BREAKING
StoryMem: Consistent Multi-Shot AI Video
0
min
video length
0
%
consistency gain
Memory-to-Video Pipeline
1
Extract keyframes
↓
2
Store in memory bank
↓
3
Condition next shot
↓
4
Build full video
Adapting Wan2.2 with LoRA
Base Wan2.2
single-shot
●
Each shot independent
●
Looks drift across cuts
StoryMem
M2V
●
Explicit visual memory
●
LoRA fine-tuning only
Open-Sourced with ST-Bench
Consistency for Storyboarding
AI NEWS BLITZ
NTU and ByteDance have released StoryMem for consistent multi-shot AI video.