I am testing the use of neural networks to generate creatives for brands and products. So far, no big budgets, just experiments, but there are already interesting results.

I started by uploading videos with an AI avatar - the first videos gave 1,000 views each, despite the fact that the content was made literally "on my knee". Then the traffic dropped — some videos received only 10 views. It felt like YouTube was filtering content with synthetic faces. But after a couple of days, the videos began to gain from 100 to 1000 views. The algorithms are still difficult to understand, but there is growth.

There is a new version of the neural network from Google Deepmind — VEO 3. This is no longer just video generation, but actually the creation of realistic production-level videos. Moreover, you can set the language (including Russian), style, diction, behavior, and even the "visual role" of the hero. Videos with virtual reporters and "grandmothers" are currently being played — some of them are collecting millions of views.

My test video (https://youtube.com/shorts/fyRsCVszzKM ) generated via VEO — collected 30,000+ views without attachments. This is a great signal that the theme can bring traffic cheaper than classic channels.

The idea: To use neural networks to create video creatives for hypotheses on a daily basis. If something comes in, we'll adapt it to the product. If it comes in steadily, we scale through a team of creators.

If you do it right, you can:

• get viral organic traffic without a budget;
• adapt to specific products/services;
• scale through low-cost performers (there are already many cool creators in advanced groups, but they lack a marketing vision).

The potential is huge. As long as the market is not oversaturated, there is a chance to occupy a niche at the entrance.

Tools:

• Avatars — Heygen (≈$70/month for unlimited).
• Video from text — VEO 3 (via proxy, you need a non-Russian card, 10 generations for free).