At the start, a single LLM subscription and pay-as-you-go generations through an aggregator like fal.ai will suffice: one key and dozens of image and video models.

Step-by-step work

1. Pain points from reviews. Collect 10-20 reviews of a product or similar product and give the LLM this task:

Here are reviews of product [X]. Write down only what is in the text:

1) 10 problems before purchase — customer quotes;

2) 10 results after purchase;

3) Top 3 objections and how to overcome them.

Don't make anything up.

2. Scenarios. For each angle from the matrix:

Offer: [product], GEO: Romania, Language: Romanian, Price 149 lei, Cash on delivery.

Angle: Comparison with a cheaper similar product.

Write 5 scenarios, 15 seconds each: 0-2 seconds — hook (visual + caption up to 6 words),

2-10 seconds — demonstration, 10-15 seconds — price and call to action.

For each scene, provide a prompt to generate an image in English.

3. Static. Generate a scene using a real product photo from your affiliate program as a reference—otherwise, the model will "add" a different product, and your calls will be rejected. You can now safely entrust captions to Nano Banana Pro.

4. Video. Create each scene from a finished image in image-to-video mode, 5 seconds long, with simple camera movement. Three scenes plus a native speaker's voiceover equals a finished 15-second video. Make subtitles large and above the bottom third of the frame, where they are obscured by the interface: many viewers watch without sound.

5. Uploading. Using Meta Ads MCP, the neural network uploads files and creates ads in batches. They appear paused, so you review each one and manually activate it before launching.

Additionally, before uploading, each creative undergoes a short human review: does the product match the real thing, are the hands and faces of the people depicted correct, is the text error-free, is the price the same as on the landing page, and so on. According to colleagues who have set up similar AI-based creative factories, rejecting 1/3 of unsuccessful creatives is a normal part of the automation business process.

At the start, a single LLM subscription and pay-as-you-go generations through an aggregator like fal.ai will suffice: one key and dozens of image and video models.

Step-by-step work

1. Pain points from reviews. Collect 10-20 reviews of a product or similar product and give the LLM this task:

Here are reviews of product [X]. Write down only what is in the text:

1) 10 problems before purchase — customer quotes;

2) 10 results after purchase;

3) Top 3 objections and how to overcome them.

Don't make anything up.

2. Scenarios. For each angle from the matrix:

Offer: [product], GEO: Romania, Language: Romanian, Price 149 lei, Cash on delivery.

Angle: Comparison with a cheaper similar product.

Write 5 scenarios, 15 seconds each: 0-2 seconds — hook (visual + caption up to 6 words),

2-10 seconds — demonstration, 10-15 seconds — price and call to action.

For each scene, provide a prompt to generate an image in English.

3. Static. Generate a scene using a real product photo from your affiliate program as a reference—otherwise, the model will "add" a different product, and your calls will be rejected. You can now safely entrust captions to Nano Banana Pro.

4. Video. Create each scene from a finished image in image-to-video mode, 5 seconds long, with simple camera movement. Three scenes plus a native speaker's voiceover equals a finished 15-second video. Make subtitles large and above the bottom third of the frame, where they are obscured by the interface: many viewers watch without sound.

5. Uploading. Using Meta Ads MCP, the neural network uploads files and creates ads in batches. They appear paused, so you review each one and manually activate it before launching.

Additionally, before uploading, each creative undergoes a short human review: does the product match the real thing, are the hands and faces of the people depicted correct, is the text error-free, is the price the same as on the landing page, and so on. According to colleagues who have set up similar AI-based creative factories, rejecting 1/3 of unsuccessful creatives is a normal part of the automation business process.

At the start, a single LLM subscription and pay-as-you-go generations through an aggregator like fal.ai will suffice: one key and dozens of image and video models.

Step-by-step work

1. Pain points from reviews. Collect 10-20 reviews of a product or similar product and give the LLM this task:

Here are reviews of product [X]. Write down only what is in the text:

1) 10 problems before purchase — customer quotes;

2) 10 results after purchase;

3) Top 3 objections and how to overcome them.

Don't make anything up.

2. Scenarios. For each angle from the matrix:

Offer: [product], GEO: Romania, Language: Romanian, Price 149 lei, Cash on delivery.

Angle: Comparison with a cheaper similar product.

Write 5 scenarios, 15 seconds each: 0-2 seconds — hook (visual + caption up to 6 words),

2-10 seconds — demonstration, 10-15 seconds — price and call to action.

For each scene, provide a prompt to generate an image in English.

3. Static. Generate a scene using a real product photo from your affiliate program as a reference—otherwise, the model will "add" a different product, and your calls will be rejected. You can now safely entrust captions to Nano Banana Pro.

4. Video. Create each scene from a finished image in image-to-video mode, 5 seconds long, with simple camera movement. Three scenes plus a native speaker's voiceover equals a finished 15-second video. Make subtitles large and above the bottom third of the frame, where they are obscured by the interface: many viewers watch without sound.

5. Uploading. Using Meta Ads MCP, the neural network uploads files and creates ads in batches. They appear paused, so you review each one and manually activate it before launching.

Additionally, before uploading, each creative undergoes a short human review: does the product match the real thing, are the hands and faces of the people depicted correct, is the text error-free, is the price the same as on the landing page, and so on. According to colleagues who have set up similar AI-based creative factories, rejecting 1/3 of unsuccessful creatives is a normal part of the automation business process.

At the start, a single LLM subscription and pay-as-you-go generations through an aggregator like fal.ai will suffice: one key and dozens of image and video models.

Step-by-step work

1. Pain points from reviews. Collect 10-20 reviews of a product or similar product and give the LLM this task:

Here are reviews of product [X]. Write down only what is in the text:

1) 10 problems before purchase — customer quotes;

2) 10 results after purchase;

3) Top 3 objections and how to overcome them.

Don't make anything up.

2. Scenarios. For each angle from the matrix:

Offer: [product], GEO: Romania, Language: Romanian, Price 149 lei, Cash on delivery.

Angle: Comparison with a cheaper similar product.

Write 5 scenarios, 15 seconds each: 0-2 seconds — hook (visual + caption up to 6 words),

2-10 seconds — demonstration, 10-15 seconds — price and call to action.

For each scene, provide a prompt to generate an image in English.

3. Static. Generate a scene using a real product photo from your affiliate program as a reference—otherwise, the model will "add" a different product, and your calls will be rejected. You can now safely entrust captions to Nano Banana Pro.

4. Video. Create each scene from a finished image in image-to-video mode, 5 seconds long, with simple camera movement. Three scenes plus a native speaker's voiceover equals a finished 15-second video. Make subtitles large and above the bottom third of the frame, where they are obscured by the interface: many viewers watch without sound.

5. Uploading. Using Meta Ads MCP, the neural network uploads files and creates ads in batches. They appear paused, so you review each one and manually activate it before launching.

Additionally, before uploading, each creative undergoes a short human review: does the product match the real thing, are the hands and faces of the people depicted correct, is the text error-free, is the price the same as on the landing page, and so on. According to colleagues who have set up similar AI-based creative factories, rejecting 1/3 of unsuccessful creatives is a normal part of the automation business process.