SD arch trained from scratch on Creative Commons dataset
Generate videos from text prompts with optional images