Kling 3.0 vs Gemini Omni Flash vs Grok Imagine 1.5 vs Seedance 2.0. Another battle for the best AI video tool. This time I tested an extremely difficult stunt scene: a bridge jump, landing on a moving truck, then jumping onto a car and taking it over. No model... handled it perfectly. Not even Seedance 2.0. The scene itself was very hard, but I kept the prompt relatively simple, without stuffing it with too many complex stunt terms. After Veo 3.1’s terrible results in previous rounds, I didn’t even include it this time. Seedance needed 4 attempts to give me a decent result. Gemini Omni Flash needed even more: 7 generations, and I picked the best one. Kling and Grok got 4 attempts each. None of them looked promising enough to justify more retries. My take: Grok Imagine 1.5, the new “revolutionary” release, once again showed that it still can’t handle complex action scenes properly. Two tests in a row now, and the result is objectively weak. Kling 3.0 feels outdated, but in this test it actually looked slightly better than Grok Imagine 1.5, which says a lot. Gemini Omni Flash gives decent movement, but the physics and action logic are clearly off. It’s better than Grok and Kling here, but still very far from the super-generator Google markets it as. Seedance 2.0 was the best again, as expected. Still not perfect. It also has a clear physics issue, because pulling a driver out of a moving car while standing on the road makes no physical sense. But overall, Seedance still produced a much more intense and cinematic scene than all the others. My ranking for this test: 1. Seedance 2.0 — clear winner, even with flaws 2. Gemini Omni Flash 3. Kling 3.0 4. Grok Imagine 1.5 — unfortunately, the weakest right now in my opinion Do you agree? #AIVideoshow more

Грок только пиздеть может какой крутой, а на деле баганая пиздятина! Скоро месяц будет как его нахуй весь сломали и с каждым днем все сука хуже и хуже. Исчезла подписка, лимиты перестали обновляться полностью. Зато хваисаются моделью новой. Уебки блять

yeah, I’ve seen a lot of people saying the same thing. The worst part is when they hype a “new model” while the actual product experience keeps getting worse: broken limits, missing subscription stuff, console issues. for creators, that matters more than marketing

Grok imagine it’s just a joke :)

after two tests in a row, it's getting harder to argue with that 😅

SD2 wins again! but i liked kling.

Yeah, I like Kling too. It’s still useful, but for complex action scenes Seedance feels stronger to me.

Crazy. Good job

Thank you! Really happy you liked it

👀💫

I'm surprised of how well kling keeps delivering decent results, is my workhorse, cheap and reliable.

Totally fair. Kling is not the strongest anymore, but it still gives usable results more often than people expect. For complex action I’d still pick Seedance first, but Kling is definitely not dead.

Kling é o melhor... Seedance é basicamente mais frenético e só.

Kling é uma boa ferramenta, concordo. Mas para cenas de ação mais complexas, pra mim o Seedance ainda funciona melhor.

Pode ser.. é que realmente toda vez que vejo alguém postar Seedance é com cena de ação frenética. Mas, pode ser só percepção minha.

Sim, faz sentido. Seedance aparece muito em cenas de ação porque é justamente onde ele mais impressiona. Mas seria interessante testar também cenas mais calmas e realistas.

wtf happened to grok imagine… god. I really think they are keeping everything nerfed so they can sprinkle good news during IPO launch

something definitely feels off. I don’t know about the IPO theory, but the current Grok Imagine experience is nowhere near the hype. The gap between marketing and real creator use is pretty obvious.

Hi! 👋 Just scrolled through your feed and I’m obsessed with your videos. I’m reaching out from @renoiseai. We’re a new AI video tool. We’d love for you to join our Creator Program. The perks are pretty straightforward: 1. Welcome Bonus: 5,000 credits just for joining. 2. Weekly Challenges: Create and post based on our themes to earn cash rewards based on your engagement. 3. Passive Income: Upload your assets to our site and earn a commission every time another user reuses them. Check out the full details here: [ Let me know if you’re interested :)

@renoiseai Thank you so much! I really appreciate the kind words and the invitation 😊 I’ll take a closer look at the details and think it over.

@renoiseai 🫶🫶

That's a tough challenge! It's interesting to see how each tool handles complex scenes like that.

Yeah, this scene was brutal for all of them. That’s why I like this kind of test — it shows the real limits very fast.

why is the swarthy woman identical across all the models? some fuckery is afoot.

