2026-05-01
Solidifying the sub-projects while building the main one. Rt2112 Development Log | April 29 - May 1, 2026
What I did this time
Rt2112 Vol. 10 Kindle release started (English, Japanese, German)
Migration to DistroKid completed. Initial release settings configured
CTR improvement strategy for Lo-fi 2112 thumbnails (badge design)
Kindle description templates prepared in 7 languages
Review of the Kindle multilingual cover production workflow at Nanobanana
Design of Spotify / Apple Podcast distribution schedule
Uploading posters for each episode to Pinterest (semi-automated)
Numerical snapshot (as of 4/30)
Main Channel: Road to 2112
A Bench That Talks (2 days 11 hours): Views 7 / Imp 172 / CTR 4.1%↑ / AVD 9:07↑
MIT's Skin Patch (20 hours): Views 12 / Imp 445 / CTR 2.7% / AVD 1:32
Hindi-speaking content appearing in recommendations for MIT's video → Initial signal for full auto-dubbing release
A Bench~ has 14.3% search traffic. New title structure is working for search
Lo-fi 2112 (28 days)
Imp 23k (+1,673% vs previous 28 days) / CTR 1.3% / AVD 15:38
Total watch time 153 hours (+770%)
Top 3 Shorts: 384 / 356 / 353 views
Titles with "Study" have the highest views (Dusk Study Vibes: 60 views)
Impressions have surged but CTR is low → Need for thumbnail improvement
TikTok music usage
April peak exceeded 20,000 views. Estimated monthly views over 30,000 (completely organic).
👉 Things remain unchanged, and while I continue to fly low with sluggish impressions, there are interesting movements, such as searches from Hindi-speaking regions that hadn't appeared before.
Also, while distributing to various platforms via a distributor, it is somehow spiking on TikTok.
If you ask what this leads to, it leads to nothing! However, since I am operating under the name "Sassa Mahha," I would be very happy if someone picked it up on some social media platform.
Future policy
Apply badge-adorned thumbnails to new videos and verify CTR effects.
Kindle 🟡 entry versions: release steadily without holding back.
Decided to release one episode per day on Spotify.
Started releasing a new album on DistroKid.
👉 The Kindle versions for lower and middle elementary school students are quietly falling behind in publication (or rather, I'm slacking off), so
I want to clear them during my spare time over Golden Week. Since there is a limit of 3 books per day, it becomes a tedious daily task that fluctuates with my daily motivation (bitter laugh).
Lo-fi is a battlefield where the mainstream is so dominant that CTR drops, but it's a place I can't concede from a branding perspective, so I will try to improve it with thumbnail adjustments.
Rt2112 Vol. 10 Kindle released (English, Japanese, German)
New book releases gradually starting from the end of April.
The Japanese title is "Namida no Shindo" (The Vibration of Tears). It's a story about self-driving cars.
After trial and error from volumes 1-5, I switched to Claude from volume 6, changed to a YouTube Audio-based production from volume 8, and this is the first production after establishing the general workflow.
It's quite smooth.
(Although Spotify releases still have some manual work left due to trial and error, the priority is low.)
Now, for the first episode of volume 10, "The Vibration of Tears," the YouTube side starts on 5/2.
I uploaded it around mid-April, went through 2 weeks of auto-dubbing, and distributed it in 22 languages? Naturally, the Japanese audio is terrible, so you don't need to listen to it.
The cited news has been kept to the past month or so, and the timeline has aligned, so May is the turn to become a stacking machine.
Also, this turn involved a lot of improvements and AI utilization despite being tedious work, so I'll talk about that.
Specific settings for DistroKid migration
Reason for migrating from TuneCore: AI song rejection (100% AI-generated content cannot be distributed)
Instagram / Facebook → Both set to "No". Even on TuneCore, I haven't passed the Facebook Music review
TikTok → "Yes" only. I won't stop viral potential
Not using the SNS pack. Avoiding the risk of Content ID claiming copyright on my own videos
Leaving the existing TuneCore catalog as is (will decide at the one-year renewal mark)
👉 While I could edit the music that was rejected by TuneCore, the quality of the Suno output is so high that tweaking it would just be superficial and feel like a waste of time.
I want the speed to be able to distribute it as is if possible.
I chose DistroKid as the migration destination.
After actually using it, it suits me so well that I think, 'Claude, why didn't you introduce this to me from the start?'
It covers all the major platforms, and more importantly, it's a huge plus that I can push to Tier 3 music streaming platforms in Southeast Asia and Africa with a single click.
I was already using an API on YouTube to insert titles in various languages, and it's translated into 26 languages. Since the countries driving traffic are Thailand, Vietnam, and Malaysia, it's a very helpful situation in the sense that they can find me just by linking to YouTube and searching for 'Lo-fi 2112'.
One thing I was quietly happy about after migrating was being able to distribute 20 songs at once.
TuneCore only allows 15 songs per album, while this allows 20.
The work of scrutinizing this has disappeared, and the browser input process is also compatible with Claude in Chrome. Distribution work has become faster and can now be done in the background.
