The higgsfield-supercomputer repo is Kevin’s open-source starter kit for Higgsfield AI: one self-contained HTML dashboard plus four Higgsfield Claude Skills already installed in the project. You clone it, add your Higgsfield API key, and the skills run inside Claude Code. No build step, no package manager, no server. It is a fork-and-go starting point, not a hosted product.
higgsfield-supercomputer/
|
+-- dashboard.html open in a browser, no build step
|
+-- .agents/skills/ four skills, pre-installed
| +-- higgsfield-generate
| +-- higgsfield-soul-id
| +-- higgsfield-marketplace-cards
| +-- higgsfield-product-photoshoot
|
+-- skills-lock.json pins the four to higgsfield-ai/skills
|
+--> Claude Code reads the skills
and calls the Higgsfield API
with your keyWhat Is the Higgsfield Supercomputer Repo?
It is a personal command center for Higgsfield AI generations, published as source on GitHub by Kevin.
The repo description is short and accurate: a Higgsfield AI dashboard plus a Claude Skills bundle. Fork it, plug in your API key, run. That is the whole idea. You get a local workspace where Claude Code already knows about the Higgsfield models, and a browser page that shows your generations without any tooling to install.
Two things make it useful rather than another demo folder. First, the dashboard is a single HTML file, so opening it is the entire setup for the visual side. Second, the four skills are committed into the repo at .agents/skills/, so a fresh clone already has them in the right place. You are not following a blog post to copy folders around.
The source lives at github.com/kevinbadi/higgsfield-supercomputer. It is Kevin’s own project, built in public, and you can watch him work through builds like it on the KevBuildsApps YouTube channel.
What’s Inside: dashboard.html, .agents/skills, and skills-lock.json
Three items make up the whole repo, and each one has a single job.
| Path | What it is | What you do with it |
|---|---|---|
dashboard.html |
A self-contained dashboard with no dependencies and no build step | Open it in a browser. Nothing to compile. |
.agents/skills/ |
Four Higgsfield Claude Skills, pre-installed | Claude Code picks them up from the project folder. |
skills-lock.json |
A version manifest pinning the four skills to higgsfield-ai/skills |
Read it to see which upstream version you have. |
The lock file is the part people skip, and it is the part that matters later. Because the skills are pinned to a specific upstream commit or version, your clone behaves the same way next month as it did the day you forked it. When the upstream ships a change you want, you decide when to take it instead of finding out mid-project that a skill gained a new required parameter.
The four skills are:
higgsfield-generate: image and video generation across the Higgsfield model catalog.higgsfield-soul-id: Soul Character training for a face or identity.higgsfield-marketplace-cards: listing card generation.higgsfield-product-photoshoot: product photography.
How to Set Up the Higgsfield Supercomputer Repo (Fork, CLI, API Key)
Setup is three commands and one account, exactly as the README lists them.
Clone the repo first:
git clone https://github.com/kevinbadi/higgsfield-supercomputer.git
cd higgsfield-supercomputerThen install the Higgsfield CLI with the install script the README points to:
curl -fsSL https://raw.githubusercontent.com/higgsfield-ai/cli/main/install.sh | shThird, authenticate. You need a Higgsfield account at higgsfield.ai, and then you sign the CLI in with your API key so the skills can call the models on your behalf. Read the README section on authentication for the exact command; it is the step where a copy paste from the wrong place will waste ten minutes.
After that, open dashboard.html in your browser and start Claude Code from inside the repo folder so it sees .agents/skills/. Nothing else is required. There is no server to run, no environment file to fill in beyond the key, and no database.
One practical note for anyone who works across a few machines: the key lives in your local Higgsfield configuration, not in the repo. Do not commit it. If you fork publicly, fork before you authenticate.
The generate Skill: Image and Video Across the Higgsfield Model Catalog
The generate skill is the one you will use daily, because it reaches every model in the catalog through one conversation.
That means Soul, Nano Banana, Seedance, Kling and Marketing Studio are all available without you remembering which endpoint or parameter set each one wants. You describe what you need, Claude Code picks the model and calls it. A typical prompt inside the repo looks like this:
Generate 3 product shots of the ceramic mug on a walnut desk,
soft window light from the left, 4:5, then make a 5 second
video of the same scene with a slow push in.The skill turns that into the right model calls, and the outputs land where the dashboard can show them. The useful part is not that it can make one image. It is that you can move from a still to a short clip in the same thread without switching tools or re-explaining the subject.
If you want to see how the same kind of agent loop behaves end to end, the launch video is a good place to start.
The Other Three Higgsfield Claude Skills: Soul ID, Marketplace Cards, Product Photoshoot
Each of the remaining three skills is a narrow tool for one repeatable job, and together they cover most of the visual work a small shop actually needs.
Soul ID
Soul Character training fixes a face or an identity so you can generate that same person again in new scenes. If your content depends on a consistent on-camera persona, this is the skill that keeps the face from drifting between generations. You train once, then reference the character.
