Kling 2.0 Review: State of the Art AI Video Quality

Kling 2.0, an important upgrade to the state-of-the-art ai-videoerator that was released by the Chinese technology company Kuaishou, came on the market last week to a stream of breathtaking reactions from makers, who quickly burned hundreds of dollars to test its capacities.

“AI video quality only 10x at night. I am speechless,” tweeted AI filmmaker PJ Ace, who already claimed to have spent $ 1,250 on credits that explore the limits of the tool. “I have never seen this liquid movement or asks this accurately.” The post collected more than 757,000 views, which emphasized the buzz around this release.

The new version marks an important leap forward from Kling 1.6, offers an improved fast understanding, more liquid character movement and improved visual aesthetics that users describe as “filmed, not generated”. Kling 2.0 in particular can generate videos Up to 2 minutes longAs a result of which competitors such as OpenAi’s Sora remain in the dust when it comes to extensive narrative options.

“In general, Kling retains the first place in the rankings,” said the YouTuber Tim Simmon, who specializes in assessing generative AI models, in his judgement. He believes that it is the clear winner in generating image-to-video, where the competition comes closer when it comes to a direct text-to-video generation.

This new version arrives in an increasingly drank AI-Video generation market. Competitors include Runway, known for High-Fidelity-Output-Die recently released its V4 model, aimed at cinematic results and Google’s VEO2, with its robust text-to-video options and aesthetically pleasant results.

Until now, the model has yet to be seen on artificial analysis’ Video Generator Leaderboard-Those the best generative video models therefore ranks its predecessor, Kling 1.6 is already the leader in image-to-video and is in second place on text-to-video based on blind tests.

Kling 2.0 has a multi-element editor, with which users can add, exchange or remove video content using text or image input.

The platform also introduces two specialized components: Kling 2.0 Master for video money and Kolors 2.0 for making images not to be confused with another open-source Chinese AI image generator that was released Under the same “color” name– Makers of makers more control over their outputs.

The focus of the cinematic quality tool makes it particularly attractive for filmmakers, marketers and content makers. The model is extremely powerful in terms of sources, with generations that take hours in the free plan and up to 16 minutes for almost 5 seconds video on online platforms.

The prices start at $ 29 per month for the standard plan, which includes professional mode, videos of 8 seconds and a reimbursement of 30 videos per day. A free plan offers 6 daily generations with limits and water brands of 4 seconds. The professional plan, for $ 89 per month, provides a high resolution, advanced movement controls and priority processing.

Testing the model

We have tried the new model in five categories of dynamism, illustration, text-to-video, structural coherence and multi-subject coherence. This is what we have found.

Dynamic

All video drivers still treat well scenes, but usually struggle with fast movement, complicated scenes and dynamic setup. This reflects real video or animation break your TV during a “Tom & Jerry” yacht or a campaigned war scene, and you will spot weird frames everywhere.

We tested the model with a stationary image of a man who flew through a city and asked to generate the scene.

Kling 2.0 turned out to be extremely sensitive to small rapid changes. Our first attempt used: “Dynamic Tracking Shot: A Man flies with extremely high speeds in a bustling Stadsstraat. The camera follows closely behind, the current of buildings and traffic that is looking for, improves the feeling of speed and excitement after he takes a sharp turn.”

Unfortunately, the promptly generated the illusion of a subject that was upholstered back on the street. This was probably due to our choice of words in the prompt.

So we only removed one word: “Behind.” That changed the result and produced a much better video that flies the subject ahead, opposite the camera.

Kling laid the most important scene elements solid-dynamic and rapid movement, although the body of the subject changed strangely when changing direction, and some elements missed a uniform structure. Other models such as the VEO2 trading dynamics of Google for realism, creating slower, static, but more coherent scenes.

Illustration

Prompt: “360-degree horizontal pan: a bustling city complicated around a huge tree, filled with houses and bridges. The camera moves smoothly from the front to the back of the tree, who catch children play, people who have daily activities, and flying cars that land on branches and perform, all under a warm, invited atmosphere.”

The model excels with imaginative styles such as comics and illustrations, but is struggling with small details. It gives priority to coherence over detail, with respect for the most important fast elements with a smooth camera movement and a liquid scene.

Object structure remains solid without the Wiebel that is seen in other generators, although some children (those small details would be past the original structure of the entire composition – a tree and the hustle and bustle) lose coherence and flying cars occasionally disappear.

Nevertheless, this test yielded the best results that we have seen of each video drum.

Text-to-Video

Prompt: “A blonde woman in a red dress and an Asian man in a black pack of chat in a Starbucks. Medium shot.”

Text-to-Video presents unique challenges for AI generators. The model must make a first frame (essentially a text-to-image task) and use that as a reference for all following frames. Ideally, you want a specialized image generator for that first frame – and ideally also for the last frame if you want the best coherence.

Kling 2.0 does not seem special here – but it is not bad either. The scene has the characteristic airbrush style that is common for many image generators, but bodies retain the right structure, fingers seem accurate and there are no noticeable artifacts that disturb the scene.

It is an improvement compared to Kling 1.6, but not what the model was designed for.

Structural coherence

Prompt: “Air reproduction: shot of a complicated, abstract architectural structure that rotates.”

Although Kling can struggle with small details in busy scenes, it excels in maintaining coherence and details in Schoten with one subject.

We shared an image of a complicated piece and asked the model to rotate it. Kling 2.0 treated this almost flawlessly – the Enlightenment remained consistent, movement was uniform, no artifacts appeared and the structure maintained its integrity.

This possibility makes it potentially valuable for 3D modeling, which makes it possible to object and scene examples from different angles.

Multi-subject coherence

Prompt: “Five gray wolven puppies are playing and haunting each other around a remote gravel road, surrounded by grass. The puppies run and jump, hunt each other and buttons to play, play.”

This remains the Achilles’ heel of all video models, including Kling 2.0. Since OpenAi has shown that Sora could not generate a lot of baby animals that played together, all videoerators have tried this challenge with mixed results. No model consistently achieves perfect results.

Kling 2.0 generated a lively, realistic enough scene, but the wolves merged together, appeared and disappeared between frames. If the only coherence analyzed is, then there is not much difference between Kling 2.0 and Kling 1.6.

One remarkable improvement: The irregularities usually occur in the background, with foruming animals that usually maintain better coherence.

Kling 2.0 is accessible via Kling AIFreepik, Pollo AI And other providers.