Lo-fi 2112 Thumbnail Badge Design
Hypothesis for 1.3% CTR: It's not clear from the thumbnail that 'this is background music'
Design to place three Splatoon-style gear power icon badges in the bottom right of the thumbnail
Badge 1: Time
Badge 2: Purpose (STUDY / WORK / RELAX / SLEEP / CAFÉ)
Badge 3: Character (Ko-Ne, AkaGe...)
Do not include purpose keywords in the title. Since the premise is to sell the worldview of the main channel, the roles are divided: thumbnail = functional display, title = worldview
Create character icons
👉 Regarding the Lo-fi 2112 channel, while impressions are increasing every time, the CTR is stuck at the 1.0% to 1.5% level, which I suppose is unavoidable, but I do feel that some reinforcement is necessary.
I decided to add three badges to the thumbnail.
Time and purpose are common in background music thumbnails. I think there are naturally people who click just by looking at this, so this is a countermeasure for that.
The third badge is a DJ character badge.
I'm placing it there with the hope that it will grow as a medium-to-long-term IP, and it's also great for differentiation.
While it's a standard genre of music, I want to gradually pick up a share from the major players.
Refreshing the Kindle description template
Old: 'In a quiet future between 2035 and 2112'
New: "In a near future not so far from now"
Unified with the YouTube description. Removed the date specification.
Added "Each story is inspired by today's real news." to clarify the USP.
Created templates for 7 languages (EN/JA/ES/PT/DE/FR/IT) in one go.
👉 Minor improvements. The description I wrote around volumes 6-7 hadn't been updated to match the YouTube workflow, so I'm aligning it with the current state.
Nanobanana multilingual cover workflow
Generate poster art with English titles using Gemini (existing workflow).
Erase the title area in Photoshop Nanobanana: "Continue the background illustration naturally. No text, no letters, no characters."
Generative fill for each language title: "Large decorative stylized title text "{TITLE_TEXT}" in cream white and warm gold color, matching the existing typography style of this image..."
Do not fix the font style; specify it to match the style of the original image.
If the color isn't explicitly specified, it blends into the background and visibility drops → Fixed to cream white and warm gold.
Issue with background boxes appearing → Suppressed with "no background box, no text box, no banner, no rectangle behind text."
Confirmed that accent characters (ó, á, ñ, ã, etc.) can be rendered in Nanobanana.
👉 Talking about Kindle covers.
Kindle uses the main story of each volume (usually episode 1 or 28) for the cover.
I create posters for each episode of each volume using Gemini, and since they are all in English, I use Photoshop's Gemini (Nanobanana) when changing them to each language.
I've struggled with this in the past, and if you look closely at the covers around volumes 6 and 7, there are some rough parts.
Since the images aren't that large, I usually just publish them as is.
This time, by splitting it into two steps, the output has become stable.
Photoshop is good at content-aware fill, so I add one step to fill in the English text and assimilate the background.
By inserting the title name in each language into the cinematic-style text I always use for poster generation on top of that, I can achieve a layout similar to the English poster.
Honestly, if I were to obsess over quality, there's still more I could do, but as I mentioned at the beginning, it's not a large image to begin with, so it's at a 'good enough' level.
Spotify distribution schedule design
Distribute one episode at a time to Spotify, offset by one week from the YouTube release.
Start with a one-episode-per-day pace with a one-week offset, then switch to a two-episode pace midway.
28 episodes (one volume) are cleared in about 14 days. Fits within a 3-week production cycle.
Can create two waves of interest with the same content on YouTube and Spotify.
In the future, when the Spotify Japan Partner Program is opened, monetize English content for overseas listeners (not in Japanese).
👉 This is low priority in terms of monetization. In Japan, neither Spotify for Creators nor Apple Podcasts allows for monetization, so uploading there is essentially free work.
In the first place, I had been uploading copies of my YouTube videos to Spotify, but the play count was always zero, so I deleted them all.
With that in mind, I am rethinking my approach.
Since having a Spotify RSS feed allows for distribution to Apple Podcasts as well, I thought simple, episode-based MP3 audio stories would be fine, so I tweaked my Python script to allow direct uploading of MP3s generated by Fish Audio.
Specifically,
Title (2.5-second pause), Source (5-second pause), Body text.
The idea is to output these three as MP3s and combine them using ffmpeg.
This eliminates the need for the work I was doing on the YouTube side, where I was vaguely cutting clips in Premiere to extend the duration.
For thumbnails, I am just reusing the posters.
Since they are vertical, they don't fit perfectly, but spending time on that is a waste, so I've decided it's good enough.
Uploading episode posters to Pinterest (semi-automated)
This task is surprisingly significant.
Even though there was an official flow for bulk uploading via CSV, it never worked properly, leaving about 500 images, including those from other projects, sitting idle.
While doing this, I decided to ask Claude to handle just the task of entering text after uploading the images to Pinterest, and it worked perfectly; uploads are now steadily progressing in the background.
Since there were nearly 100 posters for Rt2112 alone, this is a huge help.
In fact, one of the images on Pinterest spiked and impressions have increased quite a bit, so it's not to be underestimated. However, I don't want to spend too much effort here. Since that's the kind of SNS it is, automating this task is exciting.
As we have entered May, a month for building things up, the goal is to minimize manual effort and keep automating whatever can run in the background.
Although results are not yet visibly apparent, there are signs of improvement, so I will continue to work hard this month while focusing on how to turn those spikes into a breakthrough.