Twice a week, the most important stories in image and video generation - new models, notable research, and meaningful product releases - distilled into a 2-minute read. No hype, no filler.
Free. Unsubscribe any time. No spam, ever.
Google has updated its Vids app with the Omni 1.2 Flash model, allowing users to generate longer AI video clips of up to 15 seconds at no cost. The update also brings improved consistency across scenes, making it easier to produce cohesive video content. This marks a meaningful step forward in Google's effort to integrate generative video tools into its Workspace productivity suite.
Upload a photo, pick a trending dance template, and Groovely animates it into a vertical clip synced to the beat. No camera, choreography, or editing required.
NVIDIA has released Cosmos 3, a new set of open models designed for interactive world simulation. The models are hosted on Hugging Face and represent a step forward in physically grounded, controllable video generation for robotics and autonomous systems research.
Google DeepMind has made Gemini Omni 1.2 Flash generally available, bringing longer video clips and a notable price reduction to its AI video generation model. The update extends maximum clip length to 15 seconds and drops the cost to $0.07 per second at 720p - 30 percent cheaper than its predecessor. Conversational editing and a draft-then-upscale workflow offer additional flexibility for users watching their budgets.
Gods Don't Give Gifts, an AI-generated sci-fi film featuring pop star Grimes, is set to reach theaters on December 4th. Grimes is contributing both her likeness and vocal performance to the project, marking one of the more prominent celebrity involvements in AI filmmaking to date. The film is being positioned as a test case for whether AI-produced features can find a commercial audience.
Google has updated its Gemini Omni 1.2 Flash model with native video generation support, extending clip length to 15 seconds and introducing editing capabilities alongside improved consistency. The release also comes with a 30% reduction in pricing, making it a more accessible option for developers building video-focused applications.
An AI-generated feature film called "Gods Don't Give Gifts" is mounting a surprisingly serious awards campaign, complete with an AI-generated Grimes cameo, Times Square advertising, and official Oscar qualifying screenings. Creator Gossip Goblin argues the project deserves to be taken on its own terms rather than dismissed as a publicity stunt. The film is shaping up as one of the more visible tests of how the industry will respond to fully AI-generated feature content.
Runway has detailed how it serves its 57-billion-parameter video model - the one powering Runway Characters - at 24 frames per second for live, interactive conversation. The engineering work cuts the hardware requirement in half, from eight H100 GPUs down to four. The post offers a rare look at the infrastructure tradeoffs involved in running large generative video models at real-time latency.
OpenAI has updated its image generation lineup with two new API models - GPT-Image-2.6 Flare and Sunburst - alongside a consumer-facing Images 2.6 release. The update introduces GIF output, SVG export, markup-based editing, and a complexity parameter, with Flare capable of returning results in as little as six seconds.
OpenAI has updated its image generation platform with a new version, Images 2.6, adding animated GIF output and image reference support alongside a set of other capability improvements. The release also introduces two new API models - Flare and Sunburst - and brings faster generation speeds and more precise editing controls. The update signals a broader push to make OpenAI's image tools more flexible for both developers and end users.
OpenAI has introduced GIF generation to ChatGPT via its new Images 2.6 model, allowing users to create short animations from text prompts or uploaded photos. The model maintains enough visual consistency across frames to keep subjects recognizable, enabling basic flipbook-style animation. Users can also adjust playback speed and edit individual frames after generation.
OpenAI has added native GIF generation to ChatGPT, letting users create animated images from a text prompt or an uploaded photo. The feature arrives alongside the new Images 2.6 model and includes controls for animation speed and frame-by-frame editing. Previously, producing a GIF required multiple steps outside of ChatGPT's native workflow.
OpenAI has updated its image generation system to Images 2.6, adding faster output, GIF creation, and more precise editing tools. The update rolls out across ChatGPT, Codex, and the API, with new controls available to developers. It marks a steady incremental step in OpenAI's image tooling rather than a wholesale overhaul.
Edited photos of Jeff Bezos and actor Jason Statham depicting both men with full heads of hair have been circulating on X, Threads, and Bluesky. The images are fabricated - produced through photo editing, AI-assisted or otherwise - and do not reflect any real change in either man's appearance. The Verge looked into the spread of these posts and found them appearing frequently enough to reach multiple members of their own team through social feeds.
Fal, an AI inference startup focused on generative media workloads, has raised $1.3 billion in a new funding round led by Bessemer Venture Partners, doubling its valuation to $16 billion. The five-year-old company has emerged as a key piece of infrastructure for developers building on top of image and video generation models. The raise reflects continued investor appetite for the compute and serving layer underneath generative AI applications.
