Teardown No. 48 · Short-form video

How Vinayak Ramesh made over 300 pins in 90 days for three Substack subscribers, and shelved his Claude Code video pipeline

Vinayak Ramesh gave a brand new Pinterest account 90 days, posting three pins a day from mid-May to see if it could drive traffic to his Medium and Substack. About a month in, he built a pipeline with Claude Code and the Gemini API. With one command, it turned an old article into a short vertical video with a ready-to-paste caption. Each video cost between 80 cents and a dollar. He stopped the videos because they started to look the same and traction never showed up, then went back to three static pins a day. Over the three months, impressions rose from 4,885 in month one to 5,370 in month three. He reports "Three Substack subscribers directly attributed to Pinterest over three full months." He is keeping the pins going with lower expectations.

Vinayak RameshVerified Oct 22, 2026Markdown copy

The problem

He started pinning “purely to see if it could drive traffic back to Medium and Substack newsletter.” In the guest post, he describes sitting on 87 articles that “had done their job once and then just… stopped.” Turning one of them into a video meant a new script, new visuals and new captions every time. He writes: “I don’t have three extra hours a night for that. Not with a full-time job.”

The blueprint (as he describes it)

  1. Start a new account and pin daily. In mid-May he started a brand new Pinterest account and posted three pins a day. He committed to 90 days before deciding whether to continue.

  2. Add a one-command video pipeline. About a month in, he built a pipeline in Claude Code. Running npm run create -- --daily takes 8–15 minutes and, he says, “needs zero input from me.” The steps run like this:

    • Claude Code reads one of his articles.
    • Gemini writes a 50–60 second script with 9–11 scenes.
    • Imagen 4 generates an illustration for each scene.
    • Remotion renders a 1080×1920 MP4 with animated captions, Ken Burns motion and scene transitions.
    • A metadata generator writes the TikTok caption, hashtags split across the caption and first comment, and a Pinterest title and description.

    Two files come out: an .mp4 and a .txt post package. He puts his active time at “30 seconds to check the output.”

  3. Pick articles by rule. Articles come from his Obsidian vault. The picker takes never-used articles first, then articles used more than 14 days ago, and recently used ones only as a last resort. Reused articles get a “freshAngle” flag so the script takes a different angle. Selection is weighted 70% Medium and 30% Substack.

  4. Fix what broke. He forces JSON output from Gemini 2.5 Flash and raises the output token limit to stop truncated scripts. He never puts hex codes in image prompts and ends every prompt with a zero-text instruction. Images are converted to base64 before Remotion renders, because file paths produce black frames.

  5. Rotate in product mentions. A history tracker returns true every fifth video, and on those runs the caption includes a Gumroad product mention.

  6. Try to stop the videos looking alike. The first version cycled 10 visual types in a fixed rotation. He expanded it to 20 types, chosen so adjacent scenes differ, and added camera-angle and background layers to every image prompt.

  7. Post mainly to Pinterest, by hand. He copies the Pinterest title and description from the .txt and uploads the video himself. He chose Pinterest because “a video is basically dead within 4 hours” on Instagram and TikTok, while “a decent pin keeps sending traffic back to an article for months.” He also recommends keeping a manual notes file on what worked, since “Nothing in the code does this part for you.”

  8. Shelve the videos. He says the plan of one video a day “worked”, but posting one every day “started to prove costly”. The bigger issue, in his words: “the videos started looking like the same. Same visual beats. Same energy every single time.” Traction never showed up, so he stopped and went back to three static pins a day.

  9. Check the export and keep going at lower stakes. At the 90-day mark he pulled the analytics export “instead of trusting the chart in my head.” He is keeping three pins a day but is “done treating this as the centerpiece of my content strategy.” Next, he wants to test pinning outside his usual niche.

Stated results

  • Monthly impressions: 4,885 in month one, 5,096 in month two and 5,370 in month three. He calls it “A small climb, about 3% then another 5%.”
  • Most recent 30 days:
    • 5.75k impressions, up 14%.
    • Total audience of 2.01k, up 31%.
    • 261 saves.
    • Outbound clicks down 27% to 24.
  • Boards: His Digital Products, Pinterest Marketing Tips and Substack Growth Tips boards convert saves at between 5.5 and 5.9%. His two biggest boards by impressions, general content creation and side hustle ideas, convert under 4% each.
  • Newsletter: “Three Substack subscribers directly attributed to Pinterest over three full months, plus 20 to 30 profile visits.”
  • Output: Over 300 pins, posted every day for three months.
  • Video pipeline: Each video cost between 80 cents and a dollar. The build took one session of about 2–3 hours. No views, clicks or subscribers from the videos are stated. He says “Traction never showed up.”
  • Medium traffic from Pinterest: None stated.

Caveats

  • The primary post is on his Buy Me a Coffee page and asks readers to buy him a coffee. He also sells templates, prompt packs and playbooks on Gumroad, and the pipeline adds a Gumroad product mention to every fifth video caption.
  • The pipeline write-up is a guest post on The Digital Creator, the newsletter run by Sharyph, a past Postby teardown subject. The issue opens with an intro by the host and is sponsored by Revid.AI, an AI video tool that is not part of his pipeline. The host also promotes his paid playbook and a premium upgrade.
  • The guest post was written while the pipeline was running. Its claim that “Every run looks different from the last one” conflicts with his later reason for stopping.
  • The figures do not line up across the two posts:
    • Video length: The experiment post says 40 to 50 second videos. The guest post says 50–60 second scripts, and the first-run example was 62 seconds.
    • Active time: The guest post gives 30 seconds in one place and “under fifteen minutes” in another.
  • His manual baseline of 45–90 minutes per video is his own estimate.
  • Outbound clicks fell even as impressions rose. He notes people are saving pins “far more than they’re clicking through.”
  • His series includes a post on “The mid-summer platform-wide traffic crash everyone was seeing.” The supplied source does not say how that affected his numbers.
  • The experiment post does not say how the static pins were made. His Pinterest account is not named in the sources.
  • Handles: Substack @vinayakramesh1 and Medium @thevinayakramesh (verified 2026-10-22: linked from each other and from his Buy Me a Coffee page). No X, LinkedIn, Bluesky, Threads, Instagram, Mastodon or TikTok profile could be verified.

Sources

Want agents like these working your distribution? See the Agent Stack Teardown. Get new teardowns weekly: subscribe (double opt-in).

Get one teardown a week.

Same format, a new operator every issue. Double opt-in.