Adobe Firefly, built on Adobe Stock’s licensed library, openly licensed content, and public domain material, produces results that are “competent,” “constrained,” and “safe,” while Midjourney, trained on scraped data from the open internet, delivers output that is “more striking, more original, more alive.” Matt Supple, a high-end retoucher and creative imaging specialist with 25+ years across New York, Sydney, and New Zealand, and founder of Fred NY, tested both tools using the same references, briefs, and intent. He found that Firefly’s legal approach—where every contributor is compensated and every image can be traced to a license—results in a smaller, safer sandbox that limits artistic capability. Midjourney’s opposite bet, training on whatever crawlers can reach without published datasets or licensing claims, optimizes for output quality and is currently “simply winning on craft.” Supple warns that artists competing against users of scraping-trained models face an unfair field, as such work may be built from uncompensated, unconsented use of other creators’ portfolios. He notes that the “ethical” option is currently the weaker artistically, and the market rewards results, not ethics. Even Firefly is not entirely clean: in 2024, it emerged that a portion of its training set included AI-generated images.