BrandDiffusion: Multimodal Personalized Marketing Visual Content Generation

Creating visual content such as product advertisements for marketing purposes has attracted research attention recently. Traditionally, such visuals showcase the product against a specific backdrop while adhering to a consistent corporate style to maintain brand visual identity. We present BrandDiffusion, a marketing image generation framework that leverages the power of pre-trained Stable Diffusion models while allowing designers to generate ideas by adapting prompt-controllable models to capture the intended marketing style. In addition, it is also capable of automatic product placement which incorporates harmonization of the product image with the generated background, guided by saliency detection, BLIP captioning, and SDEdit denoising process. We showcase the capabilities of the proposed framework through quantitative and qualitative evaluations, highlighting the quality of the generated content, its relevancy to user prompts, and a strong brand coherency.