This is an Act I tutorial on AI image and creative generation for marketing: the single asset, end to end, from brief to export. We split the work into three jobs and match a tool category to each, then spend most of the time on the part that actually bites you, commercial safety. **The three jobs.** Fast on-brand graphics live in hosted design tools like [Canva Magic Studio](https://www.canva.com/newsroom/news/magic-studio/) and Adobe Express. Original hero art comes from image models: [Midjourney](https://docs.midjourney.com/hc/en-us/articles/27870375276557-Using-Images-Videos-Commercially) (Version 8.1), OpenAI's GPT Image 1.5, [Adobe Firefly](https://www.adobe.com/products/firefly.html), [Ideogram](https://ideogram.ai/), [Google Imagen 3](https://developers.googleblog.com/imagen-3-arrives-in-the-gemini-api/), and [Flux](https://bfl.ai/licensing) from Black Forest Labs. Ad creative with legible baked-in text is its own job, best done by generating a clean hero then adding type in Photoshop or Canva. **Prompting.** Use one structure: subject, medium, style, lighting, framing, mood, palette. Choose aspect ratio before you generate, and leave negative space for copy. Reference images and tools like Firefly Custom Models or Midjourney's character reference handle brand and character consistency. **Commercial safety, the heart of it.** Pure AI images aren't copyrightable; [Thaler v. Perlmutter](https://www.mayerbrown.com/en/insights/publications/2026/03/supreme-court-denies-review-in-ai-authorship-case) is now settled law. [Firefly](https://www.adobe.com/legal/licenses-terms/adobe-gen-ai-user-guidelines.html) is the only major tool with transparent training plus indemnity. Watch trademark and celebrity likeness, follow [FTC disclosure](https://humanadsai.com/blog/ftc-ai-generated-content-disclosure) plus New York's new synthetic-performer law, and understand [C2PA content credentials](https://editorsweblog.org/2026/03/28/ai-watermarking-standards-c2pa-content-credentials-provenance) and the EU AI Act. Lawsuits to track: [Getty v. Stability AI](https://www.ropesgray.com/en/insights/viewpoints/102lvxe/getty-image-loses-copyright-infringement-claim-against-stability-ai-in-uk) and [Andersen v. Stability AI](https://knowingmachines.org/knowing-legal-machines/legal-explainer/cases/andersen-v-stability-ai) (trial September 2026). **The slop trap.** [Visual tells](https://www.opus.pro/blog/ai-slop-aesthetic-12-tells) read as cheap. Fix them with a ten-to-thirty-minute human-in-the-loop edit pass, the same discipline we teach for copy. Plus the [copyable workflow](https://www.digitalapplied.com/blog/social-media-image-sizes-2026-every-platform) from brief to export sizes, and a light cost touch. News up top: Cannes Lions, Canva AI 2.0, Google AI Max, and the OpenAI Partner Network.