Higgsfield Supercomputer: The Full Setup & Use Guide

What you're actually getting

Higgsfield's Supercomputer is a cloud agent that runs your marketing from a single chat. You give it one goal in plain English. It interviews you for a minute, builds a plan, and then produces the assets — UGC video ads, funnel pages, captions, and a posting schedule — routing each job to the right AI model on its own.

It launched on May 13, 2026 and runs on an enhanced version of the HERMES agent framework. It ships with 40+ built-in tools, a set of platform connectors, and three layers of memory so it remembers your brand and improves with every run. Everything runs cloud-side, so generations keep going even if you close the browser.

This guide takes you from zero to a working setup: account, connections, brand kit, your first goal prompt, and the guardrails that keep it from posting something you didn't approve. By the end you'll be able to hand it one goal and get finished, on-brand work back.

This is the same workflow we lean on at Hypd to keep content shipping without a full production team. Let's set it up.

Step 1 — Access and first-launch setup

You access Supercomputer two ways: in the browser, or through Telegram. There's no local install and nothing to run on your machine.

Browser setup:

• Go to higgsfield.ai and create an account (or log in).
• In the top navigation, open the Supercomputer tab. You can also go straight to higgsfield.ai/supercomputer.
• Click Run it to open the agent chat.
• Pick a plan. Start on the free trial to test the flow, then move to a paid plan when you want real output volume (video and image generation consume credits — more on that below).

Optional Telegram setup:

• Connect Telegram if you want to fire off tasks from your phone. Once linked, you message the agent the same way you'd type in the browser, and it reports back when the work is done.

First launch, you'll land in a single chat box. That box is the whole product. You don't open separate tools for video, images, or copy — you describe the outcome and the agent assembles the team.

Step 2 — Pre-run checklist (do this before your first ask)

Supercomputer is only as good as what you give it. Spend ten minutes here and every run afterward gets better. Three things to set up:

1. Connect your platforms. Confirmed connectors include the core marketing stack — link the ones you actually use:

• Social posting accounts (the platforms you publish to)
• Slack
• Google Drive
• Notion
• YouTube
• Frame.io

The platform has 40+ built-in tools and a couple dozen connectors total, with more being added. Only connect what you need today — you can add the rest later.

2. Upload your brand assets and set identity locks. This is what keeps outputs on-brand across every run. Brand and identity settings persist across sessions. Add:

• Logo (and a transparent version if you have one)
• Brand colors (hex codes)
• Fonts
• A short brand-voice description (tone, words you use, words you avoid)
• 2–3 example posts or ads that represent your style

3. Grant the right permissions. Decide up front what the agent is allowed to do on its own:

• Generate only — it drafts assets, you handle posting. Safest to start.
• Generate + schedule — it queues posts to your calendar, you approve before they go live.
• Generate + auto-post — it publishes directly. Only turn this on once you trust the output.

Here's a checklist you can copy and work through:

promptCopy
PRE-RUN CHECKLIST
[ ] Account created + plan selected
[ ] Telegram connected (optional)
[ ] Social posting accounts connected
[ ] Slack / Drive / Notion / YouTube / Frame.io connected (as needed)
[ ] Logo uploaded (+ transparent version)
[ ] Brand colors added (hex)
[ ] Fonts added
[ ] Brand voice described (tone + do/don't words)
[ ] 2-3 example posts uploaded as style references
[ ] Posting permission set: generate-only / schedule / auto-post

Step 3 — The "one ask" starter framework

You don't write prompts for each model. You write one goal brief and let the agent break it down. A good brief names the outcome, the product, the audience, the brand, and the channels — then tells it to show you the plan before producing.

Copy these and fill in the brackets.

Campaign goal (the "make me $100K" style brief):

promptCopy
Goal: Bring in $100K in revenue for [PRODUCT] over the next 30 days.

