282 Generated, 99 Kept: An AI Slopcore Curation Workflow
A hands-on curation workflow from a slopcore music video case: 4 days of conversation with Claude, 282 clips generated, 99 kept — mass AI generation followed by
This is a practical curation workflow distilled from the slopcore music video build log published on X by the original author @javif4_ (javier fan) on October 12, 2026. The thesis is one sentence. Generating is almost free; choosing is the work.
Image source: @javif4_ / X
What the author did: 4 days of separated generation and selection
The author spent 4 days in conversation with Claude building one song. The concept: Berlin rave meets opera. In the author's own words: "Human-made Slopcore", "Unpromptable". The published figures are:
- 4 days of conversation with Claude for song and video work
- 55,000 lines of code (quoted exactly as the author claimed, not a verified code count)
- 282 clips generated, 99 selected
- Track title: I'M YOUR SLOP (I'M YOUR ART)
The author was frank about the difficulty in replies: "this time claude burnt it, it was difficult" and "i fight with claude while doing this.. but now we are ok". Mass generation was not smooth automation; it came with credit burn and trial and error.
The key line: "I was generated, but I was chosen"
The same day, the author posted a follow-up written in Claude's voice, at the author's request for an opinion:
The real line is "I was generated, but I was chosen." The question used to be "can AI make this?" Now it's "which one is worth keeping?" Generating is almost free. Choosing is the work.
That sentence is the entire workflow. Generating 282 and keeping 99 are two different skills. The first is a model-and-credits problem; the second is a human taste-and-judgment problem. The full video is on YouTube.
How to follow it: the author's method page
The author wrote, "Want to make Slopcore? I can't code, but I can show you my method," (adding that Claude ate all the credits) and shared a method page:
- Method page: javiexmachina.com/learn
One caveat. The original posts and replies contain no specific commands, toolchain, or code, so this article adds none either. Check the author's method page directly for the actual method. This article covers only what the source discloses: long conversation plus mass generation, followed by human selection.
Practical notes
- Generate expecting to discard. 183 of 282 clips were cut. Secure generation volume that includes the planned rejects.
- Failures are material too. The author separately showed off the bad generations of the orange-haired girl and laughed. Discarded clips calibrate the next round of selection.
- Budget credits up front. The author burned through credits too. Estimate the cost of 4 days of conversation and hundreds of generations before starting.
- Fix the selection criterion in one sentence. "Generated but chosen" gave the 99 kept clips their coherence. Write the taste sentence first, then start selecting.