@scumbagdotdll that is actually another weakness of current video models. Same prompt should not mean almost the same face every time. I was mainly judging motion, physics and action logic here, but the character similarity is definitely something worth noticing too.

I like tests like this because they expose the real bottleneck: physics and temporal consistency. Generating beautiful shots is one thing. Maintaining believable action across multiple interactions is much harder.

That’s exactly why I use complex action scenes for these tests. Single shots can look amazing, but once several movements have to connect, the weaknesses show up immediately.

So big disproving your work just more curious, when testing these did you try a pencil style storyboard? That for me tends to make the action sequences better

Good point. I didn’t use a pencil-style storyboard here. For real projects, I often use GPT Image to help the model understand the scene better. But for this test, I wanted to see how each video model handles the prompt on its own.

If you can test it with pencil style storyboards that would be very interesting !!

That’s a really good idea. I’ll definitely try that in future videos and see how the models handle it.

Seedance still wins despite flaws.

Yep. It still makes mistakes, but the overall scene feels way stronger than the others.

Hey@YourAlphaMom, loved your test of those AI video models! I’d love to invite you to create videos with them on our platform — I’ll give you free credits. Please DM me!

Thank you! I’m really glad you liked the test, and I appreciate the offer. I’ll think about it, but for now I’m pretty comfortable with my current workflow and tools, so I’m not really looking to switch platforms yet. Still, thank you for the free credits offer!

In Gemini Omni, the woman appears to be walking in the air, and the image looks extremely artificial. There is also a serious inconsistency in the size and scaling of the characters. Despite this, placing Gemini Omni in second place shows that your comparison is not really objective.

You’re right about those issues. I still placed Omni second because, to me, the other results had even bigger problems. But I respect your take, and thanks for the detailed feedback.

Seedance always wins 🏆

Hard to argue right now 😄 It keeps surviving the hardest prompts better than the rest.

This is amazing

Thank you so much! Really glad you liked it

Omni is going crazy with the camera angles!

Yeah, Omni’s camera work was honestly the best part of its result. Some shots looked really cinematic, even when the action logic started breaking.

Are you sure you are using Grok 1.5 and not Grok 1.0?

Unfortunately, you might be right. I checked again and it looks like I may have made a mistake here. From what I can see now, the new Grok model doesn’t seem to be rolled out on the official site yet, only through the API. So yeah, this one may be on me.

@Helicoptarion well you should take down your video then

I get your point. I did make a mistake with the grok version, and I already acknowledged that in the replies. But I don’t think deleting the whole video makes sense, because it still compares the other models and the grok version available to regular users on the official site. Better to leave the discussion visible than pretend it never happened.

seedance2>>keling>=gemini omini>>grok 1.5

Pretty close to my ranking too. I’d still keep Seedance clearly first, then Omni/Kling depending on the scene, and Grok last for complex action.

seedance 2.o is crazy

It still has the biggest edge in complex action scenes. The others are catching up, but not quite there yet.

这种连续动作现在还是太刁钻了

是的，这种连续动作太容易把模型打崩了。 单个镜头还好，一旦要连续完成几个动作，物理和动作逻辑马上就露馅。

have you work with vidu ?

not yet. I haven’t tested Vidu in these comparisons, but maybe I should add it in a future round.

Fantastic!! Thanks for the breakdown on which to use. I'm just learning all the different types of platforms and it's wild.

The opening camera movement in that Google Omni demo was absolutely insane! 🤯🎬

Yeah, that opening shot was easily Omni’s strongest moment. The camera move looked really good, even if the physics fell apart later.

Seedance still feels ahead, but none of them are really nailing real physics yet

Exactly. Seedance wins the scene, but real physics is still the wall every model keeps hitting. It just hides the problems better than the others.

Hi, your AI work is really impressive! We’re hosting a Story Sprint event at Anamana for AI storytellers — a chance to try Anamana Studio and win cash prizes. Link in bio / details here: 🙌

Thank you! Appreciate the invite. I’ll take a look when I have some time.

Great! Glad to hear you’re interested. Let me know if you have any questions — would love to see you apply😊

tested a few of these for actual client work. kling 3.0 won on physics but grok imagine 1.5 nailed the landing consistency. the real bottleneck isn't the model , it's how fast you can iterate. i'd pick the one that lets you batch 10 takes in under 5 mins. speed > specs for rev...