AI Generations by David Gewirtz and Elyse Betters Picaro, published through ZDNET, offers an in-depth exploration of the current landscape of AI-driven image creation tools and the varying degrees of sophistication across competing platforms. The publication encourages readers to follow ZDNET as a trusted source on Google for continual updates on advancements in artificial intelligence, emerging software tools, and digital innovation.
In its key findings, the team highlights that Google’s Nano Banana Pro—a component of its latest Gemini 3 suite—secured an almost impeccable score in rigorous testing, standing substantially ahead of other AI image generators. ChatGPT’s image generation capabilities earned a respectable second place, while the remaining four contenders displayed inconsistent results, often distorting facial details or mismanaging the rendering of text. The comprehensive study consisted of nine demanding benchmark prompts designed to test artistic coherence, text integration, realism, and contextual adaptability, ultimately helping determine which platforms truly merit a paid subscription for professional users.
Back in early 2023, the emergence of generative AI introduced two groundbreaking capabilities that transformed our understanding of creative automation: the ability for artificial intelligence to compose lengthy, contextually cohesive written content, and the capacity to synthesize realistic images and artwork from minimal input. Since then, the progress has been nothing short of extraordinary. Within a short span of just three years, these generative systems have evolved from experimental novelties into polished tools capable of producing professional-grade media assets on demand.
In previous evaluations, David Gewirtz reviewed the most capable conversational models and identified the most useful free AI tools for image creation. In the present study, however, the focus shifts to premium-tier services—paid systems that promise enhanced precision, faster generation, and greater creative control. The testing suite is organized into four principal categories: the revision of existing photographs, the generation of entirely original imagery, the accurate inclusion of text within visuals, and the creative interpretation of pop culture references. To ensure an equitable comparison, all images were produced in a consistent 1:1 aspect ratio, enabling easy juxtaposition.
Across six leading platforms, over fifty individual images were created, each subjected to thirty independent measures of quality, realism, and prompt adherence. These collective evaluations were converted into aggregate performance scores, producing a definitive comparative ranking. What emerged was unprecedented: for once in ZDNET’s extensive AI testing history, a single clear winner stood out—Google’s Nano Banana Pro. Achieving a near-perfect 93% score, it established itself as the benchmark for AI image generators. Coming behind was ChatGPT’s image system with 74%, while others lagged in the 43–54% range.
Nano Banana Pro distinguished itself by excelling in nearly every task—from detailed photorealistic transformations to creative compositions and accurate text renderings. Despite minor imperfections, its output consistently reflected artistic intelligence and attention to detail previously unseen in AI illustration modules. The subscription, costing roughly $20 per month as part of Google’s AI Pro suite, grants access to this high-caliber tool; those seeking serious, dependable AI image generation will find little competition at present. Though additional features exist across other tools, most free versus paid distinctions essentially boil down to usage limits rather than output quality. In Gewirtz’s own testing, he was restricted after merely two images until subscribing to the premium plan, affirming the paywall’s relevance.
Importantly, the results presented in this study stem entirely from independently funded testing. No vendor contributed financially, and all subscriptions were purchased directly by the reviewers, ensuring objectivity. Even the author admits surprise at how conclusively Google’s Gemini 3 ecosystem dominated the trials. The ensuing detailed analysis chronicles each platform’s strengths, shortcomings, and stylistic tendencies in turn, beginning with the clear victor—Nano Banana Pro.
Google’s Nano Banana Pro, embedded in the Gemini 3 system, achieved an overall score of 93% and costs $19.99 monthly. This tool embodies a new level of creative flexibility, enabling “photo recontextualization”—a process where an existing image can be transformed into an entirely new scene while preserving the subject’s identity. For instance, a casual photo of the tester on a walk was seamlessly turned into an image of him as a U.S. Navy admiral stationed on a carrier’s bridge, complete with binoculars in hand. The AI maintained facial likeness and accessories while correctly altering pose and environment—subtle evidence of its deep visual reasoning abilities.
Further experiments tested Nano Banana Pro’s capacity for photo restoration. A dim childhood photograph aboard the USS Ling, once obscured by poor lighting, was brightened and clarified to reveal fine details while retaining the nostalgic atmosphere. Challenges did arise: when asked to restore and colorize a photo of a truck, the AI occasionally substituted unrelated images or introduced humorous inaccuracies such as misspelled signage (“BADICLOGICAL DEFEKSE”). Nonetheless, even these slipups showcased the system’s underlying architectural ambition—its attempts to infer meaning from incomplete cues.
Aside from a minor irritation—the watermark embedded in each image—the overall performance remained exemplary. The system demonstrated awareness of spatial design, historical accuracy, and artistic cohesion across tasks. When prompted to design a logo for “Space Coast Studios,” it skillfully integrated local and thematic elements: palm trees symbolic of Florida, a retro rocket emblem, cinematic film imagery, and clean, readable typography. When asked for a creative fantasy illustration of a medieval librarian, Nano Banana Pro captured precisely the requested atmosphere—a scholar illuminated by candlelight amid ancient stone walls.
