Custom forcing enables fast personalized video generation from images
Custom Forcing: Training-Free Subject Customization for Autoregressive Video Generation
Computer Vision and Pattern Recognition
Summary
Autoregressive video models can create videos quickly but usually only generate generic characters described by text. The authors introduce Custom Forcing, a method that uses example images to keep the subject in generated videos consistent without needing extra training. They fix problems where the identity might drift or the video defaults to generic subjects by adjusting how much the reference images influence the video. This method works much faster and better preserves the unique subject over long videos compared to prior methods.
What this means in practice
- •For video content creators: Generate long personalized videos from a few example images without slow model retraining or loss of subject identity.
- •For augmented reality developers: Produce real-time video streams with stable, user-specific subjects for AR applications requiring causal, low-latency generation.
Authors
Yunseung Ok, Hyunsoo Kim, Minseo Kim, Suhyun Kim
Abstract
Autoregressive video models can generate minute-long videos in real time, but they produce generic subjects from text rather than specific subjects from user-provided images. Existing customization methods either require costly per-subject optimization or use pretrained conditioning networks that jointly process all video frames with bidirectional attention. Neither approach is designed for causal streaming. We present Custom Forcing, a training-free method that stores reference-based anchor frames in the persistent KV cache of a frozen autoregressive video model. However, fixed anchors face two limitations: simple conditioning allows identity to drift, and the text prompt continues to favor a generic subject. To address these problems, drift-adaptive value amplification (DVA) scales reference influence with the degree of identity drift, while anchor contrast guidance (ACG) steers generation away from the generic class prior. Over two-minute rollouts, fixed anchors fall from 0.58 to 0.42 in DINO-I, while Custom Forcing keeps it between 0.58 and 0.62 without reducing motion. Custom Forcing also achieves higher subject similarity than bidirectional customization methods and better preserves identity over 30s than causal image-to-video and reference-to-video models, while generating each frame 9.5--28.5 times faster than these long-video baselines.