I've been playing around with lots of different video AI tools recently, in particular for the videos I made for Jicarilla's nascent career as a singer...
One of the really interesting tools out there is Google Veo 3, which is available with some of the Google One packages that also give other stuff like Google Cloud storage. I saw a video recently made in Google Veo where a box in a warehouse pops open to assemble an IKEA-style room, and it gave me the idea of using this for the "Building in a Box" concept we have in Splinterlands (the imagery was always hilarious to me). I'll get into some of the pros & cons of Google Veo 3, but first, here's the output:
Google Veo 3 has made the rounds on X for a few specific video styles it renders very well, such as the "interview", the vlogger, and these unpackings. One of the most compelling features of Google Veo 3 is that it automatically generates audio for a clip, and frankly the results are pretty amazing. Look up examples of the interviews or vloggers and you'll find spoken audio also works quite well, and can be added directly into prompts (though spoken words sometimes get rendered as subtitles, so like a lot of generative AI it's somewhat hit or miss). With something like the video I created above, it adds relevant background sounds that make the videos seem much more real and generally works quite well (though it's still not perfectly in sync with some of the motions).
The biggest downside of Google Veo 3 is the cost. Each generation is quite costly and the amount of credits you get with the Google One plans isn't really sufficient to produce that many videos. When taking into account that you still need to do some experimentation because of the unpredictable nature of generative AI, it makes it really hard to use it without overspending. When I'd previously used Runway, I picked a plan that let me do an unlimited number of "relaxed" generations (slow, but free) which would allow me to test concepts around before then committing to something and using higher rendering options for the final render. Can't do that (currently) with Google Veo 3.
The UI is also quite disorganized. It's much less user friendly than most of the other established tools like Kling or Runway or aggregators like Dzine and OpenArt. Speaking of the aggregators, they do have Google Veo 3 as an option but suck even more credits (understandably) when those are used, so I haven't found any way of getting more affordable access to it.
Hope you enjoyed the video above and let me know if you have other ideas for things you'd like me to try, and also if you have experience or thoughts on other AI video tools.