Quicksand Fans

Today, KlingAI launched their new model, 2.0 "Master" for its animation and image generation tools. I've been eagerly awaiting an updated model given what I've been making with Kling 1.6. The new model boasts better dynamic interaction and better prompt adherence.

I am, however, disappointed. I've attached a brief sample of testing. For non-QS scenes (such as filler or setup scenes, as shown in "Storm20"), the dynamic effects are really, really good. However, the new prompt adherence is tough to work with. It tries TOO HARD to make everything more dynamic, My original prompt had "rain makes her hair whip over her face" during a "violent storm", and it splashed water everywhere, threw lightning bolts every few seconds and wrapped her head in hair. The one I uploaded removed all the extra details and produced a pretty good dynamic scene... that I didn't actually ask for.

For QS scenes, I've have very, very poor luck. It does better at things like coating arms and clothing in mud. However, it still has the random success rate depending on the image prompt. In this case, I actually get worse results more often than with 1.6. "Bra20" is the typical output, but it struggles with dunking characters face-first, as it did before in 1.6. "Kling20 Bikini" is my best output, but that uses an image that already consistently had good results in 1.6. The dynamic movement is pretty good and it kept the mud texture together in this scene.

However, that brings it down to one key problem: cost. Master 2.0 costs nearly 3x as 1.6 generations, and takes about 3x as long. A 10s animation at Professional quality on 1.6 is 75 credits - about 90 cents (US), with Standard quality being half the price. 2.0 Master only has one setting, which costs 100 credits for a 5s clip or 200 credits for a 10s clip (about 2.6 US dollars). Given that each generation can now take about 10 minutes (as opposed to 2 minutes) and the high failure rate, this is far too expensive to reliably create AI animations with.

As much as I'd love to push the new model to see what I can make with it, it's not feasible with its current pricing.

Biggest problem I've had with Kling AI is the ability to try it.

They want you to sign up only with a "popular email service". No, I'm not going to create a gmail account just so I can use kling. It's a paid for service, why can't they just use the email I've been using for years?

Yeah, I have found Kling to be quite sporadic in its results. Here is two things I have made recently. The foot one was done using the motion brush tool which I really like.

I love the look of the Storm one!!
I’d love to see a whole scene of that ending up in quicksand fatality!

Acidtester wrote:I love the look of the Storm one!!
I’d love to see a whole scene of that ending up in quicksand fatality!

Oh yea, that output turned out really nicely. Again, it's like a wild horse - I had to temper some of the prompts to get the scene to relax a little bit, but 2.0 is capable of some very dynamic scenes - just at a really high cost. It may actually be good for those 5-second clips that set up a scene well, whereas the 1.6 renders are sufficient (if not better) for the QS segment itself at lower cost.

mjw wrote:Biggest problem I've had with Kling AI is the ability to try it.

They want you to sign up only with a "popular email service". No, I'm not going to create a gmail account just so I can use kling. It's a paid for service, why can't they just use the email I've been using for years?

I used a burner iCloud email address to sign up.


As for 2.0 in general, don't have much to add to what Viridian already said; the results as far as mud texture and character movement are great, but actual sinking seems to be even more of a crapshoot than it already was. I have way too many "faceplant" generations compared to 1.6.

And yeah, it's expensive. I have a monthly subscription and I'd only be able to generate about eight five-second videos, counting the ones that don't turn out the way I want, before I'm out of credits.

To add, 2.0 renders in 720p, while 1.6 renders in 1080p.

It costs 3x as much, takes 5x as long, lacks the functionality (no Elements, no camera motion) and the interpretation of prompts is unpredictably wacky.

To Kling's credit, 1.6 is just that good. But instead of experimenting with 2.0, I'm finding better ways to use 1.6.