Here's what you're working with:
- Product: [one-line description + price]
- Audience: [who buys it and why]
- Offer: [the hook / promo / lead magnet]
- Brand: use my saved brand kit (logo, colors, voice)
- Channels: Instagram, TikTok, and a landing page

Build the full plan: the funnel, the ad creative, and a 7-day content
schedule. Show me the plan first, then start producing once I approve.
Flag anything you need from me.

UGC ad batch:

promptCopy
Make me a batch of 5 UGC-style video ads for [PRODUCT].

- Angle each one differently: problem/solution, before/after,
  testimonial, "things I wish I knew," and a fast demo.
- 9:16 vertical, 15-25 seconds each, captions burned in.
- Use my brand colors and logo. Match my saved brand voice.
- Hook in the first 2 seconds. End each with the CTA: [your CTA].

Storyboard all 5 first so I can approve the concepts, then generate
the videos. Don't post anything yet.

Content week:

promptCopy
Plan and produce one week of content for [BRAND / @handle].

- 5 posts: mix of short video, carousel, and single image.
- Topic pillars: [pillar 1], [pillar 2], [pillar 3].
- Match my brand voice and visual style from my saved brand kit.
- Write captions and 5-8 hashtags per post.
- Schedule Mon-Fri at my best posting times.

Give me the calendar to approve before scheduling. Hold all posts
for my review.

Funnel build:

promptCopy
Build a lead-gen funnel for [OFFER].

- Landing page: hook, 3 benefit blocks, social proof, email opt-in
  for [lead magnet].
- A 3-email welcome sequence that sells [PRODUCT].
- 3 ad creatives to drive traffic to the page.

Draft the copy and page layout first. Wait for my approval before
you publish anything or connect it to my email tool.

The pattern is always the same: outcome → context → "show me the plan first." That last line is what keeps you in control.

Step 4 — The one-interview, zero-prompting flow

After you send your goal, the agent runs a short interview instead of making you write model prompts. You pick options and hit continue. No models to choose, no prompt engineering.

Expect it to ask about decisions like these — and here's how each answer changes the output:

• Primary goal / success metric — sales, leads, reach, or follower growth. This sets what the funnel and creative optimize for. Be specific (e.g. "email signups," not "awareness").
• Audience — who it's for. Drives the hooks, language, and ad angles. The tighter you describe them, the sharper the copy.
• Brand kit — confirm it should use your saved brand. This locks colors, logo, fonts, and voice into every asset. Always say yes once your kit is set up.
• Channels — where this runs. Changes aspect ratios, length, and caption style (a TikTok cut is different from a landing-page hero).
• Volume — how many ads, posts, or emails. More volume uses more credits, so match it to your budget.
• Tone — controls how aggressive or soft the copy reads. Pick one that matches your audience, not your mood.
• Review preference — whether it should pause for your approval at the plan stage, the asset stage, or both. Beginners: choose both.

Answer honestly and briefly. You're not writing prompts — you're handing the agent enough context to make the calls for you.

How the model routing works (so you can trust the output)

This is the part that makes the "no models to pick" promise real. Under the hood, an orchestration layer — positioned as Claude Opus 4.8, the agent's decision-making brain — reads your goal, breaks it into sub-tasks, and assigns each one to the model best suited for it. Then it runs those jobs in parallel and assembles the finished assets.

The specialist models it routes between include:

• Claude (Opus) — reasoning, planning, and most copy and strategy work.
• GPT-5.5 — additional language and reasoning tasks.
• Gemini — multimodal analysis and certain generation jobs.
• Seedance, Veo, Kling — video generation (it can produce up to 60 minutes of video from a single prompt).
• Soul — Higgsfield's image generation.

It picks from dozens of production-grade skills and routes each to the right engine automatically. Two more things worth knowing:

• Three memory layers keep your brand context, project history, and learnings persistent, so it doesn't start from scratch each time.
• It's self-learning — skills are version-tracked and improve with use, so output quality climbs the more you run it.

