Bidding for Influence: Auction-Driven Diffusion Image Generation

Lillian Sun, Henry Huang, Fucheng Warren Zhu, Giannis Daras, Constantinos Costis Daskalakis

Keywords: Diffusion Models, Online Advertising

Abstract: Motivated by online auctions for banner ads, we propose auctions that fractionally allocate the creation of a banner to bidders according to their preferences. Our mechanism elicits bids and textual prompts from the advertisers, and composes them into a score function that drives a reverse diffusion process that generates the banner. Then, it implements Monte Carlo sampling to calculate approximate VCG-based payments to incentivize high-welfare images. Extensive experiments on a diverse 20-prompt dataset with up to 3 agents demonstrate key economic properties. Our mechanism achieves: (1) bid monotonicity; (2) efficiency improvement of up to 20.7% higher welfare than a single-winner VCG baseline; and (3) approximate incentive compatibility, with average regret as low as 7% when deviating from truthful bidding. These benefits are achieved while preserving high image quality. Our study establishes a principled and scalable bridge between auction theory and controllable image diffusion, laying a foundation for economically aligned, multi-stakeholder image generation in advertising and beyond.

Primary Area: alignment, fairness, safety, privacy, and societal considerations

Submission Number: 14031