Mixed feelings and certainly don't want come across over critical. I think with the tools you use it's probably the bet you can get and look great. The main problem is the limitations of veo3 and other models as it stands. The physics is still off. It's really something that needs to be finetuned and I reckon a good Wan2.2 trained model would cope better but then you have to work within the other limitations of that model i.e. shorter clips, no audio etc.
I found it very impressive. I especially liked the very explosive effect of the thrown pies. It's not 100% realistic, granted, but I find I actually enjoy the exaggerated pastry constellations that impact briefly creates. Since you don't show any of those in the still images above, I made some screencaps from your video to demonstrate what I'm talking about
Gurph Morlicks said: I found it very impressive. I especially liked the very explosive effect of the thrown pies. It's not 100% realistic, granted, but I find I actually enjoy the exaggerated pastry constellations that impact briefly creates. Since you don't show any of those in the still images above, I made some screencaps from your video to demonstrate what I'm talking about
Yeah, I like tose moments too. Actually the cream explosion is oftentimes pretty good, it needs little timing tweak here and there in post production, cause there sometimes are timing issues with the pie hitting and after a standstill suddenly explodes. what's more difficult is to have the pie actually be tilted correctly, so it does not hit the target horizontally.
messg said: Mixed feelings and certainly don't want come across over critical. I think with the tools you use it's probably the bet you can get and look great. The main problem is the limitations of veo3 and other models as it stands. The physics is still off. It's really something that needs to be finetuned and I reckon a good Wan2.2 trained model would cope better but then you have to work within the other limitations of that model i.e. shorter clips, no audio etc.
No worries, I am happy about any feedback! There are certainly limitations and Veo3 still lacks consistency, so in order to achieve good result you oftentimes have to create several outputs and tweak the prompt here and there. There's no one prompt that works for any scenario. But it is still amazing how good AI video has become in such a short period of time. I still remember my first Dall-E images that were barely recognizable as human beings just a few years ago.
There's not only limitations to physics - though I have to say especially with liquids Veo has become very good recently. But it's also certain body movements some models cannot yet reproduce. Im sure there are already better ways out there for AI video creation, but I'm quite happy with myworkflow at the moment.