devs pls fix
Really poor. In the last shot, looks more like smoke.
The fact the director even included this item in the reel speaks volumes.
But it represents a coming price pressure that will face the leading foundation models. Open source will prevent runaway costs.
Moreover, it prevents the hair-trigger platform safety checkers from shutting down creative work. Video models are notoriously bad at shutting down a huge number of requests.
Creatives will prefer to work on cloud or private GPU clusters. Waiting 10 minutes for a few seconds of 480p is unacceptable. Hobbyists will have fun, but most actual production work is happening in the cloud.
Artist's time is worth money, and they like to spin up dozens of concurrent generations at a time to more quickly explore the generation state space and make progress on completing work.
According to one user preference leaderboard, MiniMax H3 is already ahead of Seedance 2.0 based on thousands of A/B votes: https://artificialanalysis.ai/video/leaderboard/text-to-vide...
I haven't seen any user preference comparisons between Seedance 2.5 and MiniMax H3. As an upper bound, H3 cannot be more than 6 months behind Seedance 2.5 since H3 is already ahead of where Seedance was 6 months ago.
SOTA a year ago was Kling 2.5, and H3 does not look or perform at that level.
The Artificial Analysis rankings are whack. They rank Omni first, which is incredulously wrong. Google's models broadly suck, and there they all are - right at the top.
Artificial Analysis has notoriously ranked models such as Grok Imagine highly and continues to rank Happy Horse as a good model despite the model being absolute garbage.
Could it be because they are subject to broad based statistical attacks? It's easy to encode information about the origin of media in either its metadata or output frames. Or maybe there's simply no overlap between creatives and people who click on ELO scores.
I've spent thousands upon thousands of dollars generating video. I will stand by the claim that nothing touches Seedance 2.0 / 2.5.
It's good that we're getting better open weights. It puts price pressure on the foundation model companies. But these open models are not a substitute for Kling or Seedance yet. Not even close.