A photographer has called out a fellow photographer for allegedly passing off AI-generated images as genuine photographs in their portfolio, with the dispute centered on images of Whitehaven, a historic harbor town on the Cumbrian coast of England. The case highlights growing tensions within the photography community as AI image generation becomes increasingly convincing. It raises difficult questions about disclosure, authenticity, and professional ethics in an era when synthetic imagery can be
Adobe has announced partnerships with Universal Music Group and Warner Music Group to bring licensed music from major artists into Adobe Premiere mobile. The move gives mobile video editors access to familiar tracks for use in their projects, with new browsing and editing tools to match. Tracks will be available for personal, non-commercial use when the feature rolls out.
A new study from e-commerce UX research organization Baymard finds that a majority of U.S. online shoppers feel less confident purchasing a product when its photos appear to be AI-generated. The findings raise questions about how retailers should approach AI image tools as they become more accessible and widely used in product photography workflows.
Google has expanded Dreambeans, its AI-powered personal storytelling experiment, to India with support for 13 Indian languages. The move signals Google's interest in bringing generative AI narrative tools to one of the world's largest and most linguistically diverse markets. It marks one of the broader regional rollouts for the product since its initial launch.
Runway has released Dev plugins for three popular coding environments - Claude Code, OpenAI Codex, and Cursor - allowing developers to access Runway's generative media API directly from their editor. The integration lets coding agents build and invoke video and image generation features without leaving the development workflow. It is a practical step toward making generative media capabilities a routine part of software projects.
Google is quietly testing a Comics artifact feature inside Gemini Notebook that would let users generate comic-style visuals directly within the tool. The feature appears to offer four distinct visual styles along with controls for language and page orientation, though no release date has been announced.
Adobe Research and UC San Diego have developed EditAgent, an AI system that approaches image editing by selecting specialized tools for each task - much as a skilled Photoshop user would. Rather than repeatedly applying a single general-purpose model, it delegates decisions to a language model that evaluates results using measured image data. The trade-off is significant: the approach demands roughly ten times the compute of simpler methods.
Google researchers have built StreamVibe, a video AI agent designed to observe people doing hands-on tasks and offer assistance only at the moments it is actually needed. A new benchmark reveals that current models struggle significantly with this timing problem, especially as videos grow longer and errors accumulate.
Meta has quietly rolled out a new AI-powered fashion app called Style Sketch in three Southeast Asian markets - Indonesia, Singapore, and Thailand. The app allows users to virtually try on clothing using generative AI. The limited regional launch suggests Meta is testing the waters before any broader rollout.
Fashion photo studios are shutting their doors as clothing brands shift toward generative AI tools to produce product imagery, drawn largely by the cost reductions the technology offers. A Reuters report highlights the growing real-world impact this trend is having on traditional commercial photography businesses. The change marks a notable structural shift in how the fashion industry produces visual content at scale.
China's first fully AI-generated feature film has hit theaters nationwide, but early returns suggest audiences are not warming to the format. Despite an opening day that included 10,000 screenings, the film is struggling to find a meaningful box office foothold. The release marks a notable moment for generative video AI, even as the commercial reception raises questions about audience readiness.
Google DeepMind has released Genie 4 as a research preview, a real-time world model that now generates native audio alongside interactive environments. The update integrates live dialogue, text-based steering, and Gemini 3.8 Flash as a planning component inside the simulation loop. It marks a meaningful step toward AI-generated worlds that are not just visual but fully audible and conversationally responsive.
Apple has rolled out a significant update to its Creator Studio subscription suite, with the headline addition being MotionVFX content built directly into Final Cut Pro for Mac. The update also brings Pixelmator Pro integration, Logic Pro improvements, and expanded AI features across Apple's productivity apps.
Runway has made its Ads product generally available across all paid plans, giving marketers a dedicated tool for generating on-brand creative variants from existing high-performing ads. The company reports that internal use of the tool grew weekly ad volume tenfold while reducing cost per subscriber by 41%. The release marks Runway's clearest move yet into applied advertising workflows.
Google DeepMind has unveiled Genie 4, a real-time world model capable of simulating interactive 3D environments complete with sound and talking characters. The system uses Gemini models in the background to maintain consistency across the simulation. It remains an early research model, with longer sessions still producing noticeable visual artifacts.
Liquid AI has published a guide on building specialized multimodal models using LFM3, its latest generation of foundation models. LFM3 moves away from the transformer-only architecture, offering a different approach to handling multiple input types. The post, hosted on Hugging Face, walks through how developers can adapt these models for targeted use cases.
Google has unveiled Genie 4, the latest version of its generative world model, bringing audio support, improved scene consistency, and real-time interactive steering. The update doubles the coherent scene duration from one minute to two, while also delivering more realistic physics and more humanlike character rendering.
