Jun Sasaki / Log

2026-03-19

Road to 2112 Handover Memo v4 (2026-03-19)

Since Claude's brainstorming chats become sluggish, I replace them periodically. This is a handover memo for dumping project text at those times.
By keeping it contained within the project, I have been able to maintain the context almost entirely, with almost no negative side effects.


Biggest progress since last time

  • v6 format (58 minutes/4 episodes/auto-dubbing in 20 languages) has been fully established

  • Production flow, costs, and posting operations are all finalized

  • The channel has started riding the algorithm wave (impressions are surging)

  • Monetization: Initial forecast of 6-8 months → potential to move up to 2-3 months

v6 format finalized items

  • Duration: 58 minutes (safety margin for auto-dubbing within 60 minutes)

  • Number of episodes: 4 episodes/video

  • Positioning: Cozy Sci-Fi Relax Stories (foreground reading BGM for work)

  • Ad slots: Pre-roll + 2 mid-rolls + end = 4 slots/video (2.7x improvement)

  • Channel name: Road to 2112: Cozy Sci-Fi Relax Stories

→ Changed from 2 hours to 1 hour to trigger auto-dubbing. Impressions increased significantly after receiving a comment from Ukraine. It seems YouTube's testing period has ended. While I think the monetization forecast is optimistic since it's based on feeding analytics into Gemini Deep Research, I also feel it's a realistic possibility.

Auto-dubbing status

  • 20 languages confirmed active (Vol. 11 test successful)

  • Standard TTS languages (AR, NL, HE, JA, KO, ML, PL, RU, UK, etc.) are published same-day

  • Expressive Speech target languages (ES, PT, DE, FR, IT, HI, ID) take several days to process

  • Expressive Speech is high-quality AI dubbing that includes emotion and intonation

  • Shorts (15 seconds) are published immediately in all languages → The engine itself is fine; long-form content is just heavy to process

  • Manually uploading subtitles (Auto Translate) may unfreeze the process. Currently under verification

→ A breakthrough will occur when major languages are published. While I hope for improvements on Google's side, based on the evolution of AI in 2025, I am confident that it will definitely be improved.



Technical specifications confirmed via Gemini

  • Processing flow: Speech detection → Automatic source separation → STT → Translation → TTS → Remix with original BGM

  • A single mixed track is the best practice

  • AI voice (Fish Audio) is fine

  • Language setting during upload must be set to "English"

Cost confirmation

  • ElevenLabs: Cancelled ($99/month → $0)

  • Fish Audio API: Approx. ¥515/volume (approx. ¥1,030 for 2 volumes per month)

  • Suno: ¥1,200/month (Annual contract already paid. Until February 2027)

  • Monthly total: ¥2,230 (85% reduction from v5's ¥16,350)

→ Since the character count has decreased from the next volume, it is sometimes about ¥365/volume. Veo, which creates video assets, runs at 3 videos per day, and if an output error occurs, it becomes a bottleneck, so I am considering wan2.2 on Runpod.

Production Flow (v6 Final Version)

  • Manuscript: md as is (do not touch)

  • Audio: rt2112_voice.py → Fish API → mp3

    • Automatically remove ◇/———/##, delete the entire original source block

    • Direct HTTP call (no SDK required, Python 3.9 compatible)

    • pip install requests + export FISH_API_KEY=key

  • Editing: Proofreading in Premiere (fixing line breaks/howling) + BGM adjustment

  • BGM dB: Decide by feel according to the song (prioritize listening comfort of the EN version)

  • Loudness: Essential Sound loudness is off. -14 LUFS in export effects

  • Video: Gemini/Veo + Premiere (variations using 8-second asset playback and reverse playback patterns)

  • Upload: YouTube (Language: English) → Scheduled for 23:00 JST (= 12:00 PM NY time)

  • Multilingual: Auto-dubbing enabled + assisted by Auto Translate subtitles

→ The sensory part is the BGM. Just because I changed the decibels doesn't mean auto-dubbing will trigger, and since it triggered for all languages in Shorts, I concluded that the BGM volume is not the direct cause. In the first place, since I am aiming for a genre as background music for work, I prioritize maintaining the worldview through the volume balance between the music and the reading.


Posting Operations

  • Long (58 min): 3-4 videos per week. Currently building up at a pace of 1 per day

  • Shorts: Reading clips failed with a 92.4% swipe rate. Switching to single-panel manga illustrations

  • Posting time: 23:00 JST (= 12:00 PM NY time)

  • Goal: Reach 30 videos by the end of the month (algorithm inflection point)

→ I was originally making an entry version (for children's independent reading) and creating illustrations for English manga, so I think I will repurpose those for short videos. The priority is building up long videos, so I'll do it if I feel like it. Pinterest is the same.

