Runway Gen-4.5: AI That Creates Realistic Videos from Text Like No Other!

Runway Gen-4.5: AI That Creates Realistic Videos from Text Like No Other!

Ondřej Barták
Ondřej Barták
Entrepreneur and Programmer
5. 12. 2025
5 minutes reading
Runway Gen-4.5: AI That Creates Realistic Videos from Text Like No Other!

What Is Runway Gen-4.5?

Runway, a startup founded in 2018, has just unveiled its new Gen-4.5 model. This tool allows users to create high-resolution videos based solely on written descriptions. For example, when you enter a prompt such as "The camera moves across a kitchen table with various objects, the light is dim and comes through a window, casting shadows and patches of light," the model creates a video in which the objects move naturally, with realistic shadows and details. Gen-4.5 focuses on understanding physics, human movement, camera movements, and cause-and-effect relationships. According to the Video Arena leaderboard, managed by Artificial Analysis, the model achieved an Elo score of 1,247 and took first place. In this evaluation, people anonymously compare model outputs and vote for the better ones.

Top 10 text-to-video generators
Top 10 text-to-video generators.

Runway has a team of only 100 people, yet it has managed to outperform giant companies. CEO Cristóbal Valenzuela said that their focus and attention to detail made this possible. The model is code-named "David" after the biblical story of David and Goliath, reflecting how a small company is taking on industry giants. Runway specializes in AI research and world models that learn from videos and real-world data to better simulate physics. Its customers include media organizations, studios, brands, designers, creatives, and students. The company has reached a valuation of $3.55 billion, which is approximately CZK 81.65 billion at the current exchange rate.

Improvements in Gen-4.5

Gen-4.5 delivers major improvements over previous models, such as Gen-1, which two years ago became the first model to allow the public to generate videos from text. The model can now handle complex scenes with many elements, with everything moving naturally. For example, given the prompt "Water flows from a garden faucet and fills a rusty metal bucket; beneath it is a perfectly folded white paper boat with a small yellow flower; the boat floats on the stream of water into a house and stops in front of a broken television," it creates a video without cuts in which the water flows realistically and the boat moves smoothly.

The model excels at detail: objects have the correct weight and momentum, liquids flow naturally, and surfaces feature fine details such as strands of hair or material textures that remain consistent throughout movement. It can also handle complex compositions, such as "A polar bear sits curled up inside a transparent block of ice attached to a small silver trailer; the camera follows the trailer being pulled by a rope at eye level, then pans right to reveal a goat pulling the trailer." The video shows the bear inside the ice being pulled by the goat, with natural movement.

In terms of physics, the model simulates real-world collisions and movement. One example is the prompt "An anthropomorphic cactus character hugs a red balloon character, and the balloon character bursts," in which the balloon realistically pops during the hug. Another example is: "A diver on the ocean floor fires a machine gun; the camera is handheld and imperfect, occasionally off-center" – here, underwater movements behave naturally, with bubbles and momentum.

For characters, the model captures emotional nuances, natural gestures, and facial details. For example, "A young woman with straight blonde hair and freckled skin sits quietly, looking up through her eyelashes with a vulnerable expression" creates a video with subtle expressions. The model can also handle dramatic scenes, such as "A brunette woman with an intense and serious expression approaches the camera, then raises her hand to reveal a pistol and fires several times toward the camera."

Different Styles and Model Limitations

Gen-4.5 handles a wide range of styles. In photorealistic mode, it creates videos that look like real footage, such as "An uninterrupted underwater shot with no cuts; the camera moves forward underwater past various floating objects; the light is dim and filters through the water's surface." Non-photorealistic styles include animations such as "An elegant country fox wearing a straw hat and a suit rides a motorcycle along a dusty road through the countryside, in a hyper-detailed style inspired by stop-motion cinema."

The model can also handle ordinary scenes from everyday life, such as "The scene shows a television with a cartoon dog from the 1990s; a hand turns it off; the camera moves down to reveal 'Runway' spelled out in cereal in a bowl of milk, then moves back up to a refrigerator with 'Whisper Thunder' written in magnets." Or cinematic scenes: "A child on the back of a mandrill beside a group of mandrills running across a plain, cinematic style with fast motion blur, handheld documentary style, low angle."

Nevertheless, the model has limitations: sometimes effects appear before their causes, such as a door opening before the handle is pressed; objects may disappear or appear unexpectedly, such as a cup vanishing after being covered; and actions often succeed even when executed poorly, such as a badly aimed kick that still scores a goal. Runway is working on these issues, particularly for world models intended to accurately simulate the consequences of actions.

Comparison with Competitors and Availability

Gen-4.5 outperforms models from Google (Veo 3 in second place) and OpenAI (Sora 2 Pro in seventh). Cristóbal Valenzuela emphasized that this achievement follows seven years of work and that AI should not be monopolized by a handful of companies. The model was developed on Nvidia GPUs, including the Hopper and Blackwell series. Collaboration with Nvidia helped optimize both training and speed. Jensen Huang, president and CEO of Nvidia, said that this is an exciting time for video and world models, and that Nvidia is helping Runway revolutionize the industry.

Gen-4.5 is gradually being rolled out to all Runway customers at a price similar to previous models. It will be available through the Runway platform, its API, and partners. The company plans to add modes such as image-to-video, keyframes, and video-to-video. Runway works with partners in commerce, marketing, broadcasting, and gaming.

The model maintains the speed and efficiency of Gen-4 without sacrificing quality. It is the first of several major releases Runway has planned.

Category:AI
Did you enjoy this article?
Discover more interesting posts on our blog
Back to blog

Related posts

Altman Announced the Singularity Days After His Models Escaped the Lab on Their OwnAltman Announced the Singularity Days After His Models Escaped the Lab on Their Own
OpenAI chief Sam Altman declared on the Relentless podcast that humanity has already entered the singularity. “We’re like, in the singularity now,” he said verbatim. For decades, the term belonged more to science-fiction literature
6 min read
28. 7. 2026
AI Remixed a Madonna Song—and Now It Tops the Charts in Australia. Musicians Are Furious.AI Remixed a Madonna Song—and Now It Tops the Charts in Australia. Musicians Are Furious.
Since April, Australian radio has been playing a dance remake of Madonna’s hit Like a Prayer on repeat. Released by Queensland DJ Josh Fawaz, it tops the radio airplay chart and has 35 million Spotify streams.
6 min read
28. 7. 2026
Claude Opus 5 Built a Shooter from Scratch. What Can Claude of Duty Do?Claude Opus 5 Built a Shooter from Scratch. What Can Claude of Duty Do?
A first-person shooter that runs directly in the browser, with its own physics and eleven separate code modules. Around 55,000 lines in total, split across eleven subsystems and built on Thr
4 min read
28. 7. 2026
Přihlaste se k odběru našeho newsletteru
Zůstaňte informováni o nejnovějších příspěvcích, exkluzivních nabídkách, a aktualizacích.
CodedTrip

Operated by CodedTrip LLC, USA.

YouTube
TikTok