Marketplace Cards
Listing card generation produces the formatted product image a marketplace listing expects. Instead of prompting a model from scratch for every item, you hand the skill the product and get a card back. For anyone listing dozens of variants, that consistency is worth more than any single hero shot.
Product Photoshoot
Product photography means the staged, lit, styled shot. This skill generates that look without a studio. The three skills stack naturally: generate a base image, style it as a photoshoot, then turn it into a marketplace card with a consistent Soul character if the listing needs a model.
Here is a worked chain that uses three of the four in one session:
1. Train a Soul character from these three photos.
2. Photoshoot the linen shirt on that character, studio white.
3. Make a marketplace card from the best shot, square crop.Can You Use Higgsfield Claude Skills for Social Media and Content Creation?
Yes for the asset side, and that is where this repo stops.
The skills produce images, video, character references and product cards. What they do not do is publish. There is no scheduling, no caption writing for a specific platform, no comment inbox and no analytics in the Higgsfield bundle. That is by design: it is a generation toolkit, not a distribution tool.
So the honest workflow is two steps. Make the assets in the repo with Claude Code, then move them somewhere that can post. Creator OS is that second step, and it is a different product from a different codebase. Creator OS is one place to publish, schedule, reply and read analytics on 11 platforms plus Skool, usable from the web app, the iOS app, a REST API, a CLI and a hosted MCP server. Media you upload once returns a med_ id you can reuse on every platform, which fits this handoff well: generate in the repo, upload once, schedule everywhere.
If you want the publishing half written out, the post on posting to every social platform from Claude covers the agent path, and the Social Agents repo post covers Kevin’s other open-source harness for running accounts. There is also a piece on Claude skills for social media marketing if you are still mapping the skills landscape.
How the Higgsfield Supercomputer Repo Compares to Other Claude Skills Repos
Most skills repos you will find are collections: a folder of prompts you copy into place and wire up yourself. This one is a workspace.
The difference is packaging. The dashboard ships in the same repo as the skills, the skills ship pre-installed rather than as something you merge, and the lock file records the version so the bundle is reproducible. A collection asks you to assemble the parts. This repo hands you the assembled parts and one file to open.
That trade has a cost worth naming. A pinned bundle is opinionated. If you already have your own skills layout, or you want to pull Higgsfield skills selectively into a larger setup, a plain collection gives you more freedom. The supercomputer repo is for the case where you want a Higgsfield workspace today and do not want to design one.
Kevin’s other open-source work follows the same pattern of shipping a working setup rather than a snippet. OPEN VIDEO EDIT is a set of Claude Code skills that turn a raw talking head clip into an animated short, rendered locally with Python and ffmpeg. If you generate video in the Higgsfield repo and then want captions and b-roll cards on it, that is the natural next repo to clone.
FAQ
Do I need a paid Higgsfield account to run the higgsfield-supercomputer repo?
You need a Higgsfield account at higgsfield.ai and an API key, because the skills call Higgsfield models on your behalf and those calls are billed by Higgsfield, not by this repo. The repo itself is free source code. Check Higgsfield’s own pricing for what your account includes, since that is their product and their terms.
What is the difference between the Higgsfield Supercomputer dashboard and Higgsfield.ai itself?
The dashboard is a single local HTML file that comes with this repo, and it has no build step and no dependencies. Higgsfield.ai is the company’s own product and website. The dashboard is a personal view for your own generations inside the repo you forked; it is not a replacement for the Higgsfield app, and the models and account still live on Higgsfield’s side.
Can Claude Skills for content creation run without an API key?
Not the Higgsfield ones. These four skills exist to call Higgsfield models, so the key is what authorizes those calls, and without it the generate, Soul ID, marketplace card and photoshoot skills cannot do their job. The HTML dashboard is just a file and will open without a key, but it has nothing to show until generations exist.
How do I update the Higgsfield Claude Skills when higgsfield-ai/skills ships a new version?
The versions are recorded in skills-lock.json, which pins the four skills to the upstream repo. To move to a newer version, update the pinned reference and pull the matching skill files, then test one generation before you rely on it. Read the README for the exact steps rather than guessing, since the lock format is the repo’s own.
Is the higgsfield-supercomputer repo good for social media marketing workflows?
It is good for the production half: stills, short video, consistent characters and product cards. It has no scheduling, publishing, inbox or analytics, so pair it with a tool that does that. Creator OS covers those four things on 11 platforms plus Skool, and its API, MCP server and CLI are included in every plan.
Get started
Start by forking the repo at github.com/kevinbadi/higgsfield-supercomputer and following the three setup steps above. Then open dashboard.html and run one generation through Claude Code to confirm the key works.
When you have assets you want to publish, create a Creator OS account at creatoros.ca/sign-up. Plans start at $14.99/month for 11 platforms plus Skool, with the API, the MCP server and the CLI included in every plan. New subscribers start at $0 with a free trial and can cancel anytime.
For the publishing side in detail, the docs are at creatoros.ca/docs. And if you would rather watch a build than read one, the walkthroughs are on the KevBuildsApps YouTube channel.
Get started for $0
Free trial for new subscribers. Cancel anytime.