Type it. Watch it appear. Runway's real-time trick.

Runway demoed real-time HD video generation at NVIDIA GTC — first frame in under 100ms. Plus: Anthropic's 81K-person global survey, an AI short film that sparked a real debate, and a JSON prompt for hyper-realistic portraits.

Carat EditorMar 23, 20264 min read

Type it. Watch it appear. Runway's real-time trick.

Hey — it's the Carat team. 😊 Runway just unveiled real-time HD video generation at NVIDIA GTC. Anthropic went and surveyed 81,000 people in 159 countries about what they actually want from AI. Lots to dig into today.

🔥 Type it. Watch it appear. Runway just pulled this off.

Runway showed a real-time HD video generation model at NVIDIA GTC. First frame renders in under 100ms — faster than a single eye blink (100–400ms).

The model runs on Runway's world simulation model, GWM-1. It was demoed on NVIDIA's Vera Rubin servers (72 GPUs, 20.7TB of memory). Unlike traditional video generation, it streams frames as they're generated — like a game engine building the world in real time.

Still research-stage. Getting this to actual consumer products is a long way off. But the direction is clear: video generation is moving from 'wait for it' to 'it's already there.'

If you make ad content, YouTube thumbnails, or short-form clips — this is the direction things are heading. Runway Gen4 Turbo is already on Carat, alongside 10 other video models. Worth getting comfortable with AI video before real-time goes mainstream.

📌 3 Stories Today

1️⃣ This Spider-Man AI clip is shockingly convincing

AI creator kasumov made a Spider-Man 'wazzupp' clip that went viral. Spider-Man turns, hits the pose — motion and expression are so smooth you don't clock it as AI at first glance. The whole thing took a few minutes.

Whether it's a licensed IP character or a game character, putting whoever you want into whatever scene you want is getting fast and easy. Try it on Carat with Kling or Sora 2.

2️⃣ Anthropic asked 81,000 people in 159 countries. Here's what they said.

Anthropic ran its largest-ever survey of Claude users — 80,508 responses across 159 countries and 70 languages. 67% said they feel positive about AI.

The #1 fear? Not job loss. AI reliability — wrong or hallucinated answers — topped the list at 26.7%. In Africa and Southeast Asia, people see AI as economic opportunity. In North America and Western Europe, regulation and surveillance dominate the worry list.

The top thing people want from AI: handle repetitive tasks for me (18.8%). Generating images, making videos, pulling content ideas. That's exactly what Carat is building toward.

3️⃣ One person made a short film with AI. The internet can't agree if it counts as art.

A 2.5-minute short film using cooking as a metaphor for AI creation. Director Simon Meyer — an official Kling AI creative partner — made it solo using only Kling 3.0. Three days after posting to Reddit: 1,300 upvotes, 304 comments.

The film asks whether making something with AI counts as creating it — so the comments turned into a real debate. 'Curating isn't creating.' 'AI is an automated kitchen, not a chef.' 'No — this is exactly what new creation looks like.' The discussion itself became part of why the film mattered.

One person. No team. A finished short film. It's not a technical question anymore — it's a question of what you want to make. Kling is on Carat if you want to try it yourself.

🧪 Today's Prompt Tip

JSON prompts for hyper-realistic portraits in Nano Banana 2

AI creator @KeorUnreal shared a JSON-format portrait prompt for Nano Banana 2 on X — it got 852 likes. The format specifies composition, lighting, and outfit in structured detail.

Nano Banana 2 handles structured JSON better than plain text. More control over pose, expression, and background. You can use this directly on Carat.

{
  "style": "photorealistic portrait",
  "composition": "upper body, front-facing close-up, natural gaze",
  "subject": {
    "age": "late 20s",
    "ethnicity": "East Asian",
    "expression": "natural smile, lively eyes",
    "hair": "dark hair falling softly"
  },
  "outfit": "white linen shirt, two buttons undone",
  "background": "pure white studio, soft gradient",
  "lighting": "left softbox, soft natural light feel",
  "quality": "8K resolution, sharp, editorial ad quality"
}

The key is the 'lighting' field. Specific directions like 'left softbox, soft natural light feel' make a real difference in output quality. Adding a reference like 'editorial ad quality' works well too. Nano Banana 2 is on Carat — give it a try. Copy-paste the JSON if it's new to you. The results will surprise you. 🤗