Channel Data (as of March 18)

  • Number of published videos: approx. 18

  • Impressions surging (hockey stick growth)

  • Average view duration 6:30 (excellent for a 58-minute video)

  • 682 total views with 54.3 hours of watch time

  • CTR 2.8% (temporary decline due to rapid impression expansion; normal)

  • Comments and views from Ukraine (proof that auto-dubbing is working)

  • Vol. 15 appearing in recommendations with '5 more than usual'

Notes on Fish Audio API

  • Audio quality is somewhat unstable (occasionally drifts from the specified voice)

  • Transitions across line breaks are more natural with the API than the Web UI (processes line breaks as pauses)

  • Source material blocks (after ———) are automatically deleted, so no need to edit the manuscript

  • Approximately ¥400-500 per volume of 28 chapters

Vol. Handling (Notes on production order)

  • Vol. 11-17 = Book 009 (in 4-chapter increments)

  • Vol. 18-24 = Book 008 (filler; volumes produced after 007 and 009. Process these first)

  • Vol. 25 and later = Book 010 onwards

  • Vol. 1-10 of v5 are the old 2-hour versions. Scheduled for remake into 60-minute versions at a later date

→ Since the source material has been removed, the audio side prioritizes immersion over novelty. Kindle retains the news source material. Difference in target audience.

KDP

  • Successfully generated all 14 EPUBs for Book 008 (7 languages x blue/yellow)

  • KPF not appearing → Kindle Previewer has been uninstalled. Reinstallation required

  • KDP is the destination for traffic from YouTube. Currently, mass-producing YT content is the top priority

  • Organic traffic is almost non-existent (normal)

→ Use Python from the directory to execute the epub to kpf conversion in one go. For the Kindle description table of contents, also use Python to extract the headings for each chapter and print a list of 28 chapters. This workflow no longer requires any improvements.

List of Tools

  • rt2112_voice.py: Fish API voice generation (direct HTTP call) ✅ Completed

  • mix60.py: 58-minute audio mixer (ffmpeg) ✅ Completed. Manual Premiere work is primary

  • mk_all_epub.sh: Batch generation of EPUBs for all languages ✅ Operation verified

  • mk_epub.py: EPUB/KPF generation ✅ KPF is waiting for Previewer reinstallation

  • kdp_toc.py: KDP table of contents list generation ✅ Operation verified

Deliverables (created in this chat)

  • rt2112_v6_revenue_sim.md: v6 revenue simulation summary

  • rt2112_autodub_spec.md: Auto-dubbing specifications and test results

  • rt2112_voice.py: Fish API voice generation script v6

  • rt2112_voice_setup.md: voice.py setup guide

  • mix60.py: 58-minute version mixer

  • rt2112_shorts_titles.md: Shorts title pattern collection

  • rt2112_vol_map.md: Correspondence table for Vol ↔ Volume ↔ Chapter

  • rt2112_value3_note.md: For note logs

Goal

  • 500,000 yen/month (must achieve): 670,000 monthly views. Goal within 1 year

  • 1 million yen/month (stretch goal): 1.33 million monthly views

  • Monetization (YPP): Potential to reach in 2-3 months (ahead of initial forecast) → Feels optimistic.

Next steps

  • Mass production of long-form content. Focus entirely on reaching 30 videos by the end of the month

  • Auto Translate subtitles → Check results of auto-dubbing regeneration

  • Shorts: Switch to single-panel manga illustrations

  • Reinstall Kindle Previewer → Restore KPF output

  • Fish API audio generation for Volume 008 (Confirmed completion at 377 yen. Awaiting production execution)

Side note

I've been thinking about how to create export content since last year, and I've concluded that YouTube is the right answer. Even if you can easily create various businesses with generative AI, the fact remains that marketing challenges can only be solved by throwing money at them or by becoming a celebrity. Therefore, since I built the business foundation in 2025, I am going all-in on the Rt2112 expansion in 2026. If I can monetize, I can do the former, and even if I can't, it serves as insurance for the latter, which is fine either way. In any case, it's my nature to keep running costs at a bug-level minimum and fight a battle I can't lose. Right now, I'm a sole proprietor and going all-in with my time, but even if I were to return to being a salaryman, I could maintain operations, so time risk is not an issue. In this kind of cumulative business, you never know when or how things will turn out even if you make a plan, so staying alive is the top priority.





Advertisement