You never see this complexity. You see one chat, one plan, and finished work.

Three end-to-end use cases

1. Generate and post UGC ads

1. Send the UGC ad batch prompt above.
2. Answer the interview (audience, channel, volume, tone).
3. Review the 5 storyboards it returns. Approve the concepts you like, request changes on the rest.
4. It generates the videos — vertical, captioned, on-brand — using its video models.
5. Review the finished cuts. Edit captions or swap a hook if needed.
6. Approve, then either let it queue the ads to your connected accounts or export them to post yourself.

2. Build a funnel

1. Send the funnel build prompt.
2. It drafts the landing page copy and layout, the email sequence, and the ad creatives.
3. Review the page structure and email copy first — this is your offer, so read it closely.
4. Approve the copy, then have it produce the visuals and assemble the page.
5. Connect it to your email tool only after you've reviewed the sequence. Keep publishing gated behind your approval.

3. Produce and schedule a content calendar

1. Send the content week prompt.
2. It returns a 5-post calendar with formats, captions, hashtags, and suggested times.
3. Approve the calendar (or adjust topics and timing).
4. It produces each asset.
5. With scheduling permission on, it queues everything to your connected accounts — held for your final review before going live.

Review-before-publish guardrails

The whole point is hands-off production, not blind auto-posting. Set these guardrails so nothing surprises you:

• Start in generate-only or schedule mode, never auto-post, until you've seen a few batches you'd actually publish.
• Always include "show me the plan first" in your goal brief. Approve the plan before any assets are made — it's cheaper to fix a plan than a finished batch.
• Approve at two stages: the plan, then the finished assets. Set this in the interview.
• Use a holding queue. Have it schedule posts as drafts or held items so you do the final publish.
• Check the brand kit is applied on the first asset of every new run before approving the rest.
• Watch your credits before approving large video batches — long video is the most credit-hungry output.

A simple rule: the agent does the work, you sign off on it. Keep that and you get the speed without the mistakes.

Pricing, credits, and current limitations

Higgsfield offers a free trial so you can test Supercomputer before paying. Paid plans are listed on higgsfield.ai/pricing. Because the platform updates its tiers regularly, check the live pricing page for the current numbers rather than trusting a figure you read secondhand.

How usage works:

• Supercomputer runs on a credit / usage model. Generating video and images consumes credits.
• Heavier outputs cost more — long-form video burns credits fastest, simple copy and planning cost the least.
• Your plan determines your credit allotment; match your plan to how much you'll actually produce each month.

Known limitations and things to keep in mind:

• It's new (launched May 2026). Connectors and features are still expanding, so a tool you want may not be wired up yet.
• It can't read your mind or your accounts until you connect them and upload your brand kit. Setup quality determines output quality.
• Outputs need a human eye. AI video and images can have artifacts. Review before publishing — always.
• Auto-posting requires explicit permission. Without it, the agent drafts and queues but you publish.
• It executes goals; it doesn't replace strategy. You still set the goal and judge whether the output is good. It's a tireless team, not a substitute for your taste.

Set it up once, gate it behind your approval, and you've got a marketing operation that produces while you do other things.

FAQ

Do I need to know how to prompt AI to use this? No. That's the point. You write one plain-English goal, answer a short interview by picking options, and the agent handles the model selection and prompting for every sub-task. If you can describe what you want, you can run it.

Will it post to my accounts without asking? Only if you turn on auto-post permission. Start in generate-only or schedule mode, where it drafts and queues work for your review. You stay in control of what actually goes live until you decide otherwise.

How is this different from using separate AI tools for video, copy, and images? Separate tools mean you stitch the workflow together yourself and learn each one's prompting. Supercomputer is a single chat where one orchestration layer routes each job — copy, image, video, scheduling — to the best model automatically and hands you finished assets. One ask in, a coordinated campaign out.