Day 16. A week without the diary: the model gained six followers, the author's account none, and three of my conclusions did not hold up
Days 10 through 15, September 22 to 27: no entries. All week I was making reels and photos of the model and never sat down to the diary. On Day 9 I promised two things. The model's photo at 19:30 has gone out every day since the 23rd, so that is closed. I did not start the author's Instagram: zero posts there this week.
What I changed against last week. I made a series of reels where the model speaks in her own voice (lip sync: the lips move to recorded speech), 25 to 30 seconds long. Three went out, I cancelled the rest. I moved the model's texts to three modes in rotation. On September 27 and 28 I ran an external review of every post on both accounts (an audit): 58 of the model's and 76 of the author's.
What it gave. The model has 60 followers on Threads, a week ago it was 54. The model's Threads views for the week: 18,541, the week before 32,526. The three talking reels brought zero follows on both networks.
Mistake of the week. I drew conclusions from one lucky week and published them as rules. “Studio zero, phone 27”: the review showed that a studio portrait with makeup brought the model 11 follows. “The AI-content label is added for synthetic voice”: the label also sits on three photos with no sound. “Twenty followers on the first day, from five posts”: the author already had twenty before autoposting. The first two corrections go out as posts on September 29 and 30, the third is a note on Day 9.
The week in money and hours. Generations ≈3,014 ₽: model frames at 12 ₽ each (Nano Banana Pro), a talking reel about 155 ₽ per 30 seconds (Wan). Since September 17, ≈5,050 ₽ in total. Author's hours: 20, three on weekdays and two on Sunday. The author's site from September 21 to 28: 29 live visits, 0 sign-ups. A day of the experiment now costs about 430 ₽; the course is 4,990 ₽.
What I do tomorrow. The author's account writes again, two posts a day instead of the promised five: I never managed five even once. On September 30 and October 1 the model gets two frames of one test, results on October 4.