Google DeepMind has unveiled Genie 4, the latest version of its world simulation model, promising a significant step forward in real-time interactive environments. The system aims to generate more realistic and controllable simulated worlds than its predecessors. This release continues DeepMind's ongoing research into generative world models as a foundation for training agents and enabling creative applications.
Google has announced Genie 4, the latest version of its AI world model for generating interactive experiences. The update brings more realistic visuals, real-time natural language editing, audio generation, and a doubled maximum experience length of two minutes. The model runs on Gemini 3.8 Flash under the hood to power its simulations.
Blackmagic Design has released DaVinci Resolve 21.1.2, bringing up to 14x faster IntelliSearch analysis on Apple M6 hardware and 4x faster OGraf playback in the Edit and Fusion pages. The update also expands Linux graphics support and refines timeline editing behavior. More than a dozen fixes and enhancements round out the release.
ASAP Rocky is facing accusations from two video artists and a parkour athlete who claim that AI was used to replicate their original work in his latest music video without permission or credit. The case adds a high-profile name to a growing list of disputes over generative AI and creative ownership. It raises pointed questions about accountability when AI-assisted production pipelines obscure the line between inspiration and copying.
vLLM Semantic Router has released d3-ultra, a 27-billion-parameter open-weight model designed to handle text, image, and video inputs in a single pass and output probability distributions across possible answers. The model scored 87.2 on the team's SR-Text benchmark. Its open-weight nature makes it accessible for researchers and developers building multimodal routing and decision systems.
Topview's Film Studio 2.0 includes a growing suite of over 25 cinematic tools aimed at giving filmmakers more structured control over AI-generated video. No Film School highlights five standout tools from the platform's Cinematic Suite, ranging from a 3D scene builder to an actor expression editor. The piece offers a grounded look at where AI filmmaking tooling currently stands and what workflows are becoming practical for working artists.
Runway has introduced App Studio, a no-code tool that lets users build custom AI applications for generating images, video, and audio. The new feature integrates with Runway's existing Brand Kits and Workflows, making it easier for teams to deploy consistent, branded generative AI tools without writing any code.
Elon Musk has followed through on his earlier claim that xAI's Grok would produce an AI-generated feature-length retelling of Homer's The Odyssey, apparently prompted by dissatisfaction with Christopher Nolan's live-action version. The result is a Grok-produced film of sorts, though the execution appears to fall well short of a conventional cinematic release. It marks one of the more prominent public attempts to use generative AI tooling to produce long-form video narrative.
Grok has released the first seven minutes of its AI-generated adaptation of Homer's Odyssey, offering an early look at where long-form AI video production currently stands. The footage is technically coherent but falls short in the areas that typically make storytelling compelling - artistry, emotional resonance, and a sense of human perspective. It marks a notable milestone in AI video length and ambition, even if the results remain uneven.
Kyutai has released CADRE, a family of seven open causal diffusion video autoencoders designed specifically for interactive world models. The models achieve 20 frames per second on a single H100 GPU with a reaction latency under 230 milliseconds - a threshold considered important for interactive applications to feel responsive.
Canva has launched its first in-house image generation model, with a notable distinction from most competitors: it was trained on design templates rather than photographs. The approach reflects Canva's design-first identity and sidesteps some of the copyright concerns that have surrounded photo-trained models. It positions the tool as purpose-built for graphic design output rather than photorealistic imagery.
Canva has unveiled Craft, its first internally developed image generation model, marking a significant step away from the design platform's previous reliance on third-party AI providers. Craft is built to integrate natively with Canva's existing suite of tools, aiming to give users a more cohesive creative experience. The move signals Canva's intent to build deeper control over its AI capabilities rather than depending on outside models.
Canva has introduced Craft, its own image generation model capable of producing 2K resolution outputs with planned layouts, multilingual text rendering, and editable elements. The model is accessible through Canva's Dream Lab interface as well as a public API. Craft appears designed to close the gap between raw AI image generation and practical design workflows.
Canva has launched its first in-house image generation model, called Canva Craft, which uses a two-stage architecture to plan a design before producing a final image. The model supports up to 2K resolution and more than 30 aspect ratios, targeting a wide range of use cases from social media content to digital art. It puts Canva in more direct competition with image generation offerings from OpenAI, Microsoft, and Google.
Canva has launched its first proprietary image generation model, opening it up to all users on the platform. The model is pitched as offering stronger performance and greater customization options compared to the third-party models Canva has previously relied on, while also reducing costs.
Food delivery platform Just Eat has issued an apology after AI-generated menu images appeared on its app showing clearly inaccurate food pairings - including bacon and eggs suspended in a fizzy drink. The incident has drawn attention to the risks of deploying generative image tools in contexts where visual accuracy directly affects consumer trust. Just Eat has not yet detailed the specific process that allowed the images to go live.
