There is a whole genre of YouTube video right now claiming you can point Claude Code at an image model and clone businesses earning $100k a month. We watched one of the popular ones so you don't have to. The workflows in it are real and worth understanding. The income claims deserve a lot more skepticism than they get, and we'll get to why.
So here are the three claude side hustle ideas from that video, what the workflow actually looks like, and the math nobody puts in the thumbnail.
One note before the workflows: every earnings number below comes from the video's own claims or from third-party sales estimators. Estimators guess from bestseller ranks. They are frequently off by a wide margin, and the person presenting the numbers was sponsored by one of the tools in the stack. Treat all of it as marketing until you've tested a niche yourself.
1. Kids' educational books on Amazon KDP
The pitch: black-and-white "questions kids ask" encyclopedia books sell well on Amazon. One example book's bestseller rank suggests somewhere between 4,800 and 7,900 copies a month per a sales estimator, at roughly $13 royalty per copy through Amazon KDP, Amazon's print-on-demand self-publishing program. If the estimate is even half right, that's a serious number. It is also one book out of thousands that sell nothing, which the video does not dwell on.
The workflow:
1. Use Claude (chat) to brainstorm 100 kid-friendly questions for the book. 2. Manually generate a few style-reference images with an image model until one looks right. 3. Open Claude Code, connect it to an image-generation tool via that tool's agent integration or MCP, and hand it the question list plus the reference image. 4. Claude Code generates a test image, compares it against the reference, then batch-produces all 100 pages in a consistent style. 5. Assemble in Canva at 8.5x11 with print margins, upscale the images, export a print-ready PDF, upload to KDP.
The part the video skips: KDP's content policy requires you to disclose AI-generated content when you publish, and Amazon has been actively culling low-effort AI books. Check the current policy before you build anything, because this is the kind of rule that changes quarterly. Also, "the images generate themselves" does not mean the book is good. The bestselling examples work because the questions are well-chosen and the layout is clean, which is editorial work AI only partly does for you.
2. Clip art packs on Etsy
The pitch: stores selling AI-generated clip art packs (20 to 60 themed images, sold as digital downloads) are racking up sales fast. The video cites a store under a year old with about 14,900 sales. Digital downloads have near-zero marginal cost, so the margin story is real if the sales are.
The workflow is nearly identical to the book one: generate a style reference manually, brainstorm 25 image ideas per pack with Claude, then let Claude Code batch-generate the pack against the reference. The moat, such as it is, comes from shipping many packs across many niches so you catch Etsy search traffic.
Caveats worth knowing: Etsy requires sellers to disclose AI involvement in creation, buyers increasingly filter for it, and the niche is exactly as easy to enter for the next thousand people as it was for you. Our read is this market is in a land-grab phase that will compress hard. Early stores with review momentum keep winning; new entrants mostly won't replicate that 14,900-sale chart.
3. "This or that" videos for YouTube and TikTok
The pitch: long-form videos of AI-generated images on themes like "build your dream house," monetized through YouTube ads or TikTok's Creator Rewards. The video points at one example with 11 million views and estimates $3 to $7 RPM on YouTube, $0.50 to $2 on TikTok.
The workflow: brainstorm image ideas with Claude, batch-generate square photorealistic images through Claude Code, optionally animate a few into short video clips with an image-to-video model, then assemble everything in Canva using its ready-made "this or that" templates. The creator claims under 10 minutes of editing per video once the assets exist, which we actually believe, since the template does most of the work.
The honest problem: monetization gates. YouTube wants 1,000 subscribers and 4,000 watch hours before you earn anything, and both platforms have policies against mass-produced repetitive content that they enforce unevenly but increasingly. The example channel with 11M views is the lottery winner you're shown. The channels that posted the same format and got 400 views are not in the video.
The math the tutorials skip
Run the pessimistic case, not the highlight reel. If your first 10 KDP books average 3 sales a month each, that's maybe $120 a month for several weekends of work. If your first Etsy packs sell twice a week at $4, similar story. The realistic early hourly rate on all three of these is bad, probably worse than minimum wage, and it only gets good if you hit product-market fit in a niche and scale production. Some people do. Most quietly stop after month two. From what we can tell watching this space, the consistent winners treat it like publishing (many bets, fast iteration, real attention to what sells) rather than like a machine you switch on.
The actual skill being sold in these videos isn't prompting. It's niche selection and distribution. The AI batch-generation part is now nearly free and available to everyone, which means it can't be the edge.
Where to learn this properly
You can piece the whole stack together from YouTube, and plenty of people do. If you'd rather have it sequenced, structured courses cover this ground faster. Full disclosure: Mindwand is our course, so weigh that recommendation accordingly. Coursera and Skool communities cover similar material. Format matters more than brand. Pick one. Finish it.
For the broader picture of what's working across AI side hustles, and how these three stack up against service-based options, see our ranked breakdown of 5 proven AI side hustles.