AI-Powered Ad Creation with ChatGPT 4o Image Generation

Reference images matter more than complex prompts

Instead of crafting elaborate text prompts, provide clear inspiration ads and product images. The model's multimodal understanding makes prompting straightforward.

Why: ChatGPT 4o is a language model first, so it has better intent understanding and creativity than pure image models like Midjourney

"the important thing I found in creating good ads, are actually the inspiration images and the product images that you provide"

Well-known brands don't need reference images

For established brands like Nike, Adidas, or Ridge wallets, you can prompt directly without reference images because the model has extensive training data.

Why: The model has seen millions of examples of major brands through internet training data

"it has a really good understanding of well-known concepts, ideas, brands. So it knows Nike really well. It knows Adidas really well"

Start new chats for prompt iterations

When adjusting prompts, always start a fresh conversation instead of continuing the same thread to avoid quality degradation.

Why: The model uses the previous generated image as reference, which contains inherent flaws that compound over iterations

"if you want to adjust a prompt, start a new chat and make the adjustments to your prompt in that new chat"

Use Sora's explore page for prompt inspiration

Browse Sora's public gallery to see successful prompts and remix existing concepts for your own use cases.

Why: Provides proven prompt structures and aesthetic ideas that shorten the creative process

"I highly recommend people check out Sora... it's a really good way to understand how people are prompting"

The ad is the targeting

Create multiple variations of ads with different demographics (age, gender, ethnicity) to find what resonates with specific audiences.

Why: Small visual differences can dramatically impact ROAS - the difference between 1.6x and 2.8x return on ad spend

"marketers have these saying nowadays that the ad is the targeting"

Use 'photorealistic' keyword for realistic results

Explicitly include 'photorealistic' or 'ultra realistic' in prompts to avoid cartoonish or unrealistic outputs.

Why: Without this keyword, the model may default to stylized or artistic interpretations

"using the ultra realistic or photorealistic keyword is really important for creating photorealism images"

Generate image-only outputs with clear instructions

Use phrases like 'generate the image' to ensure you get visual outputs rather than text-based mockups.

Why: Prevents the model from creating text representations instead of actual images

"if you just say generate the image, you're going to consistently get image results"

Ethical remixing of proven ad concepts

Using inspiration from successful ads is standard practice - AI just automates what agencies already do manually.

Why: Ad inspiration and template usage was already widespread; AI makes it accessible to everyone

"this was already happening. It was just happening physically. People were looking for inspiration. They were using templates"