Nice summary! I missed the mention of EQ-VAE when it comes to generation quality. Tiny trick, huge impact! Have you tried it?
This seems like a great model to experiment fine tuning with original art, given it’s relatively small and with open license. Is that a fair assessment?<p>Thanks for the great write up and making it available to us all.
Hi HN, I’m one of the two authors of the post and the Linum v2 text-to-video model (<a href="https://news.ycombinator.com/item?id=46721488">https://news.ycombinator.com/item?id=46721488</a>). We're releasing our Image-Video VAE (open weights) and a deep dive on how we built it. Happy to answer questions about the work!
I take my children to school to learn them how to use English grammar.