ngram blog
Product updates, best practices, and insights on AI-powered video creation for modern teams.
Pika shipped four standalone audio models, SFX, Speech, Soundtrack, and Music, priced far below ElevenLabs and Hunyuan Foley on paper. Here's what that does to the cost of voicing, scoring, and sound-designing an AI video, and to the vendors selling audio on its own.
LTX-2.5 is a 22B-parameter open-weights video model that generates synchronized audio and video faster than Veo 3.1, Kling 3.0 Pro, and Gemini Omni Flash, and runs on a 16GB consumer GPU. Here's what that means for AI video's build-versus-buy math.
Alibaba's Tongyi Lab announced a public beta for Wan 3.0 on August 6, 2026, gated to invited accounts - the first major AI video model that reads PowerPoint decks, PDFs, and spreadsheets directly. Here's what that changes in production workflows, and what it doesn't.
MiniMax H3 ships 2K AI video with native stereo audio in a single pass. Here's what that collapses in a video production pipeline, what still doesn't work, and why the model layer is racing toward audio-native generation.
Runway's new Media Router auto-picks the right video, image, or audio model for a job, the same idea OpenRouter brought to text models three and a half years ago. Here's why it's arriving now, and why Runway is willing to route paying customers to its own competitors to make it work.
Black Forest Labs launched FLUX 3, its first multimodal frontier model unifying video, audio, and robotic action. Here's why the first production customer is a car factory, not a marketing team.
NVIDIA's Synthetic Video Detector, a NIM microservice announced at SIGGRAPH on July 20, 2026, is now embedded in Wowza's livestreaming infrastructure across 35,000+ deployments in 170+ countries. Here's what NVIDIA's own accuracy and latency numbers actually show.
PixVerse closed a $439 million Series C extension on July 13, 2026, crossing a $2 billion valuation and pivoting from clip generation toward real-time AI world models with its new PixVerse Game Engine. Here's what the round, the valuation math, and the world-model pivot mean for AI video.
Goldman Sachs ranked ByteDance's video-generation models above Zhipu, DeepSeek, and every other Chinese AI developer it evaluated, the first standalone investable ranking of AI video quality from a bulge-bracket bank. Here is what the ranking, the Zhipu coverage initiation, and the numbers behind Seedance actually show.
Runway shipped Agent 2.0 on June 25, 2026 and Agent Skills on July 2, letting one command turn a brief into a localized, multi-format ad campaign. Luma shipped the same shift first, in March. Here's what it means that AI video suppliers no longer compete on generation quality alone.
Google Photos Video Remix lets anyone restyle an existing clip with a text prompt, powered by Gemini Omni. Here is what the July 2026 launch reveals about AI video editing becoming a default consumer feature, five weeks after Google made AI video generation free on YouTube.
TwelveLabs raised $100M and signed a multiyear AWS Trainium chip deal on July 1, 2026, while 2026's biggest AI video headlines keep going to generation models. Here is what the funding map reveals about the video understanding layer.
Kuaishou's Kling AI raised $2B+ at an $18 billion valuation, the largest funding round in AI video generation's history. Here is what the ARR growth, peer valuations, and the 2027 IPO clock actually reveal about the economics of the model layer.
Google's June 30 launch of Nano Banana 2 Lite and Gemini Omni Flash public preview puts a real price tag on image-to-video for the first time: $0.034 per 1,000 images to draft, $0.10 per second to animate. Here's what a 15-second product video actually costs to generate, and what it means for commerce video economics.
In June 2026, A24 and Lionsgate each took equity stakes in AI video providers 11 days apart. Neither framed it as cost reduction. Here is what the deals reveal about where AI filmmaking tools have landed.
ByteDance announced Seedance 2.5 on June 23, 2026, the first major AI video model to generate a native 30-second 4K clip in a single pass. Here is what that means for production teams dealing with the stitching problem.
OpenAI shut Sora down in April 2026. ByteDance froze Seedance 2.0's global rollout in March after Hollywood cease-and-desist letters. Within 90 days, two of the most credible enterprise AI video models exited the international market. Alibaba launched HappyHorse 1.1 on June 22 into that gap, the only enterprise-grade model with a global API, native lip sync, and audio in a single pass. Here is the full analysis.
xAI's Grok Imagine 1.5 debuted May 31, 2026 and hit #1 on the Artificial Analysis Image-to-Video Arena with 1404 Elo, beating HappyHorse, Seedance 2.0, and Google Veo. Built on the Aurora autoregressive architecture with native audio in a single pass, it reached the top in 10 months from zero. Here's what the pace means.
OpenAI killed Sora on April 26, 2026. The app burned roughly $1M per day, earned $2.1M in total lifetime revenue, and held a 1% 30-day retention rate. Here is what the numbers reveal about the real economics of AI video infrastructure in 2026.
Avataar launched Varya, an India-built video model distilled from Wan 2.2 that generates video at about $0.005 per second. Here is what the launch says about collapsing AI video generation costs.
New York's Synthetic Performer Disclosure Law is live as of June 9, 2026, and EU AI Act Article 50 enforcement arrives August 2. Here's what both laws actually require, who is exposed, and a practical compliance checklist for the next 54 days.
Google just embedded AI video generation into YouTube for free. Here's what that means for the 2.7 billion people who already use the platform, for content creators, and for where the AI video industry goes from here.
NVIDIA Cosmos 3 is an open world model for physical AI. Here's why it matters for AI video, image-to-video workflows, and the move from clips to simulation.
The Remotion skill gives coding agents current guidance for building programmatic React videos. See its August 2026 status, install command, and launch history.
