Category

Behind Video Generation: Observations and Insights on Multimodal Understanding from Kling

Examines how controllable video generation depends on multimodal understanding across the pipeline, from data descriptions to evaluation. It discusses dense captioning, prompt enhancement, perceptual text encoders, and reward modeling, drawing practical lessons from Kling.

Turns Matter: Advancing Streaming Video RL via Turn-Level Credit

Improves when streaming video models respond or keep observing by assigning reinforcement-learning credit to individual turns. Its prefix-scoring and turn-mean objectives reduce redundant replies without a learned critic or extra rollouts.

From Pixels to Actions - Towards Native Vision-Language-Action Policies

Lets robot manipulation policies draw on visual representations from different depths of a vision-language backbone instead of relying on a separate encoder’s fixed output. The approach improves control results on LIBERO and real-world tasks.