WRSWorld Research StateContribute
WORLD RESEARCH STATE

JanusFlow: decoupled visual encoders with rectified-flow and representation alignment

Contribution grade

Unified multimodal generative modeling · Human-designed research method or resource

Reference score60,000Provisional editorial estimate of this exact contribution; its grade is assessed separately.

What changed

Integrated rectified-flow image generation with an autoregressive language backbone, retaining separate understanding/generation encoders and aligning intermediate representations during training.

Scientific scope

Integrated rectified-flow image generation with an autoregressive language backbone, retaining separate understanding/generation encoders and aligning intermediate representations during training. Inherits Janus encoder decoupling, rectified flow and unified transformer predecessors. Delta is the stated flow/alignment integration, not invention of autoregression-plus-diffusion or representation alignment. Benchmark advantages are source-specific, not independently replayed.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Inherits Janus encoder decoupling, rectified flow and unified transformer predecessors. Delta is the stated flow/alignment integration, not invention of autoregression-plus-diffusion or representation alignment. Benchmark advantages are source-specific, not independently replayed.

Contributors

Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.

Full scientific record · Browse contribution grades · Request a correction