I Tested Kling 4.0 on Artlist

Artlist is my favorite platform for AI creative work, so when Kling 4.0 landed there this week, I tried it right away. Kling has been one of my go-to video models for a while, and I was curious how big the jump from Kling 3.0 really is.
For my tests I used my usual “prompt 1”. I’ve been using it for image and video generations for 3 years now, and I extend it a tiny bit every time a model can do more. It’s a handshake between a woman and a robot in space, and it’s a nice way to compare models over time. I ran two tests and changed how I set up the references each time.

What’s new in Kling 4.0

These are the changes that stood out to me:

- Up to 30 seconds per clip. That’s double what Kling 3.0 could do in one generation.

- More references. On Kling’s spec sheet you can combine up to 10 images and up to 5 videos (30 seconds combined), with a maximum of 15 references per generation.

- A new 21:9 format next to 16:9, 1:1 and 9:16.

- Up to 10 keyframes, so you can plan how a clip unfolds from one image to the next.

- Up to 4K resolution with stereo audio.

For audio, Kling 4.0 only supports voice references, so you can’t upload a music track as a reference. I learned that after trying to force it, but more on that later.

Named references are a real time saver

You can now save references with a name and a category: character, location or prop. I uploaded a woman as “@Zana”, a cyborg as “@Cyborg” and a purple galaxy scene as “@Purple Space”. Once they’re saved, you just type the name in your prompt and can reuse them in the next video. If you make content for your business regularly, this is handy. You can keep the same mascot, product or setting across a whole series without uploading and describing everything again. I also wanted my own background music in the video. Since audio files can’t be added as a reference, I put my track under a plain white screen and uploaded it as a video reference (@vid1). The clip was too long, but I could trim it right on Artlist. Imagine having to cut it in another program, export it and upload it again just for a reference.

Test 1: One reference and lots of room to fill

For my first test, I used only one reference, @Purple Space, and gave the model the full 30 seconds. I didn’t describe the woman or the cyborg at all, because I wanted to see how much Kling 4.0 picks up from the image on its own.

My prompt:

> Animate a warm handshake between the humanoid robot and the woman from @Purple Space in slow motion, emphasizing the connection. The robot’s LED lights softly glow in sync with its movements, while the woman’s expression shows joy and curiosity. The Milky Way moves fast from right to left behind them, as if time stopped when they met. The woman says warmly: “Nice to meet you.” The robot responds in a robotic voice: “The pleasure is mine. Lili says hi.” Use multiple camera angles and shots throughout the scene. The music from @vid1 is the background music for this video.

The generation took a bit longer than usual. I guess a lot of people wanted to try Kling 4.0 that day.
The result was hilarious. A handshake and two lines of dialogue is not much for 30 seconds, so Kling filled the time with its own ideas. The robot mixed up the whole conversation, and at some point the woman pulled out a microphone and started talking gibberish.
So if you pick the full 30 seconds, give the model enough action to fill them. Otherwise it gets creative.

Test 2: Named characters, location and music

For the second test, I shortened the clip to 15 seconds and pointed to every reference by its name. I still let Kling 4.0 choose the camera angles itself. My prompt:

> A warm handshake in slow motion between @Cyborg and @Zana, set in @Purple Space. @Cyborg keeps its exact look from the reference, and its LED lights glow softly in sync with its movements. @Zana keeps her exact look from the reference, including her blond hair, and her expression shows joy and curiosity. Behind them, the Milky Way moves fast from right to left, as if time stopped when they met.
>
> @Zana says warmly: “Nice to meet you.” @Cyborg responds in a robotic voice: “The pleasure is mine. Lili says hi.”
> Use multiple camera angles and shots throughout the scene.
> Audio: Use only the music from @vid1 as the background music for the entire video. Do not show any visuals from @vid1. No other music.

This one I like. Kling 4.0 followed my @Zana character well and added two cuts on its own. It did mix up the references a little, though: the cyborg came from the @Purple Space image instead of my @Cyborg upload. And it ignored my background music again, so the white-screen trick didn’t work in either test.

Where you’d use 21:9

21:9 is the ultrawide cinema format. Most Hollywood movies are shot in a similar ratio, which is why you see black bars on your TV when you watch a film. It’s also the format of ultrawide monitors and is supported on YouTube, Netflix and Prime Video. For a small business, I’d use it for brand films or ads that should feel like a movie, wide hero videos on a website, YouTube intros or big screens at events and in shops.

My tips for Kling 4.0

- Name every character, location and prop in your prompt, and use the exact names you saved.

- Start with 10 or 15 seconds. Only go to 30 seconds when your prompt has enough happening.

- Check whether a reference image shows more than you want. If your location image also has a character in it, the model might use that one.

- Add your own music afterwards in your editor.

My verdict

Kling 4.0 is a big step forward. Getting 30 seconds in one generation is impressive, and the videos look great. Kling will stay my favorite model for start and end frames. With references it isn’t 100% there yet, and in my tests it mixed them up now and then. Still, I really enjoyed testing it, and I’ll keep using Kling 4.0 for my own videos.