Hey guys,
As the title suggests, I'd like to know people's thoughts on AI video generation.
The purpose for which I intend on using it for is completely non-music related.
But there are so many out there I have no idea what is good and what is not.
If anyone here has had any experience with any platforms, please share your experiences in this post.
I can't really afford to spend a ton of money on another subscription. I'm broke enough as it is. But I have an idea I want to explore. So more interested in platforms that give bang for your buck.
Cheers

 

What do you want to generate and for what purpose—your music? What are your goals? Windows 11?

 

Windows 10.
Absolutely nothing to do with music.
My wife and I have had an idea for a gothic vibe channel.
Just wanted to use an AI generator to use for short clips, promo, etc etc. The majority of the content will be filmed on set. Just wanted to actually upload my wife's face and have AI generate videos using her features and have videos for promo etc being in a castle for example.
The Goals are to have a specific genre interested in that style who would subscribe through sites like Patreon.

 

for exploration you want local AI, so you'll need rather beefy PC (ideally at least 24GB nvidia gpu and 64GB ram),

as UI frontend you can use wan2gp or ComfyUI,
as backend models, it's Wan 2.1, Wan 2.2, Hunyuan, LTX-2 and others you can choose from

example: https://civitai.com/images/117458803

 

Another option is to generate images with AI and then string these generated
images together using video editing software, perhaps adding captions.

 

that's actually very relevant advice still, best AI video quality is achieved by using reference images like storyboards, combined with video/controlnet motion guides

 

following this ..

 

Kling if you want higher resolution cinematic widescreen clips. But there's a fail rate with any gen subscription. Only maybe 50% of clips will be usable.

Veo 3.1 is free if you want vertical mobile phone ready clips, but it'll have their watermark on it (you can remove that with a good video editor)

 

LTX-2 is more resource friendly than Wan for local inferences atm, there are destilled versions that can run from 8 to 16gb VRAM. And it renders audio as well, is a crazy good model for it's size.

But ComfyUI has a little learning curve, is not like closed source web applications.