SlotMem framework improves character consistency in long video generation

Researchers have developed SlotMem, a novel internal memory framework designed to improve character consistency in long narrative video generation. This system addresses the challenge of maintaining recurring character identities across scene transitions and temporal gaps by using a Character-Semantic Probe to identify character-relevant visual tokens and a Memory Encoder to compress these into role-wise memory slots. A Memory Writer then updates character memories with new observations, and Character-Wise Cross-Attention retrieves and injects this role memory into localized tokens of the same character. Experiments on video generation benchmarks demonstrate that SlotMem enhances long-range character consistency while preserving video quality. AI

IMPACT Enhances character consistency in AI-generated videos, potentially improving narrative coherence and realism.

RANK_REASON This is a research paper detailing a new technical framework for AI video generation. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• Character-Semantic Probe
• Character-Wise Cross-Attention
• Memory Encoder
• Memory Writer
• SlotMem

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →