I Ran the Viral AI‑Novel Workflow on a CLI: 800 Words for Half a Cent

Description and use

A step‑by‑step CLI workflow for creating a short novel: first generate a 100‑word skeleton, then a 798‑character opening, followed by human editing. Cost is about ¥0.022 (≈ $0.003).

Problem

Generate a short novel skeleton and opening quickly and cheaply using a CLI, while avoiding pitfalls in message construction and model timeouts.

Outcome

The pipeline produced a 100‑word skeleton and a 798‑character opening in under 30 seconds for roughly ¥0.022 ($0.003). The author noted that delegating the rewrite step timed out, confirming the need for human editing.

Workflow

• 1. Generate skeleton: bl text chat --model qwen3.8-flash --message "<premise> output a skeleton under 100 words, structure only, no prose"
• 2. Generate opening: create a JSON messages file (avoid shell heredoc pitfalls), then run bl text chat --model qwen3.8-flash --system "web-serial author, six years: colloquial, fast, core conflict inside 300 chars, one reversal per 800 words" --messages-file msgs.json
• 3. Human editing of key spots (opening hook, reversal logic, emotional landing).

Evidence

Skeletons are diffable and reviewable; prose is not. That is the first reason this pipeline survives contact with a production process.

Source

The third step of the original workflow is human editing, and my attempt to delegate it timed out twice — which turned out to be the point.

Source