Why isn't it more popular?

[zweifisch]

I've created several videos that seem to match the quality of those generated with SVD. Even though this project has a more permissive license, it's far less popular than SVD. I wonder why this could be? Is it the computational efficiency?

[Doubiiu]

Thanks for your recognition of our work. I really appreciate that. Personally, umm some reasons could be:

1. StabilityAI is more widely recognized for open-source AIGC projects, like Stable Diffusion, DeepFloyd, and is well-supported by the community (thrid-party integration, optimization, etc.).
2. Currently we are focusing more on different kinds of research projects (different applications) instead of computational efficiency. So yeah the computational efficiency is a reason.

[AdaptiveStep]

They should focus on making it faster, so it can be generated in realtime (turboXL) using only a 4090. That is key to making it spread out. If they require industrial GPU's such as A100, it wont be spreading as much. Just look at grok, a free 300B parameter model but nobody cares since its requirements are too large.