All articles tagged with #image generation
Google rolled back its Nano Banana 2-powered AI image-generation feature in Google Earth a day after launch due to misinformation concerns, as users created realistic but false imagery of war zones, refugees, and other sensitive scenes; Google says it will implement stronger guardrails and has added a SynthID watermark to AI-generated images to aid identification.
Google Earth on the web now supports Nano Banana 2, the latest AI image-generation upgrade from Google Gemini, letting users prompt aerial imagery to create historical reconstructions, future visions, or planning-related visuals. Use cases include infographics about landmarks, visualizing personal projects, or imagining Pompeii-era scenes; the feature is rolling out globally to all Google Earth on the web. As with AI, results may be imperfect or not fully authentic, and there may be no disclosed usage limits.
Meta is discontinuing Muse Image, an AI tool that used public Instagram content as reference for generative imagery and defaulted to enabling it. The move comes after strong backlash from actors and other public figures, with the SAG urging users to protect their likeness. Meta said the feature missed the mark and will no longer be available, highlighting ongoing concerns about AI image rights and privacy.
Meta's Muse Image introduces an invisible Content Seal watermark and a detector, but Reuters found the detector correctly flags full-size Muse Image images yet drops to about 55% accuracy after cropping, highlighting a gap between generation and detection. Privacy concerns over using public-profile photos led Meta to remove that feature, as the company pursues future tools like Muse Video.
Meta unveiled Muse Image, a feature within Meta AI that lets users generate images using the likenesses of people with public Instagram accounts by tagging them; users aren’t notified when their likeness is used and accounts must opt out to prevent usage. Privacy advocates and groups like Public Citizen, CAA, and SAG-AFTRA criticize the approach and call for affirmative consent or stricter controls, while Meta says protections exist and is evaluating next steps. No policy changes have been announced yet as the discussion continues.
Meta's Muse Image lets users generate AI images from other people's public Instagram profile photos without notifying them, triggering privacy and consent concerns from advocates and regulators. Meta says users can opt out in settings, and the feature will expand across its apps, all amid a crowded AI image market and ongoing scrutiny of non-consensual imagery.
Meta has launched Muse Image, the first AI image generator from its Superintelligence Labs, powering image tools across the Meta AI app, Instagram, and WhatsApp (with Facebook and Messenger coming soon). Described as 'agentic' and working with the Muse Spark language model to reason, search, and plan before generating, Muse Image also allows you to @mention Instagram accounts to pull in likenesses (with privacy controls), transform images, redesign rooms from web images, and draw edits on photos—plus it will power 30 new AI effects for Instagram Stories.
Google’s Nano Banana 2 Lite, the fast, low-cost member of the Gemini 3.1 family, generates images in about 4 seconds (vs ~20 seconds for the standard Nano Banana) and costs roughly $0.034 per 1K images, making it ideal for rapid prototyping. It trades some fidelity for speed—text accuracy, infographic data, and character consistency can suffer—and outputs include SynthID watermarks. Available in Google AI Studio and the Gemini API (Flash-Lite; Omni Flash for video expanding), Nano Banana 2 Lite offers cheap, quick visuals while higher-fidelity Pro/standard models remain for more exact work.
Google is expanding its personalized intelligence feature in the Gemini app to all eligible US personal accounts, allowing image-generation requests to reference data from Gmail and Google Photos (in any Gemini-supported language). However, Google Photos connectivity to Gemini isn’t available in Texas and Illinois due to state privacy laws.
PCMag's Ruben Circelli tested Microsoft's four MAI models announced at Build 2026—MAI-Thinking-1, MAI-Image-2.5, MAI-Transcribe-1.5, and MAI-Voice-2—now in limited preview and free on the Playground, with mixed results. MAI-Thinking-1, a first-of-its-kind reasoning model, underwhelmed vs. Claude’s Sonnet and lacks internet access; MAI-Image-2.5 is an incremental upgrade but text issues keep it behind Gemini Nano Banana Pro; MAI-Transcribe-1.5 is adequate but not best-in-class; MAI-Voice-2 sounds robotic and unconvincing. Overall, the lineup is competent but not compelling enough to warrant a prime-time spotlight, though Microsoft expects rapid improvements.
In a 30-factor, real-world test, OpenAI's ChatGPT Images 2.0 scored 97% (150 points) to Google's Gemini Nano Banana's 85% (131 points), with Images 2.0 delivering stronger context-aware image generation and better face fidelity, while Nano Banana struggled with text prompts, rendering details, and privacy-related concerns raised by unintended inclusion of prompt history in some results.
OpenAI’s Images 2.0, branded as the first image model with thinking capabilities, promises advanced text rendering and strategic design, but its demos—especially AI-generated football posters—have reignited social‑media chatter that graphic design is dead, with critics arguing outputs tend toward sameness and lack human soul, while proponents contend AI can augment designers rather than replace them.
Gary Marcus argues that ChatGPT’s new image engine is visually impressive but does not demonstrate true understanding. He points to labeling errors in bike diagrams and odd results from a custom tandem-bike prompt as evidence that the system can imitate understanding without grasping how parts function. The piece emphasizes that regurgitating images isn’t the same as real comprehension in AI.
Axios tests ChatGPT Images 2.0, which adds wider aspect ratios, a standard mode for all users and a paid 'thinking' mode with built-in reasoning. It produced a range of visuals—from personalized sympathy cards and old-style photo albums to trading cards and infographics—demonstrating strong capabilities but with occasional polish gaps and longer render times due to reasoning; a faux newspaper task briefly pulled older headlines before updating to current stories, signaling promising but imperfect results for business use.
OpenAI released ChatGPT Images 2.0, a more capable image-generation model that can output multiple images from a single prompt, render text more accurately, and handle non-English languages, using ChatGPT’s reasoning to fetch recent information and support customizable aspect ratios; English outputs are strongest, with improvements expected for other languages.