Day 6. The first followers came in one night, and I saw where I lose them
In the morning the model's follower counter showed thirteen. Yesterday at the same hour it was zero. Not an accidental zero: this account had a previous version of the character with 83 followers, and I removed them myself before the countdown started, which is why the 30-day chart shows minus 83. What worked was not what I had spent a week preparing: of the five posts the autoposting published on its own yesterday, four text posts together brought 1,632 views and not a single follow. The fifth, a photo in makeup captioned “put on makeup at eleven at night, not going anywhere”, by evening brought 4,980 views, 439 likes, 127 replies and eleven follows. The algorithm showed it to strangers: 93% of the views came from the feed, not from the profile.
I pulled every statistic down to the number and saw a thing I would not have noticed by eye. From this post 2,037 people visited the model's profile. Eleven of them followed. That is half a percent. By evening the total was twenty: the rest came from other posts and from the profile itself. For the accounts from my own niche report, the ones with a photo feed and a clear recurring theme, that share is 4 and 13 percent. So we have learned to get reach, but the door does not work: two thousand people knocked and left. For the rest of the day I fixed the door, not the posts.
The avatar came first. In it the model hid her face in a sweater collar, while people arrived from a close-up portrait in makeup and did not recognise her. I made ten new ones, and none fit: in a forty-pixel circle a light face on a light background turned into a pale blob. What helped was not a pretty frame but a simple principle: a light face needs a dark frame around it, hair or background. I checked not by eye but with a simulation of the feed's real size, and that was the only thing that worked. I also rewrote the bio so the first line answers the most frequent question in the comments right away: “is this a neural network?”. Eight people wrote that yesterday, and “Neural net” collected more likes than any compliment.
The second thing I saw in the comments: four different people asked about the freckles. Whether they are real, whether it is makeup, why the hair is not red. Not one reel in a month had raised a single question, and one photo raised four. So in the afternoon I generated a series of portraits with different makeup, ten looks, and got an outside breakdown from a makeup artist on what actually suits this face. It knocked out one of my frames: a green liner on green eyes, which had seemed a good idea, kills the iris. Her colours are the opposites: plum, burgundy, copper. I would never have learned that from the statistics.
The day's mistakes also cost money, and they are mine. Three frames in a row came out with someone else's face, because I added a third reference, a finished portrait, and it pulled the model away from the base. The frame “washing off makeup in the bathroom” came out with neat hair and a fresh liner, because I built the scene out of pretty words rather than logic: wet, made up and groomed at the same time does not happen. Two frames came out with frightened eyes, because the prompt said “caught off guard”, and the model drew exactly that. The rule I wrote down for myself: describe muscles, eyelids and brows, not a state.
In the evening a new reel went out, the first one set in locations: a taxi at night and a table laid for two. The makeup is the same in both frames, because the first version I assembled had the model bare-faced in the taxi and made up at the table, and I was called on it. Spent today: about 1,300 rubles, more than on any day of the run, and most of it went to trials that ended up in the “rejects” folder. Earned: 0 rubles. The model has twenty followers on Threads; on Instagram still zero, even though the same photos go there. Hence the decision for the coming days: the weight of content and of my time moves to Threads, and Instagram stays a showcase for whoever arrives from there. The first taxi reel got 264 views and five replies in an hour, for comparison on the screenshots. Tomorrow afternoon is the first honest measurement of a location reel against that very photo.