
python-executor
A throwaway Python machine in the cloud with a hundred libraries already installed on it — you send code, it runs, you get the files back.
python-sdk
The Python library for inference.sh: run a model, or build an agent that calls your own functions, all from inside a script.
qwen-image-2
Alibaba's quick, general-purpose image generator, and the faster half of the two Qwen models.
qwen-image-2-pro
Most image generators garble any writing they try to draw; this is Alibaba's answer to that.
related-skill
A table of contents for everything else inference.sh ships — what exists for images, video, audio or search, and the command that installs it.
remotion-render
Video made out of code instead of footage — you write a React component, and an MP4 comes back.
seedance
ByteDance's video model that makes the sound as well — four to fifteen seconds of footage arriving with its own audio.
seo-content-brief
The homework you do before writing an article that is meant to show up on Google.
social-media-carousel
The swipeable stack of slides people post on Instagram and LinkedIn, which is really just seven pictures in a row.
speech-to-text
Three transcription models side by side — a fast one, a very accurate one, and a premium one — and when each is worth using.
storyboard-creation
The drawing of every shot before anyone turns a camera on, so the crew knows what they are filming.
talking-head-production
A production guide for videos of a person delivering your script — choosing the portrait, choosing the model, and getting past the one-short-clip limit.
technical-blog-writing
A blog post written for developers, who will close the tab the moment it starts sounding like marketing.
text-to-speech
Nine different voice engines in one place — the fast cheap one, the premium one, the one built for podcasts — and notes on which suits which job.
tools-ui
When an AI assistant goes off to run something, these are the little status cards that show it happening — and the approval prompt when it needs permission first.
twitter-automation
Posting, liking, retweeting and direct-messaging on X from a script, without ever opening the site.
twitter-thread-creation
Writing for a place where every tweet has to stand on its own and still pull the reader into the next one.
video-ad-specs
Every platform wants a different video ad — a different shape, a different length, sound on or sound off — and this is the rule sheet for all of them, together with the hook-to-call-to-action structure that goes inside the ad.
video-prompting-guide
An AI video generator gives you roughly what you described, so this teaches the describing: which words set the shot, the camera move, the lighting and the overall look, and what each of the main video models responds to best.
web-search
Ask a question and an answer comes back with the sources behind it; point at a web page instead and the page comes back as clean readable text. The searching and reading are done by Tavily or Exa, called from the inference.sh command line.
widgets-ui
Describe a card, a form or a row of buttons as plain JSON, and a ready-made React component draws it on screen as a working piece of interface. It is how an assistant's answer can come back as something you click rather than something you read.
youtube-thumbnail-design
Most people first see a YouTube thumbnail at about the size of a postage stamp. These are the rules for one that still works that small — which colours to pair, what expression the face should wear, how few words of text, and which corners YouTube covers up.
No skill in this package matches that.
HOW TO GET IT
WHOLE PACKAGEInstall the whole package (85 skills)
npx skills add inference-sh/skillsSINGLE SKILLOr install a single skill
npx skills add inference-sh/skills --skill <name> --full-depthPick the skill name from the Skills tab — each entry there installs independently.