Archive, page 3

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.

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 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.

Director Cassandra Brooksbank resurrected a shelved AXE commercial - originally pitched to The Martin Agency in 2021 and killed by budget constraints - using the AI filmmaking platform Absurd. The resulting 60-second spec spot, "For Days That Run Hot," was built using the same production disciplines she applies to live-action work: written treatment, character casting, location scouting, storyboarding, and an editorial pass in Premiere. The project offers a detailed look at how a working commerc

NVIDIA has released Cosmos3-H-Dreams, a generative simulation model designed to produce realistic video of clinical surgical environments. The model is hosted on Hugging Face and represents a focused application of NVIDIA's Cosmos platform toward medical and procedural training scenarios. By synthesizing plausible surgical footage, the system opens possibilities for training both human practitioners and AI models without requiring access to live operating rooms.

Runway has joined Anthropic's Claude Marketplace, giving enterprise users access to Runway's video, image, and audio generation tools directly within the Claude ecosystem. Teams can now apply their committed Anthropic spend toward Runway Creative capabilities.

Adobe has added a generative AI tool to Lightroom on iOS, iPadOS, and Android that can open closed eyes in photos with a single tap. When the app detects closed eyes, it surfaces an Open Eyes Quick Action directly over the image for quick access. Unlike older approaches to the same problem, this feature relies on generative AI rather than borrowing replacement eyes from a separate photo.

Google has launched a public-facing website for SynthID, its AI watermarking and detection tool, allowing anyone to check whether an image has been created or significantly altered by AI. The tool works by reading watermarks embedded by Google's own image generation systems, which limits its coverage but makes it a practical first check for content from those pipelines.

Sound stages built during the streaming boom are sitting vacant across the US and UK, and private equity firms are increasingly converting them into data centers rather than waiting for Hollywood production to recover. The shift reflects a broader reallocation of capital away from film infrastructure and toward AI computing - driven by post-strike production slowdowns, tighter streaming budgets, and the rising demand for server space. Several high-profile studio projects have already been halted

Kling AI has updated its video generator to version 4.0, bringing stereo audio, multilingual lip sync, clips up to 30 seconds, and support for up to ten keyframe images. A lighter Flash variant accompanies the release for users who prioritize speed over maximum quality. Benchmark comparisons suggest Kling 4.0 is competitive with peers like Wan 3.0 and Seedance 2.5, though it does not clearly outpace them.

New research finds that AI-edited photos and videos can cause people to form false memories of events - and that exposure to manipulated imagery tends to make viewers more confident in those incorrect recollections. The findings raise fresh concerns about the psychological impact of AI-generated and AI-altered visual media as it becomes increasingly difficult to distinguish from authentic footage.

Google's SynthID detector offers a straightforward way to check whether an image was generated or altered by AI, though the tool comes with some important limitations worth understanding. Here's how to use it and what to expect from the results.

Apple's Image Playground app brings on-device generative image creation to iPhones and Macs, and it received a set of notable updates this fall. The tool lets users generate images from text prompts and personal photos without sending data to the cloud. Here is a look at what the app does and how to get started with it.

OpenAI appears to be testing a new image generation model internally codenamed "Aurora," with three alpha variants briefly appearing in the company's API before being pulled. The models have since shown up in Arena-style blind comparison tests, suggesting OpenAI is gathering quality benchmarks ahead of a wider release.

Artlist, the music and media licensing platform, produced a short sci-fi series called "The Arc" using a budget of just $38,000 - leaning heavily on generative AI tools throughout production. The project serves as a real-world test of how far AI-assisted filmmaking has come, and what it still cannot do.

Adobe and NVIDIA are deepening their ongoing partnership, bringing NVIDIA RTX Spark integration to Photoshop, Premiere Pro, and Substance 3D with a promise of up to twice the AI processing speed. The collaboration also extends to cloud infrastructure, with accelerated computing applied to Frame.io for faster media search and transcoding. Further work on personalized marketing tools is also underway through joint development of Brand Concierge technology.

Documentary filmmaker Alex Gibney used a CGI avatar of Elon Musk to simulate an interview that Musk himself declined to give. Speaking on IndieWire's Screen Talk podcast, Gibney discusses how the production team built the simulation and what it meant for the finished film. The approach raises broader questions about AI-generated likenesses in nonfiction filmmaking.

Elorian has released its first AI model, Glimpse, which is designed to reason with images as active components of a thought process rather than treating them as passive inputs to describe. According to the company's own benchmarks, Glimpse outperforms GPT-6 Astra on scientific chart reasoning and short-form video captioning. The result is a model that positions visual understanding as central to reasoning, not an afterthought.

President Trump shared a set of White House renovation photos on Truth Social, but one image comparing an old column to a restored one appears to have been edited using AI. The photo was flagged by Google's AI detection tool as "made or edited with OpenAI," and sharp-eyed Reddit users noted structural inconsistencies in the image itself.

NVIDIA has released Nemotron 3.5 Super VL, a 120-billion-parameter open multimodal model built on a mixture-of-experts architecture with only 12 billion parameters active at any given time. The model supports a one-million-token context window and ships with NVFP4 quantized weights, allowing it to run on a single DGX Spark system. NVIDIA also released two smaller Bolt edge variants alongside the flagship model.