Six parts laid down; the finale has only two questions left: what does one episode actually cost, and how does one person reach 20 episodes a month? The first assembles the cost fragments from five parts into a complete ledger; the second moves the entire pipeline into an agent workflow — and you’ll see that Part 3’s JSON storyboard table was designed for exactly this from day one. Finally, the whole series compresses into a one-page checklist for the wall above your desk.
The cost of one episode: the complete ledger
At Part 2’s ¥500/month toolchain tier, the direct cost of one 3-minute episode:
Image-to-video: 180 seconds × ¥0.43/s ≈ ¥77, times Part 4’s reroll coefficient of 1.16 ≈ ¥90; TTS around ¥10–30; stills amortization and misc — total direct cost ≈ ¥100–150 per episode, plus about 8 hours of your time. Two reference frames make the number meaningful: the market pays ¥300–1,500 per minute for the same episode as custom work (3 minutes = ¥900–4,500 gross); and the ¥500 monthly fee covers 15–20 episodes of direct cost — the toolchain pays for itself the moment any client order or any guarantee lands.
The hidden line of the ledger is your time: 8 hours × your hourly rate is the biggest cost item. That is exactly why agent-ification exists.
Two ROI calculations, one combination
Calculation one: custom-production cash flow. 20 episodes/month × 3 minutes × ¥300/minute (the floor of custom quotes) ≈ ¥18,000 monthly gross, minus roughly ¥2,000–3,000 of direct cost and tool fees — positive and predictable cash flow, at the price of linear income and zero asset accumulation.
Calculation two: revenue-share assets. By the July rules: a 60-minute title ≈ 20 finished episodes, with a ¥40k guarantee at heat 5000+ and ¥200k at 7000+, plus continuing share beyond the guarantee — a single title’s ceiling is an order of magnitude higher, but Part 1’s 0.6% hit rate means enormous variance.
The combination is the shape of most teams that survive: custom cash flow pays the toolchain and the rent, while a fixed slice of monthly capacity goes into your own revenue-share titles — the last line of Part 1’s map lands here: the first two streams feed cash flow, the third builds assets, and cash flow is what lets you afford to bet on assets.
Agent-ification: the pipeline as a workflow
Twenty episodes a month can’t be stacked out of overtime — it comes from removing the human from the repeatable stages. Look back at the pipeline: the script cascade is LLM calls, image and motion generation are API calls, QA is a second LLM — the machine-doable portion of the pipeline happens to be exactly a chain of tool calls. And that is the shape of an agent:
An orchestrator reads the JSON storyboard, calls four tools row by row (script LLM → text-to-image → image-to-video → TTS), a QA sub-agent scores each batch and loops failures back, and the two human gates remain untouched — topic selection and final review are human territory, marked off in Part 3, and agent-ification never trespasses. That’s the same principle as this site’s Foundation Boundaries for Agent Systems: across the three layers of context, tools, and evaluation, the human holds the last ring of evaluation.
And the key insight: agent-ification isn’t new work — it’s cashing an old dividend. When Part 3 insisted on the JSON-structured storyboard, it looked like a team-collaboration format; through the agent lens, it is the API contract between machines. Every field written into that schema is, today, a parameter of a tool call.
The one-person weekly cadence (~20 episodes a month, weekends off): Monday — batch scripts ×4 episodes (cascade + Gate 1); Tuesday–Wednesday — images (Bible + Reference Lock in batches); Thursday — motion and voice (native A/V models + the sync ladder); Friday — edit, review, ship.
The one-page decision checklist
Seven parts, tens of thousands of words, compressed into seven sentences. Print it:
- Read both halves of the window — demand is real, brute force is dead (P1).
- Pay monthly, keep a backup per layer — lists rot, principles hold (P2).
- Topic > script > production — the only correct priority (P3).
- Bible → Reference Lock → LoRA — escalate with episode count (P4).
- Voice-first; measure δ, climb the sync ladder (P5).
- Read levers, not rules; sign exclusivity with leverage (P6).
- Quality is the only input the matrix amplifies (P7).
Closing words, and the last action list
This series opened with “is the money still there,” and after seven parts the answer hasn’t changed: yes — and it concentrates, more every month, in the hands of people who can make content. Tools improve monthly, which is exactly why tools stopped being the moat; rules change quarterly, which is exactly why reading them became one. Your assets are three: taste in topics, the pipeline SOP, and a personal template library that keeps compounding (Bible, prompts, storyboard tables, voice tables).
The last action list has one item, because it is the graduation project: within 30 days, use these seven parts to produce your first 60-minute title and walk one complete revenue-share cycle — from entity registration to the guarantee application. Finish it, and every part of this series becomes muscle memory. When you do, come back and write your own numbers into your own retrospective — that’s the only content in this track that never expires.
Happy shipping.