AI Side Hustles Half-true — works only if you do the unspoken work

By the Income Reality Check Editorial Team · Updated 2026-06-14 · 7 min read

Claude Fable 5 built a “$10K website” in minutes — the part the demo skips is selling it

Verdict: Half-true — works only if you do the unspoken work. The website really does get built for a couple of dollars; the $10,000 is a sales number, not a build number.

Zubair Trabzada’s video “Claude Fable 5 Built a $10K Website in Minutes” walks through a genuinely impressive trick: feed Claude Fable 5 a screenshot of an award-winning site, pair it with the Higgsfield MCP for AI images and video, and watch a polished, animated landing page appear on localhost. His framing is that design agencies charge thousands for these sites, and now you can generate one “for the cost of one or two dollars.” The build part is real. The “$10K” part is doing a lot of quiet lifting, and most of it happens off-camera.

What the video actually claims

The pitch has two layers. The first is technical: using Claude Fable 5 (Anthropic’s newest model, released June 9, 2026, with a 1-million-token context window) inside Claude Code, plus a connected Higgsfield MCP server, you can clone or invent a “beautiful, award-winning” website from a single prompt and a screenshot. Higgsfield handles the visuals — generating images with Nano Banana Pro and short cinematic clips with models like Seedance 2.0 or Kling, then slicing those clips into the scroll-triggered frames that give the site its 3D feel. The creator says each full iteration of a site costs him roughly $1 to $2 in Higgsfield generation credits.

The second layer is the money. He repeatedly contrasts that $1–$2 cost with “websites that agencies charge thousands of dollars” for — and the video’s title attaches a specific number, $10,000, to the output. Midway through, he mentions a paid community where he teaches people “how to build and sell Claude Code projects to actual businesses,” including “how to reach out to clients, how to find your clients, how to sell.” He’s run an agency for two years, he says, and selling is “the important part.”

That last admission is the whole story. The free video shows you the building. The $10,000 lives in the selling, and the selling is what’s behind the paywall.

What the method actually requires

Start with the real costs, because “one or two dollars” is the per-iteration figure, not the monthly one. Claude Code isn’t free: it ships with a paid Claude plan, which runs $20/month for Pro or $100–$200/month for the Max tiers, per Anthropic’s published pricing. Higgsfield is a separate subscription on top. Its plans start around $15/month for 200 credits (Starter), $49/month for 1,000 credits (Plus), and roughly $99–$129/month for 3,000 credits (Ultra), according to Higgsfield’s own pricing page. Video generations are the expensive part — premium clips can cost 40–70 credits each — so a handful of “iterate the whole site” passes burns through a Starter plan fast. Call it $50–$150/month in tooling before you’ve earned a dollar.

Now the bigger gap. What the demo produces is a localhost mockup — a gorgeous hero animation and a few scroll sections. What a paying client buys is a finished, deployed product: working forms, a content management system they can edit, mobile and cross-browser testing, accessibility, page-speed optimization, analytics, hosting, a domain, and revisions when they decide the blue is wrong. The award-winning sites he clones from Awwwards usually win for that engineering and custom interaction work — not for the look of a single spinning car. A screenshot clone captures the surface and skips the part that takes the hours.

There’s also a legal seam the video glosses over. Cloning a real, named site (he points at phantom.land) and reselling a near-identical copy isn’t a clean foundation for a client business — it’s someone else’s design and, often, their trademarked brand. Inspiration is fine; pixel-for-pixel reproduction sold for money is not.

Then there’s the work that produces the income: finding someone willing to pay. Web designers who do cold outreach typically see first conversations in two to four weeks, and those who rely on content and personal branding wait three to six months before momentum builds — and that’s after assembling a small portfolio first. None of that is in the video, because it can’t be automated by an MCP.

Is the $10,000 number real at all?

Yes and no, and the distinction matters. Custom agency sites genuinely do sell for five figures — so “$10K website” isn’t invented. But $10,000 is what a client pays for a complete engagement, not what lands in your account from running a prompt. Out of that figure come your tooling costs, the hours of revisions, deployment and testing, and the unpaid time you spent landing the deal. The build being cheap doesn’t make the project cheap. It moves the cost from production to sales — and sales is the harder, slower line item.

This is also the genre the U.S. Federal Trade Commission has been watching closely. In September 2024, the agency announced “Operation AI Comply,” five cases targeting deceptive AI claims; three were business-opportunity schemes promising fast income “powered by artificial intelligence.” One, FBA Machine, cost consumers more than $15.9 million on earnings claims that, the FTC said, rarely materialized. To be clear, nothing suggests this creator is running a scam — he’s selling a course and a real workflow, not a fake one. But the FTC’s point applies broadly: an income claim has to be backed by what typical people actually earn, not by a best-case demo. (For U.S. readers, that’s the standard sellers are held to.)

Who actually wins this game

The people who turn this workflow into real money already have the second skill. They can sell. They have a niche, a short list of businesses with outdated websites, a portfolio, and the patience to send fifty cold emails for three replies. For them, Claude Fable 5 and Higgsfield are a genuine speed boost — the production bottleneck that used to eat a weekend now takes an afternoon, which means more proposals out the door and better margins on each job.

Who doesn’t win? The complete beginner the video keeps addressing — the person with no clients, no niche, and no idea how to price or scope a project. For that person, a beautiful localhost demo is not a business. It’s a screenshot. The CNBC profile of a Fiverr web designer who built a real side income is instructive: she cleared over $1,000 a month and landed her first client within two days — but on a marketplace, against reviews and competition, after putting her services in front of buyers. The tool didn’t find the buyer. She did.

What you’d realistically earn

If you already freelance, this stack can plausibly add a few hundred to a few thousand dollars a month by letting you take on more or better-looking projects. If you’re starting cold, a more honest first-year arc looks like this: weeks of unpaid practice and outreach, your first small job somewhere in the $300–$1,500 range, and steady four-figure months only after you’ve built a referral base and a repeatable sales habit. The skill itself is worth real money — the Bureau of Labor Statistics puts the median wage for web developers at about $90,930 a year and for web and digital interface designers at about $98,090 (May 2024). But that pay reflects the whole job, not the hero animation. A “$10,000/month, hands-off” reading of this video is the part the calculator doesn’t survive.

Want a fuller picture of how these AI-business pitches tend to play out? See our breakdowns of launching a $10K AI business solo and starting a one-person business with Claude in 30 days.

Who this is (and isn’t) for

This makes sense if you’ve got existing design or development chops, a few hours a week for outreach, $50–$150/month for tooling, and the stomach to hear “no” a lot before you hear “yes.” It’s a production accelerator for someone already in the game. It does not make sense as a passive-income plan for someone with no audience, no portfolio, and an expectation that the prompt does the selling. The cloning angle in particular — copying named sites and reselling them — is a liability, not a strategy.

What to remember

Claude Fable 5 and Higgsfield genuinely compress website production from days to minutes for a couple of dollars per pass. That’s the half that’s true. The other half — the $10,000 — isn’t a build outcome; it’s a sales outcome, and the video quietly routes that lesson to a paid community. Treat the demo as proof the tools work, not proof the money is automatic.