OMR Education · OMR Education / Rolf Hermann / Andre Alpar / Tarek Müller

Profi-Prompting statt AI-Slop: KI-Bildgenerierung im Marketing mit Jan und Karoline Gericke von Awed Studio

·49 min·3 clips

Jan and Karoline Gericke explain how their agency combines customer, creative, and tech perspectives to tackle AI image generation in marketing.

1. OMR Education hosts Jan and Karoline (Caro) Gericke from Awed Studio, a Hamburg-based creative agency specializing in AI-driven marketing production. 2. Jan Gericke has 20 years of advertising industry experience across agencies and brand marketing, including a lifestyle company on the client side; Caro was formerly a marketing manager at a smaller tech company. 3. The episode's core argument is that marketing teams fail with AI image generation not because the technology is weak but because they layer it onto old processes instead of redesigning workflows from scratch. 4. Caro defines AI slop as the 2025 U.S. word of the year meaning content produced without a concept, often through click workers entering mistranslated prompts into image generators purely for engagement-farming attention. 5. She distinguishes AI slop from intentional AI art like Italian Brainrot, which creates something new rather than poorly imitating reality, and argues brands can use even slop aesthetics deliberately if they fit the brand and audience. 6. The first decision framework discussed is when to use real-life production versus AI versus a hybrid: experience-forward campaigns (a motorcycle ride through the Eifel) work better in real life; fantastical or technically impossible scenarios (a car on the moon) are ideal for AI. 7. The Coca-Cola Christmas AI spot is cited as a case where the technology was well-executed but the creative decision was questionable — the brand is built on conveying real human feeling, and audiences noticed. 8. Premium brand backlash is a concrete risk: Caro gives Gucci as an example where AI-produced social media posts generated negative reactions because audiences felt the brand was cutting corners. 9. Jan notes that Otto, a major German e-commerce retailer, has launched an initiative to produce all e-commerce product images with AI, with a dedicated internal team — but acknowledges the initial infrastructure investment is high. 10. The scalability argument is nuanced: AI does not automatically save money; the initial workflow setup is expensive, and for mid-sized companies the investment may not pay off unless the output volume is high enough. 11. Jan presents the technical architecture of diffusion models: they are trained on tagged image datasets, learn to transform images into noise and back, and generate new images by de-noising a random seed — which is why every output is unique and why copyright is structurally addressed. 12. The five-finger problem is explained: diffusion models do not count fingers, they form probabilities — since each finger makes the next finger more likely, hands end up with too many. Newer models like Nanobanana Pro check output before delivery, which partially solves this. 13. Jan recommends stopping image iteration after three to four rounds because each recompression cycle introduces pixel errors and color drift, making it better to restart from scratch. 14. Caro presents the 5-W prompting method — What medium, Who is the subject (with age, ethnicity, specific features), Where (setting), How (lighting, time, century), Which style — as the core framework for structured AI prompting aligned to a brand. 15. The system-prompt layer is introduced: placing an LLM (such as ChatGPT, Gemini, or Claude) in front of the image AI to translate brand inputs into correctly structured prompts, enabling scalable high-volume production. 16. Jan explains the three types of prompts: the system prompt (a persistent briefing that tells the AI model its role and constraints), the user prompt (the 5-W structured input), and the context prompt (for editing specific details in an existing image). 17. Inpainting is explained as the technique of erasing a specific area of a generated image and re-generating only that region — useful for small fixes but not guaranteed to improve the overall image. 18. The episode has a workshop-style format, with the hosts walking through live examples including a one-eyed cat on a birch tree, making the technical concepts concrete and demonstrating the iterative prompting process in real time. 19. Marketing professionals, creative directors, and in-house brand teams evaluating AI image generation for campaign or e-commerce production will find the most actionable content in this episode. 20. Listeners without marketing or creative production context, or those already deeply technical in generative AI, will find the material either too introductory or too brand-strategy-focused for their needs.

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from OMR Education