ROXy (@VK_ROXy)

17 checkboxes. 2 vehicles. 1 sketch. The order Claude reads them in decides who's at fault. What 24 minutes of Anthropic actually buys you: 0:00 — Hanna and Christian open, neither speaks Swedish 0:52 — The Swedish insurance claim scenario, unpacked 2:01 — Naked prompt fails: Claude calls it a skiing accident on Schabongatan 3:51 — Anatomy of a prompt that survives production 5:31 — Task and tone context do the heavy lifting 8:43 — Static background data belongs in the system prompt 9:36 — XML tags, the structure Claude was fine-tuned on 12:50 — Few-shot examples for the gray-area cases 15:50 — The final task reminder that kills hallucinations 20:00 — Output formatting and hard constraints 22:11 — Prefilled responses put words in Claude's mouth 23:17 — Extended thinking as a prompt-engineering crutch The baseline at 2:01 doesn't just underperform. Same Claude 3.5 Sonnet at temperature 0, a one-line prompt, and the model decides it's a skiing accident on a Swedish street called Schabongatan. Ship that into a claims pipeline and every misrouted file costs an adjuster's hour to unwind. The turn comes at 15:50. Hanna forces read-order: form first, then sketch. Reading the 17 checkboxes first gives Claude the vocabulary to parse a stranger's scribble. That single instruction moves a hallucinating model to "vehicle B at fault," box 1 for vehicle A, box 12 for vehicle B. Sequence killed the hallucination. Not a bigger model. Not a longer prompt. Just the order. At 22:11 Christian prefills the response with an opening XML tag. No preamble. No apology. Just the structured verdict your SQL database can swallow. That's the whole gap between demo and production. Skip the 24 minutes and you keep paying a claims adjuster to clean up after a chatbot that guessed skiing.👇

Agencies charge $2,000/mo for AI ad creative. Higgsfield + draft mode discipline = $29. The full 32-minute build, chaptered: 0:00 — Intro: the sister-trap premise and the storm-at-sea kill 4:28 — Stage 1: the script Claude actually writes for you 5:24 — Stage 2: the asset factory inside Cinema Studio 10:36 — The naming trick that saves 400 wasted generations 11:25 — Stage 3: the pirate cabin, shot by shot 14:01 — The talking parrot fix that made the scene alive 15:41 — Directing a cannonball through a sea battle 18:01 — The ship blast that transitions into the desert 18:46 — Full pirate scene playback: wet-deck fall into the ocean transition 19:45 — Desert scene: ostrich riders and dust transitions 30:00 — Full jungle sequence playback 31:30 — The twist ending that reframes the whole film The frame this post opens with (agencies at $2,000/mo for AI ad creative, a $29 stack on the other side) is the pitch. What Adil actually shows is one operator building a 32-minute cinematic short across three settings, solo, in Krea Cinema Studio with Claude writing every prompt. The whole build is on screen, timestamp by timestamp. The turn comes at 10:36. He names every asset, tags them the same in Krea Cinema Studio and Claude, and the scene generator auto-matches inputs by those tags. That's why the pirate captain keeps the same face 20 minutes later. Skip this step and you burn 400 generations wondering why the model keeps drifting. At 17:19 he stops typing and draws. He sketches where the cannonball lands and where the crew stands, then feeds the picture to Claude. One drawing beats ten sentences. Every failed batch you avoid is credits back in your pocket. The whole thing runs on draft-first discipline. Batch cheap in low-res, iterate the prompt in Claude, commit to 4K once. That's how one person out-produces the retainer. He did it alone. In Krea. With Claude.

$10/month Suno Pro. Built on 135,811 hours of YouTube Music nobody paid for. That's the trade. You pay $10. Suno keeps 0% of your royalties. UMG, Sony, and Warner sue Suno in Massachusetts federal court instead. The leaked source code lists the ingest by the hour: → 113,879 hours of YouTube Music → 152,162 hours of YouTube Music tagged → 62,117 hours of Pond5 → 17,615 hours of Genius HQ Residential proxies to dodge YouTube's flags. 2,013,545 clips in one file. Suno's own filing calls it "essentially all music files of reasonable quality" on the open internet. The $50,000 studio stack didn't get replaced. It got laundered through a scraper and resold to you for the price of a lunch.👇