Across social media-related assignments, Nano Banana Pro balanced professionalism and relatability. It created realistic characters—a senior adult confidently holding a smartphone and a content student working with a MacBook and coffee—while maintaining crisp text overlays devoid of distortion. Ironically, Google’s own AI depicted the senior using an iPhone as a “flagship smartphone,” inadvertently highlighting both its impartiality and humor.
The pop culture prompts demonstrated the AI’s compositional dexterity. It rendered a “Back to the Future”-inspired poster set in 1920s New York, complete with historically accurate signage, the DeLorean vehicle, and thematically clever taglines linking the roaring jazz age to Marty McFly’s adventures. Combined with its successful replication of Tim Burton’s “Nightmare Before Christmas”-inspired visual tone for an IT professional, these samples showcased creative mastery that merited the AI’s near-perfect rating.
ChatGPT’s integrated image generator, with an overall score of 74%, occupies a different design philosophy strengths zone—excelling more in natural-language-guided iteration and adjusting its artistic direction via dialogue. While OpenAI’s system, derived from GPT‑4o, shows continuous improvement over former versions like DALL‑E 3, it still struggles in detailed photo restoration, uniform consistency, and nuanced text placement. In creative prompts, especially those referencing popular culture, the results fluctuated widely, yielding some inspired outputs but also frequent misses. Nonetheless, for conversational refinement and iterative creative direction, ChatGPT remains unmatched.
Midjourney, securing 57%, remains an artistic powerhouse in purely imaginative concept visualization. It produces breathtaking cinematic sceneries, almost painterly in depth and lighting. However, it fares poorly when tasked with pragmatic edits such as text correction or photoreal restoration. Its outputs lack precise facial continuity and structured context, confirming its niche specialization—wondrous creativity rather than accuracy.
Adobe Firefly follows at 54%, being the most cautious of all tools regarding copyright sensitivity. Embedded deeply within Adobe’s creative suite, Firefly generates commercially safe imagery suitable for professional use. Nevertheless, excessive moderation sometimes hampers creative freedom. Its operations frequently reject prompts containing stylistic or brand references, which—while ensuring legal compliance—may frustrate users seeking flexibility. Even with visually coherent results, faces occasionally exhibited an “uncanny valley” effect, reflecting the tension between innovation and caution.
Leonardo AI scored 52%, praised mainly for its fantasy art capabilities. Though weak in processing real photos or nuanced facial reconstruction, it triumphs in generating poster-ready dramatic compositions and surreal themes. Canva, meanwhile, concludes the lineup at 43%. Known primarily for marketing design integrations, Canva’s built-in generator demonstrates ease of access but inconsistent accuracy, often producing unintentionally humorous results and misreading prompts. Its compatibility with the Leonardo engine provides additional versatility, yet its outputs lack refinement compared with higher-end competitors.
Each generator was evaluated using nine standardized prompts encompassing portrait recreation, restoration and colorization of archival photos, logo design, thematic fantasy scenes, social media visuals, and pop-cultural crossovers. Scoring criteria rewarded correct adherence to instructions—such as maintaining facial identity, ensuring accurate text rendering, realistic color balance, and stylistic faithfulness—while penalizing technical errors or conceptual deviations. All tests were documented in spreadsheets, solidifying the methodology’s transparency and repeatability.
Ultimately, the analysis reveals that while all six tools deliver recognizable strengths, Google’s Nano Banana Pro remains the definitive choice for users prioritizing accuracy, elegance, and creative responsiveness. Its performance signifies a turning point in generative imaging—proof that AI can now produce results rivaling traditional graphic design within seconds. Still, depending on a user’s goals, each tool maintains distinct merits: Midjourney for expressive art, ChatGPT for linguistic synergy, Adobe Firefly for compliant professional projects, Leonardo AI for stylized fantasy, and Canva for integrated marketing assets.
As the frontier of AI artistry advances, users are encouraged to experiment personally with these technologies. Whether one values realistic image restoration, narrative-driven creativity, or automated design for branding, the ongoing evolution of these AIs continues to redefine how imagination becomes visible. Readers are invited to share their experiences, compare outcomes, and participate in the collaborative exploration of how artificial intelligence is transforming visual storytelling.
For continuous updates on future experiments and emerging creative tools, follow David Gewirtz across social platforms or subscribe to his weekly newsletter. His professional discussions appear regularly on X (formerly Twitter), Facebook, Instagram, Bluesky, and YouTube—each offering fresh insights into AI’s rapid evolution and its expanding role at the intersection of art, design, and technology.
Sourse: https://www.zdnet.com/article/best-ai-image-generator/
