Background: Alcohol is the 3rd leading cause of cancer in the US, linked to seven types, but only 30% of Americans are aware. Health warning messages can effectively communicate this risk, though the mechanism remains unclear. Objective: This study examined if Al-generated pictorial messages (graphic, neutral) surpass text messages and explored mediators affecting motivation to reduce alcohol use. Methods: In March 2025, a national online crossover experiment with 596 U.S. adult alcohol users (49.3% female; aged 21 +) evaluated 8 alcohol-cancer risk messages (4 general, 4 cancer-specific), each at 3 intensities: text, neutral pictorial, and graphic pictorial (n = 24). Each participant viewed two randomly assigned messages out of the 8, each at 3 intensities (n = 6), in a counterbalanced order. The primary outcome was motivation to reduce alcohol in the next 30 days. Mediators were attention, perceived learning, believability, fear, and avoidance. We used multilevel structural equation modeling to estimate parallel and serial mediation pathways, comparing graphic vs. text, neutral vs. text, and graphic vs. neutral, while adjusting for age, sex, alcohol level consumption (moderate, heavy), alcohol social norms, and message order and specificity (general vs. cancer-specific). Results: Graphic and neutral pictorials drew more attention, perceived learning, believability, and fear than text messages (p<0.01 for all). Graphic pictorials also increased avoidance relative to both neutral and text messages. In parallel mediation models, graphic pictorials had a significant total effect on motivation (beta = 0.91, SE = 0.12), with 93.6% of the effect mediated through indirect paths, primarily through attention (47.2%) and fear (24.3%). Neutral pictorials also increased motivation compared to text messages (beta= 0.47, SE= 0.10), with 99.3% of the effect mediated, mainly by attention (56.8%) and believability (16.7%). Graphic pictorials outperformed neutral ones (beta= 0.439, SE = 0.092, p < 0.001 for all), with 87.6% of the effect mediated mainly through attention and fear. Serial mediation analyses confirmed attention as the primary upstream driver. For graphic vs. text comparisons, sequential paths through attention and believability (31.8%), attention and fear (22.9%), and attention and learning (17.6%) significantly predicted motivation. Similar patterns were observed in comparing neutral vs. text and graphic vs. neutral. Across all models, the direct effects of message intensity on motivation were not significant. Conclusions: Al-generated pictorial alcohol warnings, particularly graphics, substantially increase motivation to reduce alcohol consumption by enhancing attention, fear, perceived learning, and believability. These findings highlight Al's role in improving alcohol-cancer risk communication.