NotebookLM Data Tables: I Fed It a Messy Spreadsheet
NotebookLM Data Tables promises to turn messy data into clean tables. I tried it with a real messy spreadsheet. The catch is at the input, not the AI.
Create
July 16, 2026 · 4 min read
I sat down today to make my first Midjourney image and got as far as a search bar that doesn't accept a single keystroke. Step one, before you type a single prompt, is a $10 subscription. Midjourney is a text-to-image AI tool, similar to DALL-E or Stable Diffusion, except you can't try it without paying first. This piece is for anyone who typed "what is Midjourney" into Google and wants the real mechanics, not a marketing page. By the end you'll know what your first $10 buys, what happens the moment you type a prompt, and one thing the sample gallery won't tell you. I didn't expect the hardest part of trying Midjourney to be the part before the trying.
No, and that surprised me more than it should have. Midjourney dropped its free tier in March 2023, after a wave of abuse, and it never came back. When I opened midjourney.com without logging in, the prompt bar at the top of the page was visible but frozen: the placeholder text reads "Log in to start creating..." and typing into it does nothing.
Midjourney's own Getting Started Guide confirms it in writing. Step one is literally titled "Subscribing to Midjourney." Step two, "Making Your First Image," only exists after step one is done. There's a narrow refund path if you use fewer than 20 GPU minutes, but that's not enough generations to actually judge whether the tool suits you.
Money first, then patience, in that order. Midjourney bills in GPU time, not per image, split across four tiers: Basic at $10 a month for 3.3 fast GPU hours (about 200 minutes), Standard at $30 for 15 fast hours plus unlimited slower "relax mode" generations, Pro at $60 with a private "stealth mode" added, and Mega at $120 for the most fast hours.
Paying annually knocks the monthly rate down on every tier. These figures come from Midjourney's pricing structure as cross-checked in July 2026, since the pricing page itself blocks automated access and the numbers below are matched across two independent sources instead. Set aside an afternoon, not a whole day, to learn the interface once you've picked a tier.
Once you've subscribed, the mechanics are almost boring in their simplicity. You land on the Create page, type a plain-English description into the Imagine bar, and press enter. Midjourney generates four images at once from that single prompt, and its own onboarding video walks through exactly this sequence: type, send, watch four thumbnails fill in.
From there you pick a favorite and either upscale it, generate variations, zoom out, pan, or hand it to the built-in editor for local changes. None of that needs Discord anymore. The bot commands that used to be mandatory now live inside a full web app, though the Discord server is still where most troubleshooting conversations happen.
Because the images in Midjourney's public Explore feed carry more baggage than a plain prompt does. You can browse Explore without an account, and see for yourself: every image there links to a page showing its exact prompt and parameters. I checked three of the top-of-day images, and every single one used a --profile code, a personalization fingerprint Midjourney builds from an account's own generation history.
Two of the three also stacked three or four --sref codes, style references pulled from other images entirely. Here's the weird part: even Midjourney's own four-image "welcome to midjourney!" demo, shown to brand-new visitors on the Create page, uses two profile codes.
A fresh account doesn't have a trained profile yet.
It can't. So the first plain-text prompt you type is never going to match what the front page shows you, and nobody selling the subscription mentions that part.
Illustration of an anthropomorphic cat doing wakeboard --raw --hd --profile pw1w793
Ten dollars buys 3.3 fast GPU hours a month, billed by generation time rather than by image count, with no relax mode, no stealth mode, and no priority queue. Standard, at $30, removes the ceiling that matters most: unlimited relax-mode generations once your fast hours run out, which is why it's usually the realistic starting tier for regular use rather than a one-off test.
I'd still start on Basic, for one reason: it's the cheapest way to find out whether you even like how Midjourney renders your ideas before committing to a bigger bill.
Skip Midjourney if you want to try before you buy. There's no meaningful free path, and the 20-GPU-minute refund window doesn't leave enough time to form a real opinion. If a zero-cost option matters more to you than Midjourney's specific look, our roundup of AI tools that are actually free is a better first stop.
Skip it too if your business earns more than $1 million a year, since Midjourney's terms require the Pro or Mega tier for commercial use at that size, not the cheaper plans. And skip it if you need photorealistic product shots with exact logos and text: Midjourney leans toward stylized, illustrative output by default, a different job than matching an existing photo pixel for pixel. For store-ready product images specifically, our AI product-photo walkthrough covers tools built for that job.
--profile data the showcase images use.Yes, but only on Basic, and only because I want to see for myself whether a brand-new account without a single --profile code produces anything close to what the Explore feed promises. That's an honest question I still can't answer from outside the paywall. It's the one thing about Midjourney that no article, including this one, can settle for you without you paying the ten dollars yourself.
NotebookLM Data Tables promises to turn messy data into clean tables. I tried it with a real messy spreadsheet. The catch is at the input, not the AI.
I gave five AI writing tools one brief: write my bio. Each broke differently. Where ChatGPT, Jasper, Copy.ai, Grammarly and Sudowrite fall short.
A Sora 2 test review that skips the hype: a verified map of where its physics promise breaks (hands, flips, text) and why the app closed in 2026.