Creators quote $250 per video—she self-filmed + used Claude, made 46 in 4 days for under $10

Creator quote: $250 per video. Nadia Zueva didn't pay. She spent one hour self-filming in her bedroom and handed everything else to Claude: finding hooks, writing scripts, selecting footage, editing, scheduling. Four days, 46 videos, scraping costs under $10.

She runs Aesty, an AI styling app that generates outfit ideas from clothes you already own. The team is two people, both engineers, with no marketing department. She chose FGC (founder-generated content) for iteration speed: no waiting for creators to read a brief, shoot, and redo; the same day you can see which hook works and change immediately. She had previously written about AI carousels and AI UGC, but the videos weren't AI-generated—trust is dropping, organic reach isn't moving, and AI video is better suited to paid ads.

The process is as follows:

Trend monitoring. Claude scraped 249 viral app promos from the past six months, sourced from TikTok and YouTube Shorts, pulled with Apify and yt-dlp. Most TikTok videos had no captions, so Claude captured the first frame and used OCR to read the on-screen hook. The 249 were ultimately grouped into 15 hook formats: wdym...???, i'll never forget the girl who..., POV:..., girl to girl:, hot take:, before/after, and so on. It also read the comments. Under a competitor's POV video, the top comment asked: do you have to photograph every piece of clothing? Aesty doesn't, and that question became the main hook directly.

Filming. Claude rewrote the hooks into Aesty versions and generated a shooting page: lines for each scene, on-screen copy, two viral references, and a button to upload footage directly from the phone. She filmed for about an hour, 20–30 reaction and outfit-change shots, including bending toward the camera in a bathrobe and then standing up in a full outfit for a transformation match cut. Claude generated a contact sheet for each piece of footage and picked the best 2–3 seconds, with reasons. Remotion cut the hooks and demo together, with the drumbeat landing on the edit points. Multi-screen demos had step captions; versions without captions were rated not top by her—bare screen scrolling explains nothing.

Before publishing, it checks frame by frame. It has caught a private photo that flashed by, and a friend's name appearing in a share panel—one at the opening, one in the last half second.

Decision-making. She only judges top or not top. The selected ones get variants; the rest are discarded.

Scheduling. Postiz is integrated via API. Claude maintains the plan for each account and builds a review page; she reviews and approves each item, and anything unapproved never reaches Postiz. Before scheduling, it pulls content already published by other accounts so nothing repeats on the same platform. It then asks Postiz which time slots are already taken and only fills the gaps. Variants are spread across different platforms and different time slots. YouTube titles limited to 100 characters, marking content as not made for kids, and TikTok brand disclosure are all set through the API.

First batch production data: 46 videos in 4 days, scraping costs under $10, creator costs $0. At that creator's quote, 30 videos would be worth $5–7k.

Someone pointed out in a reply: that creator is selling her face, not the hook. Her test conclusion was: the face matters far less than the hook. Someone asked her what to do if the same face appears 40 times in the feed. Her response: the videos are distributed across 15+ accounts, and no single feed sees the same face densely. On X, however, in 227 consecutive days of build-in-public videos, the face is exactly why viewers stay.

For account safety, she summarized a set of rules. What triggers throttling: duplicate fingerprints, including identical files, copy, and hashtag sets; high-frequency posting on new accounts, more than 3 per hour or more than 3 per day; the same cover image on the same day; repeated use of the same audio; spammy copy such as 'link in bio' and 'follow me.' Countermeasures: make every caption and pinned comment unique; rotate hashtags without repeating the previous set; re-encode files and strip metadata; on new accounts post 1–2 per day, 4–6 hours apart, staggered across platforms; spread variants across different days and platforms; use audio from the original viral video in the corresponding format; mention the brand name once, at most one emoji, 2–4 hashtags, and end with a real question.

Using a scheduler, posting the same video across platforms, and a single low-view video did not trigger throttling in her tests. Her only shadowban happened when she accidentally posted the same content 5 times in a row. Views dropped sharply, recovered on their own after two days, and there was no appeal.

How do you tell throttling from a bad video? TikTok first serves each video to a small batch of users; only if those people watch and save it does it get pushed to the next round. She judges by view rounds: R0 (initial batch) below 200, R1 200–500, R2 500–2k, R3 2k–20k, R4 above 20k. At the low-view stage, an engagement rate above 1% usually means throttling: the people who saw it liked it, but the platform isn't pushing it out. An engagement rate below 1% means the content failed. Same symptoms, opposite solution. For throttling, pause for 48–72 hours, then post one per day with long gaps. For content failure, change the first 1.5 seconds and copy a format that has already worked.

This week, beyond the Aesty account, her own founder account also scheduled 16 Instagram posts and 24 YouTube uploads (6 full videos + 18 Shorts), completed within minutes. New videos get 1–2k views each; the best one got 12.2k on Instagram and 20k on YouTube Shorts. Not explosive, which is why they make 59 videos instead of 5. The 20k founder-on-camera short is being turned into 10 variants, with the same joke and audio, each with a new hook.

Tool stack: Claude Code (Opus 5.5), Apify + yt-dlp, Apple Vision OCR, Remotion + FFmpeg, Whisper to distinguish music from speech, Postiz API. All related skills are open source, with code in the awesome-prompts repository: ugc-hooks, trendwatch, tiktok-shadowban-check.

She notes in the article that this is a paid partnership with Postiz, which she uses every day. Her closing sentence is a direct quote: Stop paying people to read your brief.