New models and features in AI image, video, voice and music generation, from Midjourney, Sora, Veo, Suno and more.
Google just rolled out its flagship text-to-image model, Imagen 3, to all U.S. users via ImageFX alongside a detailed research paper. While users praise its photorealistic textures and crisp typography, the model's ultra-strict safety guardrails are already triggering frustration over blocked harmless prompts.
xAI just rolled out AI image generation to Grok for X Premium subscribers, and it is instantly causing chaos across the platform. Unlike OpenAI or Google, Grok freely generates recognizable politicians doing drugs, wearing Nazi uniforms, or holding weapons, refusing almost nothing except outright full nudity. Why it matters: With U.S. elections looming and EU regulators already investigating X, Musk's hands-off approach turns mainstream AI safety norms completely upside down.
Rendora just launched a web platform that turns written scripts into fully animated 3D avatar videos in seconds. Users can pick characters and environments to generate footage complete with realistic lighting and cinematic angles, all fully customizable afterwards. It basically cuts out the need for real actors or complex 3D animation pipelines.
OpenAI just unlocked its DALL-E 3 image generator for free ChatGPT accounts, giving non-paying users up to two creations per day. Previously restricted to $20-a-month Plus subscribers, the feature is rolling out now, letting casual users turn conversational ideas into visuals without paying a dime.
ByteDance just launched its Jimeng AI app on the Apple App Store in mainland China, following its desktop and Android rollout. The tool lets users generate images and videos directly from text prompts, offering up to 80 images and 26 videos for free before charging 69 yuan ($9.70) per month. Why it matters: ByteDance is moving quickly to scale generative video to everyday creators, even as access remains limited to China.
Amazon just launched Titan Image Generator v2 on AWS Bedrock. The upgraded model adds reference-image guidance, background removal, color palette conditioning, and intelligent foreground segmentation to keep brand visuals consistent. AWS also maintains its copyright indemnification policy for enterprise users.
Stability AI just launched Stable Fast 3D, a rapid reconstruction model that creates complete 3D assets from a single image in just 0.5 seconds. Built on TripoSR, it slashes inference time from SV3D's 10 minutes down to sub-second speeds while generating meshes and material parameters. Weights and code are open-sourced on Hugging Face and GitHub under a community license allowing commercial use for sub-$1M revenue teams.
Runway just announced Gen-3 Alpha Turbo, a new video model that generates clips seven times faster than the original Gen-3 Alpha. CEO Cristóbal Valenzuela noted it can render a 10-second clip in just 11 seconds at a lower cost, rolling out to users over the coming days.
Midjourney just launched V6.1 as its new default image model. The update delivers cleaner hands, faces, and textures, smarter text rendering, and a 25% boost in standard generation speeds, alongside an upgraded 2x upscaler. Inpainting and zooming still temporarily fall back to V6.0.
NVIDIA just showcased James, an interactive digital human built on its ACE platform and ElevenLabs voice tech that connects with users using humor and emotion. The company also previewed Maxine 3D, which turns 2D video feeds into real-time 3D avatars for telepresence. Virtual brand ambassadors are quickly shifting from uncanny bots into expressive, lifelike agents.
Runway just rolled out Image-to-Video capabilities for its Gen-3 Alpha model, allowing creators to anchor any static image as the first frame alongside text prompts. The update delivers much tighter artistic control and visual consistency, giving creators a reliable starting point instead of rolling the dice with pure text.
Shutterstock and Getty Images just rolled out major visual AI upgrades powered by Nvidia’s Edify architecture. Shutterstock launched a commercial beta for Generative 3D that creates editable assets and 16K HDRi lighting maps from prompts. Meanwhile, Getty doubled its generation speed to four images in six seconds and added fine-tuning, both leveraging commercially licensed data.
Runway just got caught in a major training data leak after 404 Media obtained an internal spreadsheet showing the company scraped thousands of YouTube channels and pirated films to train its flagship Gen-3 video model. Why it matters: Despite raising $141M from Google and Nvidia while dodging questions about data provenance, Runway now faces intense scrutiny as AI copyright tensions boil over.
Kuaishou just rolled out Kling AI 1.0 worldwide, opening its Sora competitor to anyone with an email address. Users receive 66 free daily credits for text-to-video and image-to-video generation, complete with multi-format outputs and custom camera controls. While OpenAI keeps Sora behind closed doors, Kling is giving creators real access right now.
Stability AI just launched Stable Video 4D, an open-source model that turns a single video into eight novel-view videos. It generates 5 frames across 8 camera perspectives in about 40 seconds, streamlining dynamic 3D optimization for game development and video editing. The weights are live on Hugging Face.
Suno just launched a new "Stems" feature for Pro and Premier subscribers, allowing users to split AI-generated tracks into separate vocal and instrumental files. Each stem generates as an individual clip that can be downloaded or extended independently. It gives creators far more control over mixing, editing, and repurposing their AI tracks.
EA Sports just built 11,000 virtual college football players for College Football 25 using in-house AI that turns 2D headshots into 3D avatars. The studio bypassed the logistical nightmare and massive cost of full-body scans for thousands of student athletes, establishing a scalable workflow for future sports game franchises.
Luma Labs just rolled out a new Loops feature for its Dream Machine video model, allowing users to generate seamlessly repeating video clips. It makes creating endless motion graphics and animated backgrounds significantly easier without jumping into a video editor.
Sakana AI just launched Evo-Ukiyoe and Evo-Nishikie, two image models dedicated to Japan’s historic ukiyo-e woodblock art. Fine-tuned on over 24,000 archival pieces from Ritsumeikan University, they can create traditional prints from Japanese text prompts or colorize antique monochrome illustrations. Both models are openly available on Hugging Face for research purposes.
Oliver Cameron just unveiled Odyssey, a new venture building Hollywood-grade visual AI aimed at films, TV, and video games. Rather than just generating scenes, the system lets creators actively direct scenery, characters, lighting, and motion. Why it matters: Adding genuine director-level control could transform generative video from an unpredictable gimmick into a serious production tool.
