Temporal Specification Guidance for Text-to-Video Generation

Text-to-video models often generate the visual concepts requested in natural language prompts but fail to satisfy their intended ordering, timing, and persistence requirements.