I have started looking at what options there are for text to video Ai, short clips with movement has an appeal alongside still images.

What options have you had success with, quality of result, simplicity of use, ease of signup and not heavily restrictive are key.

I have tried Luma Labs, and from what I've seen everything else available to the casual user is the same quality or worse.

Frankly, I don't recommend it. It doesn't really work for wam videos because it hasn't been trained on what wam is. Frames don't really follow on from the previous frames, sometimes you might get something close to what you want but not really. Prompting actions doesn't work, it will do its own thing. Sometimes the results are nightmarish body horror. I compiled some of my better results into this short video. The is a pick from 150 generations (which were prompted from images) and it cost me about £30 to make.

https://umd.net/profile/i/thereald-3/section/videos/video/splat-the-brat--video-test

I know there is some open source stuff about, that if you have a good GPU, and patience, and time you can get some better results from, if you're prepared to train it. But it's still a bit rubbish.

Another one that's starting to be discussed is klingai.com. It's hard to judge since they don't offer a paid sub and you can only do a couple gens per free account but it's already blowing Luma and Runway out of the water in terms of prompt adherence and model consistency while also being very light censor wise.

I'd recommend using a burner account as the website looks a bit unsafe though.

Bumping this thread to add that Kling now understands image to video descriptions with pie throwing / slime etc. It even works as a prompt for non-wam images.

Of course the physics are still wonky and the results are hit or miss but this is already a gigantic improvement compared to even three months ago.

https://youtube.com/shorts/1qFdgHY8el0?si=RB22QDcXBqxPeG_s

Hello,

I've recently been introduced to Hailuo and found their results to be amazing. Having a number of fetishes and passions, I've been trying a variety of combinations in the videos. It's still a bit hit and miss - especially where the mess ends up - but some of the results are incredible. The scenes are only 6 seconds long but prompts have been pretty good, although if you ask for too many things not all of them will make the scene.

https://hailuoai.com/video

However, while I've had some incredible results with "foam" and "creamy pudding", I'm still struggling to get pies to look convincing, be held properly, or be thrown. Any advice or suggestions are greatly welcomed.

GP

I really don't know how you get Hailuo to work with pies. Don't get me wrong, it's clearly the most advanced video AI option right now but it still feels really far from creating a realistic throwing movement.

Can you share any prompt or prompt component you have been using regarding pies?

That's the thing: I haven't got pies work convincingly yet. Gungings have been good but still working on the right phrase of the liquid and how it falls and sticks to the victim's body and clothing, be something about the throwing mechanics aren't working yet. I have an approach to try - "slapping" might work, or "clapping" for delivering pie sandwiches - but it is just a hypothesis and not yet tested.