2026-04-21
Upward trend after a slump. Still a long way to go, but a good sign. Rt2112 Development Log | April 18 - April 21, 2026
With YouTube demonetization becoming more prominent, some AIs say my content is safe while others say it's at risk, so I spent these four days focusing on monetization with diversification in mind for the medium to long term (though that was the plan all along).
If I am to focus on B2C creator work, the environment where YouTube acts as a salesperson on my behalf is quite valuable, so continuing to build it up while treating it as an advertising tower seems to be the best policy for risk hedging.
What I did this time
[Rt2112, Lo-fi 2112] Renewed image generation prompts for thumbnails
[Lo-fi 2112] Completed usage mapping for 10 DJ characters
[Lo-fi 2112] Redesigned playlists into 3 categories based on user behavior
[Rt2112] Discovered traction in the KDP German market
Cluster analysis of the algorithm
Designed the monthly note summary (scheduled to be published as a free article at the beginning of the month)
Numerical snapshot (as of 4/21)
Main Channel: Road to 2112 (28 days)
Views 828 (+7%)
Impressions 27k (+58%)
CTR 2.0%
AVD 3:12
Monthly viewers 551 (98.2% new)
Recorded 11.1% CTR and 27:39 AVD on the latest video
👉 After the strategic shift (from Sleep-focused to Intellectual Tech-focused), visitors temporarily decreased, but CTR and AVD are starting to show an upward trend. In addition to the above, I have been making minor adjustments, such as quietly changing the pinned comment to a KDP funnel.
Lo-fi 2112 (28 days)
Views 533
Impressions 20k
CTR 1.4%
AVD 15:31
Monthly viewers 345, +8 subscribers
Shorts peak 385 views, +1 subscriber from Shorts
Dusk Study Vibes is the top (58 views, AVD 33:36)
👉 Overall, it is gradually growing. Views have dropped recently, but that was the case with the main channel too, so I'm not worried. It's like an analysis phase after the initial honeymoon period.
KDP (April)
Market share: Germany 73%, Japan 24%, USA 3%
Spanish version had 19 pages read
First Kindle read in the USA
👉 Quietly, KDP itself has been a painful area after doing it for a year. Aiming for the US but not getting read in the US. In the midst of that, persistence paid off.
From here on.
Future plans
Start story production (I had stopped this because I was turning existing stock into videos)
Production with new thumbnail prompts (a small improvement, but a move that will pay off)
Get the German version to 15 volumes
Continue mass-producing Lo-fi Shorts via remixing (just going with the flow.)
Start operating 3 playlists (Sleep, Study, Road) (just going with the flow, but a solid and smart move)
👉 Ultimately, the first month of opening is a honeymoon period, and after that, a slump seems inevitable while the algorithm learns.
In fact, even doing the same things, the metrics I'm testing are lining up—hitting browse features, hitting suggested videos. For 90 days, I need to make small improvements while building up in line with this movement.
Even if the view count isn't rising, the underlying numbers are improving, so I will just keep building.
The rest is paid content (180 yen / free if shared)
I mainly write about the process of why I did the above.
Revamping the thumbnail prompt
I rebuilt the master prompt for thumbnails that I feed into nanobanana (Gemini).
There are three key points. I fixed the time of day to "evening to late night" (since 97% of the data is night thumbnails). I limited the placement of characters and small robots to 20-30% of the screen to make the background stand out. I added all elements that cause failures in Veo's loop animations (steam, fire, flowing water, fabric swaying in the wind, drinks) to the negative prompt.
Since this list was created from actual trial and error, the hit rate for the gacha has increased. It rarely fails now, and even if it does move strangely (like a tablet suddenly appearing), it usually happens in the last second of the 8 seconds, so I can just cut it in Premiere.
In addition, I realized once again that whether or not a thumbnail comes out that I think is "good" is everything for CTR. The thumbnail that hit 11.1% CTR was one that I thought was "this is good." My intuition is directly reflected in the numbers. In a sense, this means my intuition is on target, so "vibe" is important.
Mapping of 10 DJ characters
The DJ robots for Lo-fi 2112 have expanded to 10 characters.
