AI-generated images for ads — Midjourney, Sora, Flux
A year of generative ads. Patterns that work and the visible AI smell that ruins campaigns.
Ad campaigns now routinely include AI-generated images. The naive approach — generate, run, fail — burned a lot of budgets in 2024-2025.
What we've learned:
- AI-generated faces still feel uncanny in display ads. Use silhouettes or partial faces.
- Text in images is generated wrong 80% of the time. Add text in post.
- Specific products: AI gets details wrong. Use AI for backgrounds and atmosphere, real photos for the product.
- Style consistency across a campaign needs strong reference images and same model version.
What works best: hero banners with abstract or stylised imagery, where photorealism would feel cold.
Tools we use: Midjourney for stills, Flux for photoreal, Sora for short video clips under 8 seconds.