Multimodal Agent MemorySubmit a paper

VideoGen-of-Thought

VideoGen-of-Thought: Step-by-step generating multi-shot video with minimal manual intervention

Mingzhe Zheng, Yongqi Xu, Haojian Huang, Xuran Ma, Yexin Liu, Wenjie Shu, et al. · 2025-10-02

Read the paperPDF

Management
T2 Dynamic Memory
Memory changes during deployment through a stable management policy.
Primary modality
Video & streaming
Also: text
Representation
Internalized latent & parametric

How experience becomes memory

Each path is one memory component, traced from what is stored to how it is found again.

  1. Identity-preserving (IPP) tokens

    1. Represented asPersistent multimodal latent stateInternalized latent & parametric
    2. Stored inFlat sequential records
    3. Retrieved byLatent / parametric activation
    4. Source evidenceSource not retained
    IPP identity tokens; Latent boundary blends (FIFO of previous shot's last frame). Identity tokens reused per shot; boundary latent blended from the previous shot's last frame.Full-paper method / memory representation and retrieval · PDF p. 3 · Figure 1 / method overview

How memory changes (T2 operations)

  • Update · Rule-based update

    Boundary latents are blended and reset across shots (mutation by a fixed rule). For VideoGen-of-Thought, this specifically supports fixed-rule insertion or replacement of Identity-preserving (IPP) tokens.

Related systems