This was indeed a difficult assignment. Details shifted (like the droplets on the bottle), text broke (like the small text underneath the logo), and especially the proportions do not seem right, as this is supposed to be only a 0.2L bottle. I tried a mixture of seedream 5 lite, nanobanana 2, and gpt image 2 for the frames. Then I tried both seedance 2 reference to video and first frame to video. My question: these generations are done with 720p (in order to spare credits), would 1080p have had a higher succes rate on text preservation?

Hi Hugo, is it ok to mix and match models when creating videos so you can take advantage of each models strengths? Or do you find it better to use just one to ensure you get more continuity? Thanks!

Hello frends! From now on we'll be sharing weekly updates on Tuesdays. But you know how this is... only if it's worth it, we are not AI influencers 😀 Today though, a couple things to share: - Maria Vinagre, my partner at HumAIn has join the group (finally has some time) and will be making some very cool things (PRO-Grade Workshops) in the next weeks. - This Thursday is the first live Q&A, check the Calendar tab and join if you can :) - Freepik is now called Magnific. Usually I wouldn't care this as worth update, but most of our training is done in Freepik as it is our go-to tool. Well, they change name but the rest remains the same. Always small here and there UI updates but what's really worth it is in the same places. Have a great day fam!