Archive, page 11

Fal's H3 Max Live model can generate video faster than real time, meaning output is produced more quickly than it would take to watch the finished clip. This crosses a threshold that has practical implications for interactive and live applications. It is early days, but the milestone signals a meaningful shift in what video generation pipelines can support.

Runway has introduced Solaris, its first Interface World Model, which generates interactive user interfaces in real time, frame by frame, without requiring any code. The system represents a new direction for generative AI, moving beyond static images or video clips toward dynamic, responsive environments. It is an early look at how AI models might eventually power the interfaces people use to interact with software.

Fintech company Chime worked with Runway to turn photographs of nearly 100 real customers into a finished national television advertisement. The production approach cut costs significantly - reportedly saving $200,000 on a single clip compared to traditional methods. The campaign offers a concrete look at how generative video tools are beginning to factor into professional broadcast workflows.

Google has updated its native multimodal video model with Gemini Omni 1.1 Flash, bringing a set of targeted improvements to scene extension, frame-level control, and output resolution. The update extends the model's ability to read prior video context from a single frame to a full 10-second window, enabling more coherent continuations. First and last frame pinning, along with 4K upscaling, round out the changes.

China's short-drama industry has moved rapidly toward AI-generated content, with 95 percent of the 128,000 short dramas released in Q1 2026 produced using AI. Actors report being asked to surrender their voice and likeness data before losing their jobs, and labor disputes tied to AI displacement are climbing. The trend offers an early, concrete look at how generative video is reshaping entertainment workforces at scale.

LAION has released its Big Video Dataset (BVD), a large open collection of 80 million videos totaling 10 million hours of footage, aimed at advancing AI video research. The dataset also includes 55 million automatically described clips, and models trained on it have outperformed those trained on the previous benchmark dataset, InternVid. Its open availability marks a notable moment for researchers who have lacked large-scale, freely accessible video data.

Midjourney has pushed an update to its V8.2 edit model, addressing image quality issues some users encountered over the past day. The fix arrives quickly after reports surfaced, and the team is encouraging users to retry any edits that fell short. Further updates are said to be on the way.

Midjourney has begun rolling out an early test of its V8.2 edit model, bringing instruction-based editing, multi-image reference generation, and inpainting to its latest generation pipeline. The update replaces the previous omni-reference system and supports up to four image references at once. It marks a notable expansion of what users can do directly within the Midjourney platform without leaving to third-party tools.

Eddie AI has introduced a pay-as-you-go pricing model, MCP server support for Claude and ChatGPT, and a native iPhone app. The updates give video editors more flexible ways to access AI-assisted editing without committing to a monthly subscription. Here is what the changes mean in practice.

Google has updated its Gemini Omni 1.1 Flash video model with broader context analysis, longer scene extension limits, and a lower-cost draft mode. The changes address practical concerns around consistency and affordability in AI-generated video. For developers and creators working at scale, the update shifts the cost-quality tradeoff in a meaningful direction.

Google is expanding its Flow video creation tool with a new set of creative control features, building on the Gemini 2.5 Flash model first introduced at Google I/O. The updates give filmmakers and creators more precise tools for shaping their video outputs. The rollout marks a continued push by Google to make AI-assisted video editing more practical for professional and creative workflows.

Google has updated its Gemini Omni 1.1 Flash model with a new set of creative controls and generative video features aimed at developers. The release expands what builders can direct and fine-tune when generating video content through the API. It marks a continued push by Google to give developers more precise influence over model outputs.

Moonbug Entertainment, the studio behind Cocomelon and Blippi, has instructed its artists to begin experimenting with AI tools as part of a broader shift in its production workflow. The company says human oversight will remain a core part of the process. The move raises questions about how AI adoption will affect the artists and animators who work on these widely watched children's properties.

Adobe is updating Photoshop with a dedicated "AI Assisted Editor" view, currently in beta, that consolidates the app's AI tools into a single toolbar. The update also introduces a "markup" feature that lets users draw directly on an image to guide the AI on what to change - no text prompts required.

Adobe Photoshop is adding a new "Markup" tool that lets users sketch directly on an image to guide AI-based generation - communicating intent visually rather than through text prompts alone. The update also brings Lightroom-style controls to Photoshop, a long-requested addition for photographers and retouchers who work across both apps.

Adobe Firefly has expanded into audio with three tools - Generate Music, Generate Speech, and Generate Sound Effects - designed to bring sound into the editing process earlier than traditional workflows allow. Rather than treating audio as a final step after the picture is locked, creators can now use generated sound to test pacing and tone while a project is still taking shape. The tools are integrated into Firefly's broader multi-model creative environment alongside image and video generation.

Z.ai has released GLM-5.3-Flash, the first model in the GLM-5 series to handle multiple modalities natively, built on a Mixture-of-Experts architecture with 320 billion total parameters and 18 billion active at inference time. The model supports a context window of just over one million tokens and is available under an MIT license. API access is priced at $0.15 per million input tokens and $0.50 per million output tokens.

Pro-Kremlin Telegram channels have circulated AI-generated deepfake videos depicting two Ukrainian lawmakers appearing to call for surrender and peace talks. The clips accumulated roughly 130,000 views within two weeks, raising fresh concerns about the role of synthetic media in active armed conflicts.

Stability AI, the company behind the open-weight image generation model Stable Diffusion, has secured $76 million in new funding, bringing its total raised to $232 million. The round signals continued investor interest in open-source generative image infrastructure despite a competitive and crowded market. Details on the lead investor and use of funds have not yet been fully disclosed.

Runway has launched Runway Dev, an API platform giving developers programmatic access to its image, video, and character generation models. The offering is positioned as enterprise-ready from the start, consolidating multiple model capabilities under a single interface. It marks a deliberate move by Runway to expand beyond its consumer and creative-professional tools into the infrastructure layer of AI media production.

Apple has refreshed the Mac Studio with new M5 Max and M5 Ultra chip options, bringing meaningful spec bumps across CPU, GPU, storage, and memory bandwidth. While Apple's marketing leans heavily on AI performance figures, the underlying hardware gains are relevant for video editors and other creative professionals as well. The M5 Max starts at $2,499 and the M5 Ultra at $5,499.

Upscaling a video to 4K is more nuanced than simply increasing a file's resolution setting. Modern software uses increasingly sophisticated techniques to fill in missing detail, but the results are not always perfect. Here is a look at how the process actually works and what limitations remain.

Samsung's new My FanCam feature uses AI to automatically track and center a selected subject within a video, keeping them in focus even as the action moves around. The tool is aimed at everyday moments like school performances or sports events, where a wide shot is necessary but you still want a clean close-up. Here is how to get started with it.

Alibaba has released Wan3.0, a video generation model capable of producing clips up to 30 seconds long from a range of inputs including text prompts, images, PDFs, and PowerPoint files. A 30-second clip at 1080p is priced at $6. The release comes as Alibaba reported a sharp 75 percent year-over-year drop in quarterly profit, reflecting heavy investment in AI infrastructure.