Ko-Ne (cat), AkaGe (ginger cat), Lo-Hoo (owl), KUMA (bear), RABI (rabbit), MERINO (sheep), La-ale (whale), Rune (husky), Lark (jellyfish), and Tanuki-O (tanuki).
I mapped a home ground and purpose to each character. Ko-Ne is for Sleep & Relax, Lo-Hoo is for Study & Focus, Rune is for Road & Journey, and so on. Like this👇
What changes with this mapping is that the viewer chooses a playlist based on "I'm in the mood for Sleep today," and various DJ characters appear within it. It's not "I want to listen to that cat robot one," but "I want to sleep, so the Sleep list."
However, I left some flexibility so that while they have a "main responsibility," they can appear anywhere. It's music, after all, and the vibe is that it's boring if you restrict it too much.
Reason for narrowing down playlists to 3 based on actions
Initially, there were 5 candidates: Sleep & Relax, Study & Focus, Sunset & Chill, Road & Journey, and Zen & Wabi-sabi.
I dropped "Sunset & Chill." The reason is that "idling in the evening" is ambiguous as an action. It gets absorbed into Sleep & Relax or Road & Journey. The actions for which humans need BGM can be almost entirely covered by three categories: "sleeping/resting," "concentrating," and "traveling/changing mood."
If there are too many playlists, viewers get confused. With three, they can choose in an instant. For example, if I were to do something inspired by Japan in the future, Zen & Wabi-sabi would be a buzzworthy word. However, that is also Calm & Focus. Zen is expressed through the atmosphere of the animation. The point is that it's action-based.
KDP German market and the A10 algorithm
When researching KDP, I found that Amazon's A10 algorithm raises organic search rankings when it detects high-quality external traffic. The flow of YouTube -> sassamahha.me/books (book list) -> Amazon is exactly the kind of "high-quality external traffic" that A10 evaluates.
The number of publications required for stable monetization in the SF genre is 12-18 volumes. I'm at 9 volumes now. I'll reach 15 volumes with the production of Vol. 36 and beyond. The strategy is to break through the "15-book barrier" first with the German version and get the flywheel spinning. Germany has an ebook market of €505M, a low supply of Cozy Sci-Fi, and a high affinity for environmental themes. It might actually be a blue ocean.
Clusters identified by the algorithm
Since the last update, I have completely revamped the content from sleep-focused to what I call businessman's bedtime (relaxation) content. The results are starting to show. Related videos on the main channel now include Midsomer Murders (a British mystery drama) and sci-fi documentaries. The algorithm is delivering Rt2112 to the 'English-speaking x long-form x listens to stories before bed' demographic.
Come to think of it, the structure of Rt2112 is 'cozy mystery x sci-fi x emotional.' A small mystery at the beginning (a strange technological phenomenon) -> investigating it -> the truth is human emotion. To put it simply, it's similar to The Girl Who Leapt Through Time.
The algorithm is starting to see this essence. Cozy mystery is a best-selling genre on English-language Kindle. The genre affinity is high, serving as a bridge for viewers who find it on YouTube to flow over to Kindle.
For story production from Vol. 36 onwards, I plan to consciously incorporate this 'mystery at the beginning -> truth is emotion' structure.
Side note
Starting this week, I will resume story production on a cycle of releasing every three weeks. I am currently in a state of clearing out past stories in April to train the algorithm. From early May, the stories from the current production turn will start coming in, so I think that's when things will really start.
View counts are still low, but since the cluster has started to change and it's beginning to resonate with the target audience I originally wanted to reach, the probability of hitting it big is also increasing.
Impressions are being supplied constantly, so I think it's just a matter of building it up.
Since the production cycle will be 3-4 days a week, I'm even thinking about silly things like doing a walking vlog to lose some weight.
(Currently, I am making 2 videos a day, one main and one sub)
Honestly, I don't know if this will be a hit, but personally, choosing a genre I can continue with was the right decision.
In the end, the initial honeymoon period and the mid-slump period are inevitable, so I recommend a genre you can stick with for 90 days.