@Theo Shinton I've used them all and the truth is most of the time they leapfrog each other when new models are released and the cycle repeats. I realize you are on a budget so first look at Google tools in their studio for anything you can access for free. Even on a budget, I always suggest people think about time to create. If you use different tools for audio, images, graphics, etc. there's a time cost to that. When VEO 3 was released, that was the main benefit. One prompt, then you received video, audio, with good lip sync, but you only could get 8 seconds of video at a time. I will tell you that most of the AI video tools are moving towards a we'll do everything for you so keep that in mind as you are learning. Each tool requires a little different prompting to achieve the best result. Right now, Heygen is probably my favorite AI video tool. They have a brand-new video agent model released a few days ago which does everything for you and the results are pretty good. It's not the cheapest but shows you the current capabilities for an all in one. I've attached a 3 1/2 minute video about the dangers of AI. (ironic) The video was entirely created as a one-shot prompt. (no editing) It's certainly not perfect, but right now I would argue it's better than most AI video creation tools because the creation is seamless and easy. All you need is a really good script and it does the rest.

@Theo Shinton If you create your Avatar own using images, to get the best results you need different looks, head positions, angles, etc. If you are using one of their own in-house avatars, they typically already have those, and it appears their new system can easily create it if they don't. All I did for the video I uploaded, was tell Heygen to use the character (who was standing in a previous video generation) to place him in a seated position and it created the background image you see. It's understood the context pretty well so far, so you get something that typically works. There is also an editor now where you can adjust the video output, but of course the goal is to prompt well enough to not have to do that very often.