Web2App

Stack · Published 2026-07-23 · 2 min read

How I work with neural networks. Part 3 of 4: creatives

Every Monday I push all the new creatives across every account into testing. That used to mean:

  • open the folder
  • remember what I had already launched
  • assemble the campaign in Meta by hand
  • upload the creatives

The upload is the longest stage, because creatives sometimes take a terribly long time to go through.

A bot handles all of that now.

What the bot does

The bot scans the creative folders on Google Drive for each ad account. The logic is simple: "new" is anything not yet moved into the tested folder, and a creative lands in tested at the moment it is deployed. So the bot always shows exactly what I have not launched yet.

The report it assembles contains:

  • the list of new concepts for each ad account
  • how many videos and how many images there are in each

I get that in Telegram and send it into testing with a single confirmation.

Where the creatives come from

  • Editors on each project work in their own folder on Google Drive.
  • The bot has access to Veo 3 and fal.ai and generates creatives itself on request. It generates images very well and very fast. Video only if you explain the workflow to it in advance — though I think that is a matter of a couple of months, once the models get a little smarter.

A generation request looks like this: "for the sleep tracker concept make 50 images with hooks as different from each other as possible and launch them into the test ad set". Image quality: subjectively, very good looking. Objectively, on every project the AI-generated images perform better than the ones made by designers. I have some ideas about why that is; more next week.

More about the bot in the AI module of User Acquisition

Write to me if you want the same process.

Questions by direct message: @SiberianPaaanda


Original on Telegram

Contact   All articles