2026-05-26

Signs of a Signs of Revival. Rt2112 Development Log | May 22 - May 26, 2026

In the previous devlog, I wrote that '8 minutes died on YouTube, so I'm going back to 120 minutes.' This time, I'm talking about actually making and releasing that 120-minute collection video.

Why 120 minutes?

The limit for auto-dubbing (YouTube's automatic multi-language dubbing) was 120 minutes. I thought it was 60 minutes in March and abandoned the 2-hour format, but it actually supported up to 120 minutes. In other words, I could have kept the original 2-hour format. Three months of structural changes started from a misunderstanding.

What I made

7 stories (approx. 70 minutes) within 120 minutes + ambient music (approx. 50 minutes). On top of that, as an update from the beginning, I used a planning method where I organized them by theme rather than by volume order to address specific issues.

The theme for v1 is 'For People Who Think Too Much at Night.' For people who think too much at night and can't sleep. v2 is 'When the World Gets Too Loud.' For when the world is too noisy.

A story about setting up a pipeline with Claude Code so that the operation of the project can be carried out with as little effort as possible.

Production Workflow

Theme Setting — Determine a theme that resonates with the viewer's identity.
Book Selection with Claude Code — Select 7 stories from existing content that fit the theme. Write rules in CLAUDE.md and execute in bulk.
Source Generation — Generate a news source introduction to be read at the beginning of each story ("This story was inspired by...").
Audio Generation — Use voice.py v10 to send to the Fish Audio API for bulk audio generation. Order: Title → Source → Main Story.
Loop Animation — Convert poster art for each story into landscape images using Gemini. Convert to minimal motion animation using Veo and Grok.
BGM — Generate with Suno. Fixed soundscape, no melody.
Assembly in Premiere Pro — Opening loop animation → Intro narration → 7 stories → Thank you screen → Ambient. 12-second interval between each story.
Upload → Auto-dubbing permission-based (pre-approved for 8 languages supported by Excessive Speech only).

👉 While the production process itself, such as changing loop animations for each story, has increased in effort compared to the beginning, the assembly method remains largely unchanged.
Instead, I build from the purpose of the video = planning based on titles that meet the needs of the target audience.

Title Design: Level 6 Framework

I happened to come across TubeLab's 7-level framework on X, and used it to brainstorm with various AIs. I adopted Level 6, "Identity." A title that makes viewers think, "This is about me," and click.

Currently testing several patterns

“For People Who...” pattern — used in v1
“When...” pattern — used in v2
Poetic pattern — currently testing by renaming the old 58-minute video to "A Future That Remembered Us"

👉 As a rule, do not overuse "Robot," "AI," or "Sleep" in titles. Create them using emotion, memory, loss, salvation, and promises. Not too dark, with a hopeful ring at the end. The temperature of "Loneliness → but maybe there is salvation."

Intro Narration

Include a common intro for each theme at the beginning of the collection.

v1: "This is for you — the one who thinks too much at night. Here are stories imagined from today's news, set in a near future. Close your eyes, and just listen."

v2: "When the world gets too loud, you don't need silence. You need a quieter story. Here are seven, imagined from today's news, set in a near future. Let them carry you."

👉 Added narration to the first 30 seconds. Until now, it was common to only have
"Road to 2112. Four stories from a gentle future".
While this is a branding element in a sense, YouTube is project-heavy, so I decided to include an intro tailored to the video theme.

Operation of a 3-channel system

Bedtime Tech Story — Road to 2112 (Main)

Dedicated to 120-minute collections. Changed the channel name. "Use first, brand later." People searching will look for "bedtime story." They don't know Road to 2112.

Road to 2112 — Real News, Future Story (Podcast)

Automatically distributed to YouTube/Apple/Amazon/Pocket Casts via RSS from Spotify for Creators. 8-minute single episodes. Still images. No BGM. Zero production effort.

👉 Since 8-minute single episodes don't provide impressions, I'm fine with it being a stock channel that complements the main one.

Jun Sasaki | Sassa Mahha (Vlog)

Japanese process vlog.

Do not mix formats within the same channel. Having 8-minute and 120-minute videos together sends mixed signals to the algorithm.
That said, since it's just a rough screenshot log, I'm not doing anything that significant.

Change in auto-dubbing policy

Narrowed down from all 20 languages to only the 8 languages supported by Excessive Speech (EN/FR/DE/HI/ID/IT/PT/ES). The audio quality of standard TTS regions (JA/KO/UK, etc.) was poor for bedtime content and was polluting the AVD.

👉 After all, I tested it while keeping it fully open for a little over two months, and for a while, I even waited about two weeks for everything to become public before releasing it, but nothing changed.
If I'm going to have it spread out at the time of release, I concluded it's better to release it in the initial English-only stage. Both v1 and v2 are released as soon as they are ready.
Since the 8 languages finish translating about 2-3 weeks after release, I judged that releasing them after that is sufficient.

Numbers (as of 5/26)

v1: For People Who Think Too Much at Night (120 min, released 5 days ago)

Imp 108 / CTR 4.6% / Views 14 / Unique Viewers 3
AVD 9:04 (+8:17 vs normal)
Watch Time 2.1h
Traffic: Channel pages 46.1% / Suggested videos 23.1% / Browse features 7.7%
View count ranking 1/10

v2: When the World Gets Too Loud (120 min, published 3 hours ago)

Imp 97 / Views 6
AVD 19:04 (+18:54 vs. normal) — Channel record high
Watch Time 1.9h
Traffic: Browse features 66.7%
View count ranking 1/10

Poetic title test: A Future That Remembered Us (58 min, 34 hours after renaming)

Imp 55 / CTR 3.6% / Views 2
AVD 14:44 (+14:34 vs. normal)
Traffic: Browse features 100%

Reference: 58-minute video with highest CTR

Imp 51 / CTR 5.9% / AVD 12:13
Both CTR and AVD are high, but impressions are not being distributed. Channel signal issue

Channel overall (90 days)

Views 1.9K / Imp 54K / CTR 2.1% / AVD 3:52
Monthly viewers 123 / Subscribers 18

👉 The view count is by no means high, but the AVD is rising. As a result, impressions have started to gradually increase.
I was planning improvements in the order of AVD → Impressions → Watch time, so this is in line with the results.
Next, it would be ideal to increase total watch time with a sleep-aid boost.

Next steps

Set 8-minute standalone episodes to private on the main channel. Purify channel signals to long-form bedtime content → move to sub-channel.
Post 120-minute collections twice a week (local weekend + Wednesday).
Considering changing v1 thumbnails to a night-time aesthetic (to align with the bedtime context).
Production of v3 'The Family You Didn't Expect'.
June: Writing Vol. 12.

The numbers are still small. I wouldn't call it a 'hit' at the 6-14 view stage, but it is the most promising it has been in the last three months. The combination of CTR, AVD, and browsing ratio is different from any format I've used before. It's a sign of a sign.