Multimodal Agent MemorySubmit a paper

VideoAuteur

VideoAuteur: Towards Long Narrative Video Generation

Junfei Xiao, Feng Cheng, Lu Qi, Liangke Gui, Jiepeng Cen, Zhibei Ma, et al. · 2025-06-07

Read the paperPDF

Management
T2 Dynamic Memory
Memory changes during deployment through a stable management policy.
Primary modality
Video & streaming
Also: text, Image
Representation
Hybrid
Distinct operative representation paths are deliberately linked or jointly consumed by downstream reasoning, generation, or control.

How experience becomes memory

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

  1. Rolling visual-latent context (recent keyframes) conditioning successive segments

    1. Represented asRetained perceptual source evidenceSource- and modality-specific
    2. Stored inFlat sequential records
    3. Retrieved byLatent / parametric activation
    4. Source evidenceSource directly available
    Visual latents (keyframes) in a rolling window. Recent keyframes condition the next segment.Full paper: 4. Method · PDF p. 4 · Figure 1 / method overview

  2. Rolling visual-latent context (recent keyframes) conditioning successive segments

    1. Represented asPersistent multimodal latent stateInternalized latent & parametric
    2. Stored inFlat sequential records
    3. Retrieved byLatent / parametric activation
    4. Source evidenceSource not retained
    Visual latents (keyframes) in a rolling window. Recent keyframes condition the next segment.Full paper: 4. Method · PDF p. 4 · Figure 1 / method overview

How memory changes (T2 operations)

  • Update · Rule-based update

    During operation, VideoAuteur applies fixed-rule insertion or replacement to Rolling visual-latent context (recent keyframes) conditioning successive segments; the memory state changes while the governing update policy remains fixed.

  • Forgetting · Discrete eviction

    The rolling 1-3 keyframe window slides and evicts older keyframes. For VideoAuteur, this specifically supports explicit deletion or eviction of stored items of Rolling visual-latent context (recent keyframes) conditioning successive segments.

Related systems